Browse the corpus

ot significant in the multivariate model (p=0·121). PPR for pregnant women versus children also differed slightly when restricting the analysis to local studies only (PPR 1·67, 95% CI 1·46–1·92 compared with subnational or national surveys only, 1·39, 1·21–1·86), but this was not significant (meta-regression: p=0·362). A small number of studies provided enough detail to allow analysis by gravidity group.26, 28, 37, 38, 39, 40, 49, 51 The PPR of children versus primigravidae was much lower (1·16, 95% CI 1·05–1·29, 8 studies, I2 48%, figure 5) than the overall PPR, whereas the difference between children and multigravidae was higher (PPR 1·94, 1·68–2·24, 7 studies, I2 80%, figure 5). The correlation coefficients were 0·95 (p <0·0001) and 0·93 (p=0·003, figure 2), for the comparison in primigravidae and multigravidae, respectively. All studies were conducted in areas of moderate to high transmission and results of meta-regression did not show a difference in the PPR between children and primigravidae (p=0·992) or multigravidae (p=0·209) when malaria transmission level was taken in to account; however, the number of studies for this analysis was small (table 3 and appendix).

n areas of moderate to high transmission and results of meta-regression did not show a difference in the PPR between children and primigravidae (p=0·992) or multigravidae (p=0·209) when malaria transmission level was taken in to account; however, the number of studies for this analysis was small (table 3 and appendix). Discussion In this meta-analysis we compared the prevalence of malaria infection, as detected by microscopy or rapid diagnostic malaria tests, in pregnant women with the prevalence in children in the same study in the same calendar period and in the same location or region. We showed that the prevalence of malaria infection in pregnant women is lower than that in children aged 0–59 months from the same population, although prevalence estimates in both groups were closely correlated, with a strong linear relation (r=0·87) across the endemicity spectrum. The difference in prevalence between children and pregnant women was smaller when the pregnant women were primigravidae and also in areas of low malaria transmission. Our findings suggest that changes in malaria infection prevalence in pregnant women attending routine antenatal care may be considered as an alternative indicator to track temporal and spatial trends in malaria transmission intensity.

Efforts to estimate the global burden of typhoid fever can be traced to a meeting of the Pan American Health Organization in 1984 and publication of the outcome in 1986.1 Although an important first step, the 1984 study was recognised as having a number of limitations including provision of scanty methodological detail, the availability of few source data, exclusion of China from the estimate, and lack of consideration of the age distribution of typhoid fever. Subsequently the global typhoid burden was re-estimated for the year 2000, accounting for growth of the global population, new typhoid fever incidence data from population-based studies and the control groups of vaccine trials, advances in the understanding of the age distribution of typhoid fever and its relation to force of infection, adjustment for blood culture sensitivity, and formalisation of methods for assessment of disease burden.2 Since 2000, an updated review of population-based studies of typhoid fever incidence and data from notifiable disease reports from countries with advanced surveillance systems has been published.3 Incorporating these data, the Institute for Health Metrics and Evaluation (IHME) added their first estimate of disability and death associated with typhoid and paratyphoid fevers in aggregate to the Global Burden of Disease (GBD) 2010 project.4,5 The IHME GBD 2010 estimate could be criticised for insufficient methodological detail for external reproducibility, lack of disaggregation of typhoid and paratyphoid fevers, little description of the age distribution of disease, and the surprising selection of liver abscesses and cysts as the prime disease complication of interest.6

010 estimate could be criticised for insufficient methodological detail for external reproducibility, lack of disaggregation of typhoid and paratyphoid fevers, little description of the age distribution of disease, and the surprising selection of liver abscesses and cysts as the prime disease complication of interest.6 It is in this context that Vittal Mogasale and others revisit typhoid fever burden with an eye to refining estimates to inform vaccine policy.7 Theirs is not a global estimate, although most typhoid fever cases do occur in countries classified in the low-income and middle-income group. Furthermore, with monovalent typhoid vaccines in mind, the focus is exclusively on Salmonella enterica serovar Typhi, with no estimate for Salmonella Paratyphi A or for invasive non-typhoidal Salmonella. The investigators did a series of well described systematic reviews to update and improve estimates of typhoid fever incidence, including age distribution, blood-culture sensitivity, and case-fatality ratio. They also take the innovative step of adding a risk-factor-based adjustment of typhoid fever incidence that accounts for lack of access to improved water in rural areas and in urban slums. This adjustment was derived from a further systematic review of case-control studies to ascertain the contribution of waterborne transmission to typhoid fever risk. In so doing, Mogasale and colleagues estimate that 11·9 million typhoid fever illnesses and 129 000 deaths occurred in low-income and middle-income countries 2010. These numbers are lower overall by almost half compared with earlier estimates,2 and suggest higher incidence in Africa and lower incidence in Asia than previously thought. Whether these differences reflect true changes in typhoid fever epidemiology over time, methodological differences, or both is difficult to know.

. These numbers are lower overall by almost half compared with earlier estimates,2 and suggest higher incidence in Africa and lower incidence in Asia than previously thought. Whether these differences reflect true changes in typhoid fever epidemiology over time, methodological differences, or both is difficult to know. Mogasale and colleagues highlight a number of limitations. First, despite the growing number of studies on typhoid fever incidence, the amount of source data remains quite scarce. Furthermore, what constitutes a population-based study of typhoid fever incidence is open to inter pretation. Mogasale and others chose a fairly permissive interpretation to optimise the breadth of data. One consequence is the inclusion of a heterogeneous group of study types that are likely to vary considerably in the completeness of capture of cases. This can be problematic when seeking to understand typhoid fever incidence by age group, when differences in detection by age could have substantial effects on apparent age distribution. Indeed, the age distribution of cases derived from Mogasale and colleagues’ review differs from that measured by very intensive active surveillance in a high incidence setting.8

rstand typhoid fever incidence by age group, when differences in detection by age could have substantial effects on apparent age distribution. Indeed, the age distribution of cases derived from Mogasale and colleagues’ review differs from that measured by very intensive active surveillance in a high incidence setting.8 Second, although it is an important and biologically plausible refinement, risk-factor adjustment based on lack of access to improved water in rural areas and urban slums could be open to criticism, as the authors acknowledge. The imperfect relation between access to improved water and consumption of microbiologically safe water is underscored by the occurrence of massive typhoid fever outbreaks in settings with water sources that would be classified as improved.9

areas and urban slums could be open to criticism, as the authors acknowledge. The imperfect relation between access to improved water and consumption of microbiologically safe water is underscored by the occurrence of massive typhoid fever outbreaks in settings with water sources that would be classified as improved.9 Third, reliable estimates of typhoid fever complications and death remain elusive. Hospital-based studies can be biased towards severe disease, yet the early detection and treatment of cases inherent and appropriate in high-quality populated-based disease surveillance systems undoubtedly modifies patients’ outcomes.10,11 Finally, it is important to ask how the results stack up against other sources of data. Few would question that typhoid fever has declined in a number of Asian countries.12 Furthermore, there have been increasing reports of high levels of endemic13,14 and epidemic15,16 typhoid fever from some locations in Africa. However, studies of community-acquired bloodstream infections suggest that non-typhoidal Salmonella has been more common than typhoidal Salmonella in sub-Saharan Africa17 and national disease surveillance data do not seem consistent with the suggestion that South Africa is a country with a high incidence of typhoid fever.18 Indeed, as highlighted by Mogasale and colleagues, incidence estimates for sub-Saharan Africa are heavily influenced by one population-based study from an urban slum in Nairobi, Kenya.13 The recently completed multicountry study of typhoid fever incidence in Africa should go some way to providing more data and addressing these concerns.19

ed by Mogasale and colleagues, incidence estimates for sub-Saharan Africa are heavily influenced by one population-based study from an urban slum in Nairobi, Kenya.13 The recently completed multicountry study of typhoid fever incidence in Africa should go some way to providing more data and addressing these concerns.19 Burden of disease estimates are foundational to building the investment case for both vaccine and non-vaccine interventions for typhoid fever. Decisions about who would most benefit from vaccination and at what age rely on a clear epidemiological picture. Our picture of typhoid fever burden remains clouded, but Mogasale and colleagues have made refinements that challenge us to think more deeply and to value new data. Soon two new estimates of global typhoid and paratyphoid fever burden, from IHME GBD 201320 and the WHO Foodborne Diseases Burden Epidemiology Reference Group,21 will become available. The iterative process of refining and updating burden estimates for typhoid fever is now occurring both consecutively and in parallel, with multiple groups working somewhat independently. Looking to the future, it might be time to take stock of existing estimates and methods, drawing from the strengths of each approach, and striving for both methods that are transparent and results that are timely. Typhoid control would benefit from collective effort to ensure the best possible data to support policy decisions and from a clear message to the world on the scale of the problem.

s and methods, drawing from the strengths of each approach, and striving for both methods that are transparent and results that are timely. Typhoid control would benefit from collective effort to ensure the best possible data to support policy decisions and from a clear message to the world on the scale of the problem. JAC serves as a resource adviser, Invasive Salmonella infections, to the WHO Foodborne Diseases Burden Epidemiology Reference Group; an expert for the Institute for Health Metrics and Evaluation Global Burden of Disease 2013 project; and a reviewer for the Coalition against Typhoid (CaT) typhoid vaccine investment case. JAC is supported by the joint US National Institutes of Health-National Science Foundation Ecology and Evolution of Infectious Disease program (R01 TW009237) and the UK Biotechnology and Biological Sciences Research Council (BBSRC) (BB/J010367/1), and by UK BBSRC Zoonoses in Emerging Livestock Systems awards BB/L017679, BB/L018926, and BB/L018845.

Introduction In malaria transmission areas, pregnant women—in particular primigravidae—are known to be susceptible to malaria and to have higher prevalence and densities of parasitaemia than are non-pregnant women from the same population.1 The size of the excess risk varies with the age of the pregnant woman, reflecting cumulative exposure to malaria over a lifetime, and with parity, as a result of pregnancy-specific immunity acquired after exposure to malaria in previous pregnancies. The consequences of malaria infection during pregnancy will depend on maternal malaria immune status; however, infections are associated with maternal anaemia and fetal growth retardation, and can result in acute illness, pregnancy loss or preterm delivery, and even maternal mortality. The World Health Organization recommends use of insecticide-treated nets (ITNs) and intermittent preventive treatment in pregnancy (IPTp) with a dose of sulfadoxine-pyrimethamine at every scheduled antenatal care visit for the prevention of malaria in pregnancy in areas with moderate-to-high malaria transmission.2, 3, 4 However, because of rising parasite resistance to sulfadoxine-pyrimethamine and decreasing malaria transmission in some regions, alternative strategies for IPTp are now being assessed, such as screening and treatment strategies in pregnancy. This approach consists of the use of rapid diagnostic tests to screen women for malaria at the first or each antenatal visit and treatment of positive women with artemisinin combination therapies.5

regions, alternative strategies for IPTp are now being assessed, such as screening and treatment strategies in pregnancy. This approach consists of the use of rapid diagnostic tests to screen women for malaria at the first or each antenatal visit and treatment of positive women with artemisinin combination therapies.5 Data for malaria prevalence in children obtained from household surveys, such as malaria indicator surveys or school-based surveys, are used to measure transmission intensity and success of malaria control activities in a region.6, 7 Household surveys are logistically demanding and expensive. School surveys, by contrast, are cheaper to do and often include larger sampled populations;8 however, neither approach provides a simple routine real-time measure of malaria in the community. Pregnant women attending antenatal care are a potential alternative source of data for malaria prevalence.

nding and expensive. School surveys, by contrast, are cheaper to do and often include larger sampled populations;8 however, neither approach provides a simple routine real-time measure of malaria in the community. Pregnant women attending antenatal care are a potential alternative source of data for malaria prevalence. A systematic review9 showed that antenatal clinic attendance in pregnant women in most countries in sub-Saharan Africa is high, with at least 75% of pregnant women attending one or more visits in 44 countries in 2010, and at least 90% of pregnant women doing so in 21 countries. That pregnant women are easily accessible for contact at antenatal clinics especially for first visits, makes them a potential surveillance population to track malaria transmission intensity. Because women at the first antenatal clinic visit have not yet received their first dose of sulfadoxine-pyrimethamine for IPTp, malaria infection prevalence at this first visit is likely to be an indicator of malaria transmission intensity in their community. Information on the prevalence of malaria infection at the antenatal booking appointment may become more widely available if screen and treat approaches for malaria control in pregnant women were to be adopted in areas with low or reduced transmission in Africa.5

tor of malaria transmission intensity in their community. Information on the prevalence of malaria infection at the antenatal booking appointment may become more widely available if screen and treat approaches for malaria control in pregnant women were to be adopted in areas with low or reduced transmission in Africa.5 In this meta-analysis, we investigate the relation between malaria infection prevalence in pregnant women and the more standard reference population of children from the same community. We use assembled data from across Africa published since 1983 to assess how any correlation might be modified by gravidity and malaria transmission intensity.10 Methods Search strategy and selection criteria We obtained data on the prevalence of malaria infection in pregnant women from the Malaria in Pregnancy Library.11 This library is a comprehensive bibliographic database created by the Malaria in Pregnancy Consortium that is updated every 4 months with a standardised protocol to search more than 40 sources, including PubMed, Web of Knowledge, and Google Scholar.12 We used data up to January, 2015, without language restriction.12

11 This library is a comprehensive bibliographic database created by the Malaria in Pregnancy Consortium that is updated every 4 months with a standardised protocol to search more than 40 sources, including PubMed, Web of Knowledge, and Google Scholar.12 We used data up to January, 2015, without language restriction.12 Inclusion criteria were: studies in sub-Saharan Africa, based in either the community or antenatal clinics, that screened pregnant women for malaria parasitaemia by microscopy or rapid diagnostic test, irrespective of the presence of symptoms. We excluded studies that selected only women with a history of fever or malaria, and studies that diagnosed malaria at delivery, so that the data for pregnant women would be comparable with those for women attending antenatal clinic. There was no time limit for inclusion and we did not restrict study selection to those with first antenatal visit data. We undertook a systematic evaluation of studies in pregnant women and extracted data including study location, year of study, study population, inclusion and exclusion criteria used, use of malaria prevention strategies (ITNs, IPTp, or prophylaxis), type of malaria diagnostic test used, and test results. Where sufficient information was available, data were extracted by gravidity group, study site, and malaria season. Where needed, and if possible, we contacted authors of the included studies for additional information.

ion strategies (ITNs, IPTp, or prophylaxis), type of malaria diagnostic test used, and test results. Where sufficient information was available, data were extracted by gravidity group, study site, and malaria season. Where needed, and if possible, we contacted authors of the included studies for additional information. Data on the prevalence of malaria infection in pregnant women were then selected on the basis of the availability of the same prevalence data in children aged 0–59 months collected during the same study period and in the same locality as the data in pregnant women. The contemporaneous prevalence data in children and pregnant women were either extracted from studies reported in the Malaria in Pregnancy Library that also reported data in children, or obtained from surveys that collected data on pregnant women and children simultaneously. We identified these data from the large database of over 28 483 temporally and spatially unique surveys of malaria infection undertaken across Africa since 1980 and described elsewhere,6 and from nationally representative household surveys, such as Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and Malaria Indicator Surveys.13, 14, 15 An overview of the methods used in these surveys has been reported previously.9, 16 The information we extracted from the child records included study population, inclusion and exclusion criteria used, use of ITNs, type of malaria diagnostic test used, and test results.

Surveys, and Malaria Indicator Surveys.13, 14, 15 An overview of the methods used in these surveys has been reported previously.9, 16 The information we extracted from the child records included study population, inclusion and exclusion criteria used, use of ITNs, type of malaria diagnostic test used, and test results. We assessed the quality of studies after considering source population, participant selection, appropriate tests, characteristics reporting, and completeness of outcome data. Quality was classified as low-to-moderate or good. Further details of the methods used to assess quality are included in the appendix.

Surveys, and Malaria Indicator Surveys.13, 14, 15 An overview of the methods used in these surveys has been reported previously.9, 16 The information we extracted from the child records included study population, inclusion and exclusion criteria used, use of ITNs, type of malaria diagnostic test used, and test results. We assessed the quality of studies after considering source population, participant selection, appropriate tests, characteristics reporting, and completeness of outcome data. Quality was classified as low-to-moderate or good. Further details of the methods used to assess quality are included in the appendix. Statistical analysis Meta-analyses were conducted using Stata (version 13, StataCorp LP, College Station, TX, USA) using the metan command with input of numerators and denominators for pregnant women and children and the “rr” option to pool the prevalence. We expressed differences between prevalence estimates in pregnant women and children as pooled prevalence ratio (PPR) obtained by meta-analyses using DerSimonian and Laird random-effects models.17 We used random effects models because of the wide heterogeneity in study design and to minimise the effect of study size.18 The extent of heterogeneity was measured using the I2, a measure of the proportion of total variability explained by heterogeneity rather than chance expressed as a percentage,19 with 0–40% representing no or little heterogeneity, 30–60% moderate heterogeneity, 50–90% substantial heterogeneity, and 75–100% considerable heterogeneity.20 To explore determinants of the relation between the prevalence in pregnant women versus children, we examined sources of heterogeneity across studies of the PPR estimates using random-effects meta-regression.21 Regression coefficients were presented as odds ratios (ORs) and their corresponding 95% CIs. We estimated between-study variance (τ2) using the algorithm of residual (restricted) maximum likelihood, and calculated p values, and 95% CIs for coefficients using the modification by Knapp and Hartung.22 For the meta-regression, study-level predictors were considered for inclusion in the initial models if the p value for the univariate association of that variable with the endpoint was <0·2.

tricted) maximum likelihood, and calculated p values, and 95% CIs for coefficients using the modification by Knapp and Hartung.22 For the meta-regression, study-level predictors were considered for inclusion in the initial models if the p value for the univariate association of that variable with the endpoint was <0·2. We considered the effect of the following predictors: gravidity, study period, location of recruitment for pregnant women (community or antenatal clinic), coverage of antimalarial prevention (chemoprophylaxis or IPTp) in pregnant women, type of diagnostic test, malaria transmission intensity, as defined by the average malaria prevalence among children and pregnant women (as a continuous variable and stratified as <5%, 5–40%, >40%),23, 24 and ITN coverage. Because there is a high correlation between ITN use in pregnant women and children (appendix), we used data for coverage in children to represent both groups. HIV infection is known to increase the risk of malaria in pregnancy;25 however, unfortunately none of the included studies had a systematic assessment of maternal HIV status. As an approximation of maternal HIV status, we used the information from the prevalence of HIV in women aged 15–49 years in the same study, or data from a Demographic and Health Survey closest to the study date, or data from other sources by country in all people aged 15–49 years (appendix).

ent of maternal HIV status. As an approximation of maternal HIV status, we used the information from the prevalence of HIV in women aged 15–49 years in the same study, or data from a Demographic and Health Survey closest to the study date, or data from other sources by country in all people aged 15–49 years (appendix). We did a sensitivity analysis to explore the potential effect of the type of study included (regional survey versus observational study) and of study quality on the primary outcome by comparing the results of (sub)national surveys with local studies, or results from low-to-moderate studies with those from good quality studies. Role of the funding source The funding institution had no role in the design and development, data extraction, analysis and interpretation of the data, or preparation, review, or approval of the paper. AMvE had full access to all data and had final responsibility for the decision to submit for publication. Results Of 7011 records screened, we identified 18 data sources (13 national or subnational surveys and five local studies)26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50 with information in children that could be matched with studies in pregnant women, resulting in 57 substudies after stratification of information by location and study period (figure 1). Table 1 and the appendix show study characteristics and the results of the quality assessment.

41, 42, 43, 44, 45, 46, 47, 48, 49, 50 with information in children that could be matched with studies in pregnant women, resulting in 57 substudies after stratification of information by location and study period (figure 1). Table 1 and the appendix show study characteristics and the results of the quality assessment. Studies took place between 1983 and 2012; one study recruited participants from an antenatal clinic and all others were from the community.38 Five sources used rapid diagnostic malaria tests. There was no uniform reporting method on use of malaria prophylaxis or IPTp in pregnant women; four sources reported case management, and for surveys where IPTp was reported, the use varied from 3% to 94% for at least one dose of sulfadoxine-pyrimethamine. The estimated HIV prevalence in women ranged from 1% to 26%; prevalence was less than 10% in two-thirds of sources (12 of 18). Seven of 18 sources were considered good quality; the least commonly reported criterion was the number of women and children who were missing a blood test result. There was a strong correlation between the prevalence of malaria infection in children aged 0–59 months and pregnant women (Pearson correlation coefficient 0·87, p<0·0001, figure 2), with the average prevalence in children higher than that in pregnant women (PPR 1·44, 95% CI 1·29–1·62, figure 3), but with considerable heterogeneity between studies (I2=80%, 95% CI 75–84).

of malaria infection in children aged 0–59 months and pregnant women (Pearson correlation coefficient 0·87, p<0·0001, figure 2), with the average prevalence in children higher than that in pregnant women (PPR 1·44, 95% CI 1·29–1·62, figure 3), but with considerable heterogeneity between studies (I2=80%, 95% CI 75–84). Results of meta-regression identified the following effect modifiers of the overall PPR (table 2): higher PPR when the average infection prevalence was higher, and children's age group, with a higher PPR when comparing children aged 6–59 months with pregnant women than when comparing children aged 0–59 months with pregnant women (p=0·017 for the effect of age in the multivariate model). The type of malaria test used did not have an effect on PPR (rapid diagnostic tests only 1·41, 95% CI 1·18–1·69; microscopy only 1·47, 1·27–1·71; p=0·535 for the effect of diagnostic test in the univariate model). We explored the relation further for malaria transmission; in subgroup analysis, there was less heterogeneity in areas with a prevalence below 5% (I2 42%, 0–70, table 2) but in areas with a higher prevalence I2 was more than 80%. The graph of the log prevalence ratio (figure 4) showed a more consistent pattern in the areas of high malaria prevalence, but even in areas with a prevalence of over 40%, heterogeneity was high.

eity in areas with a prevalence below 5% (I2 42%, 0–70, table 2) but in areas with a higher prevalence I2 was more than 80%. The graph of the log prevalence ratio (figure 4) showed a more consistent pattern in the areas of high malaria prevalence, but even in areas with a prevalence of over 40%, heterogeneity was high. A sensitivity analysis in all studies showed that PPRs were lower when analysis was restricted to low-to-moderate quality studies (1·34, 95% CI 1·17–1·54) than when analysis included only higher quality studies (PPR 1·76, 95% CI 1·39–2·24, p=0·086) but this difference in effect was not significant in the multivariate model (p=0·121). PPR for pregnant women versus children also differed slightly when restricting the analysis to local studies only (PPR 1·67, 95% CI 1·46–1·92 compared with subnational or national surveys only, 1·39, 1·21–1·86), but this was not significant (meta-regression: p=0·362).

the pregnant women were primigravidae and also in areas of low malaria transmission. Our findings suggest that changes in malaria infection prevalence in pregnant women attending routine antenatal care may be considered as an alternative indicator to track temporal and spatial trends in malaria transmission intensity. Antenatal clinic populations are a convenient and easy-to-access group for real-time malaria infection surveillance because most women attend antenatal clinic at least once during pregnancy, even in some hard-to-reach rural areas. Women attend scheduled visits with a focus on preventive health strategies, prompt identification and treatment of illness or conditions, and birth planning. The patterns of malaria prevalence at antenatal booking (that is, before women have received any intervention) may, thus, reflect transmission intensity in their communities. An advantage of using antenatal clinic data to assess trends in malaria transmission is that in many countries pregnant women are routinely screened for HIV, syphilis, and anaemia at their first antenatal booking visit and the addition of testing for malaria would not require any additional sampling. The large difference in malaria prevalence between primigravidae and multigravidae suggest that gravidity would need to be taken into account.

n are routinely screened for HIV, syphilis, and anaemia at their first antenatal booking visit and the addition of testing for malaria would not require any additional sampling. The large difference in malaria prevalence between primigravidae and multigravidae suggest that gravidity would need to be taken into account. That the risk of malaria in pregnant women is lower than that in children in areas of moderate-to-high transmission is not surprising. Parasites can sequester in the placenta, avoiding detection by diagnostic tests, and the concomitant peripheral parasite prevalence can be lower than that in the placenta. A meta-analysis by Kattenberg and colleagues52 reported a sensitivity of peripheral maternal blood microscopy of 72% (95% CI 62–80) for detection of placental malaria, so if all placental malaria infections had been detected in the peripheral blood, in some regions the prevalence in pregnant women might have approached that recorded in children.

rg and colleagues52 reported a sensitivity of peripheral maternal blood microscopy of 72% (95% CI 62–80) for detection of placental malaria, so if all placental malaria infections had been detected in the peripheral blood, in some regions the prevalence in pregnant women might have approached that recorded in children. However, in areas of higher malaria transmission the prevalence gap between pregnant women and children increases and the lower detection level in the peripheral blood is not likely to explain the difference. Previous studies and meta-analysis showed that pregnant women with acute malaria are consistently better at clearing parasites after antimalarial treatment with chloroquine or sulfadoxine-pyrimethamine than are children.10, 53 This finding probably reflects the higher level of acquired protective malarial immunity in pregnant women, especially multigravidae, in areas of high malaria endemicity and, thus, their ability to control and suppress parasite densities when infected relative to the immunity level in young children. Primigravidae generally do not have antibodies to placental-type parasites at the onset of pregnancy, but generate these during the course of pregnancy if exposed to malaria, and some have suggested using these antibody responses as sentinel markers for malaria transmission.54

e to the immunity level in young children. Primigravidae generally do not have antibodies to placental-type parasites at the onset of pregnancy, but generate these during the course of pregnancy if exposed to malaria, and some have suggested using these antibody responses as sentinel markers for malaria transmission.54 In addition to gravidity, several other factors modified the relation between the population prevalence of malaria infection in pregnant women and children, including the age of the children used for comparison, with greater relative differences with pregnant women in the 6–59 months age group than 0–59 month old children. This likely reflects the lower risk of malaria in the first months of life compared with that later in infancy.55 Although there was a good correlation between malaria in children and pregnant women, the high heterogeneity across the malaria spectrum indicates that data in pregnant women may be more useful to assess trends than to use as an approximation of malaria transmission or to estimate malaria prevalence in other vulnerable groups. For example, for a malaria prevalence in pregnant women between 10% and 20% (12 data-points), the prevalence in children varied from 4·7% to 49·7%. The heterogeneity was less in areas of low transmission and in primigravidae.

ximation of malaria transmission or to estimate malaria prevalence in other vulnerable groups. For example, for a malaria prevalence in pregnant women between 10% and 20% (12 data-points), the prevalence in children varied from 4·7% to 49·7%. The heterogeneity was less in areas of low transmission and in primigravidae. There are important limitations to this type of secondary analysis that should be considered. First, these data might not be representative of sub-Saharan Africa because the number of studies with available data in both pregnant women and children at the same location and during the same time was small (18 sources). Second, most of the data for the comparison between children and pregnant women came from community-based surveys, and it is not yet clear whether these data are representative of the antenatal population, especially the potential target population for sentinel malaria surveillance––that is, those attending an antenatal clinic for their first booking visit. Most pregnant women in Africa have their first antenatal clinic visit before month 6 of pregnancy (appendix), when the risk of malaria is high, compared with the third trimester (van Eijk, unpublished observation); use of malaria prevention such as chemoprophylaxis or IPTp in women attending for their first antenatal visit is unlikely, so that the prevalence of malaria among first antenatal clinic attendees may be closer to that of children than reflected in our analyses.

h the third trimester (van Eijk, unpublished observation); use of malaria prevention such as chemoprophylaxis or IPTp in women attending for their first antenatal visit is unlikely, so that the prevalence of malaria among first antenatal clinic attendees may be closer to that of children than reflected in our analyses. However, women who do not attend antenatal clinics may be at greater risk of malaria given that antenatal clinic attendance can be low in some rural populations, and in women with low socioeconomic status; both of these factors have been associated with an increased risk of malaria.9, 16, 56, 57 Although this source of selection bias is likely to be small in malaria-endemic Africa where more than 90% women attend an antenatal clinic at least once,9 in countries where this is not the case––that is, where more than 10% of women do not attend an antenatal clinic––population-based surveys may be needed to assess whether the risk of malaria infection in these women is different from that in women who do attend antenatal clinics.

end an antenatal clinic at least once,9 in countries where this is not the case––that is, where more than 10% of women do not attend an antenatal clinic––population-based surveys may be needed to assess whether the risk of malaria infection in these women is different from that in women who do attend antenatal clinics. In settings where more than 10% of women do not attend ANC, the use of annealing methods should be considered that combine data from a relatively small random community survey sample with the convenience sample obtained from data that can be routinely collected in antenatal clinics, as has been done for HIV studies.58 These hybrid prevalence estimators provide more accurate information than those available from using only data derived from antenatal clinics, and are more efficient than when data are collected only through larger (and thus more expensive and only periodic) community-based random survey samples such as in Demographic and Health Surveys or Malaria Indicator Surveys.58 Examples of countries with antenatal clinic attendance rates less than 90% in a malarious country include Nigeria (61% in 2013), Mali (74% in 2012–13), Angola (80% in 2006–07), Togo (73% in 2013), and the Central African Republic (68% in 2010) (appendix).

In settings where more than 10% of women do not attend ANC, the use of annealing methods should be considered that combine data from a relatively small random community survey sample with the convenience sample obtained from data that can be routinely collected in antenatal clinics, as has been done for HIV studies.58 These hybrid prevalence estimators provide more accurate information than those available from using only data derived from antenatal clinics, and are more efficient than when data are collected only through larger (and thus more expensive and only periodic) community-based random survey samples such as in Demographic and Health Surveys or Malaria Indicator Surveys.58 Examples of countries with antenatal clinic attendance rates less than 90% in a malarious country include Nigeria (61% in 2013), Mali (74% in 2012–13), Angola (80% in 2006–07), Togo (73% in 2013), and the Central African Republic (68% in 2010) (appendix). Another limitation of this analysis is that, although average malaria prevalence among children and pregnant women was used for the assessment of malaria endemicity, the 2–9 year age group is typically used for this.59 Further, the subnational surveys used a two-stage cluster sampling design and this might have had an effect on the standard error around the prevalence estimate, but we could not take this effect into account in our secondary analysis, which might have resulted in an overestimation of the precision of the effect estimates.

her, the subnational surveys used a two-stage cluster sampling design and this might have had an effect on the standard error around the prevalence estimate, but we could not take this effect into account in our secondary analysis, which might have resulted in an overestimation of the precision of the effect estimates. In sensitivity analysis, the PPR from low-to-moderate quality studies was lower than the PPR of higher quality studies. This finding might be partly explained by differences in transmission intensity because the mean prevalence of malaria in children in low-to-moderate studies was about half that observed in the better quality studies (16% vs 31%, respectively). An alternative explanation might include different compositions of the study populations in low-to-moderate quality studies, with, for example, more primigravidae or women of young age. However, information available from the included studies was insufficient to explore this theory further.

(16% vs 31%, respectively). An alternative explanation might include different compositions of the study populations in low-to-moderate quality studies, with, for example, more primigravidae or women of young age. However, information available from the included studies was insufficient to explore this theory further. Although the biology and epidemiology of malaria and HIV differ substantially, lessons can be learned from the extensive experience with the use of antenatal data as a convenience sample for HIV-infection surveillance. For example, the use of hybrid prevalence estimators and the annealing of antenatal data with small random community samples to reduce bias.58 Overestimates have been reported when comparing estimates from antenatal clinics with community surveillance: suggested reasons included preferential antenatal attendance (for example, referral of people suspected of having HIV to certain clinics), the geographic under-representation of rural clinics (to obtain the sample size in the required period, high volume antenatal clinics are used which are more likely to be in urban areas), and cultural factors.60, 61, 62, 63 However, because of their consistent method and routine collection antenatal clinics are still the main source for trends in countries with generalised epidemics.63

sample size in the required period, high volume antenatal clinics are used which are more likely to be in urban areas), and cultural factors.60, 61, 62, 63 However, because of their consistent method and routine collection antenatal clinics are still the main source for trends in countries with generalised epidemics.63 Our meta-analysis found a strong linear relation between the prevalence of malaria infection in pregnant women and children from the same population. Routine information on the malaria infection status of pregnant women attending antenatal care might become increasingly available if countries switch from IPTp with sulfadoxine-pyrimethamine to “screen and treat” approaches. This switch could happen because of decreasing malaria transmission rates or increasing high-grade resistance to sulfadoxine-pyrimethamine, the only antimalarial currently recommended for IPTp. Antenatal surveillance for malaria infection, especially during the first antenatal booking visit, should be explored as a pragmatic and sustainable method for the real-time monitoring of malaria trends. Supplementary Material Supplementary appendix

Our meta-analysis found a strong linear relation between the prevalence of malaria infection in pregnant women and children from the same population. Routine information on the malaria infection status of pregnant women attending antenatal care might become increasingly available if countries switch from IPTp with sulfadoxine-pyrimethamine to “screen and treat” approaches. This switch could happen because of decreasing malaria transmission rates or increasing high-grade resistance to sulfadoxine-pyrimethamine, the only antimalarial currently recommended for IPTp. Antenatal surveillance for malaria infection, especially during the first antenatal booking visit, should be explored as a pragmatic and sustainable method for the real-time monitoring of malaria trends. Supplementary Material Supplementary appendix Acknowledgments We thank Patricia Graves and Jeremiah Ngondi for providing additional data. This review was in part funded by the US Centers for Disease Control and Prevention (CDC) through a cooperative agreement between the Division of Parasitic Diseases and Malaria, CDC, and the Liverpool School of Tropical Medicine held by FOtK. AvE and JH are also supported by the Malaria in Pregnancy Consortium, which is funded through a grant from the Bill & Melinda Gates Foundation to the Liverpool School of Tropical Medicine. RWS is supported by the Wellcome Trust as Principal Research Fellow (#079080 & #103602). AMN is supported by the Wellcome Trust as an Intermediate Research Fellow (#095127) and is Director of the Information for Malaria Project funded by the UK's Department for International Development, UK.

chool of Tropical Medicine. RWS is supported by the Wellcome Trust as Principal Research Fellow (#079080 & #103602). AMN is supported by the Wellcome Trust as an Intermediate Research Fellow (#095127) and is Director of the Information for Malaria Project funded by the UK's Department for International Development, UK. Contributors AMvE, FOtK, and RWS conceived and designed the study. AMvE and RWS did the literature search and acquired the data. AMvE, FotK, and RWS analysed and interpreted the data. AMvE and JH wrote the first draft of the paper. FOtK, RWS, AMN, and JH critically revised subsequent drafts of the paper. All authors approved the final version. FOtK and JH obtained funding. Declaration of interests We declare no competing interests. Figure 1 Flow diagram for the literature search Figure 2 Scatter plots for malaria prevalence in all pregnant women, primigravidae, and multigravidae versus children 0–59 months, sub-Saharan Africa, 1983–2012 Figure 3 Forest plot of prevalence ratios for malaria in children (0–59 months) versus pregnant women, sub–Saharan Africa, 1983–2012 Mx=microscopy. RDT=rapid diagnostic malaria test. SNNPR=Southern Nations, Nationalities and People's Region. Dotted line shows the pooled prevalence ratio. Studies are listed in ascending order of prevalence of malaria in children. Figure 4 Bubble plot with fitted meta-regression line of the log prevalence ratio: child-maternal malaria prevalence and average malaria prevalence, sub-Saharan Africa, 1983–2012

Mx=microscopy. RDT=rapid diagnostic malaria test. SNNPR=Southern Nations, Nationalities and People's Region. Dotted line shows the pooled prevalence ratio. Studies are listed in ascending order of prevalence of malaria in children. Figure 4 Bubble plot with fitted meta-regression line of the log prevalence ratio: child-maternal malaria prevalence and average malaria prevalence, sub-Saharan Africa, 1983–2012 Circles are sized according to precision of each estimate with larger bubbles for more precise estimates. Average malaria prevalence is the average of malaria prevalence in children and pregnant women. Figure 5 Forest plot of prevalence ratio of malaria in children aged 0–59 months versus primigravidae or multigravidae, sub-Saharan Africa, 1983–2012 Studies are listed in ascending order of prevalence of malaria in children. Table 1 Characteristics of 18 studies included in the comparison of malaria in children 0–5 years of age versus pregnant women, sub-Saharan Africa, 1983–2012

Figure 5 Forest plot of prevalence ratio of malaria in children aged 0–59 months versus primigravidae or multigravidae, sub-Saharan Africa, 1983–2012 Studies are listed in ascending order of prevalence of malaria in children. Table 1 Characteristics of 18 studies included in the comparison of malaria in children 0–5 years of age versus pregnant women, sub-Saharan Africa, 1983–2012 Country and location of recruitment Study period Design Primary objective of study Level of information* Test and species Sample size Pregnant women: antimalarial for prevention† Children: antimalarial for fever‡ ITNs or nets Age ¶HIV prevalence estimate (%) Pregnant women Children Pregnant Women Children Pregnant Women (years) Children (months) Angola MIS 2006–0726 Angola, community 2006–07 Survey Evaluation control malaria Regional (4) RDT, Pf 345 2497 SP1+ 3% 7% ITN 22% ITN 18% 15–49 6–59 2·027 Côte d’Ivoire DHS 2011–1228 Côte d’Ivoire, community 2011–12 Survey Evaluation control malaria Regional (5) Mx, any 451 3184 SP1+ 26% 4% ITN 40% ITN 37% 15–49 6–59 4·628 Deribew 201029 Ethiopia, community 2009 Survey Evaluation ITN use Local (2) Mx, any 242 2410 Case management NR ITN 63% ITN 57% Mean 26 0–59 1·030 Dicko 2003 & Dicko 2005 31, 32 Mali, community 1993–94 Survey Epidemiology malaria Local (2) Mx, any 235 2366 CQ 16% NR NR NR 15–45 (mean 28) 6–59 1·827 Equatorial Guinea MIS 2008 & 200933 Equatorial Guinea Bioko, community 2008, 2009 Survey Evaluation control malaria Regional (2) RDT, Pf 481 5045 SP1+ 46% 6% ITN 58% ITN 54% 15–49 0–59 3·934 Graves 200935 Ethiopia, community 2006–07 Survey Evaluation control malaria Regional (3) Mx, any 209 1010 Case management NR ITN 19% ITN 19% 15–49 0–59 1·936 Mabunda 200637 Mozambique, community 2002–03 Survey Evaluation control malaria Regional (11) Mx, Pf 1531 6641 NR NR NR NR 12–44 (mean 26) 0–59 10·727 Matola 198538 Tanzania, ANC and MCH 1983 Survey Evaluation use of CQ as prevention Local (1) Mx, any 196 297 CQ 24% CQ 11% NR NR 16–43 (mean 25) 0–59 mean 14 0·5 McElroy 1999 & Bloland 199939 Kenya, community 1992–96 Cohort Epidemiology malaria Local (1) Mx, any 1047 328 Case management NR Net 9% Net 9% Mean 26 0–59 25·527 Mozambique MIS 200740 Mozambique, community 2007 Survey Evaluation malaria control Regional (3) Mx, any 459 3828 SP1+ 27% 18% ITN 7% ITN 7% 15–49 0–59 13·941 Namibia MIS 200942 Namibia, community 2009 Survey Evaluation control malaria National (1) RDT, Pf 192 1977 SP1+ 8% 6% ITN 26% ITN 34% 15–49 6–59 16·427 Nyan 2009 MIS The Gambia43 The Gambia, community 2008 Survey Evaluation control malaria Regio

Regional (3) Mx, any 459 3828 SP1+ 27% 18% ITN 7% ITN 7% 15–49 0–59 13·941 Namibia MIS 200942 Namibia, community 2009 Survey Evaluation control malaria National (1) RDT, Pf 192 1977 SP1+ 8% 6% ITN 26% ITN 34% 15–49 6–59 16·427 Nyan 2009 MIS The Gambia43 The Gambia, community 2008 Survey Evaluation control malaria Regio nal (5) RDT, Pf 402 2470 SP1+ 94% 14% ITN 45% ITN 43% 15–49 6–59 1·727 Rehman 201344 Equatorial Guinea, community 2007–09 Survey Evaluation control malaria Regional (5) RDT, Pf 741 8087 SP1+ 18-45% 3% ITN 19% ITN 19% NR 12–59 10·034 Rwanda 2007–08 DHS45 Rwanda, community 2007–08 Survey Evaluation control malaria National (1) Mx, any 642 4662 SP1+ 51% 1% ITN 60% ITN 56% 15–49 6–59 3·746 Rwanda 2010–11 DHS46 Rwanda, community 2010–11 Survey Evaluation control malaria National (1) Mx, any 486 4046 Case management 2% ITN 72% ITN 70% 15–49 6–59 3·746 South Sudan MIS 200947 South Sudan, community 2009 Survey Evaluation control malaria Regional (3) Mx, any 435 2993 SP1+ 17% 13% ITN 36% ITN 25% 15–49 <20, 13% 0–59 3·227 Sudan MIS 200548 Sudan, community 2005 Survey Evaluation malaria control Regional (6) Mx, any 320 2023 CQ or SP 10% 19% ITN 6% ITN 8% 15–49 0–59 0·527 Van Eijk 200849 Kenya, community 2003 Survey Health assessment Local (1) Mx, any 672 1162 SP1+ 8% 32% ITN 69% ITN 67% Mean 26 6–59 18·350 IITN=insecticide-treated net; MIS=malaria indicator survey; RDT =rapid diagnostic testing; Pf=Plasmodium falciparum; SP1+=at least one dose of sulfadoxine-pyrimethamine.

9 0–59 0·527 Van Eijk 200849 Kenya, community 2003 Survey Health assessment Local (1) Mx, any 672 1162 SP1+ 8% 32% ITN 69% ITN 67% Mean 26 6–59 18·350 IITN=insecticide-treated net; MIS=malaria indicator survey; RDT =rapid diagnostic testing; Pf=Plasmodium falciparum; SP1+=at least one dose of sulfadoxine-pyrimethamine. DHS=Demographic and Health Survey; Mx=microscopy; NR= not reported; ANC=antenatal clinic; MCH=maternal and child health clinic.; SP= sulfadoxine-pyrimethamine. * Level of information: information used at a national level, or by region when it was part of a national or sub-national survey, or at a local level if it was a local study; number of sub-studies derived from the source by location or study period is shown in brackets. † Prophylaxis as reported in pregnant women, or in women in the same survey who completed a pregnancy in the last 2 or 5 years, for their last pregnancy; for surveys where only SP2+ was reported, SP1+ was estimated as 1·67 × SP2+.16 ‡ Antimalarial treatment for a fever episode in the previous 2 weeks. ¶ Prevalence in women aged 15–49 years. Table 2 Meta-regression of factors that might affect the prevalence ratio for malaria in children 0–59 months versus pregnant women in sub-Saharan Africa, 1983–2012

† Prophylaxis as reported in pregnant women, or in women in the same survey who completed a pregnancy in the last 2 or 5 years, for their last pregnancy; for surveys where only SP2+ was reported, SP1+ was estimated as 1·67 × SP2+.16 ‡ Antimalarial treatment for a fever episode in the previous 2 weeks. ¶ Prevalence in women aged 15–49 years. Table 2 Meta-regression of factors that might affect the prevalence ratio for malaria in children 0–59 months versus pregnant women in sub-Saharan Africa, 1983–2012 Number of surveys Pooled prevalence ratio (95% CI) I2 (%) (95% CI) for subgroup analysis Odds ratio meta-regression (95% CI) p value by level τ2 Variance explained (%) p (overall) No covariates 57 1·44 (1·29–1·62) 80 (75–84) 0·182 Place of recruitment of pregnant women ANC 1 1·34 (1·10–1·62) 0·93 (0·36–2·39) 0·877 0·190 0·0 0·877 Community 56 1·44 (1·28–1·63) 80 (75–85) 1·00 (Reference) Malaria test RDT 19 1·41 (1·18–1·69) 71 (54–82) 0·91 (0·66–1·24) 0·535 0·192 0·0 0·535 Microscopy 38 1·47 (1·27–1·71) 83 (78–87) 1·00 (Reference) Time period (year) <2000 4 1·72 (1·38–2·15) 80 (47–92) 1·26 (0·77–2·07) 0·344 0·175 0·0 0·344 ≥2000 53 1·40 (1·23–1·60) 80 (74–84) 1·00 (Reference) Average malaria prevalence * as an indicator of transmission level Continuous 57 .. .. 1·00 (0·99–1·02) 0·139 0·177 2·4 0·139 >40% 13 1·51 (1·33–1·72) 84 (73–90) 1·79 (1·03–3·10) 0·039 0·184 0·0 0·084 5–40% 31 1·53 (1·24–1·88) 83 (77–88) 1·79 (1·06–3·04) 0·030 <5% 13 0·82 (0·47–1·40) 42 (0–70) 1·00 (Reference) Antimalarial regimen during pregnancy† None 18 1·17 (0·94–1·46) 89 (84–92) 1·00 (Reference) 0·154 15·5 0·106 IPTp 29 1·64 (1·41–1·91) 75 (64–82) 1·38 (1·01–1·88) 0·042 Prophylaxis‡ 10 1·61 (1·27–2·04) 34 (0–68) 1·38 (0·85–2·25) 0·188 ITN use during pregnancy No ITN information 16 1·27 (1·07–1·51) 90 (86–93) 1·00 (Reference) 0·176 3·0 0·357 ITN use < 25% 22 1·57 (1·31–1·88) 65 (45–78) 1·18 (0·84–1·68) 0·332 ITN use ≥25% 19 1·64 (1·20–2·23) 72 (55–82) 1·29 (0·89–1·88) 0·173 Age definition of child group 0–59 months 31 1·25 (1·07–1·47) 84 (78–88) 1·00 (Reference) 0·156 14·1 0·111 6–59 months 21 1·67 (1·29–2·18) 73 (58–82) 1·36 (0·98–1·87) 0·063 12–59 months 5 1·68 (1·49–1·90) 34 (0–75) 1·38 (0·89–2·14) 0·152 Estimate of maternal HIV infection† Continuous 57 .. .. 0·98 (0·96–1·01) 0·139 0·179 1·6 0·139 >9% 16 1·40 (1·19–1·65) 89 (83–92) 0·94 (0·69–1·28) 0·676 0·187 0·0 0·676 ≤ 9% 41 1·47 (1·24–1·75) 74 (65–81) 1·00 (Reference) Multivariate analysis 0·149 17·9 0·025 Average malaria prevalence as an indicator of transmission level* >40% 17 .. .. 2·03 (1·12–3·66) 0·020 5–40% 26 .. .. 1·97 (1·17–3·31) 0·012 <5% 14 .. .. 1·00 (Reference) Age definition of child group 0–59 months 31 .. ..

0·676 ≤ 9% 41 1·47 (1·24–1·75) 74 (65–81) 1·00 (Reference) Multivariate analysis 0·149 17·9 0·025 Average malaria prevalence as an indicator of transmission level* >40% 17 .. .. 2·03 (1·12–3·66) 0·020 5–40% 26 .. .. 1·97 (1·17–3·31) 0·012 <5% 14 .. .. 1·00 (Reference) Age definition of child group 0–59 months 31 .. .. 1·00 (Reference) 6–59 months 21 .. .. 1·49 (1·08–2·07) 0·017 12–59 months 5 .. .. 1·30 (0·81–2·11) 0·270 ANC=antenatal clinic. RDT=rapid diagnostic test. IPTp=intermittent preventive treatment in pregnancy. ITN= insecticide treated nets. * Average malaria prevalence in children and pregnant women. † Not significant in multivariate analysis. ‡ Any dose for any time period of prophylaxis, not IPTp. Table 3 Subgroup analysis of pooled prevalence ratio of malaria in children versus malaria in pregnant women by gravidity and by average malaria prevalence in children and pregnant women, sub-Saharan Africa, 1983–2012 Number of studies Pooled prevalence ratio (95% CI) I2 (%) (95% CI) Odds ratio meta-regression 95% CI p Primigravidae >40% 5 1·16 (1·02–1·32) 66 (12–87) 0·99 0·68–1·46 0·992 5%–40% 3 1·16 (0·92–1·47) 0 (0–90) 1·00 Reference Multigravidae >40% 4 1·81 (1·54–2·12) 85 (63–94) 0·77 0·48–1·23 0·209 5%–40% 3 2·38 (1·63–3·48) 69 (0–91) 1·00 Reference

The Article by James Platts-Mills and colleagues (September, 2015),1 highlights the burden of diarrhoeal disease in young children in developing countries and also demonstrates the effect that seasonal variations have on multiple causative pathogens. Infectious gastroenteritis contributes significantly to the 1 billion episodes of diarrhoea and 3 million deaths in children under 5 years, and is the fifth leading cause of death worldwide.2 Pathogens causing diarrhoea frequently show seasonality, suggesting that climate and enteric disease are inextricably linked. This link has important implications if we accept the very real threat of climate change on human health. WHO quantified the impact of global warming on diarrhoea, reporting that warming by 1°C was associated with a 5% increase in diarrhoea.3 Increased rainfall has been associated with higher incidence of norovirus4 whereas rotavirus often peaks in colder months. As such, extreme weather events associated with climate change are likely to alter patterns of gastroenteritis. The increased replication rate of some bacterial and viral pathogens in warm conditions,5 combined with poor water and sanitation infrastructure, means that people in developing nations are particularly vulnerable.

h, extreme weather events associated with climate change are likely to alter patterns of gastroenteritis. The increased replication rate of some bacterial and viral pathogens in warm conditions,5 combined with poor water and sanitation infrastructure, means that people in developing nations are particularly vulnerable. Predicting the potential effects of climate change on the incidence and distribution of infectious gastroenteritis can assist public health providers to control and prevent severe outbreaks in the future. Implementation of programmes for rotavirus vaccination clearly shows benefit and should be an adaptation strategy to help cope with climate change. More importantly, however, Platts-Mills and colleagues1 identified multiple pathogens contributing to diarrhoeal disease, highlighting the necessity for a broader mitigation plan. I declare no competing interests.

Introduction Yaws is a neglected tropical disease caused by Treponema pallidum subspecies pertenue.1 This bacterium causes a chronic relapsing non-venereal treponematosis, characterised by highly contagious primary and secondary cutaneous lesions and non-contagious tertiary destructive lesions of the bones. The infection can become latent at any time, with only serological evidence of infection, and relapses can occur for up to 5–10 years. The ratio of clinically apparent to latent cases has been estimated to be as high as 1:6.1 In 2012, WHO launched a new initiative to eradicate yaws by 2020.2 Undertaking surveys and mapping the disease at a community level and immediately treating the entire endemic community with single-dose azithromycin3 is recommended.2 The efficacy of this approach has been shown in a study of mass treatment in Papua New Guinea.4 A key principle inherent in an eradication campaign is the need to intervene everywhere the disease occurs. However, the present geographic extent of yaws is incompletely known, because yaws is not a notifiable disease in many affected countries. To guide the WHO eradication programme, a better knowledge of yaws epidemiology is needed.

nherent in an eradication campaign is the need to intervene everywhere the disease occurs. However, the present geographic extent of yaws is incompletely known, because yaws is not a notifiable disease in many affected countries. To guide the WHO eradication programme, a better knowledge of yaws epidemiology is needed. Data that can be used to identify the burden of yaws in a community include the prevalence of active infectious yaws (ie, ulcers or papilloma), which shows the intensity of yaws transmission, and the prevalence of latent yaws (ie, seropositivity in healthy individuals), which shows the extent of latent or hidden infection in the community. Clinical surveys for active yaws lesions can be done without any sophisticated laboratory test through interviews and physical examinations, whereas serological tests measuring yaws antibody (treponemal and non-treponemal) are needed for surveys of latent disease.5 Another important source of information is national routine surveillance data, which allow estimation of the incidence of yaws at country and regional levels; countries report the number of cases at the first administrative level. In this study, we undertook a systematic review of published and unpublished work to improve our understanding of yaws epidemiology stratified by country, and to provide an update on the number of people with active yaws to estimate at-risk populations in endemic countries.

Data that can be used to identify the burden of yaws in a community include the prevalence of active infectious yaws (ie, ulcers or papilloma), which shows the intensity of yaws transmission, and the prevalence of latent yaws (ie, seropositivity in healthy individuals), which shows the extent of latent or hidden infection in the community. Clinical surveys for active yaws lesions can be done without any sophisticated laboratory test through interviews and physical examinations, whereas serological tests measuring yaws antibody (treponemal and non-treponemal) are needed for surveys of latent disease.5 Another important source of information is national routine surveillance data, which allow estimation of the incidence of yaws at country and regional levels; countries report the number of cases at the first administrative level. In this study, we undertook a systematic review of published and unpublished work to improve our understanding of yaws epidemiology stratified by country, and to provide an update on the number of people with active yaws to estimate at-risk populations in endemic countries. Methods Search strategy and selection criteria We did a systematic review to identify all relevant studies that examined yaws prevalence and incidence. We searched PubMed and WHO databases for (“yaws” OR “treponematosis” AND “prevalence” OR “incidence”) OR (“yaws” AND [each individual previous and current yaws-endemic country]6). We consulted the Department for the Control of Neglected Tropical Diseases at WHO regarding previous and present yaws-endemic countries.6 We limited the search to studies published between Jan 1, 1990, and Dec 31, 2014. This period covers studies published since the last systematic review of yaws epidemiology, which was published in 1992.7 No language restrictions were set for searches. We hand-searched the reference lists of all recovered documents for additional references. We also searched for ongoing or recently completed but unpublished studies from the WHO yaws surveillance network.

tematic review of yaws epidemiology, which was published in 1992.7 No language restrictions were set for searches. We hand-searched the reference lists of all recovered documents for additional references. We also searched for ongoing or recently completed but unpublished studies from the WHO yaws surveillance network. We included studies if they investigated active or latent yaws prevalence or incidence. Studies on active yaws had to meet the surveillance case definition provided by WHO:8 a person with a history of residence in an affected area who presents with signs of clinically active yaws, consisting of chronic skin ulcers, multiple papillomata, squamous macules, bone or joint lesions, or plantar hyperkeratosis. For latent yaws seroprevalence studies, we deemed serological test rapid plasma reagin titres of at least 1:2 and venereal disease research laboratory titres of at least 1:2 as acceptable evidence of untreated latent infection. Use of the treponemal test (T pallidum haemagglutination assay, T pallidum particle agglutination assay, and the fluorescent treponemal antibody absorption) alone was not sufficient evidence of latent infection because people who have had yaws at any time will test positive for life, even after successful treatment.

n. Use of the treponemal test (T pallidum haemagglutination assay, T pallidum particle agglutination assay, and the fluorescent treponemal antibody absorption) alone was not sufficient evidence of latent infection because people who have had yaws at any time will test positive for life, even after successful treatment. Procedures We calculated the number of people with active disease at the first administrative level (eg, province, region, and prefecture) between Jan 1, 2010, and Dec 31, 2013. First, whenever possible, we obtained the country estimates of yaws cases at the first administrative level from the latest national reporting figures provided to WHO.6 Second, for countries for which no recent data were available, we contacted yaws control programme managers to request official national routine surveillance data. To estimate the maximum population at risk of yaws, we made calculations at the second administrative level (eg, district, department, and regency). We contacted yaws control programme managers to request data on the proportion of second-administrative level regions that reported yaws cases in 2012. We summed the population living in endemic districts using the 2012 reported populations.

ns at the second administrative level (eg, district, department, and regency). We contacted yaws control programme managers to request data on the proportion of second-administrative level regions that reported yaws cases in 2012. We summed the population living in endemic districts using the 2012 reported populations. Statistical analysis For all qualifying studies, we extracted data on study country, sample size, diagnostic test used, number of people with latent or active yaws, and age range. We undertook descriptive analyses of the extracted data. Prevalence estimates are presented for each study with 95% CIs on the basis of binomial distribution. We did not undertake quantitative meta-analyses because the studies we identified did not sample populations at random and hence the estimates are not representative for a broader geographical area. All statistical analyses were done using Stata version 13.1. Role of the funding source There was no funding source for this study. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Statistical analysis For all qualifying studies, we extracted data on study country, sample size, diagnostic test used, number of people with latent or active yaws, and age range. We undertook descriptive analyses of the extracted data. Prevalence estimates are presented for each study with 95% CIs on the basis of binomial distribution. We did not undertake quantitative meta-analyses because the studies we identified did not sample populations at random and hence the estimates are not representative for a broader geographical area. All statistical analyses were done using Stata version 13.1. Role of the funding source There was no funding source for this study. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Our systematic review identified 103 records, from which we identified 23 eligible published articles9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 that described 27 studies that met our inclusion criteria (figure 1). We included data from an additional four studies identified from other sources (personal communications with country managers and yaws experts: Tabah EN, personal communication; Boua B, personal communication; Nsiire A, personal communication; Ayelo G, personal communication). The included studies covered 18 countries. Three of these countries—Guyana, Nigeria, and Wallis and Futuna—were classified by WHO as previously endemic countries with unknown status in 2012. Two countries—Ecuador and India—were reported to have eliminated yaws.29, 31 The remaining 13 countries were classified as known endemic countries in 2012.6

d 18 countries. Three of these countries—Guyana, Nigeria, and Wallis and Futuna—were classified by WHO as previously endemic countries with unknown status in 2012. Two countries—Ecuador and India—were reported to have eliminated yaws.29, 31 The remaining 13 countries were classified as known endemic countries in 2012.6 Among the 31 studies, 16 reported data on active yaws prevalence (table 1; Tabah EN, personal communication; Boua B, personal communication; Nsiire A, personal communication).9, 10, 13, 14, 16, 18, 19, 20, 21, 24, 28, 29, 30 Patients with suspected yaws skin lesions were further tested with syphilis serology, except in four studies in which diagnosis was made on the basis of clinical criteria only (Tabah EN, personal communication; Boua B, personal communication).16, 19 After excluding one study from Ecuador29 in 1998 in which no clinical cases were detected, prevalence of active yaws lesions ranged from 0·31% in Sumatra, Indonesia,18 to 14·54% around the city of Port Moresby, Papua New Guinea.21 High prevalence rates were also noted in surveys done in tropical forests in Central Africa that were inhabited by indigenous populations (ie, Pygmies), including 9·03% in Cameroon (Tabah EN, personal communication), 11·34% in the Central African Republic (Boua B, personal communication), 4·77% in the Democratic Republic of the Congo,14 and 2·95% in the Republic of Congo.10

tropical forests in Central Africa that were inhabited by indigenous populations (ie, Pygmies), including 9·03% in Cameroon (Tabah EN, personal communication), 11·34% in the Central African Republic (Boua B, personal communication), 4·77% in the Democratic Republic of the Congo,14 and 2·95% in the Republic of Congo.10 Overall, eight studies reported data on the prevalence of latent yaws (table 1; Ayelo G, personal communication).9, 25, 26, 27, 28, 29, 31 After excluding one study from India31 in which no seropositive cases were detected, prevalence of reactive serology ranged from 2·45% in Benin (Ayelo G, personal communication) to 31·05% in Tanna Island, Vanuatu.26 Seroprevalence estimates were high in all three studies from the western Pacific region.25, 26, 27 Other studies reporting high seroprevalence were done in Lobaye, Central African Republic (19·72%).9 In Ecuador, after the implementation of a yaws surveillance and treatment programme, serological surveys done in 1998 showed a low prevalence of reactive serology (3·54%),29 and a survey in India in 2005 reported no sero-reactors among 3821 children younger than 5 years.31 Table 2 summarises health-facility-based incidence studies that used passive case finding.11, 12, 15, 17, 21, 22, 23 In the study in Nigeria, the results for skin diseases were reported in 2001, mainly in adults, but no cases of yaws were noted.17 Among the remaining studies, incidence of yaws ranged between 0·15 cases per 1000 population-years in Côte d'Ivoire11 to 25·56 per 1000 population-years in a highly endemic area of Papua New Guinea.22

y in Nigeria, the results for skin diseases were reported in 2001, mainly in adults, but no cases of yaws were noted.17 Among the remaining studies, incidence of yaws ranged between 0·15 cases per 1000 population-years in Côte d'Ivoire11 to 25·56 per 1000 population-years in a highly endemic area of Papua New Guinea.22 During the 4-year period between 2010 and 2013, 256 343 yaws cases were reported to WHO from 11 countries and territories (figure 2). Togo and Timor-Leste are judged by WHO to be endemic, but did not report any case in the study period. Large-scale yaws control programmes have recently resulted in disease elimination in two countries (Ecuador and India).29, 31 Figure 2 shows the annual number of yaws cases in all countries with ongoing transmission. The reported number of active infections was below 300 per year in Benin, Cameroon, Central African Republic, Republic of Congo, and Democratic Republic of the Congo, but data were probably under-reported from all of these countries. 215 308 (84%) of 256 343 cases reported to WHO were from three countries—Papua New Guinea, Solomon Islands, and Ghana. Table 3 summarises the estimates of the number of people at risk of yaws, stratified by region. We estimated that, in 2012, 8944 8862 people were living in yaws-endemic areas: about 46·7 million people in Africa, 35·8 million in southeast Asia, and 7·0 million in the western Pacific. At-risk population estimates for Ghana, Côte d'Ivoire, and Indonesia might be revised down because not all communities in each endemic district in these countries are endemic for yaws.

ing in yaws-endemic areas: about 46·7 million people in Africa, 35·8 million in southeast Asia, and 7·0 million in the western Pacific. At-risk population estimates for Ghana, Côte d'Ivoire, and Indonesia might be revised down because not all communities in each endemic district in these countries are endemic for yaws. Figure 3 shows the cumulative number of yaws cases from 2010 to 2013 in the WHO Africa region, shown by subnational regions. Six subnational regions in Ghana were very highly endemic (ie, >5000 cases within the 4-year reporting period), including the Eastern, Central, Volta, Western, Ashanti, and Brong-Ahafo regions. In Côte d'Ivoire, the regions of Fromager, Sud-Bandama, Haut-Sassandra, and Bas-Sassandra were highly endemic (ie, 1000–4999 cases within the 4-year reporting period). The East region in Cameroon, Likouala department in the Republic of Congo, and Lobaye prefecture in Central African Republic, which are close to one another, were all moderately endemic (ie, 100–999 cases within the 4-year reporting period). Data from Central African Republic were limited to surveys in two regions and the situation of the rest of the country remains to be investigated.

Congo, and Lobaye prefecture in Central African Republic, which are close to one another, were all moderately endemic (ie, 100–999 cases within the 4-year reporting period). Data from Central African Republic were limited to surveys in two regions and the situation of the rest of the country remains to be investigated. Figure 4 shows the cumulative number of yaws cases in the WHO western Pacific and southeast Asia regions within the 4-year period, shown by subnational region. In Papua New Guinea, five provinces were very highly endemic (>5000 cases)—New Ireland, West New Britain, East New Britain, Madang, and Autonomous Region of Bougainville provinces—whereas seven provinces were highly endemic (1000–4999 cases). The Western province in Solomon Islands, and Tafea province in Vanuatu were also very highly endemic. In Indonesia, most cases were found in the province of Nusa Tenggara Timur, where 13 084 cases were reported during the 4-year period. No recent surveillance data have been reported from Timor-Leste, but the country is regarded as endemic according to WHO.

nt breast tumours, diabetes, and cardiovascular disease) is estimated to amount to a third of the country's public health-care budget. Unless effective glycaemic control among patients with diabetes becomes commonplace, fatal diabetic complications will continue being not only a health burden, but also an economic one. In these circumstances, to meet the SDG3 targets, Mexico will need to increase its health investments. In line with international targets, this would mean reaching a level for government funding of 5% of GDP. But Mexico will also need to improve efficiency, which requires financial and institutional reform and political will to ensure that resources are allocated in an optimum way. Optimum allocation will require changes in how resources are distributed across different parts of the country, as well as how they are allocated across institutions, interventions, and health-care providers. Attention should be focused on poor-performing states and on the emerging health crisis of NCDs and injuries. A so-called whole health systems universal health coverage approach that includes all health system functions and major actors, not limited to Seguro Popular or health insurance for financial protection alone, will also be needed.

s, and Tafea province in Vanuatu were also very highly endemic. In Indonesia, most cases were found in the province of Nusa Tenggara Timur, where 13 084 cases were reported during the 4-year period. No recent surveillance data have been reported from Timor-Leste, but the country is regarded as endemic according to WHO. Discussion Our data show that about 65 000 yaws cases per year occurred in 13 endemic countries and that in at least 19 countries the incidence of yaws is unknown; thus, there has been limited progress since the last systematic review on yaws epidemiology in 1992 (85 000 yaws cases in 33 endemic countries).7 In 1953, Hackett32 estimated there were 50–150 million cases of yaws in 90 countries. A substantial decrease in the prevalence of yaws was brought about by the implementation of mass treatment campaigns and subsequent surveillance activities in the 1950s and 1960s. In many countries, yaws control and surveillance activities stopped after 1970, with a subsequent resurgence of yaws, particularly in parts of west and central Africa and in southeast Asia.7 Little activity to control the infection has been undertaken since 1990. The scarcity of political will, inadequate funding, and weaknesses in primary health-care systems in affected countries have been the biggest obstacles to the reduction of the burden of yaws in the past two decades.

Africa and in southeast Asia.7 Little activity to control the infection has been undertaken since 1990. The scarcity of political will, inadequate funding, and weaknesses in primary health-care systems in affected countries have been the biggest obstacles to the reduction of the burden of yaws in the past two decades. The methods proposed for assessing yaws burden have not changed substantially since 1953; however, unlike in the previous review by Hackett,32 who sent a questionnaire to all countries in Africa and carefully analysed the replies, or in the review by Meheus and Antal,7 who compiled original data from country reports submitted to WHO, we also extracted and synthesised a large amount of data from published studies, and complemented this with data from grey literature.

t a questionnaire to all countries in Africa and carefully analysed the replies, or in the review by Meheus and Antal,7 who compiled original data from country reports submitted to WHO, we also extracted and synthesised a large amount of data from published studies, and complemented this with data from grey literature. An important finding of our work is that almost 85% of all infections occurred in three countries—Ghana, Papua New Guinea, and Solomon Islands.6 The results of individual studies in these countries, which showed high prevalence and incidence rates, are consistent with integrated surveillance data. An overall low number of cases have been reported in national surveillance programmes in other countries in central Africa.6 However, we have shown that focal indigenous populations (ie, Pygmies) in the Central African Republic, Cameroon, Republic of Congo, and Democratic Republic of the Congo are affected by yaws, with prevalence of active disease ranging between 3% and 11% (Boua B, personal communication).9, 10, 14 The main risk factor for these groups, as for in other settings in which yaws is highly endemic, is the scarcity of access to health care and poor personal hygiene.

ocratic Republic of the Congo are affected by yaws, with prevalence of active disease ranging between 3% and 11% (Boua B, personal communication).9, 10, 14 The main risk factor for these groups, as for in other settings in which yaws is highly endemic, is the scarcity of access to health care and poor personal hygiene. Among the 13 known endemic countries, we estimated that a maximum of about 89 million people were living in yaws-endemic areas. In view of the focal nature of the disease, the size of the population at risk, in particular in Ghana, Côte d'Ivoire, and Indonesia, is uncertain. This global estimate of at-risk individuals would probably be revised down if community-based surveys were used to guide the implementation of mass treatment. The major limitation of our study is the weakness of routinely reported data. Yaws is not a notifiable disease and the use of national routine surveillance data is likely to result in an underestimation of the real number of cases because yaws predominantly occurs in rural communities with poor access to health facilities, whereas available data are primarily from health facilities. The limited reliability of clinical diagnoses of yaws and the recognition that other organisms can cause clinically similar skin lesions in yaws-endemic countries33, 34 causes problems for clinical case reporting. The weakness of reported data shows the limitations of the present data and supports the need for surveys as per the WHO strategy.

of clinical diagnoses of yaws and the recognition that other organisms can cause clinically similar skin lesions in yaws-endemic countries33, 34 causes problems for clinical case reporting. The weakness of reported data shows the limitations of the present data and supports the need for surveys as per the WHO strategy. We did not undertake a meta-analysis for several reasons. First, the studies that we included were primarily implemented in settings where yaws is endemic and no random sampling from a general population was done. Hence, the prevalence estimates are not representative of a given district, province, or an entire country. Second, the number of studies from each WHO region was limited. Third, the inclusion and diagnostic criteria varied markedly between studies, with both children and adults and both clinical and serological definitions of yaws included. These factors make direct comparison of the survey findings difficult.

tire country. Second, the number of studies from each WHO region was limited. Third, the inclusion and diagnostic criteria varied markedly between studies, with both children and adults and both clinical and serological definitions of yaws included. These factors make direct comparison of the survey findings difficult. The results of this systematic review contribute to the epidemiological knowledge needed to guide the preliminary estimation of resources that are necessary for a successful eradication programme. The inability of several countries to undertake more active surveillance and surveys is a major obstacle to achieving the WHO 2020 eradication target. The weaknesses of routinely reported data shows the need to establish a strict and sensitive surveillance system similar to other eradication programmes (eg, for Guinea worm and poliovirus) in a way that enables regionalisation of cases to make the decision about which communities need mass treatment and other control interventions. This online publication has been corrected. The corrected version first appeared at thelancet.com on June 8, 2015

The results of this systematic review contribute to the epidemiological knowledge needed to guide the preliminary estimation of resources that are necessary for a successful eradication programme. The inability of several countries to undertake more active surveillance and surveys is a major obstacle to achieving the WHO 2020 eradication target. The weaknesses of routinely reported data shows the need to establish a strict and sensitive surveillance system similar to other eradication programmes (eg, for Guinea worm and poliovirus) in a way that enables regionalisation of cases to make the decision about which communities need mass treatment and other control interventions. This online publication has been corrected. The corrected version first appeared at thelancet.com on June 8, 2015 Acknowledgments MM is supported by a Wellcome Trust Clinical Research Fellowship (WT102807). QB has received a fellowship from the programme Miguel Servet of the ISCIII (Plan Nacional de I + D + I 2008–2011, grant number CP11/00269). ZZ, LSV, and KA are staff members of WHO. The authors alone are responsible for the views expressed in this article and they do not necessarily represent the decisions or policies of the institutions they represent. We are grateful to Alexei Mikhailov for image editing.

de I + D + I 2008–2011, grant number CP11/00269). ZZ, LSV, and KA are staff members of WHO. The authors alone are responsible for the views expressed in this article and they do not necessarily represent the decisions or policies of the institutions they represent. We are grateful to Alexei Mikhailov for image editing. Contributors OM, MM, and KA had the original idea for the study. DJPK, GA, BB, WH, YK, ENT, AN, DO, FT, RD, and ZZ were involved in data gathering and analysis. OM, MM, and CG-B wrote the first draft of the report, with revisions and input from JU, LSV, QB, and KA. All authors contributed to revisions and approved the final version. Declaration of interests We declare no competing interests. Figure 1 Selection of eligible articles Figure 2 Annual absolute number of yaws cases by country Incidence given in cases per 100 000 population-years in 2010–12. Figure 3 Cumulative number of yaws cases by subnational regions in the WHO Africa region Figure 4 Cumulative number of yaws cases by subnational regions in the WHO southeast Asia and western Pacific regions Table 1 Characteristics and outcomes of the 24 included studies of active and latent yaws prevalence

Incidence given in cases per 100 000 population-years in 2010–12. Figure 3 Cumulative number of yaws cases by subnational regions in the WHO Africa region Figure 4 Cumulative number of yaws cases by subnational regions in the WHO southeast Asia and western Pacific regions Table 1 Characteristics and outcomes of the 24 included studies of active and latent yaws prevalence Year of study Country Location Schoolchildren or community survey Case ascertainment Cases (sample size) Prevalence, % (95% CI) Africa Active yaws assessment Tabah et al (2012; Tabah EN, personal communication) 2012 Cameroon Lomié, Zoubalot, Messok Community Clinical 97 (1075) 9·02 (7·38–10·90) Herve et al (1992)9 1990 Central African Republic Lobaye School children VDRL and TPHA 12 (213) 5·63 (2·94–9·63) Boua et al (2012; Boua B, personal communication) 2012 Central African Republic Lobaye, Sangha-Mbaeré School children Clinical 230 (2030) 11·33 (9·98–12·79) Coldiron et al (2013)10 2012 Republic of Congo Bétou, Ebyellé Community RDT 183 (6215) 2·94 (2·54–3·40) Konan et al (2007)13 2004 Côte d'Ivoire Adzopé Community RPR 11 (2182) 0·50 (0·25–0·90) Gerstl et al (2009)14 2005 Democratic Republic of the Congo Wasolo Community RPR and TPHA 56 (1176) 4·76 (3·62–6·14) Nsiire et al (2011; Nsiire A, personal communication) 2011 Ghana Volta Region School children ND 3159 (125 364) 2·52 (2·43–2·61) Akogun (1999)16 1998 Nigeria Garkida Community Clinical 64 (1523) 4·20 (3·25–5·33) Latent yaws assessment Ayelo et al (2012; Ayelo G, personal communication) 2012 Benin Toffo, Zé, Allada School children RPR 22 (900) 2·44 (1·54–3·68) Herve et al (1992)9 1990 Central African Republic Lobaye School children VDRL and TPHA 42 (213) 19·72 (14·60–25·70) Western Pacific Active yaws assessment Backhouse et al (1998)20 1988 Papua New Guinea Karkar Island School children VDRL, FTA-Abs, and TPHA 26 (632) 4·11 (2·70–5·97) Manning and Ogle (2002)21 2001 Papua New Guinea Port Moresby–NCD School children VDRL and TPHA 33 (227) 14·54 (10·22–19·81) Harris et al (1991)24 1989 Vanuatu Tanna Island Community VDRL 464 (20 200) 2·30 (2·09–2·51) Latent yaws assessment de Noray et al (2003)25 2001 Vanuatu Santo Island Community VDRL 57 (273) 20·88 (16·21–26·19) Fegan et al (2010)26 2008 Vanuatu Tanna Island Community VDRL and TPHA 95 (306) 31·05 (25·90–36·56) Guerrier et al (2011)27 2010 Wallis and Futuna Wallis and Futuna Community RPR and TPHA 27 (264) 10·23 (6·85–14·53) Southeast Asia Active yaws assessment Noordhoek et al (1991)18 1988 Indonesia Sumatra School children VDRL, TPHA, FTA-Abs, TmpA EIA

focused on poor-performing states and on the emerging health crisis of NCDs and injuries. A so-called whole health systems universal health coverage approach that includes all health system functions and major actors, not limited to Seguro Popular or health insurance for financial protection alone, will also be needed. Dealing with NCDs and injuries presents substantial operational and financial challenges for the health system, as well as opportunities to achieve SDG3 following a 40×30 strategy. Mexico will need to expand health insurance packages beyond WHO's suggested best-buy interventions for NCDs to include a much wider range of early detection, treatment, and management of these diseases. Strengthening primary care would be the best route for the system to maintain a focus on both maternal and child disorders while dealing with NCDs and their associated risk factors. The effective treatment of NCDs will require restructuring service delivery in a way that enables not only a shift of attention towards prevention, but also a dramatic move from episodic care to continuous care for chronic disorders, including survivorship and palliative care. Mexico's path to SDG 40×30 should be anchored in both deepening entitlements to a broader package of care services with financial protection while at the same time extending the gains achieved in MNCH with core efforts in primary care. This holistic approach echoes the SDGs in maximising the synergies between health gains and poverty reduction.

anuatu Tanna Island Community VDRL and TPHA 95 (306) 31·05 (25·90–36·56) Guerrier et al (2011)27 2010 Wallis and Futuna Wallis and Futuna Community RPR and TPHA 27 (264) 10·23 (6·85–14·53) Southeast Asia Active yaws assessment Noordhoek et al (1991)18 1988 Indonesia Sumatra School children VDRL, TPHA, FTA-Abs, TmpA EIA , and WB 114 (37 000) 0·31 (0·25–0·37) dos Santos et al (2010)19 2007 Timor-Leste Oecusse, Bobonaro, Cova Lima, Atauro Island Community Clinical 6 (1535) 0·39 (0·14–0·85) Latent yaws assessment WHO India (2006)31 2005 India Ten states School children RPR and TPHA 0 (3831) 0·00 (0·00–0·00) The Americas Active yaws assessment Anselmi et al (1995)28 1993 Ecuador Santiago basin Community VDRL and FTA-Abs 16 (1118) 1·43 (0·82–2·31) Anselmi et al (2003)29 1998 Ecuador Santiago basin Community VDRL and FTA-Abs 0 (1926) 0·00 (0·00–0·19) Scolnik et al (2003)30 2000 Guyana Bartica School children MHA-TP 52 (1020) 5·10 (3·83–6·63) Latent yaws assessment Anselmi et al (1995)28 1993 Ecuador Santiago basin Community VDRL and FTA-Abs 53 (1118) 4·74 (3·57–6·16) Anselmi et al (2003)29 1998 Ecuador Santiago basin Community VDRL and FTA-Abs 68 (1926) 3·53 (2·75–4·45) FTA-Abs=fluorescent treponemal antibody–absorption. MHA-TP=microhaemagglutination assay–Treponema pallidum. NCD=National Capital District. ND=not documented. RDT=rapid diagnostic test. RPR=rapid plasma reagin. TmpA EIA=enzyme immunoassay with TmpA antigen. TPHA=T pallidum haemagglutination. VDRL=Venereal Disease Research Laboratory. WB=western blot with T pallidum subspecies pallidum as antigen.

ion assay–Treponema pallidum. NCD=National Capital District. ND=not documented. RDT=rapid diagnostic test. RPR=rapid plasma reagin. TmpA EIA=enzyme immunoassay with TmpA antigen. TPHA=T pallidum haemagglutination. VDRL=Venereal Disease Research Laboratory. WB=western blot with T pallidum subspecies pallidum as antigen. Table 2 Characteristics and outcomes of health-facility-based active yaws incidence studies Period of study Country Location Target population Case ascertainment New cases (at-risk population) Incidence, cases per 1000 population-years (95% CI) Africa Toure et al (2007)11 2000 Côte d'Ivoire Nationwide Children and adults Clinical 9212 (15 882 758) 0·58 (0·57–0·59) Konan et al (2013)12 2011 Côte d'Ivoire Nationwide Children and adults Clinical 3343 (22 594 212) 0·15 (0·14–0·15) Edorh et al (1994)15 1991 Togo Nationwide School children Clinical 3750 (3 787 000) 0·99 (0·96–1·02) Nnoruka (2005)17 1999–2001 Nigeria Enugu Hospital Children and adults Clinical 0 (2871) 0·00 (0·00–1·28) Western Pacific Manning and Ogle (2002)21 2000–01 Papua New Guinea Port Moresby Children and adults RPR and TPHA 494 (20 000) 24·70 (22·59–26·95) Mitja et al (2011)22 2009 Papua New Guinea Lihir Island School children RPR and TPHA 138 (5 400) 25·56 (21·51–30·12) Ministry of Health, Solomon Islands (2013)23 2012 Solomon Islands Nationwide Children and adults Clinical 12 372 (515 870) 23·98 (23·57–24·40) RPR=rapid plasma reagin. TPHA=Treponema pallidum haemagglutination. Table 3 Estimates of at-risk populations living in districts judged to be endemic (second administrative level; 2012)

Period of study Country Location Target population Case ascertainment New cases (at-risk population) Incidence, cases per 1000 population-years (95% CI) Africa Toure et al (2007)11 2000 Côte d'Ivoire Nationwide Children and adults Clinical 9212 (15 882 758) 0·58 (0·57–0·59) Konan et al (2013)12 2011 Côte d'Ivoire Nationwide Children and adults Clinical 3343 (22 594 212) 0·15 (0·14–0·15) Edorh et al (1994)15 1991 Togo Nationwide School children Clinical 3750 (3 787 000) 0·99 (0·96–1·02) Nnoruka (2005)17 1999–2001 Nigeria Enugu Hospital Children and adults Clinical 0 (2871) 0·00 (0·00–1·28) Western Pacific Manning and Ogle (2002)21 2000–01 Papua New Guinea Port Moresby Children and adults RPR and TPHA 494 (20 000) 24·70 (22·59–26·95) Mitja et al (2011)22 2009 Papua New Guinea Lihir Island School children RPR and TPHA 138 (5 400) 25·56 (21·51–30·12) Ministry of Health, Solomon Islands (2013)23 2012 Solomon Islands Nationwide Children and adults Clinical 12 372 (515 870) 23·98 (23·57–24·40) RPR=rapid plasma reagin. TPHA=Treponema pallidum haemagglutination. Table 3 Estimates of at-risk populations living in districts judged to be endemic (second administrative level; 2012) Population of country* Health districts reporting yaws (n/N [%]) Population living in endemic districts Africa Benin† 9 364 619 2/34 (5·9%) Minimum 632 488. Total not known Cameroon 22 128 420 22/179 (12·3%) 2 360 944 Central African Republic‡ 4 600 125 2/17 (11·8%) Minimum 434 521. Total not known Republic of Congo 4 001 831 16/84 (19·0%) Minimum 1 555 513 Côte d'Ivoire 23 261 022 56/81(69·1%) 18 000 000 Democratic Republic of the Congo 75 507 000 ND/36 Not known§ Ghana 24 658 823 160/170 (94·1%) 23 178 000 Togo 6 191 155 2/35 (5·7%) 545 729 Western Pacific Papua New Guinea 7 146 240 75/89 (84·3%) 6 201 393 Solomon Islands 515 870 10/10 (100%) 515 870 Vanuatu 234 023 6/6 (100%) 234 023 Southeast Asia Indonesia 241 692 190 106/497(21·3%) 34 588 881 Timor-Leste 120 1500 13/13 (100%) 120 1500 ND=no data.

0 (94·1%) 23 178 000 Togo 6 191 155 2/35 (5·7%) 545 729 Western Pacific Papua New Guinea 7 146 240 75/89 (84·3%) 6 201 393 Solomon Islands 515 870 10/10 (100%) 515 870 Vanuatu 234 023 6/6 (100%) 234 023 Southeast Asia Indonesia 241 692 190 106/497(21·3%) 34 588 881 Timor-Leste 120 1500 13/13 (100%) 120 1500 ND=no data. * From 2012, except Ghana (2010) and Vanuatu and Indonesia (2009). † Accurate data were only available for two districts. The prevalence of yaws in the remaining 32 districts was not known. ‡ Accurate data were only available for two districts. The prevalence of yaws in the remaining 22 districts was not known. § District-level data were not available to allow an accurate calculation of the population at risk.

The Royal Society of Tropical Medicine and Hygiene is holding a 1-day meeting in London, UK, on Sept 25, 2015, to discuss “The disease elimination agenda: the role of science, policy and advocacy”.1 Important meetings such as this are a welcome forum at which to discuss progress and challenges and to reflect on important milestones, particularly with regard to neglected tropical diseases (NTDs). Such NTD milestones include the London declaration2 and the WHO roadmap.3 However, the success of these elimination initiatives is contingent on inclusive programming and on addressing the problem in its entirety. Most of the targets for elimination focus on interrupting transmission and infection cycles, yet many of the NTDs cause severe morbidity including disabling lymphoedema, massive hydrocele, disfigurement, and blindness.4 Despite the huge burden of morbidities, there are no clear targets towards their elimination, and, with the exception of trachoma, there are no morbidity indictors to measure the success of elimination.5 A more inclusive approach to addressing morbidity in the elimination of NTDs should focus on the following points.

4 Despite the huge burden of morbidities, there are no clear targets towards their elimination, and, with the exception of trachoma, there are no morbidity indictors to measure the success of elimination.5 A more inclusive approach to addressing morbidity in the elimination of NTDs should focus on the following points. First, elimination targets should clearly include indicators related to morbidity. Such indicators should go beyond measuring access to care and should bind success to the extent of morbidity alleviation. WHO’s trachoma elimination target of a prevalence of active trachoma of less than 5% among children aged 1–9 years and a prevalence of trachoma trichiasis of less than one case per 1000 population5 successfully combines both prevention of new infections and reduction of morbidity, and should be replicated across the different diseases. Second, resources should be clearly committed to the morbidity management aspect of these NTDs. Funding such as the USAID’s support of Helen Keller International’s Morbidity Management and Disability Prevention for Blinding Trachoma and Lymphatic Filariasis Project is welcome.6 Given the scale of the problem, more resources to address the morbidity challenge are required. As resources are directed towards preventing new infection, equally resources should also be targeted to improving the quality of life of the people suffering from the consequences of the diseases.

iasis Project is welcome.6 Given the scale of the problem, more resources to address the morbidity challenge are required. As resources are directed towards preventing new infection, equally resources should also be targeted to improving the quality of life of the people suffering from the consequences of the diseases. Third, operational research into optimising the delivery of morbidity management services is also important. One of the challenges of scaling up such services is a dearth of evidence on how to integrate them into the existing health systems and ongoing NTD programmes. Operational research focusing on integration of services, surveillance, and barriers to the existing services will be important. To achieve the challenges of elimination, morbidity management is essential, not optional. Strong advocacy and awareness-raising for donors is important, but the change should start from within by including morbidity targets in some of the NTD elimination targets. I am supported by a Wellcome Trust Fellowship in Public Health and Tropical Medicine (grant number 099876). I declare that I have no competing interests.

Introduction The Sustainable Development Goal for health (SDG3), adopted in September 2015, poses enormous challenges for all countries. Underpinning the overall objective to ensure healthy lives and promote wellbeing for all at all ages is a list of nine proposed targets (appendix).1 Previous analyses undertaken to inform the SDG3 targets provide evidence and proposals on target measurement and on what could be achievable at a global level based on observed country-specific mortality.2, 3, 4 These analyses show that country roadmaps to meet the SDG3 targets require a deeper analysis of specific mortality trends and a pragmatic approach to linking targets with cost-effective health interventions.3, 5 Norheim and colleagues3 proposed an overarching quantitative target to support SDG3, namely to avoid 40% of premature deaths by 2030 (premature deaths defined as those of individuals younger than 70 years), which in turn would be the result of achieving a series of cause-specific mortality subtargets (appendix).3 This ambitious undertaking can be reached only by clearly understanding the main causes of death by age group and the effective interventions available to affect those causes and their respective modifiable risk factors. Countries have the opportunity to apply specific and unique policy strategies targeted to their national health priorities and health system capabilities. Therefore, there are different ways to achieve a so-called 40 by 30 (40 × 30) overall target.

le to affect those causes and their respective modifiable risk factors. Countries have the opportunity to apply specific and unique policy strategies targeted to their national health priorities and health system capabilities. Therefore, there are different ways to achieve a so-called 40 by 30 (40 × 30) overall target. The Lancet Commission on Investing in Health6 showed that a grand convergence in health (a reduction in avertable infectious, child, and maternal mortality down to universally low levels) could be achieved by 2035. Since publication of its report titled Global Health 2035, the Commission on Investing in Health working group has embarked on a series of consultations with donor agencies and ministries of health and finance to explore the implications for investments in health. One of these engagements has been with Mexico's Ministry of Health on how to achieve SDG3 and get on track to converge to the better performing high-income countries (eg, among the Organization for Economic Cooperation and Development [OECD] countries of which Mexico is a member country).7 Mexico achieved major improvements in preventable maternal, newborn, and child health (MNCH), and substantial gains in life expectancy over the past 30 years. Despite this, Mexico is still behind most other OECD countries in several key areas ranging from comparatively high mortality from MNCH causes to the rising rates of non-communicable diseases (NCDs) and fatal injuries. As a result, life expectancy increase has slowed and is far from convergence with OECD countries (panel 1).8

this, Mexico is still behind most other OECD countries in several key areas ranging from comparatively high mortality from MNCH causes to the rising rates of non-communicable diseases (NCDs) and fatal injuries. As a result, life expectancy increase has slowed and is far from convergence with OECD countries (panel 1).8 Research in context Evidence before this study Our work builds on the Lancet Commission on Investing in Health and Norheim and colleagues' work. These sources provide a robust general framework for global convergence to explore how Mexico can reach Sustainable Development Goal 3 (SDG3). We build upon these analyses using United Nations Population Division age-specific mortality estimates and Mexican National Institute of Statistics and Geography age-specific and cause-specific mortality data, calibrating our results to the United Nations life expectancy projections for international comparability. We identified best-practice interventions based on the recommendations of international organisations and initiatives including WHO, the Organisation for Economic Co-operation and Development, and Disease Control Priorities, Third Edition. Added value of this study Our findings shed light on the underlying causes of mortality by age group that are limiting Mexico's potential to achieve SDG3 and thus, convergence in health status. Our study provides a country roadmap to identify and prioritise the health interventions that can support achievement of SDG3 and reduce the gap in life expectancy by 2030. Our approach can be used by other countries to operationalise SDG3.

ting Mexico's potential to achieve SDG3 and thus, convergence in health status. Our study provides a country roadmap to identify and prioritise the health interventions that can support achievement of SDG3 and reduce the gap in life expectancy by 2030. Our approach can be used by other countries to operationalise SDG3. Implications of this study Our study can assist national authorities to prioritise areas of action with the greatest health impact and economic returns. Effective priority setting is important in the face of budgetary constraints, and in evidence-based dialogue with other governmental authorities, such as the Ministry of Finance, and with other non-health sectors whose actions have an effect on health (eg, transport, education, and violence prevention). In this study we project shortfalls, relative to a target of a 40% reduction of premature deaths, in age-specific and cause-specific mortality in Mexico by comparing three different mortality scenarios in 2030. Following the Commission on Investing in Health6 notion of grand convergence, we use these results to forecast life expectancy at birth and analyse Mexico's potential to meet SDG3 and reach substantial convergence to better performing high-income countries.

xico by comparing three different mortality scenarios in 2030. Following the Commission on Investing in Health6 notion of grand convergence, we use these results to forecast life expectancy at birth and analyse Mexico's potential to meet SDG3 and reach substantial convergence to better performing high-income countries. Methods Annual population estimates for 1990–2014 and projections based on the medium fertility variant for 2015–30 were taken from the United Nations Population Division (UNPD) World Population Prospects 2015 revision. The UNPD 5-year periods for mortality estimates were averaged to obtain midpoint estimates; these in turn were averaged to obtain estimates for every fifth year between 1990 and 2030. Intermediate years were then constructed assuming linear trends between these 5-year estimates. Population and deaths were grouped into five age groups: 0–4 years, 5–19 years, 20–49 years, 50–69, and 70 years or more.

estimates; these in turn were averaged to obtain estimates for every fifth year between 1990 and 2030. Intermediate years were then constructed assuming linear trends between these 5-year estimates. Population and deaths were grouped into five age groups: 0–4 years, 5–19 years, 20–49 years, 50–69, and 70 years or more. We used cause-specific mortality estimates from the vital statistics reported by the National Institute of Statistics and Geography (INEGI). The INEGI cause-specific mortality data were combined with UNPD mortality and population estimates and projections to construct three mortality scenarios. INEGI annual reported deaths between 1990 and 2014 were grouped into 15 major categories representative of the most important disease clusters to obtain annual distributions of deaths by category across each of the five age groups (appendix). Deaths with an ill-defined or otherwise unspecified cause (International Classification of Diseases, 10th Revision [ICD-10] codes R00–R99; 2·3% for 1990 and 1·7% for 2014) or unspecified age (0·7% for 1990 and 0·5% for 2014) were redistributed using the observed cause and age distributions, respectively. The resulting cause-specific distributions for the five age groups were applied to the UNPD deaths and population figures to produce adjusted figures for deaths and death rates by age group and disease category from 1990 to 2014.

for 2014) were redistributed using the observed cause and age distributions, respectively. The resulting cause-specific distributions for the five age groups were applied to the UNPD deaths and population figures to produce adjusted figures for deaths and death rates by age group and disease category from 1990 to 2014. We constructed three scenarios with age-specific and cause-specific mortality estimates for 2030. The baseline scenario reflects changes in mortality exclusively associated with population growth and aging, and does not account for changes in mortality rates. Mortality rates were assumed to remain at the 2010 levels. Therefore, the 2010 age-specific and cause-specific mortality rates were applied to the 2030 UNPD population prospects. We used 2010 as reference year to ensure consistency with the SDG3 and the work of Norheim and colleagues.3

t for changes in mortality rates. Mortality rates were assumed to remain at the 2010 levels. Therefore, the 2010 age-specific and cause-specific mortality rates were applied to the 2030 UNPD population prospects. We used 2010 as reference year to ensure consistency with the SDG3 and the work of Norheim and colleagues.3 The so-called inertial scenario projects cause-specific mortality for 2030 assuming recent (2000–14) trends will continue so that expected progress is limited to merely letting policies and mortality trends follow their present path. With the statistical software R, we produced mortality rate projections for 2015–30 by disease category for each age group within the 0–69 years age range using a linear regression model based on the INEGI–UNPD adjusted 2000–14 mortality rates. The projected mortality rates were applied to the UNPD population structure to produce estimates of deaths for 2015–30. These distributions across disease categories for all years and age groups were then adjusted to the UNPD 2015–30 medium-variant expected premature deaths for each age category to obtain cause-specific UNPD compatible estimates for deaths and death rates across the entire 1990–2030 time period.

s of deaths for 2015–30. These distributions across disease categories for all years and age groups were then adjusted to the UNPD 2015–30 medium-variant expected premature deaths for each age category to obtain cause-specific UNPD compatible estimates for deaths and death rates across the entire 1990–2030 time period. We used trends for the 2000–14 period (instead of the 1990–2014 period) to show the more recent evolution of mortality and to avoid potential biases from coding in 1998, when the ICD-10 was introduced in Mexico. Linear trends show a particularly good fit for most causes for the 2000–14 period. In the case of homicides, however, the use of this timeframe leads to an overestimation of the expected number of deaths in this inertial scenario because of the large increase in the homicide rate between 2005 and 2011.34 Nevertheless, we used a single timeframe for purposes of consistency. The underlying trends used to estimate this scenario are shown in figure 1.

is timeframe leads to an overestimation of the expected number of deaths in this inertial scenario because of the large increase in the homicide rate between 2005 and 2011.34 Nevertheless, we used a single timeframe for purposes of consistency. The underlying trends used to estimate this scenario are shown in figure 1. The SDG 40 × 30 scenario uses the baseline scenario as the departure point and projects the number of deaths and the resulting death rates consistent with a target of a 40% reduction in overall premature mortality by 2030. We assumed a flat 40% rate of reduction across each age group in the 0–69 years age range and for each of the 15 causes. We compared deaths and death rates between the SDG 40 × 30 estimated target and the inertial scenario to identify disease and age groups and resulting policy areas that are off-track and need scaling up of interventions to accelerate mortality reductions, and thus meet SDG3 targets. We calculated life expectancy based on the inertial and SDG 40 × 30 scenarios to assess convergence, with Japan as the best performer and the USA as an underperformer in the OECD. The USA is a particularly relevant reference for Mexico given its large population of Mexican origin and because it faces similar health challenges, most notably rising levels of overweight and obesity, as well as an ageing population.

ence, with Japan as the best performer and the USA as an underperformer in the OECD. The USA is a particularly relevant reference for Mexico given its large population of Mexican origin and because it faces similar health challenges, most notably rising levels of overweight and obesity, as well as an ageing population. In order to develop a complete profile of mortality in all age groups over time, we included results for the population aged 70 years and older. For this age group, the baseline and inertial scenarios follow the same logic as for the 0–69 years age group, except that the cause-specific analysis was not undertaken. Baseline figures reflect the demographic effect of an aging population by keeping the 2010 mortality rate constant. Figures for the inertial scenario reflect the number of deaths as projected by UNPD, whereas the SDG 40 × 30 scenario assumes the same mortality as the inertial scenario.

ific analysis was not undertaken. Baseline figures reflect the demographic effect of an aging population by keeping the 2010 mortality rate constant. Figures for the inertial scenario reflect the number of deaths as projected by UNPD, whereas the SDG 40 × 30 scenario assumes the same mortality as the inertial scenario. Role of the funding source This study was partly supported by the Ministry of Health in Mexico, the Bill & Melinda Gates Foundation, and the University of California, San Francisco. The funders of this study had no role in the study design, or conduct, data collection, data management, data, analysis, data interpretation, or writing of the report. The corresponding author, as Under-Secretary of Health of Mexico from 2014 to March, 2016, was responsible for the General Directorate of Health Information, which provides information on vital statistics in coordination with INEGI. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

of Mexico from 2014 to March, 2016, was responsible for the General Directorate of Health Information, which provides information on vital statistics in coordination with INEGI. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results According to UNPD estimates, the Mexican population will age substantially in the coming decades and grow from 118·6 million people in 2010 to 148·1 million people in 2030. Under the baseline scenario, total mortality is projected to increase by 82% between 2010 and 2030 (table 1). Nearly half of this increased mortality (46%) is a product of population aging, which is projected to result in more than 810 000 additional deaths by 2030. The remaining almost 202 000 additional deaths are the result of population growth. Most of the increased mortality occurs in older populations. Premature mortality (at age 0–69 years) is projected to increase by 49% and mortality at age 70 years and older is projected to increase by 122%.

00 additional deaths by 2030. The remaining almost 202 000 additional deaths are the result of population growth. Most of the increased mortality occurs in older populations. Premature mortality (at age 0–69 years) is projected to increase by 49% and mortality at age 70 years and older is projected to increase by 122%. The SDG 40 × 30 scenario seeks to avert 40% of yearly premature deaths in 2030 compared with the baseline scenario. However, UNPD projections for 2030 show that premature mortality will fall by only 26%, which is almost 64 000 deaths short of the SDG 40 × 30 target. Although mortality rates decreased for all age groups between 1990 and 2010 and are expected to drop further by 2030, the overall rate of decrease has slowed substantially and masks varying trends in mortality across age groups. From 1990 to 2010 the death rate for children aged 0–4 years decreased by 48%, whereas for older adults (aged 50–69 years) it decreased by only 29%. From 2010 to 2030, reductions in mortality across all ages will be slower than in the previous two decades; however, mortality reductions in children and adolescents (aged 0–19 years) will continue to outpace gains in adults (aged 20–69 years). Under the inertial scenario, the decrease in premature mortality will be 42% for children aged 0–4 years and 33% for young people aged 5–19 years. By contrast, the decrease for adults aged 20–49 years will be 25% and for those aged 50–69 years it will be 23%.

will continue to outpace gains in adults (aged 20–69 years). Under the inertial scenario, the decrease in premature mortality will be 42% for children aged 0–4 years and 33% for young people aged 5–19 years. By contrast, the decrease for adults aged 20–49 years will be 25% and for those aged 50–69 years it will be 23%. The slower rates of decrease in adult mortality are of particular concern. Under the baseline scenario, 85% of premature deaths will occur in people aged 20–69 years, precisely where the least health gains are expected in the future. In fact, nearly all (98·5%) of the predicted shortfall in mortality reduction comparing the inertial with the SDG 40 × 30 target will be in those aged 20–69 years (about 17 000 excess deaths for those aged 20–49 years and 47 000 excess deaths for those aged 50–69 years). Mortality trends by major cause of death further explain the mortality dynamics seen across age groups. Trends used to construct the inertial scenario are shown in figure 1. In the baseline scenario, 13% of total premature mortality is associated with MNCH disorders and communicable diseases (table 2). By contrast, 70% is associated with NCDs, and 17% with injuries (see combined results by cause and age group in the appendix).

used to construct the inertial scenario are shown in figure 1. In the baseline scenario, 13% of total premature mortality is associated with MNCH disorders and communicable diseases (table 2). By contrast, 70% is associated with NCDs, and 17% with injuries (see combined results by cause and age group in the appendix). Childhood infectious diseases and adverse events associated with childbirth continue to represent an unfinished health agenda. Nevertheless, mortality from communicable, perinatal, maternal, and nutritional causes shows the most impressive downward trends up to 2030. Under the inertial scenario, the SDG 40 × 30 targets for MNCH disorders, as well as mortality targets for children and adolescents (aged 0–19 years), are expected to be met by a good margin (table 1). However, sustained efforts to ensure that recent trends continue for this cluster will be necessary to attain the target.

der the inertial scenario, the SDG 40 × 30 targets for MNCH disorders, as well as mortality targets for children and adolescents (aged 0–19 years), are expected to be met by a good margin (table 1). However, sustained efforts to ensure that recent trends continue for this cluster will be necessary to attain the target. A cluster of four major non-communicable disease groups, vascular disease (including cardiovascular diseases, stroke, and hypertensive diseases), diabetes, renal disease, and liver disease, will account for more than 40% of total premature deaths in the baseline scenario, maintaining their present position as a leading cause of premature mortality. From 1990 to 2010, the share of mortality from this cluster, relative to total premature deaths, increased from 72 714 (23·1%) to 103 573 (34·1%), and deaths from these causes were 1·4 times higher in 2010 than in 1990. Deaths from cancer also increased during this period, though their contribution to premature mortality increased less, from 28 640 (9·1%) to 37 669 (12·4%).

ter, relative to total premature deaths, increased from 72 714 (23·1%) to 103 573 (34·1%), and deaths from these causes were 1·4 times higher in 2010 than in 1990. Deaths from cancer also increased during this period, though their contribution to premature mortality increased less, from 28 640 (9·1%) to 37 669 (12·4%). The largest challenge and the greatest opportunity to meet the SDG 40 × 30 target therefore lies in reversing the trends for mortality from NCDs. The inertial scenario will fall short of this goal by about 54 000 deaths in 2030. Half of this shortfall results from the very small projected reduction in diabetes mortality. Under present trends, expected mortality from diabetes will nearly double by 2030, and will be 1·6 times higher than the SDG 40 × 30 target. Similar challenges exist for renal disease and a diverse set of other NCDs (such as non-congenital thyroid and other endocrine disorders, mental and behavioural disorders, or other nervous, eye, ear, respiratory, and digestive system illnesses). However, the expected death rates from vascular disease, cirrhosis and other chronic liver diseases, and cancers are projected to be more in line with the SDG 40 × 30 target.

hyroid and other endocrine disorders, mental and behavioural disorders, or other nervous, eye, ear, respiratory, and digestive system illnesses). However, the expected death rates from vascular disease, cirrhosis and other chronic liver diseases, and cancers are projected to be more in line with the SDG 40 × 30 target. Mortality due to injuries has remained constant as a share of premature mortality over the past two decades. In 2010, non-road traffic and road traffic injuries, homicides, and suicides accounted for 53·1% of all deaths in individuals aged 5–19 years, and homicides were the leading cause of premature deaths for young adults (aged 20–49 years; appendix). However, owing to large variations in mortality rates across all age groups, homicide remains the most difficult cluster for which to predict trend behaviour over the next 15 years. Nevertheless, we find that homicides account for most of the projected shortfall in this cluster relative to the SDG 40 × 30 target. Projected deaths under the inertial scenario for road traffic accidents are in line with a 40% reduction in mortality, but will certainly fall short if we consider the SDG3 target for road safety, which is a 50% reduction in road traffic deaths and injuries by 2020.

l in this cluster relative to the SDG 40 × 30 target. Projected deaths under the inertial scenario for road traffic accidents are in line with a 40% reduction in mortality, but will certainly fall short if we consider the SDG3 target for road safety, which is a 50% reduction in road traffic deaths and injuries by 2020. Mexico has the lowest life expectancy of all OECD countries and its present path is far from converging with better performing countries. From 2000 to 2013, the longevity gap between Mexico and the OECD average widened from 4 years to almost 6 years. Figure 2 shows past and projected trends for Japan (the best performer), the USA (underperformer), and Mexico, including the projected life expectancy under the inertial scenario (an additional 3·8 years) and the SDG 40 × 30 scenario (4·7 additional years), reaching 81·4 years by 2030. In 2010, the difference in life expectancy between Mexico (76·7 years) and Japan (84·2 years) was 7·5 years. To close this gap in life expectancy with Japan, Mexico would need to substantially accelerate the annual rate of increase in life expectancy. This is an ambitious goal because the increase in life expectancy between 2000 and 2010 (2·3 years) was only two-thirds of that observed between 1990 and 2000 (3·5 years). A more reasonable goal for Mexico is to reach the United Nations estimate for life expectancy in the USA of 82·3 years by 2030.

in life expectancy. This is an ambitious goal because the increase in life expectancy between 2000 and 2010 (2·3 years) was only two-thirds of that observed between 1990 and 2000 (3·5 years). A more reasonable goal for Mexico is to reach the United Nations estimate for life expectancy in the USA of 82·3 years by 2030. By reviewing the age-specific gains in mortality projected under the SDG 40 × 30 scenario, we can assess the causes and trends in life expectancy in Mexico and identify the major shortfalls in mortality reduction that will need to be addressed for Mexico to converge with the USA. The differences in mortality rates by age group compared with the USA are presented in figure 3. The age-specific gains in mortality show that convergence can be achieved under the SDG 40 × 30 strategy if efforts are focused on the population aged 20–69 years. For older adults aged 50–69 years, mortality can even improve beyond the 2030 projection for the USA. Convergence for the young population would require substantial efforts that go beyond the SDG 40 × 30 target. The population aged older than 70 years is near convergence under UNPD-projected mortality, our assumption for this group for the SDG 40 × 30 scenario. However, given the high share of deaths in this group relative to total mortality, a greater reduction in mortality rates for this population group could result in a major shift in life expectancy. As expected, younger populations are still most affected by infectious diseases and childbirth mortality; homicides, non-road traffic injuries, and road traffic accidents affect mostly adults aged 20–49 years, and NCDs, such as diabetes, take their highest toll in older adults (aged 50 years and older).

ife expectancy. As expected, younger populations are still most affected by infectious diseases and childbirth mortality; homicides, non-road traffic injuries, and road traffic accidents affect mostly adults aged 20–49 years, and NCDs, such as diabetes, take their highest toll in older adults (aged 50 years and older). Discussion To reduce premature mortality in Mexico by 40% by 2030 will require implementation of cost-effective health interventions, focusing on those age groups and health disorders in which inertial trends are projected to fall short of the SDG 40 × 30 target. Maternal, neonatal, and child mortality gains are a success story and an important contributor to past increases in life expectancy, yet there is room to ensure that the expected favourable trends in this area are sustained to meet the SDG3. Mortality for these disorders is very concentrated in low-income and underserved populations who need to be reached with quality services (panel 1). Present coverage rates for antenatal care (98·6%) and facility-delivery (94·5%) are high, but the quality of these services will need to be improved and standardised to address the wide variation in state practices and reduce mortality.29, 39, 40, 41 Efforts should focus on ensuring full adherence to WHO recommendations and clinical guidelines for antenatal and delivery care, including continuous care from early stages of pregnancy and compliance with caesarean best practices, especially focusing on the poorest areas of the country where deaths are highest.42, 43, 44

forts should focus on ensuring full adherence to WHO recommendations and clinical guidelines for antenatal and delivery care, including continuous care from early stages of pregnancy and compliance with caesarean best practices, especially focusing on the poorest areas of the country where deaths are highest.42, 43, 44 Improved access to neonatal care is also needed to reduce mortality in preterm babies. Focus should be placed on expanding adequate use of thermal care, continuous positive airway pressure therapy, surfactants for respiratory distress syndrome, oxygen therapy,43 breastfeeding, and skin-to-skin care.45, 46 To accelerate the reduction in mortality for the cluster of NCD deaths, we propose a three-pronged approach organised around the life cycle and stage of disease continuum, in line with WHO recommendations and universal health coverage goals.47

Improved access to neonatal care is also needed to reduce mortality in preterm babies. Focus should be placed on expanding adequate use of thermal care, continuous positive airway pressure therapy, surfactants for respiratory distress syndrome, oxygen therapy,43 breastfeeding, and skin-to-skin care.45, 46 To accelerate the reduction in mortality for the cluster of NCD deaths, we propose a three-pronged approach organised around the life cycle and stage of disease continuum, in line with WHO recommendations and universal health coverage goals.47 First, scaling up healthy lifestyle interventions that target underlying risk factors, such as economic incentives through taxation, is the most cost-effective strategy, although its full effect can only be seen in the long run. More rigorous fiscal and regulatory policies for tobacco, sugar, salt, and alcohol consumption should be introduced, building on existing policies and knowledge. These policies are powerful levers for reduction of NCDs, increase of revenue that can be allocated to the health sector, and can also be in favour of poor or low-income populations.6, 48, 49 In Mexico, the tobacco tax rate is still low compared with other countries and could be doubled to avert new smokers, especially young people.50, 51 In 2014, Mexico introduced a 10% tax on sugar-sweetened non-alcoholic beverages and an 8% tax on high-calorie processed foods. Although the tax was not earmarked to health and thus the budget for preventive care did not increase as a result, taxes increased prices and early evidence shows that consumption patterns are starting to change as a result.52, 53 Improving the availability and consumption of drinking water could help to decrease consumption of sugar-sweetened beverages. Finally, reducing sodium consumption, such as through regulating processed food, can be a cost-effective intervention to reduce cardiovascular mortality.24, 47, 54

g to change as a result.52, 53 Improving the availability and consumption of drinking water could help to decrease consumption of sugar-sweetened beverages. Finally, reducing sodium consumption, such as through regulating processed food, can be a cost-effective intervention to reduce cardiovascular mortality.24, 47, 54 A second set of NCD-related priorities are associated with more effective primary care-based prevention and early detection interventions combined with disease management to avoid further progression, ensure access to treatment and disease control, and ensure post-treatment follow-up to prevent recurrence. The mortality related to diabetes shows the present situation and the urgency of action. Within OECD countries, Mexico has one of the highest rates of hospital admissions due to uncontrolled diabetes, twice the OECD average.17 Poor access to, and poor performance of, primary care are to blame, including limited access to drugs and supplies and inadequate monitoring of clinical markers. Only 24% of patients with diabetes are considered to be under adequate control, and 49% of the diabetic population is unaware of their condition.29, 55 However, diabetes tends to be overused as the underlying cause of death for cardio-metabolic disease and its complications, thus the need for better classification of all causes of deaths is another pending challenge. Similarly, estimates suggest that more than 47% of adults with high blood pressure are unaware of their condition, and among adults aware of being hypertensive, only 74% were under treatment and only 51% had their blood pressure under control.56

etter classification of all causes of deaths is another pending challenge. Similarly, estimates suggest that more than 47% of adults with high blood pressure are unaware of their condition, and among adults aware of being hypertensive, only 74% were under treatment and only 51% had their blood pressure under control.56 A third set of priorities around NCDs that can deliver substantial short-term improvements in mortality are related to improved access to quality care for hospital-based treatment for patients, and timely emergency care. Even with great improvements in primary care, patients with vascular disorders or diabetes will continue to need access to hospital treatment and emergency care. There is insufficient access to hospital care, aggravated by the inefficient allocation of scarce hospital beds across patient needs.16, 17 Hospital admissions for congestive heart failure in adults are the lowest in OECD countries.16 Furthermore, uptake of effective procedures that are standard practice in other countries is comparatively low, such as coronary artery bypass grafting.24 Increased use of appropriate hospital-based interventions is therefore strongly recommended.

r congestive heart failure in adults are the lowest in OECD countries.16 Furthermore, uptake of effective procedures that are standard practice in other countries is comparatively low, such as coronary artery bypass grafting.24 Increased use of appropriate hospital-based interventions is therefore strongly recommended. Chronic kidney disease prevention, including the control of associated risk factors, is feasible. In the short term, priority should be given to slow the progression towards end-stage renal failure in patients with chronic kidney disease. For more advanced disease stages, increased access to dialysis and kidney transplants is the only option. Both interventions are quite costly although transplants are more cost-effective, and evidence suggests that promoting transplants is desirable.24 Treatment for end-stage renal disease in Mexico is still restricted and insurance coverage for treatment varies. Increased access needs to be paired with standardised practice, especially across the different alternatives for dialysis treatment. Donor programmes also need to be scaled up to meet the demand for kidneys.

Treatment for end-stage renal disease in Mexico is still restricted and insurance coverage for treatment varies. Increased access needs to be paired with standardised practice, especially across the different alternatives for dialysis treatment. Donor programmes also need to be scaled up to meet the demand for kidneys. Finally, although projected mortality from cancer is in line with the SDG3 target, further reductions in premature mortality are both desirable and feasible. In particular, there are several cancers for which cost-effective interventions are available and that are highly preventable or curable when detected early. Access to treatment needs to be aligned with strengthened early detection and prevention at the primary care level to avoid heavily spending on very costly interventions that are likely to be ineffective due to the advanced stage of detection. In addition to interventions to reduce cancer risk factors, particularly smoking, priority interventions should include expanded coverage of human papilloma virus vaccination and the scale-up of screening, early detection, and treatment in early stages of cervical, breast, and colorectal cancer.57 Reduction of injury-related mortality will result in the greatest health gains in older children (aged 5–19 years) and young adults (aged 20–49 years). High mortality rates in children and young adults help to explain the present life expectancy difference between Mexico and other OECD countries, and are a contributing factor to the decreases in life expectancy gains observed between 2000 and 2010.

older children (aged 5–19 years) and young adults (aged 20–49 years). High mortality rates in children and young adults help to explain the present life expectancy difference between Mexico and other OECD countries, and are a contributing factor to the decreases in life expectancy gains observed between 2000 and 2010. With increased motorisation associated with economic development, large efforts will be needed to meet the SDG3 target for road safety (a 50% reduction in road traffic deaths and injuries by 2020). Several effective policy options exist that can substantially reduce mortality, including setting and enforcement of speed limits, seatbelt requirements, drunk-driving laws, motorcyclists' helmet use and child safety restraint use, and improved vehicle safety standards and safer roads that account for the needs of all users (including pedestrians).58, 59 However, Mexico falls behind other countries in most of these measures.35, 59 Road safety legislation and enforcement is largely the responsibility of state and municipal authorities, and only a small number of states have legislation supporting these core preventive measures. Jalisco and Mexico City have much stronger legislation and enforcement and offer a positive example that could be used to design a national regulatory framework.59

largely the responsibility of state and municipal authorities, and only a small number of states have legislation supporting these core preventive measures. Jalisco and Mexico City have much stronger legislation and enforcement and offer a positive example that could be used to design a national regulatory framework.59 Mexico experienced a large increase in homicide-related mortality rates during 2005–11 as a consequence of the national policy to fight drug traffic. The increase in fatalities explains the large reduction in life expectancy for men, especially in northern states such as Chihuahua, Sinaloa and Durango, and Guerrero and Nayarit, and a slowdown in the increase in life expectancy for women.34 Because many factors contribute to violence, tackling mortality from homicides requires an intersectoral approach. In terms of actions more directly amenable to health-sector leadership, and its coordination with other sectors, we propose that measures to reduce alcohol and drug consumption, as well as to promote healthy environments at the community level, are included in a broad strategy to reduce mortality from this cause.

oach. In terms of actions more directly amenable to health-sector leadership, and its coordination with other sectors, we propose that measures to reduce alcohol and drug consumption, as well as to promote healthy environments at the community level, are included in a broad strategy to reduce mortality from this cause. To achieve convergence and meet the SDG3 targets, several systemic challenges will need to be addressed. The main challenge for Mexico's health system is how to organise the finance and delivery systems around these interventions in a way that ensures effective access and high quality, and standardised performance across the country (panel 2). In addition, addressing mortality due to the leading risk factors (high body-mass index, high fasting plasma glucose, alcohol and tobacco use, and high blood pressure) and injuries will require strong collaboration between the health sector and other sectors, such as education, labour, fiscal, and more.68

ry (panel 2). In addition, addressing mortality due to the leading risk factors (high body-mass index, high fasting plasma glucose, alcohol and tobacco use, and high blood pressure) and injuries will require strong collaboration between the health sector and other sectors, such as education, labour, fiscal, and more.68 Although the aim of this study was to address shortfalls in premature mortality and thus stress timely policy actions needed to meet targets, important limitations should be noted. First and foremost, national averages will always hide important heterogeneity in population health needs. Our study disaggregated results by age group but not by sex, geography, ethnic group, or other socioeconomic variables known to be highly relevant to fully understand and assess policy options. Second, most policies will invariably be met with some degree of system-wide constraints, most notably financing and staffing restrictions. These might take years, if not decades, to overcome, and thus policy formulation might not easily move to the implementation phase in time to deliver substantial results by 2030. Finally, achieving gains in life expectancy will not guarantee healthier lifetimes free of disability and a more equitable distribution of health gains. Mental health problems as well as muscular-skeletal disabilities, many of which are highly prevalent in low-income populations, are not traditionally reflected in policies seeking to avert premature mortality. Finally, it is worth noting that the additional effort needed to avert deaths across age groups and diseases is different and changing over time as new cost-effective technologies are added to the health policy toolbox. Thus, the underlying assumption of reaching a 40% reduction across causes provides only a reference rate of decrease and not necessarily the most desirable or efficient pathways towards SDG3.

oups and diseases is different and changing over time as new cost-effective technologies are added to the health policy toolbox. Thus, the underlying assumption of reaching a 40% reduction across causes provides only a reference rate of decrease and not necessarily the most desirable or efficient pathways towards SDG3. Mexico has achieved sustained progress in child, maternal, and infectious diseases-related mortality in the last three decades. However, it needs to adapt to the new health challenges in order to accelerate mortality reductions to attain SDG3. Urgent action is now needed to control NCDs and reduce fatal injuries, making a 40% reduction in premature mortality by 2030 feasible and putting Mexico back on a track of substantial life expectancy convergence with better performing countries. Supplementary Material Supplementary appendix Acknowledgments We gratefully acknowledge the analytical support of Pierre Antoine Delice and Arturo Barranco Flores. We also thank Ole F Norheim and Stéphane Verguet for their thorough comments on a preliminary version of this manuscript. Contributors EG-P coordinated the writing group, participated in developing the conceptual framework, developed the outline, wrote text, undertook analysis, interpreted results, and prepared tables and figures for the manuscript. MB-L participated in developing the outline, wrote text, undertook literature review and analysis, collated contributions, interpreted results, and prepared tables for the manuscript.

amework, developed the outline, wrote text, undertook analysis, interpreted results, and prepared tables and figures for the manuscript. MB-L participated in developing the outline, wrote text, undertook literature review and analysis, collated contributions, interpreted results, and prepared tables for the manuscript. NB participated in developing the outline, wrote text, and reviewed and revised the text. DJ participated in developing the conceptual framework and the outline, and reviewed and revised the text for the manuscript. FK reviewed and revised the text for the manuscript. RL participated in developing the outline, reviewed and provided comments on the text. GY participated in developing the outline and the conceptual framework, wrote text, and reviewed and revised the text. JS participated in developing the outline and the conceptual framework, designed tables and figures, wrote text, and reviewed and revised text for the manuscript. Declaration of interests GY reports grants from Bill & Melinda Gates Foundation, outside the submitted work. EG-P, MB-L, DJ, FK, RL, and JS declare no competing interests. Figure 1 Mortality trends by major disease group, Mexico 1990–2014

NB participated in developing the outline, wrote text, and reviewed and revised the text. DJ participated in developing the conceptual framework and the outline, and reviewed and revised the text for the manuscript. FK reviewed and revised the text for the manuscript. RL participated in developing the outline, reviewed and provided comments on the text. GY participated in developing the outline and the conceptual framework, wrote text, and reviewed and revised the text. JS participated in developing the outline and the conceptual framework, designed tables and figures, wrote text, and reviewed and revised text for the manuscript. Declaration of interests GY reports grants from Bill & Melinda Gates Foundation, outside the submitted work. EG-P, MB-L, DJ, FK, RL, and JS declare no competing interests. Figure 1 Mortality trends by major disease group, Mexico 1990–2014 Own estimates using combined data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision, and the National Institute of Statistics and Geography, Statistics of mortality (INEGI; 1990–2014). Lines correspond to the 2000–2014 trends based on a simple linear regression. The shaded area around each fitted regression corresponds to the 95% CI. The list of ICD-9 and ICD-10 codes used to group deaths by major cause is included in the appendix. Figure 2 Mexico's potential to converge in life expectancy by 2030

Own estimates using combined data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision, and the National Institute of Statistics and Geography, Statistics of mortality (INEGI; 1990–2014). Lines correspond to the 2000–2014 trends based on a simple linear regression. The shaded area around each fitted regression corresponds to the 95% CI. The list of ICD-9 and ICD-10 codes used to group deaths by major cause is included in the appendix. Figure 2 Mexico's potential to converge in life expectancy by 2030 Japan and USA estimates are based on UNPD data. Mexico's inertial forecast takes into account linear projections of trends in cause-specific mortality rates and United Nations population projections. Mexico's SDG 40 × 30 scenario considers an overall reduction of 40% in premature mortality (for individuals aged 0–69 years) by 2030. Own estimates using combined data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision, and the National Institute of Statistics and Geography (INEGI), Statistics of mortality (1990–2014). Figure 3 Mexico's potential to converge in mortality by age group, 1990, 2010, and 2030 Own estimates using data from: United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision. Table 1 Deaths and death rates by age group in 1990, 2000, 2010, and three scenarios for 2030

Figure 3 Mexico's potential to converge in mortality by age group, 1990, 2010, and 2030 Own estimates using data from: United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision. Table 1 Deaths and death rates by age group in 1990, 2000, 2010, and three scenarios for 2030 1990 2000 2010 2030 Deaths Rate* Deaths Rate* Deaths Rate* Baseline deaths Inertial scenario SDG 40×30 target Deaths† Rate* Change (%) vs baseline‡ Deaths§ Rate* 0–69 years 314 388 377·2 279 454 280·9 303 306 266·0 452 635 335 467 243·2 −25·9 271 581 196·9 0–4 years 108 934 941·7 72 500 593·3 57 603 494·3 51 939 30 124 286·7 −42·0 31 163 296·6 5–19 years 22 145 68·7 16 241 48·2 15 170 42·9 14 424 9642 28·6 −33·2 8654 25·7 20–49 years 82 538 258·9 82 792 190·0 94 348 180·1 116 925 87 324 134·5 −25·3 70 155 108·1 50–69 years 100 771 1315·4 107 921 1079·7 136 185 933·2 269 347 208 377 722·0 −22·6 161 608 559·9 70 years and over 149 190 6567·2 199 308 5965·6 252 149 5496·8 559 556 476 063 4676·6 −14·9 476 063 4676·6 Total 463 578 541·5 478 762 465·7 555 455 468·3 1 012 191 811 530 547·8 −19·8 747 644 504·7 Own estimates using data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision. * All death rates are expressed per 100 000 population. † United Nations Population Division (UNPD) death estimates for 2030. ‡ Percentage reduction of deaths under the estimated inertial scenario based on UNPD death estimates, relative to the baseline scenario.

1990 2000 2010 2030 Deaths Rate* Deaths Rate* Deaths Rate* Baseline deaths Inertial scenario SDG 40×30 target Deaths† Rate* Change (%) vs baseline‡ Deaths§ Rate* 0–69 years 314 388 377·2 279 454 280·9 303 306 266·0 452 635 335 467 243·2 −25·9 271 581 196·9 0–4 years 108 934 941·7 72 500 593·3 57 603 494·3 51 939 30 124 286·7 −42·0 31 163 296·6 5–19 years 22 145 68·7 16 241 48·2 15 170 42·9 14 424 9642 28·6 −33·2 8654 25·7 20–49 years 82 538 258·9 82 792 190·0 94 348 180·1 116 925 87 324 134·5 −25·3 70 155 108·1 50–69 years 100 771 1315·4 107 921 1079·7 136 185 933·2 269 347 208 377 722·0 −22·6 161 608 559·9 70 years and over 149 190 6567·2 199 308 5965·6 252 149 5496·8 559 556 476 063 4676·6 −14·9 476 063 4676·6 Total 463 578 541·5 478 762 465·7 555 455 468·3 1 012 191 811 530 547·8 −19·8 747 644 504·7 Own estimates using data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision. * All death rates are expressed per 100 000 population. † United Nations Population Division (UNPD) death estimates for 2030. ‡ Percentage reduction of deaths under the estimated inertial scenario based on UNPD death estimates, relative to the baseline scenario. § Death rates for 2030 population assuming a flat 40% reduction in the baseline number of deaths for the 0–69 years age group and assuming the inertial rate for the 70 years and older age group. Table 2 Premature deaths and death rates by cause in 1990, 2000, 2010, and three scenarios for 2030

‡ Percentage reduction of deaths under the estimated inertial scenario based on UNPD death estimates, relative to the baseline scenario. § Death rates for 2030 population assuming a flat 40% reduction in the baseline number of deaths for the 0–69 years age group and assuming the inertial rate for the 70 years and older age group. Table 2 Premature deaths and death rates by cause in 1990, 2000, 2010, and three scenarios for 2030 1990 2000 2010 2030 Deaths Rate* Deaths Rate* Deaths Rate* Baseline deaths Inertial scenario SDG 40×30 target Deaths† Rate* Change vs baseline (%)‡ Deaths§ Rate* Ages 0–69 years 314 388 377·2 279 454 280·9 303 306 266·0 452 635 335 467 243·2 −25·9% 271 581 196·9 Communicable, perinatal, maternal, or nutritional causes 108 864 130·6 66 176 66·5 51 637 45·3 57 892 27 253 19·8 −52·9% 34 735 25·2 Newborn and child health (ages 0–4 years) 81 433 704·0 47 119 385·6 33 255 285·4 29 985 11 386 108·4 −62·0% 17 991 171·2 Asphyxia and birth trauma 19 572 169·2 19 028 155·7 13 574 116·5 12 239 3285 31·3 −73·2% 7343 69·9 Acute respiratory infections 16 866 145·8 6833 55·9 3949 33·9 3561 0 0·0 −100·0% 2137 20·3 Intestinal infections 18 308 158·3 4029 33·0 1450 12·4 1307 0 0·0 −100·0% 784 7·5 Perinatal and other nutritional and communicable causes 26 687 230·7 17 229 141·0 14 282 122·6 12 878 8101 77·1 −37·1% 7727 73·5 Maternal and other nutritional and communicable causes (ages 5–69 years) 27 431 38·2 19 057 21·8 18 382 18·0 27 907 15 867 12·5 −43·1% 16 744 13·1 Non-communicable diseases 147 562 177·1 163 493 164·4 191 589 168·0 316 871 244 198 177·0 −22·9% 190 123 137·8 Vascular, diabetes, and related disorders 72 714 87·3 84 816 85·3 103 574 90·8 183 561 137 831 99·9 −24·9% 110 137 79·8 Vascular 35 304 42·4 34 456 34·6 41 145 36·1 72 349 52 950 38·4 −26·8% 43 409 31·5 Diabetes 16 142 19·4 25 235 25·4 38 109 33·4 70 435 69 188 50·2 −1·8% 42 261 30·6 Cirrhosis and other chronic liver diseases 16 191 19·4 19 981 20·1 18 350 16·1 31 250 7729 5·6 −75·3% 18 750 13·6 Renal 5077 6·1 5144 5·2 5970 5·2 9527 7964 5·8 −16·4% 5716 4·1 Cancers 28 640 34·4 34 138 34·3 37 669 33·0 63 471 42 847 31·1 −32·5% 38 083 27·6 Other non-communicable diseases 46 208 55·4 44 539 44·8 50 346 44·2 69 839 63 521 46·0 −9·0% 41 903 30·4 Injuries 57 962 69·6 49 785 50·1 60 080 52·7 77 872 64 016 46·4 −17·8% 46 723 33·9 Homicides 14 972 18·0 10 709 10·8 22 618 19·8 28 946 37 828 27·4 30·7% 17 368 12·6 Non-road traffic injuries 27 138 32·6 22 114 22·2 19 265 16·9 25 208 9317 6·8 −63·0% 15 125 11·0 Road traffic injuries 13 918 16·7 13 565 13·6 13 977 12·3 18 271 10 530 7·6 −42·4% 10 963 7·9 Suicides 1934 2·3

016 46·4 −17·8% 46 723 33·9 Homicides 14 972 18·0 10 709 10·8 22 618 19·8 28 946 37 828 27·4 30·7% 17 368 12·6 Non-road traffic injuries 27 138 32·6 22 114 22·2 19 265 16·9 25 208 9317 6·8 −63·0% 15 125 11·0 Road traffic injuries 13 918 16·7 13 565 13·6 13 977 12·3 18 271 10 530 7·6 −42·4% 10 963 7·9 Suicides 1934 2·3 3397 3·4 4220 3·7 5447 6341 4·6 16·4% 3268 2·4 Ages 70 years and over 149 190 6567·2 199 308 5965·6 252 149 5496·8 559 556 476 063 4676·6 −14·9% 476 063 4676·6 Total 463 578 541·5 478 762 465·7 555 455 468·3 1 012 191 811 530 547·8 −19·8% 747 644 504·7 Own estimates using combined data from the United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, 2015 Revision, and the National Institute of Statistics and Geography (INEGI), Statistics of mortality (1990–2014). * All death rates are expressed per 100 000 population. † The estimated and adjusted number of deaths for 2030 use United Nations Population Division (UNPD) population projections and implied death rates for 2030 with linear regressions based on cause-specific and age-specific mortality rates for years 2000–14. ‡ Percentage reduction of deaths under the estimated inertial scenario based on UNPD death estimates, relative to the baseline scenario. § Death rates for 2030 population assuming a flat 40% reduction in the baseline number of deaths across age groups. The list of the International Classification of Diseases, 9th Revision and the International Classification of Diseases, 10th Revision codes used to group deaths by major cause is included in the appendix.

§ Death rates for 2030 population assuming a flat 40% reduction in the baseline number of deaths across age groups. The list of the International Classification of Diseases, 9th Revision and the International Classification of Diseases, 10th Revision codes used to group deaths by major cause is included in the appendix. Panel 1 Mexico's baseline for SDG3: progress in health status and future challenges9–38 Over the past 30 years, Mexico has achieved major improvements in maternal and child health, and gains in life expectancy. This is the result of a combination of factors, including strong health leadership, an emphasis on evidence-based policy implementation, and substantial and well targeted investment in cost-effective health interventions. These interventions range from oral rehydration therapy, universal vaccination, clean water, and vector control to HIV/AIDS prevention and conditional cash transfers to promote regular clinic visits among the poorest population. The combined effect of these measures helped to reduce maternal mortality from 88·7 deaths per 100 000 livebirths in 1990 to 38·9 per 100 000 livebirths in 2014, reaching 75% of the Millennium Development Goal (MDG) target for 2015. Under-5 child mortality fell 63·2% from 41·0 per 1000 livebirths in 1990 to 15·1 in 2014, close to the corresponding MDG target of 13·7.

educe maternal mortality from 88·7 deaths per 100 000 livebirths in 1990 to 38·9 per 100 000 livebirths in 2014, reaching 75% of the Millennium Development Goal (MDG) target for 2015. Under-5 child mortality fell 63·2% from 41·0 per 1000 livebirths in 1990 to 15·1 in 2014, close to the corresponding MDG target of 13·7. Despite these substantial gains, Mexico is behind OECD countries in several key areas. First, maternal, newborn, and child health (MNCH) performance is comparatively poor because several child health disorders and preventable diseases still account for an unacceptable share of total mortality. Neonatal mortality in Mexico is more than twice the rate of that in the USA (8·2 and 3·6 deaths per 1000 livebirths, respectively), the infant mortality rate is three times higher than the OECD average (13 and 3·8 deaths per 1000 livebirths in 2013, respectively), and the under-5 mortality rate is 16·8 compared with 6·6 in the USA. Poor MNCH performance is to a large extent a result of inequalities in income and other social determinants of health, as well as an absence of access to quality health services. Substantial inequalities in health across states persist, with poorer, southern states such as Oaxaca, Chiapas, and Guerrero falling behind in maternal and child health outcomes as well as in other indicators of health status, such as life expectancy.

health, as well as an absence of access to quality health services. Substantial inequalities in health across states persist, with poorer, southern states such as Oaxaca, Chiapas, and Guerrero falling behind in maternal and child health outcomes as well as in other indicators of health status, such as life expectancy. Second, Mexico has a high and rising incidence of many non-communicable disease (NCDs) as a result of its aging population and an increased prevalence of related risk factors. NCDs now account for more than three-quarters of total mortality. Diabetes and cardiovascular disease alone account for 33·4% of adult mortality. The diabetes prevalence rate in adults aged older than 20 years is the highest in OECD countries, and twice the OECD average. The mortality rate from ischaemic heart disease is 1·2 times the OECD average. Although the rates of the most important related risk factors, overweight and obesity, are stabilising in many high-income countries, they continue to grow in Mexico. Of all OECD countries, Mexico, along with the USA, has the highest rates of overweight and obesity; prevalence in adults stood at more than 70% in 2012. Cancers are also emerging as a more frequent cause of death. In particular, breast cancer is the second leading cause of death (after diabetes) of women aged 30–54 years. Similar to MNCH, regional inequalities in NCD rates are also substantial, although the geographical pattern of inequality is different. The northern and central regions and urban areas of Mexico have the highest rates of obesity and diabetes, and mortality from ischaemic heart disease. Death rates from NCDs tend to be higher, however, in low-income populations living in Mexico and incidence and mortality from cervical cancer is concentrated in the poorest states.

n and central regions and urban areas of Mexico have the highest rates of obesity and diabetes, and mortality from ischaemic heart disease. Death rates from NCDs tend to be higher, however, in low-income populations living in Mexico and incidence and mortality from cervical cancer is concentrated in the poorest states. Third, Mexico has experienced a large increase in deaths due to injuries, particularly those that are violence-related. Although violent deaths have begun to decrease, they remain higher than in previous years. Violence-related mortality is highly concentrated in a few regions. The high rates of homicide in states like Durango, Chihuahua, and Guerrero are a major cause of the decreasing life expectancy for young men. Mortality due to accidents, particularly road traffic accidents, is also a substantial cause of preventable mortality. As a result, the increase in life expectancy has slowed. Although Mexico gained on average 4·2 months per year in life expectancy between 1990 and 2000, from 2000 to 2015 it gained only 2 months per year, and is far from converging with OECD countries. In 2015, life expectancy in Mexico reached 77 years, whereas in Japan (the best performer among OECD countries) it was 83·7 years and in the USA (a not-so-good performer) it was 79·2 years.

ectancy between 1990 and 2000, from 2000 to 2015 it gained only 2 months per year, and is far from converging with OECD countries. In 2015, life expectancy in Mexico reached 77 years, whereas in Japan (the best performer among OECD countries) it was 83·7 years and in the USA (a not-so-good performer) it was 79·2 years. Panel 2 Systemic challenges and opportunities to meet SDG360–67 Despite great efforts to increase public health spending, Mexico still spends less on health, as a proportion of gross domestic product (GDP), than other OECD countries. In 2013, Mexico's total health expenditure was 6·2% of GDP, compared with the 8·9% OECD average. The public component of health expenditures is also substantially lower in Mexico, at 3·2% of GDP, compared with 6·5% in OECD countries, and out-of-pocket spending on health remains relatively high, making up 44·7% of total health expenditures in 2013 compared with 20% in OECD countries. Despite the expansion of health coverage through Seguro Popular (a public health insurance in Mexico) in the past decade, new efforts are neeeded not only to optimise returns on health investments but also to further improve the provision of financial risk protection.

itures in 2013 compared with 20% in OECD countries. Despite the expansion of health coverage through Seguro Popular (a public health insurance in Mexico) in the past decade, new efforts are neeeded not only to optimise returns on health investments but also to further improve the provision of financial risk protection. At the same time, the country faces challenges in cost-containment, not least because high rates of chronic illness come at substantial economic cost. Costs associated with the primary obesity-related diseases are projected to increase from US$880 million in 2013 to $1·2 billion by 2030. The costs (direct and indirect) associated with diabetes and its complications account for 2·25% of Mexico's GDP. And in 2008, the annual cost of treatment for four obesity-related diseases (colorectal malignant tumours, malignant breast tumours, diabetes, and cardiovascular disease) is estimated to amount to a third of the country's public health-care budget. Unless effective glycaemic control among patients with diabetes becomes commonplace, fatal diabetic complications will continue being not only a health burden, but also an economic one.

Introduction Inclusion of Haemophilus influenzae type b (Hib) conjugate vaccine in the routine infant immunisation programme has led to tremendous reductions in childhood H influenzae type b morbidity and mortality in both developed and developing countries.1, 2 Hib vaccine was introduced into the Kenyan childhood Expanded Program on Immunization (EPI) in November, 2001, as a three-dose series administered at 6, 10, and 14 weeks of age. Within 3 years of introduction, invasive H influenzae type b disease had decreased to 12% of its baseline level.3 A booster dose of Hib vaccine is not included in the Kenyan EPI schedule, nor in the schedules of 72 (92%) of 78 low-income and lower-middle-income countries.4 In the UK, 10 years after the introduction of the Hib primary vaccination, waning levels of antibody to polyribosylribitol phosphate (PRP)—an H influenzae type b polysaccharide capsule component—as well as persistence of H influenzae type b nasopharyngeal colonisation and rising rates of invasive disease, prompted introduction of a booster dose of Hib vaccine for children aged 12–15 months in 2006.5, 6 The Government of Mexico also introduced a booster dose of Hib vaccine 9 years after launching the primary vaccination programme, in part because of waning anti-PRP antibodies in children aged 12–59 months.7 However, persistently low incidence of H influenzae type b meningitis in the western region of The Gambia more than a decade after Hib vaccine introduction shows that the disease can be adequately controlled in the absence of a booster dose.8

in part because of waning anti-PRP antibodies in children aged 12–59 months.7 However, persistently low incidence of H influenzae type b meningitis in the western region of The Gambia more than a decade after Hib vaccine introduction shows that the disease can be adequately controlled in the absence of a booster dose.8 Research in context Evidence before this study

in part because of waning anti-PRP antibodies in children aged 12–59 months.7 However, persistently low incidence of H influenzae type b meningitis in the western region of The Gambia more than a decade after Hib vaccine introduction shows that the disease can be adequately controlled in the absence of a booster dose.8 Research in context Evidence before this study We searched PubMed with the terms “Hib”, “Haemophilus influenzae type b”, “vaccine”, “effectiveness”, “seroepidemiology”, “anti-PRP”, “booster”, “cross reactive”, “carriage”, and “colonization” for articles published in any language before May 31, 2015. To identify additional publications we searched the reference lists of retrieved articles. More than a decade after conjugate Haemophilus influenzae type b (Hib) vaccines became available, only 2% of the global H influenzae type b disease burden was being prevented by vaccination. In 2001, Gavi, the Vaccine Alliance, offered financial support for the introduction of Hib vaccine in developing countries, and Kenya became one of the first African countries to include Hib vaccine in the national immunisation schedule. Like the vast majority of low-income and lower-middle-income countries, Kenya used a three-dose primary series of Hib vaccine, without a booster dose. A three-dose schedule without a booster is highly effective in reducing the burden of H influenzae type b disease in the short term; however, whether a booster dose is required to achieve sustained disease control is unclear. Although data from some countries have prompted the addition of a booster dose, other data show good control of H influenzae type b disease in the absence of a booster. The need for a booster dose of Hib vaccine is probably affected by local epidemiology and factors such as the potential for natural boosting.

unclear. Although data from some countries have prompted the addition of a booster dose, other data show good control of H influenzae type b disease in the absence of a booster. The need for a booster dose of Hib vaccine is probably affected by local epidemiology and factors such as the potential for natural boosting. Added value of this study This study provides new data documenting the near elimination of invasive H influenzae type b disease in Kilifi, Kenya, in the 12 years after introduction of vaccine into the routine infant vaccination schedule without a booster dose. The detailed seroepidemiology work before and after vaccine introduction shows that the vaccine has led to improvements in population immunity in the youngest, highest-risk age groups without compromising immunity in older children. Implications of all the available evidence This study delivers compelling evidence of the long-term operational impact of a three-dose primary series of Hib vaccine in a low-income country and provides a clear answer to a pertinent policy question in Kenya: a booster dose of vaccine is not currently needed to control H influenzae type b disease.

Implications of all the available evidence This study delivers compelling evidence of the long-term operational impact of a three-dose primary series of Hib vaccine in a low-income country and provides a clear answer to a pertinent policy question in Kenya: a booster dose of vaccine is not currently needed to control H influenzae type b disease. The long-term effectiveness of a primary series of Hib vaccine in infancy can be inferred from incidence of invasive H influenzae type b disease, nasopharyngeal carriage prevalence, and seroepidemiological data from the general population. Hib vaccination induces serum antibody production and reduces the nasopharyngeal carriage prevalence of H influenzae type b, thereby diminishing the risk of invasive disease. Reductions in carriage also reduce transmission of Hib between individuals. This contributes to herd protection, but also limits the opportunity for intermittent natural boosting of serological immunity. The pattern of H influenzae type b serological immunity in different age groups across time and the persistence of H influenzae type b serological immunity throughout the years of highest risk for H influenzae type b disease are likely to be important determinants of vaccine effectiveness beyond the primary vaccination period.

ern of H influenzae type b serological immunity in different age groups across time and the persistence of H influenzae type b serological immunity throughout the years of highest risk for H influenzae type b disease are likely to be important determinants of vaccine effectiveness beyond the primary vaccination period. There is equipoise in the scientific community regarding the need for a booster dose of Hib vaccine to control disease in the long term.9 Herein we report vaccine effectiveness, the impact of the vaccine on nasopharyngeal carriage of H influenzae type b, and population immunity to H influenzae type b in the 13 years after introduction of conjugate Hib vaccine in infancy without a booster dose in Kenya. Methods Population This surveillance study took place in the Kilifi Health and Demographic Surveillance System (KHDSS), a rural community on the Kenyan coast covering an area of 891 km2.10 A census of the KHDSS in 2000 defined the resident population and, since 2000, fieldworkers have been monitoring migration events by visiting every participating household roughly every 4 months. The annual population was 199 732 in 2000, 239 396 in 2007, and 279 877 in 2014. The population is served by several government-funded health centres and by one government hospital, Kilifi County Hospital (KCH). Among women attending antenatal care at KCH, the prevalence of HIV infection ranged between 2·4% and 4·6% during 2005–13, with a general downwards trend. The prevalence of HIV in children in Kenya was estimated in 2012 to be 0·9% nationally.11

health centres and by one government hospital, Kilifi County Hospital (KCH). Among women attending antenatal care at KCH, the prevalence of HIV infection ranged between 2·4% and 4·6% during 2005–13, with a general downwards trend. The prevalence of HIV in children in Kenya was estimated in 2012 to be 0·9% nationally.11 On Nov 1, 2001, the Government of Kenya introduced tetanus-toxoid-conjugated Hib vaccine as part of a pentavalent formulation in which lyophilised Hib vaccine (Hiberix; GlaxoSmithKline, Rixensart, Belgium) was resuspended in the diphtheria, tetanus, whole-cell pertussis, hepatitis B vaccine (Tritanrix, GlaxoSmithKline). The first children eligible to receive a 6-week dose of this pentavalent vaccine were born on Sept 20, 2001, and would have been eligible to receive their third dose at the end of December, 2001. The protocol was approved by the Oxford Tropical Ethical Review Committee (No. 30-10) and the Kenya National Ethical Review Committee (SSC1433). Parents or guardians of all study participants provided written informed consent.

On Nov 1, 2001, the Government of Kenya introduced tetanus-toxoid-conjugated Hib vaccine as part of a pentavalent formulation in which lyophilised Hib vaccine (Hiberix; GlaxoSmithKline, Rixensart, Belgium) was resuspended in the diphtheria, tetanus, whole-cell pertussis, hepatitis B vaccine (Tritanrix, GlaxoSmithKline). The first children eligible to receive a 6-week dose of this pentavalent vaccine were born on Sept 20, 2001, and would have been eligible to receive their third dose at the end of December, 2001. The protocol was approved by the Oxford Tropical Ethical Review Committee (No. 30-10) and the Kenya National Ethical Review Committee (SSC1433). Parents or guardians of all study participants provided written informed consent. Assessment of vaccine effectiveness To assess vaccine effectiveness, we determined the prevalence of invasive H influenzae type b disease in children aged 12 years or younger admitted to KCH between Jan 1, 2000, and Dec 31, 2014. Blood samples are routinely taken for culture at the time of admission (except for trauma patients or patients admitted for elective surgery). Blood was cultured using an automated BACTEC instrument (BD Diagnostics, Franklin Lakes, NJ, USA). From 1998 to 2014, with the exception of a brief change in practice in 2004–05, the clinical indications for lumbar puncture were impaired consciousness or meningism in children younger than 5 years, prostration in children younger than 3 years, seizures (other than febrile seizures) in children younger than 2 years, and suspicion of sepsis in children younger than 60 days. Cerebrospinal fluid (CSF) was cultured on horse blood and chocolate agar. Beginning in 2003, HIV testing was done on the blood of children admitted to KCH according to the Kenya national policy for paediatric hospital admissions, using two rapid antibody tests. Treatment for all disorders was according to WHO guidelines at the time of admission.

as cultured on horse blood and chocolate agar. Beginning in 2003, HIV testing was done on the blood of children admitted to KCH according to the Kenya national policy for paediatric hospital admissions, using two rapid antibody tests. Treatment for all disorders was according to WHO guidelines at the time of admission. Isolates of H influenzae from sterile-site cultures were identified by colony morphology, Gram stain, and X and V factor dependence at the KEMRI-Wellcome Trust laboratory, located adjacent to KCH. Capsular type was identified by PCR using either the cap locus (done by the Haemophilus Reference Unit/WHO Collaborating Centre for Haemophilus influenzae, Respiratory and Systemic Infection Laboratory, Health Protection Agency Centre for Infections, London, UK, for isolates collected in 2000–04) or the bexA locus (done by the KEMRI-Wellcome Trust laboratory in Kilifi, Kenya, for isolates collected in 2005–13).12 We defined a case of invasive H influenzae type b disease as isolation of type b H influenzae from a sterile-site culture in a child aged 12 years or younger admitted to KCH. Assessment of nasopharyngeal carriage We investigated nasopharyngeal carriage of H influenzae type b by undertaking annual cross-sectional surveys of a sample of KHDSS residents of all ages, selected at random from the KHDSS population register once every year from 2009 to 2012, as described elsewhere.13 Isolates of H influenzae type b from nasopharyngeal swabs were identified in the same way as for sterile-site samples.

undertaking annual cross-sectional surveys of a sample of KHDSS residents of all ages, selected at random from the KHDSS population register once every year from 2009 to 2012, as described elsewhere.13 Isolates of H influenzae type b from nasopharyngeal swabs were identified in the same way as for sterile-site samples. Assessment of serological immunity We assessed serological immunity to H influenzae type b in five cross-sectional samples of children aged 12 years or younger within the study area, consisting of four convenience samples from the Junju, Ngerenya, and Chonyi locations in Kilifi County during 1998, 2000, 2004–05, and 2007,14, 15 and an age-stratified sample (50 children in each of ten age strata: 0 years, 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 7 years, 8–9 years, and 10–14 years) selected at random using Stata (version 10.1) from each age strata from the population register of the KHDSS in 2009.

98, 2000, 2004–05, and 2007,14, 15 and an age-stratified sample (50 children in each of ten age strata: 0 years, 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 7 years, 8–9 years, and 10–14 years) selected at random using Stata (version 10.1) from each age strata from the population register of the KHDSS in 2009. Serum samples were stored at −70°C until they were tested using an ELISA for antibodies to PRP. ELISA was done at the WHO Pneumococcal Reference Laboratory, Institute of Child Health, London, UK. Methods were as documented elsewhere,16 but with the following alteration: HbOHA antigen (National Institute for Biological Standards and Controls, Hertfordshire, UK) was used at 3 mg/mL. Test, control, and reference (lot 1983; US Food and Drug Administration) serum samples were incubated at 37°C for 1 h. The antibody-binding reaction was monitored by absorbance readings at 410 nm and 630 nm. We determined anti-PRP concentrations by referring to a standard curve generated from the reference wells using four-parameter sigmoid curve fitting. Median values were reported for test serum samples displaying non-parallelism to this curve. Values below the lower limit of quantitation (0·09 mg/mL) were reported as 0·05 mg/mL.

ined anti-PRP concentrations by referring to a standard curve generated from the reference wells using four-parameter sigmoid curve fitting. Median values were reported for test serum samples displaying non-parallelism to this curve. Values below the lower limit of quantitation (0·09 mg/mL) were reported as 0·05 mg/mL. Statistical analysis For population-based analyses, we designated Jan 1, 2000, through to Dec 31, 2001, as the prevaccine era and Jan 1, 2004, through to Dec 31, 2014, as the routine-use era, to allow time for sufficient vaccine uptake, given that the Hib vaccine was introduced without a catch-up campaign. We calculated the incidence of invasive H influenzae type b disease as the number of KHDSS residents admitted to KCH and confirmed by sterile-site culture to have H influenzae type b infection, divided by the resident population at the midpoint of each observation period. We calculated the incidence of H influenzae type b meningitis as the number of KHDSS residents admitted to KCH with culture-confirmed H influenzae type b (from any sterile site) who met a definition of probable meningitis (CSF white cell count ≥50 × 106 cells/L or a ratio of CSF glucose to plasma glucose of <0·1), divided by the resident population at the midpoint of each observation period. We calculated the incidence rate ratio (IRR) by using Poisson regression for specific age groups and observation periods. We calculated vaccine effectiveness as 1 minus the IRR, expressed as a percentage.

glucose to plasma glucose of <0·1), divided by the resident population at the midpoint of each observation period. We calculated the incidence rate ratio (IRR) by using Poisson regression for specific age groups and observation periods. We calculated vaccine effectiveness as 1 minus the IRR, expressed as a percentage. We excluded data from 2004–05 from calculations related to meningitis because a transient change in lumbar puncture clinical practice occurred during this period, and we did not analyse meningitis data after 2010 because pneumococcal conjugate vaccine was introduced in 2011, which was expected to reduce the incidence of probable meningitis. We categorised PRP antibody concentrations according to putative threshold protective concentrations, and calculated geometric mean concentrations by year and age category. Pre-2009 serosurveys did not include children aged 13 years or older, so we excluded data from children who were 13 years or older in 2009 from the serological immunity analyses. We present the decline in PRP antibody concentration and reverse cumulative distribution curves according to age category and year. We did all statistical analyses using Stata, versions 11.2 and 12.0. Role of the funding source The funders of the study had no role in study design; in the collection, analysis, and interpretation of data; or the writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

We did all statistical analyses using Stata, versions 11.2 and 12.0. Role of the funding source The funders of the study had no role in study design; in the collection, analysis, and interpretation of data; or the writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results 40 482 children younger than 13 years resident in KHDSS were admitted to KCH between 2000 and 2014, 38 206 (94%) of whom had their blood cultured. Although the number of cases of invasive H influenzae type b disease declined after vaccine introduction, the site of culture, sex of patients, and syndrome-specific mortality did not change substantially between the pre-vaccine and routine-use eras (table 1). In children younger than 5 years, the median age of infection with H influenzae type b was 10 months (IQR 5–24) in the prevaccine era and 10·5 months (4–23) in the routine-use era. HIV status was determined for 25 (68%) of the 37 KHDSS children with invasive H influenzae type b disease from 2003 to 2014. Four (16%) of the 25 children with H influenzae type b disease had HIV infection (three in 2003 and one in 2008). For comparison, HIV prevalence in all paediatric admissions to KCH in 2005–14 was 4·3% (807 of 18 767 admissions in whom HIV results were available).

invasive H influenzae type b disease from 2003 to 2014. Four (16%) of the 25 children with H influenzae type b disease had HIV infection (three in 2003 and one in 2008). For comparison, HIV prevalence in all paediatric admissions to KCH in 2005–14 was 4·3% (807 of 18 767 admissions in whom HIV results were available). The mean annual incidence of invasive H influenzae type b disease in children younger than 5 years in Kilifi was 62·6 per 100 000 (95% CI 46·0–83·3) in the pre-vaccine era (2000–01) and 4·5 per 100 000 (2·5–7·5) in the routine-use era (2004–14; figure 1), which translates to a vaccine effectiveness of 93% (95% CI 87–96; table 2). Incidences of invasive H influenzae type b disease and meningitis in the early introduction era (2002–03) were significantly higher than in later periods in children younger than 5 years, but rates of invasive H influenzae type b disease were similar in different periods of routine use (table 2). Incidence of invasive H influenzae type b disease in children aged 5–12 years was low in the prevaccine era and remained low in the routine-use era (table 2). In the routine-use era, 13 of the 17 cases of invasive H influenzae type b disease occurred in children younger than 36 months (the age before which serological immunity starts to decline and a comparable metric to that presented in the first Kilifi analysis3). The incidence of non-type-b invasive H influenzae disease (ie, serotype replacement disease) did not increase after introduction of the Hib vaccine (IRR in children aged <5 years, 2000–01 vs 2004–14, was 0·71 (95% CI 0·19–2·61; table 1).

decline and a comparable metric to that presented in the first Kilifi analysis3). The incidence of non-type-b invasive H influenzae disease (ie, serotype replacement disease) did not increase after introduction of the Hib vaccine (IRR in children aged <5 years, 2000–01 vs 2004–14, was 0·71 (95% CI 0·19–2·61; table 1). A previous study17 showed that nasopharyngeal carriage of H influenzae type b in children younger than 5 years was 1·7% in 2004, 3 years after Hib vaccine introduction. We obtained nasopharyngeal swabs from 2031 KHDSS residents in the four annual cross-sectional nasopharyngeal swab surveys we did in 2009–12. One (0·2%; identified in the 2012 survey) of 623 children younger than 5 years and two (0·1%; both identified in the 2009 survey) of 1408 individuals aged 5 years or older (mean age 34 years [SD 22·5; range 5–92]), carried H influenzae type b in the nasopharynx.

-sectional nasopharyngeal swab surveys we did in 2009–12. One (0·2%; identified in the 2012 survey) of 623 children younger than 5 years and two (0·1%; both identified in the 2009 survey) of 1408 individuals aged 5 years or older (mean age 34 years [SD 22·5; range 5–92]), carried H influenzae type b in the nasopharynx. To assess serological immunity, we tested available stored serum samples from children younger than 12 years from four convenience samples (367 samples from 1998 and 2000, 253 samples from 2004–05, and 205 samples from 2007), and a further 438 samples from our age-stratified, random sample of KHDSS children in 2009. The pattern of immunity in children in the community is shown in Figure 2, Figure 3 and the appendix. The proportion of children with an anti-PRP concentration of greater than 1 mg/mL (the putative threshold for long-term protection from invasive H influenzae type b disease) increased with age in the prevaccine surveys in 1998 and 2000, from very low proportions of infants (one [4%] of 24]) aged 4–7 months and 8–11 months (those at highest risk of invasive H influenzae type b disease) up to 35 (61%; 95% CI 48–74) of 57 children aged 9–12 years (figure 2).18 After Hib vaccine introduction, a large proportion of children had protective concentrations of antibody (92 [79%; 95% CI 70–86] of 117 children aged 4–35 months had long-term protective anti-PRP concentrations in 2009) and the proportion with long-term protective anti-PRP concentrations did not start to decline until after 36 months of age (figure 2). We noted a similar pattern when assessing the geometric mean concentrations (appendix).

70–86] of 117 children aged 4–35 months had long-term protective anti-PRP concentrations in 2009) and the proportion with long-term protective anti-PRP concentrations did not start to decline until after 36 months of age (figure 2). We noted a similar pattern when assessing the geometric mean concentrations (appendix). The proportions of children of various ages exceeding anti-PRP concentrations in the years before and after vaccine introduction are shown in the reverse cumulative distribution (RCD) curves (figure 3). The proportions of children aged 4 months to 2 years exceeding anti-PRP concentrations of 0·15 μg/mL and 1 mg/mL were almost uniformly greater in the years after vaccine introduction than in the prevaccine surveys (ie, the RCD curves shift up and to the right after vaccine introduction). In children aged 3–4 years, serological protection declined in 2004–05, but by 2007, population immunity was greater than in prevaccine years. Children aged 5–12 years had a transient decline in serological protection during 2004–05 and 2007, but the RCD curves suggest the same extent of population protection in 2009 as in prevaccine years.

ears, serological protection declined in 2004–05, but by 2007, population immunity was greater than in prevaccine years. Children aged 5–12 years had a transient decline in serological protection during 2004–05 and 2007, but the RCD curves suggest the same extent of population protection in 2009 as in prevaccine years. Discussion We report sustained control of paediatric invasive H influenzae type b disease, to the point of near elimination, in Kilifi, Kenya, in the 13 years after the introduction of Hib vaccine into the routine infant vaccination schedule without a booster dose. Kenya was one of three countries in Africa that were first to include the Hib vaccine in their routine childhood immunisation programme, and this study provides evidence of a robust and durable effect of the vaccine programme. Worldwide, 46 of 54 high-income countries give a booster dose of Hib vaccine, whereas booster doses are used by only six of 78 low-income and lower-middle-income countries and 22 of 57 upper-middle-income countries.4 This disparity is a reflection of the fact that support from Gavi, the Vaccine Alliance, for introduction of the Hib vaccine in the poorest countries of the world, which began in 2000, is aligned with the WHO recommendation for routine infant vaccination with Hib as a three-dose primary series.

iddle-income countries.4 This disparity is a reflection of the fact that support from Gavi, the Vaccine Alliance, for introduction of the Hib vaccine in the poorest countries of the world, which began in 2000, is aligned with the WHO recommendation for routine infant vaccination with Hib as a three-dose primary series. After introduction of the Hib vaccine, a marked reduction in H influenzae type b disease has been documented in many developed and developing regions; however, opportunities to examine the sustainability of the vaccine impact in the absence of a booster dose have been scarce.1 Our results are consistent with findings in western regions of The Gambia of near elimination of paediatric H influenzae type b meningitis 14 years after introduction of Hib vaccine administered at 2, 3, and 4 months of age and in the absence of a booster dose.8 As in The Gambia, our study occurred in a setting with high vaccine coverage: in the KHDSS, coverage with three doses of the Hib vaccine was 91% in children aged 12 months in 2002, 88% in children aged 9–23 months in 2004, 95% in children aged 12 months in 2007, and 93% in 2013 in those aged 12–23 months resident in the KHDSS since birth who had vaccine cards available (Scott JAG, unpublished).19, 20 Additional evidence of sustained disease control in the absence of a booster dose is provided by data from South America, where the incidence of H influenzae type b meningitis 6–10 years after vaccination introduction was similar in four countries, two of which used a booster dose and two of which did not.21

ed).19, 20 Additional evidence of sustained disease control in the absence of a booster dose is provided by data from South America, where the incidence of H influenzae type b meningitis 6–10 years after vaccination introduction was similar in four countries, two of which used a booster dose and two of which did not.21 By contrast, evidence of waning immunity has prompted three countries—the UK, Mexico, and South Africa—to add a booster dose of Hib vaccine, after they initially recommended only a primary series. In response to low levels of anti-PRP concentrations, persistence of H influenzae type b nasopharyngeal carriage, and rising rates of invasive disease, the UK introduced a booster dose of Hib vaccine for children aged 12–15 months in 2006, 10 years after introducing infant Hib vaccine with a catch-up campaign but without a booster dose.5 In Mexico, in 2006, results of a cross-sectional study showed that only 40–50% of 110 children aged 12–23 months had anti-PRP concentrations greater than 1 mg/mL.7 92% of these children had received the full Hib vaccine primary course given as the combination pentavalent vaccine, the same combination vaccine as is used in Kenya. On the basis of these results, Mexico, an upper-middle-income country, introduced Hib booster vaccine in 2007. In 2010, 11 years after introducing Hib vaccine as a three-dose primary series, South Africa also introduced a booster dose of the combination vaccine containing Hib. Although the booster dose was primarily for polio prevention, it was hoped that the booster would reduce the number of Hib vaccine failures in South African children (135 [51%] of 263 cases of invasive H influenzae type b disease in 2003–09 were classified as being the result of vaccine failures, of which 55% occurred in children aged 18 months or older).22 In 2015, a resurgence of invasive H influenzae type b disease was reported in eastern regions of The Gambia where coverage with three doses of Hib vaccine was 91% in children aged 12 months, suggesting that a three-dose primary series in the absence of a booster dose might not be providing sustained disease control in this setting.23 The reason for this resurgence was unclear, but the authors of the report speculated that it could have been related to waning immunity, continued transmission, or a change in malaria prevalence.

ree-dose primary series in the absence of a booster dose might not be providing sustained disease control in this setting.23 The reason for this resurgence was unclear, but the authors of the report speculated that it could have been related to waning immunity, continued transmission, or a change in malaria prevalence. The reason a booster dose is needed to achieve sustained control of disease in some settings but not others remains unclear. Tetanus-toxoid-conjugated Hib vaccine is used widely in developing countries—both in settings with sustained control of H influenzae type b disease and in those with evidence of waning immunity—so the vaccine formulation is unlikely to explain the different patterns of disease. Vaccinated individuals are less likely to be carriers of H influenzae type b and are therefore less likely to transmit the infection. However, reductions in carriage and transmission also result in fewer opportunities for natural acquisition of anti-H influenzae type b antibodies or for boosting of such antibodies. This might have been the reason anti-H influenzae type b antibodies declined in adults after routine use of Hib vaccine in children in the UK.24 In The Gambia from 1997 to 2002, introduction and widespread use of Hib vaccine was associated with a decline in nasopharyngeal carriage of H influenzae type b from 12% to 0·25% in children younger than 5 years.25 In 2010, oropharyngeal H influenzae type b carriage, as detected by culture, was estimated to be 0·9% in children aged 12–23 months in eastern regions of The Gambia.8 This is slightly higher than the carriage prevalence detected by culture of nasopharyngeal swabs reported herein. Culture of oropharyngeal swabs and PCR-based methods might be more sensitive for the detection of H influenzae type b than are nasopharyngeal swabs;26, 27 however, carriage in KHDSS residents was also low when PCR for H influenzae type b was done on both oropharyngeal and nasopharyngeal swabs collected from children aged 2–59 months enrolled as controls in a multisite study of pneumonia aetiology in 2011–13 (three [<1%] of 856 children; Hammitt LL and Scott JAG, unpublished). Although older children and adults can serve as a reservoir for transmission, the prevalence of H influenzae type b carriage was very low in these age groups in Kilifi. On the basis of these data, the opportunities for natural boosting in Kilifi are rare, as is the risk of exposure.

mitt LL and Scott JAG, unpublished). Although older children and adults can serve as a reservoir for transmission, the prevalence of H influenzae type b carriage was very low in these age groups in Kilifi. On the basis of these data, the opportunities for natural boosting in Kilifi are rare, as is the risk of exposure. In Kilifi, in the years after vaccine introduction, naturally acquired antibody has been replaced by vaccine-induced antibody. For infants and young children, this has meant greater serological protection, with 79% (95% CI 59–92) of children aged 4–11 months (historically at greatest risk of invasive disease) now having concentrations greater than 1 mg/mL, the threshold associated with long-term H influenzae type b-specific protection.18 Older children, who in the prevaccine era had naturally acquired immunity, went through a transitional period (2004–05 and 2007) during which both geometric mean concentrations and proportions exceeding protective thresholds were lower than in the prevaccine era. As surveys of older children started to include those who had been vaccinated, and vaccine-induced antibody was persisting into late childhood, measures for children aged 5–12 years resumed their prevaccine levels by 2009. In essence, the vaccine has led to improvements in population immunity in the youngest, highest-risk age groups without compromising immunity in older children. The low number of serosurvey participants in the youngest age groups is a limitation of these data. Although the 2009 survey participants were selected at random from KHDSS records, earlier surveys were convenience samples, which might have resulted in imbalances in representation of children in KHDSS as a whole. Additionally, the longer duration of freezing could have degraded antibody in older samples. These limitations notwithstanding, these results are promising for the prospect of continued effectiveness of Hib vaccine against invasive disease in older children and for maintenance of herd immunity in this setting. However, continued observation is needed, because the proportion of older children (aged 9–12 years) with anti-PRP concentrations greater than 1 mg/mL in Kilifi in 2009 (28 [45%] of 62 children) was similar to eastern regions of The Gambia (55% of 9–14 year-olds), where a resurgence in disease has been noted.

setting. However, continued observation is needed, because the proportion of older children (aged 9–12 years) with anti-PRP concentrations greater than 1 mg/mL in Kilifi in 2009 (28 [45%] of 62 children) was similar to eastern regions of The Gambia (55% of 9–14 year-olds), where a resurgence in disease has been noted. Our immunogenicity data are similar to other studies' data from low-income or lower-middle-income countries. In vaccine trials in Niger and Nepal, 83–88% and 100% of infants, respectively, had post-primary vaccination concentrations of anti-PRP above 1 mg/mL, declining to 67–75% and 64%, respectively, by late infancy.28, 29 In Mali, 2 years after vaccine introduction and with coverage at 81%, 82% of infants aged 6–7 months had anti-PRP concentrations greater than 1 mg/mL.30 In the same setting the following year, antibody decline did not begin until after 2 years of age.31 Our results also lend support to the previously observed findings that children in developing countries generate higher anti-PRP concentrations in response to vaccination than those in developed countries such as the UK.32 Proposed reasons for this include higher background environmental H influenzae type b exposure in developing countries and exposure to bacterial polysaccharides that cross-react with the PRP capsular polysaccharide of H influenzae type b.33, 34, 35, 36 Exposure to potentially cross-reactive organisms such as Escherichia coli or serogroup 6 pneumococci is likely to be higher in developing countries without access to improved water and sanitation or with higher overall pneumococcal carriage prevalence and density than in high-income or middle-income settings.13, 37, 38, 39 The level and effects of this exposure could change with improvements in water and sanitation and expanding use of pneumococcal conjugate vaccines. Continued surveillance will monitor whether effective control of disease persists or whether shifts in epidemiology (eg, disease occurring in older children, fewer opportunities for natural boosting) will necessitate a booster dose.

rovements in water and sanitation and expanding use of pneumococcal conjugate vaccines. Continued surveillance will monitor whether effective control of disease persists or whether shifts in epidemiology (eg, disease occurring in older children, fewer opportunities for natural boosting) will necessitate a booster dose. The findings reported herein do not address the possible benefit of a booster dose of Hib vaccine in settings with different epidemiological characteristics from Kilifi (eg, higher HIV prevalence, lower vaccine coverage, exposure to highly unvaccinated populations that might have high rates of carriage). Because the reasons why some settings require a booster dose of Hib vaccine to maintain control of disease and others do not are not well understood, local epidemiological data is vital to guide vaccine policy. In the absence of long-standing surveillance for invasive H influenzae type b disease, low H influenzae type b carriage prevalence in large studies that include both children and adults and use sensitive methods of detection, or high prevalence of protective antibody concentrations throughout the ages of highest risk for H influenzae type b disease, would provide evidence of ongoing protection.

th the lowest recurrence rate has been a research priority for many years.18 In this trial, we compared the relative effectiveness of the two most commonly used operations and found that PLTR surgery has a significantly lower trichiasis recurrence rate at 12 months than BLTR surgery, particularly for more severe cases. Considerable care was taken to ensure that the surgeons did both procedures using the WHO-described method with equal precision.4 We trained surgeons who had been previously taught PLTR to do the BLTR procedure. This approach was chosen, rather than training novice surgeons simultaneously in both procedures, to reduce the learning curve to achieve proficiency in the new procedure.26 During training and standardisation, before the commencement of the trial, each surgeon did about 100 BLTR operations and was confirmed by two assessors to be performing the procedure per protocol, using the WHO Certification process.4

disease, low H influenzae type b carriage prevalence in large studies that include both children and adults and use sensitive methods of detection, or high prevalence of protective antibody concentrations throughout the ages of highest risk for H influenzae type b disease, would provide evidence of ongoing protection. Over the past 25 years, 189 countries, including 73 countries eligible for support from Gavi, the Vaccine Alliance, have introduced a Hib-containing vaccine into their national immunisation programme for children. A booster dose of Hib vaccine is recommended in most high-income countries, whereas most low-income countries have followed the WHO EPI schedule, which does not include a booster dose. In this study, we found that use of Hib vaccine according to the EPI schedule led to near elimination of paediatric invasive H influenzae type b disease, with no evidence of resurgent disease in older children in whom immunity might be expected to wane without a booster dose. Indeed, immunogenicity data show that immunity persists through the age of greatest risk of disease for most children and that antibody concentrations in older children, although lower than in young children, are similar to concentrations reported in the prevaccine era. In sum, we found no evidence to support introduction of a booster dose of Hib vaccine into the Kenyan EPI at this time. Supplementary Material Supplementary appendix

Over the past 25 years, 189 countries, including 73 countries eligible for support from Gavi, the Vaccine Alliance, have introduced a Hib-containing vaccine into their national immunisation programme for children. A booster dose of Hib vaccine is recommended in most high-income countries, whereas most low-income countries have followed the WHO EPI schedule, which does not include a booster dose. In this study, we found that use of Hib vaccine according to the EPI schedule led to near elimination of paediatric invasive H influenzae type b disease, with no evidence of resurgent disease in older children in whom immunity might be expected to wane without a booster dose. Indeed, immunogenicity data show that immunity persists through the age of greatest risk of disease for most children and that antibody concentrations in older children, although lower than in young children, are similar to concentrations reported in the prevaccine era. In sum, we found no evidence to support introduction of a booster dose of Hib vaccine into the Kenyan EPI at this time. Supplementary Material Supplementary appendix Acknowledgments We thank the residents of the Kilifi Health and Demographic Surveillance System and the dedicated team of fieldworkers, administrative staff, clinicians, and laboratory staff who worked on this study. This report is published with the permission of the Director of the Kenya Medical Research Institute. LLH, JAGS, and SCM have received grants from the Gavi, the Vaccine Alliance. JAGS is funded by the Wellcome Trust (fellowship number 98504). RJC has received a European Society for Paediatric Infectious Diseases award, and the National Institute for Health Research funded her Academic Clinical Fellowship.

ch Institute. LLH, JAGS, and SCM have received grants from the Gavi, the Vaccine Alliance. JAGS is funded by the Wellcome Trust (fellowship number 98504). RJC has received a European Society for Paediatric Infectious Diseases award, and the National Institute for Health Research funded her Academic Clinical Fellowship. Contributors LLH, TK, SS, NM, and JAGS were responsible for the design and conduct of the study. RJC, AK, SCM, PB, and DG did the laboratory analyses. LLH, RJC, AK, AM, and JAGS analysed the data, and LLH, RJC, AM, SCM, PB, DG, and JAGS interpreted it. LLH, RJC, SCM, DG, and JAGS were responsible for the writing of the report. Declaration of interests LLH has received research funding from GlaxoSmithKline Biologicals and Pfizer, and has participated in a Scientific Input Engagement for Merck. DG has received research funding from GlaxoSmithKline Biologicals, Sanofi, and Merck. All other authors declare no competing interests. Figure 1 Incidence of invasive Haemophilus influenzae type b disease in children younger than 5 years in the Kilifi Health and Demographic Surveillance System, 2000–14 Hib vaccine was introduced in November, 2001. Error bars show 95% CI. Figure 2 Children with anti-PRP concentrations of >1 μg/mL, by age group and survey year Data for 1998 and 2000 combined (A), 2004–05 (B), 2007 (C), and 2009 (D). The proportion of children aged 4–7 months in (A) is 0%. Error bars show 95% CI. PRP=polyribosylribitol phosphate. Figure 3 Reverse cumulative proportions of children with anti-PRP concentrations that exceed thresholds, by age group and survey year

Data for 1998 and 2000 combined (A), 2004–05 (B), 2007 (C), and 2009 (D). The proportion of children aged 4–7 months in (A) is 0%. Error bars show 95% CI. PRP=polyribosylribitol phosphate. Figure 3 Reverse cumulative proportions of children with anti-PRP concentrations that exceed thresholds, by age group and survey year Data for children aged 4–11 months (A) 1–2 years (B), 3–4 years (C), and 5–12 years (D). Vertical lines indicate thresholds for short-term (0·15 μg/mL) and long-term (1 μg/mL) protection against invasive H influenzae type b disease. Table 1 Invasive Haemophilus influenzae disease in children aged 0–12 years in the Kilifi Health and Demographic Surveillance System admitted to the Kilifi County Hospital, 2000–14

refrigeration. Therefore, the translation of this result into clinical practice, given the resource limitations of the Kenyan health system and other similar settings, is realistic. Because surgical cryotherapy is not routinely available, topical fluorouracil is therefore an alternative strategy to prevent recurrence. The only other randomised study of treatment for OSSN was a placebo-controlled, crossover trial of topical mitomycin for 48 patients from Australia.10 However, that study has several distinct differences to our study, which probably limit the relevance to settings such as Kenya. First, only partial incisional biopsy samples were taken for diagnosis; the tumours were not excised. Given the advanced and aggressive disease in Africa, complete surgical removal of the lesion is the preferred approach. Second, the casemix was different: patients with squamous cell carcinoma were excluded from the Australia study, the population group was older than ours (mean age 67 years), predominantly male (75%), and probably not infected with HIV (no data were provided). Although the lesions regressed clinically on treatment with mitomycin, more than half of patients had persistent OSSN on repeat histological assessment of the lesion site 1 year after treatment.

Data for children aged 4–11 months (A) 1–2 years (B), 3–4 years (C), and 5–12 years (D). Vertical lines indicate thresholds for short-term (0·15 μg/mL) and long-term (1 μg/mL) protection against invasive H influenzae type b disease. Table 1 Invasive Haemophilus influenzae disease in children aged 0–12 years in the Kilifi Health and Demographic Surveillance System admitted to the Kilifi County Hospital, 2000–14 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Age <5 years Admissions 3660 3591 3282 2776 2716 2317 2455 2130 1986 2131 1905 1642 1494 1001 1473 With blood culture 3613 (99%) 3533 (98%) 3200 (98%) 2710 (98%) 2631 (97%) 2219 (96%) 2366 (96%) 1991 (93%) 1841 (93%) 1996 (94%) 1747 (92%) 1528 (93%) 1331 (89%) 904 (90%) 1335 (91%) With lumbar puncture 939 (26%) 866 (24%) 936 (29%) 578 (21%) 351 (13%) 482 (21%) 693 (28%) 611 (29%) 553 (28%) 548 (26%) 532 (28%) 433 (26%) 410 (27%) 260 (26%) 390 (26%) With probable bacterial meningitis 33 (1%) 40 (1%) 52 (2%) 24 (1%) 20 (1%) 24 (1%) 25 (1%) 26 (1%) 18 (1%) 16 (1%) 18 (1%) 18 (1%) 11 (1%) 11 (1%) 5 (<1%) Culture-confirmed H influenzae disease 24 (<1%) 30 (<1%) 21 (<1%) 22 (<1%) 4 (<1%) 5 (<1%) 2 (<1%) 2 (<1%) 5 (<1%) 2 (<1%) 1 (<1%) 0 3 (<1%) 2 (<1%) 3 (<1%) Types a, c, d, e, f 2 (8%) 1 (3%) 1 (5%) 0 1 (25%) 1 (20%) 0 1 (50%) 2 (40%) 0 0 0 2 (67%) 1 (50%) 1 (33%) Non-capsular 3 (13%) 1 (3%) 3 (14%) 4 (18%) 0 1 (20%) 0 0 1 (20%) 0 1 (100%) 0 1 (33%) 0 2 (67%) Type b 19 (79%) 28 (93%) 17 (81%) 18 (82%) 3 (75%) 3 (60%) 2 (100%) 1 (50%) 2 (40%) 2 (100%) 0 0 0 1 (50%) 0 H influenzae type b cultured in CSF 4 (21%) 13 (46%) 8 (47%) 8 (44%) 1 (33%) 1 (33%) 0 0 1 (50%) 1 (50%) 0 0 0 0 0 Age <24 months 16 (84%) 19 (68%) 10 (59%) 10 (56%) 2 (67%) 1 (33%) 0 1 (100%) 1 (50%) 2 (100%) 0 0 0 1 (100%) 0 Boys 7 (37%) 15 (54%) 6 (35%) 7 (39%) 2 (67%) 1 (33%) 1 (50%) 1 (100%) 0 1 (50%) 0 0 0 1 (100%) 0 Died during the episode 3 (16%) 8 (29%) 5 (29%) 6 (33%) 0 1 (33%) 0 1 (100%) 1 (50%) 0 0 0 0 1 (100%) 0 Age 5–12 years Admissions 459 398 377 416 373 425 419 369 356 395 436 413 324 283 480 With blood culture 448 (98%) 378 (95%) 361 (96%) 394 (95%) 328 (88%) 371 (87%) 376 (90%) 326 (88%) 308 (87%) 331 (84%) 380 (87%) 347 (84%) 274 (85%) 226 (80%) 413 (86%) With lumbar puncture 74 (16%) 55 (14%) 68 (18%) 54 (13%) 41 (11%) 37 (9%) 56 (13%) 44 (12%) 50 (14%) 46 (12%) 88 (20%) 65 (16%) 71 (22%) 81 (29%) 96 (20%) With probable bacterial meningitis 7 (2%) 5 (1%) 6 (2%) 10 (2%) 5 (1%) 1 (<1%) 1 (<1%) 5 (1%) 4 (1%) 3 (1%) 4 (1%) 3 (1%) 1 (<1%) 4 (1%) 1 (<1%) Culture-confirmed H influenzae disease 1 (<1%) 2 (<1%) 2 (<1%)

18%) 54 (13%) 41 (11%) 37 (9%) 56 (13%) 44 (12%) 50 (14%) 46 (12%) 88 (20%) 65 (16%) 71 (22%) 81 (29%) 96 (20%) With probable bacterial meningitis 7 (2%) 5 (1%) 6 (2%) 10 (2%) 5 (1%) 1 (<1%) 1 (<1%) 5 (1%) 4 (1%) 3 (1%) 4 (1%) 3 (1%) 1 (<1%) 4 (1%) 1 (<1%) Culture-confirmed H influenzae disease 1 (<1%) 2 (<1%) 2 (<1%) 2 (<1%) 1 (<1%) 1 (<1%) 1 (<1%) 0 0 1 (<1%) 1 (<1%) 1 (<1%) 0 0 1 (<1%) Types a, c, d, e, f 0 0 1 (50%) 0 0 0 1 (100%) 0 0 0 0 0 0 0 0 Non-capsular 1 (100%) 0 0 0 0 1 (100%) 0 0 0 0 0 1 (100%) 0 0 1 (100%) Type b 0 2 (100%) 1 (50%) 2 (100%) 1 (100%) 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 H influenzae type b cultured in CSF 0 1 (50%) 1 (100%) 1 (50%) 1 (100%) 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 Boys 0 1 (50%) 1 (100%) 1 (50%) 0 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 Died during the episode 0 1 (50%) 1 (100%) 1 (50%) 0 0 0 0 0 0 0 0 0 0 0 Data are N and n (%). CSF=cerebrospinal fluid. Table 2 Incidence (per 100 000) of invasive Haemophilus influenzae type b disease and meningitis, and probable bacterial meningitis, in children aged 0–12 years in the Kilifi Health and Demographic Surveillance System, 2000–14

2 (<1%) 1 (<1%) 1 (<1%) 1 (<1%) 0 0 1 (<1%) 1 (<1%) 1 (<1%) 0 0 1 (<1%) Types a, c, d, e, f 0 0 1 (50%) 0 0 0 1 (100%) 0 0 0 0 0 0 0 0 Non-capsular 1 (100%) 0 0 0 0 1 (100%) 0 0 0 0 0 1 (100%) 0 0 1 (100%) Type b 0 2 (100%) 1 (50%) 2 (100%) 1 (100%) 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 H influenzae type b cultured in CSF 0 1 (50%) 1 (100%) 1 (50%) 1 (100%) 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 Boys 0 1 (50%) 1 (100%) 1 (50%) 0 0 0 0 0 1 (100%) 1 (100%) 0 0 0 0 Died during the episode 0 1 (50%) 1 (100%) 1 (50%) 0 0 0 0 0 0 0 0 0 0 0 Data are N and n (%). CSF=cerebrospinal fluid. Table 2 Incidence (per 100 000) of invasive Haemophilus influenzae type b disease and meningitis, and probable bacterial meningitis, in children aged 0–12 years in the Kilifi Health and Demographic Surveillance System, 2000–14 Age <2 years Age <5 years Age 5–12 years N Incidence (95% CI) N Incidence (95% CI) N Incidence (95% CI) Invasive H influenzae type b disease 2000–01 35 117·1 (81·6–162·9) 47 62·6 (46·0–83·3) 2 4·0 (0·5–14·6) 2002–03 25 79·4 (51·4–117·2) 34 44·6 (31·0–61·8) 3 2·9 (0·6–8·6) 2004–14 11 8·6 (4·3–15·3) 14 4·5 (2·5–7·5) 3 1·7 (0·3–4·8) 2004–05 5 14·5 (4·7–33·8) 6 7·1 (2·6–15·4) 1 1·9 (0·1–10·3) 2006–09 5 6·7 (2·2–15·7) 7 3·9 (1·6–8·0) 1 1·6 (0·0–8·8) 2010–14 1 5·1 (0·1–28·3) 1 2·1 (0·1–11·5) 1 1·6 (0·0–8·6) Incidence rate ratio 2000–01 vs 2002–03 .. 0·67 (0·41–1·13) .. 0·71 (0·46–1·10) .. 0·73 (0·12–4·34) 2000–01 vs 2004–14 .. 0·07 (0·04–0·14) .. 0·07 (0·04–0·13) .. 0·41 (0·07–2·44) H influenzae type b meningitis 2000–01 14 46·8 (25·6–78·6) 17 22·7 (13·2–36·3) 1 2·0 (0·5–11·2) 2002–03 12 38·1 (19·7–66·6) 15 19·1 (10·7–31·4) 1 2·0 (0·1–11·0) 2006–10 2 5·3 (0·7–19·2) 2 2·2 (0·3–7·9) 2 1·6 (0·2–5·6) Incidence rate ratio 2000–01 vs 2002–03 .. 0·81 (0·38–1·76) .. 0·84 (0·42–1·68) .. 0·97 (0·06–15·57) 2000–01 vs 2006–10 .. 0·11 (0·03–0·50) .. 0·09 (0·02–0·42) .. 0·77 (0·07–8·54) Probable bacterial meningitis 2000–01 61 204·1 (156·1–262·2) 73 97·3 (76·3–122·3) 15 15·3 (8·6–25·3) 2002–03 67 212·9 (165·0–270·3) 76 96·5 (76·0–120·8) 17 16·6 (9·7–26·6) 2006–10 80 85·4 (67·7–106·3) 103 45·3 (37·0–55·0) 17 5·5 (3·2–8·8) Incidence rate ratio 2000–01 vs 2002–03 .. 1·04 (0·74–1·48) .. 0·99 (0·72–1·37) .. 1·08 (0·54–2·17) 2000–01 vs 2006–10 .. 0·41 (0·30–0·58) .. 0·47 (0·35–0·63) .. 0·36 (0·18–0·72) N is the number of children with the disease during the corresponding timeframe.

Introduction Trachoma, a neglected tropical disease caused by Chlamydia trachomatis, is the leading infectious cause of blindness.1 Recurrent infection drives progressive conjunctival scarring, which turns the lid and eyelashes in towards the eye (trichiasis) resulting in pain and eventually blinding corneal opacification. About 1·2 million people are irreversibly blind from this disease and about 7·2 million have trichiasis.1, 2 WHO recommends the SAFE strategy for trachoma control: Surgery for trichiasis, Antibiotics, Facial cleanliness, and Environmental improvement.3 Trichiasis surgery reduces the risk of sight loss by correcting the in-turned eyelid, thus stopping the corneal damage. Surgery involves an incision through the scarred upper eyelid, parallel to the lid margin, outward rotation, and suturing in the corrected position.4 Due to the limited number of ophthalmologists in most trachoma-endemic countries, surgery is usually done by non-physicians with limited training, equipment, and time.3 Given these constraints, the technique needs to be simple, safe, and quick to do, whereas at the same time giving consistently good results.

position.4 Due to the limited number of ophthalmologists in most trachoma-endemic countries, surgery is usually done by non-physicians with limited training, equipment, and time.3 Given these constraints, the technique needs to be simple, safe, and quick to do, whereas at the same time giving consistently good results. Unfortunately, trichiasis frequently recurs after surgery. This outcome represents a substantial limitation in preventing sight loss from trachoma. Studies have reported trichiasis recurrence rates between 10% at 3 months and up to 60% at 3 years, with an average of around 20% at 1 year.5, 6, 7, 8, 9, 10, 11, 12, 13, 14 Several factors contribute to recurrent trichiasis, including preoperative disease severity, surgeon skill, and surgical procedure.15 Among these, operation type is a major determinant of outcome and subtle variations in procedure performance probably affect results.10, 11, 16 Many different surgical procedures have been used to correct trichiasis, with some evidence that bilamellar tarsal rotation (BLTR) is better than others to which it has been formally compared.10, 11, 15, 17 However, it is important to determine which is the best of these options. Research in context Evidence before this study

Unfortunately, trichiasis frequently recurs after surgery. This outcome represents a substantial limitation in preventing sight loss from trachoma. Studies have reported trichiasis recurrence rates between 10% at 3 months and up to 60% at 3 years, with an average of around 20% at 1 year.5, 6, 7, 8, 9, 10, 11, 12, 13, 14 Several factors contribute to recurrent trichiasis, including preoperative disease severity, surgeon skill, and surgical procedure.15 Among these, operation type is a major determinant of outcome and subtle variations in procedure performance probably affect results.10, 11, 16 Many different surgical procedures have been used to correct trichiasis, with some evidence that bilamellar tarsal rotation (BLTR) is better than others to which it has been formally compared.10, 11, 15, 17 However, it is important to determine which is the best of these options. Research in context Evidence before this study Members of our study group recently published a systematic review of the management of trachomatous trichiasis (Burton and colleagues, 2015). When preparing this systematic review, we searched CENTRAL, Ovid MEDLINE, Embase, ISRCTN registry, ClinicalTrials.gov, and WHO ICTRP. We searched until May 7, 2015, using the search terms “trachoma” and “trichiasis”. See the review's appendix for full search methods for each database. We identified one previous randomised trial (Adamu and Alemayehu, 2002), which compared variants of the BLTR and PLTR procedure done by ophthalmologists in a teaching hospital environment in Ethiopia; 153 patients were randomly assigned and followed for 3 months. No evidence of a difference in outcome was found. However, this earlier study was constrained by a small sample size and short duration. The surgery was performed in a teaching hospital setting by ophthalmologists, in contrast to the health centre provision by non-physicians typical of trachoma control programmes, limiting the conclusions that can been drawn.

s found. However, this earlier study was constrained by a small sample size and short duration. The surgery was performed in a teaching hospital setting by ophthalmologists, in contrast to the health centre provision by non-physicians typical of trachoma control programmes, limiting the conclusions that can been drawn. Added value of this study Our trial was designed to compare the two most common operations used to treat trachomatous trichiasis to determine which gives the best results in terms of disease recurrence and complications in a programmatic setting. The results show that the PLTR was superior to BLTR because it had a substantially lower trichiasis recurrence rate by 1 year and fewer intraoperative and immediate postoperative complications. Implications of all the available evidence This study provides evidence of superiority of PLTR, suggesting that it could be the best procedure for the programmatic management of trachomatous trichiasis. We suggest that new surgical trainees in both established and new programmes be trained in the PLTR procedure. Another trial examining the outcomes of PLTR surgery done by surgeons previously trained in BLTR surgery should be considered.

be the best procedure for the programmatic management of trachomatous trichiasis. We suggest that new surgical trainees in both established and new programmes be trained in the PLTR procedure. Another trial examining the outcomes of PLTR surgery done by surgeons previously trained in BLTR surgery should be considered. About 20 years ago several procedures were compared with the BLTR operation in randomised controlled trials.10, 11 The findings from these trials showed that the BLTR procedure had the lowest trichiasis recurrence rate of the procedures compared (about 20% at 1 year), leading WHO to recommend it as the preferred operation.3 However, the most commonly used alternative procedure, the posterior lamellar tarsal rotation (PLTR) or Trabut operation, was not included in these earlier trials. One earlier randomised trial from Ethiopia compared variants of the BLTR and PLTR, and found no difference. However, that trial was relatively small, with only 3 months' follow-up and was done by ophthalmologists in a teaching hospital, precluding conclusions for control programmes that do the vast majority of trichiasis surgery.13

andomised trial from Ethiopia compared variants of the BLTR and PLTR, and found no difference. However, that trial was relatively small, with only 3 months' follow-up and was done by ophthalmologists in a teaching hospital, precluding conclusions for control programmes that do the vast majority of trichiasis surgery.13 There is an unprecedented effort to scale up global trichiasis surgery output and improve outcomes, to clear the huge trichiasis backlog. This effort requires training many trichiasis surgeons on the easiest, safest, and most successful operation with the least recurrence and complications. There is an urgent need to examine rigorously which of these two most frequently performed operations has the best outcomes in a programmatic setting, with an adequate sample size and follow-up period. This question was identified as a research priority several years ago by the WHO Alliance for the Global Elimination of Trachoma by 2020 (GET2020).18 The aim of our trial was to determine whether BLTR or PLTR surgery gives superior results under programmatic conditions.

ith an adequate sample size and follow-up period. This question was identified as a research priority several years ago by the WHO Alliance for the Global Elimination of Trachoma by 2020 (GET2020).18 The aim of our trial was to determine whether BLTR or PLTR surgery gives superior results under programmatic conditions. Methods Study design and participants This was a single-masked, individual-randomised, controlled trial done in Ethiopia. Adults with trachomatous trichiasis were randomly allocated to either BLTR or PLTR surgery, stratified by surgeon, and followed up for 1 year. The study was approved by the Ethiopian National Health Research Ethics Review Committee, the London School of Hygiene & Tropical Medicine Ethics Committee, Emory University Institutional Review Board, and the Ethiopia Food, Medicine and Healthcare Administration and Controls Authority. The trial was done in compliance with the Declaration of Helsinki and International Conference on Harmonisation–Good Clinical Practice. An independent data and safety monitoring committee oversaw the trial.

Institutional Review Board, and the Ethiopia Food, Medicine and Healthcare Administration and Controls Authority. The trial was done in compliance with the Declaration of Helsinki and International Conference on Harmonisation–Good Clinical Practice. An independent data and safety monitoring committee oversaw the trial. Participants had upper lid trachomatous trichiasis with one or more eyelashes touching the eye or evidence of epilation, in association with tarsal conjunctival scarring. We excluded people with trichiasis due to other causes, recurrent trichiasis after previous surgery, hypertension, pregnancy, and those under 18 years. Patients were recruited mainly through community-based screening in three districts of West Gojam Zone, Amhara Region, Ethiopia. Recruitment and surgery were performed in community level health centres. Written informed consent in Amharic was obtained before enrolment from participants. If a participant was unable to read and write, the information sheet and consent form were read to them and their consent recorded by thumbprint. Randomisation and masking Participants were randomly assigned (1:1) to either PLTR or BLTR surgery for each surgeon, with random block sizes of 4 or 6. Randomisation was stratified by surgeon because of potential intersurgeon variability. The sequences were computer-generated by an independent statistician. Separate allocation sequences for each surgeon were concealed in sequentially numbered, sealed, opaque envelopes. A person independent of all other aspects of the trial prepared these envelopes.

surgeon because of potential intersurgeon variability. The sequences were computer-generated by an independent statistician. Separate allocation sequences for each surgeon were concealed in sequentially numbered, sealed, opaque envelopes. A person independent of all other aspects of the trial prepared these envelopes. On most recruitment days, two surgeons operated simultaneously. Following baseline examination, participants were allocated to the next available surgeon. A fieldworker was responsible for implementing the intervention assignment in a dedicated area. The fieldworker and surgeon jointly confirmed the allocation and recorded this in the surgical logbook. The different surgical equipment sets for the two procedures were kept separately. The randomisation fieldworker and surgeon jointly collected the appropriate surgical set for the allocated procedure. Surgeons and patients were aware of the allocation. The two examiners (EH, SA) who were responsible for clinical observations at baseline, 6 months, and 12 months were masked to the allocation. They were not involved in the allocation process, surgery, postoperative care, or the 10 day follow-up. The intraoperative and 10 day observations were made by separate fieldworkers who could not be masked to the allocation.

e for clinical observations at baseline, 6 months, and 12 months were masked to the allocation. They were not involved in the allocation process, surgery, postoperative care, or the 10 day follow-up. The intraoperative and 10 day observations were made by separate fieldworkers who could not be masked to the allocation. Procedures At the preoperative assessment before randomisation, demographic characteristics were recorded. Presenting logMAR (logarithm of the minimum angle of resolution) visual acuity at 2 m was measured using PeekAcuity software on a smartphone in a dark room.19 For visual acuities of counting fingers or less, logMAR values were attributed as follows: counting fingers, 2·0; hand movements, 2·5; perception of light, 3·0; and no perception of light, 3·5.8 We assessed contrast sensitivity with a prototype smartphone-based test that presents calibrated grey scale spots against a white background, which are identified by touch.

, logMAR values were attributed as follows: counting fingers, 2·0; hand movements, 2·5; perception of light, 3·0; and no perception of light, 3·5.8 We assessed contrast sensitivity with a prototype smartphone-based test that presents calibrated grey scale spots against a white background, which are identified by touch. Eyes were examined by a single examiner (EH) using 2·5× binocular loupes and torch, and graded using the Detailed WHO FPC Grading System.20 Lashes touching the eye were counted and subdivided by the part of the eye contacted: cornea, lateral, or medial conjunctiva. Trichiasis subtypes were recorded: metaplastic, misdirected, and entropic.21 Clinical evidence of epilation was identified by broken or newly growing lashes, or areas of absent lashes. Upper lid entropion was graded by assessing the degree of eyelid margin inward rotation.21 Corneal scarring was graded using a previously described detailed system.20 Three standardised high-resolution digital photographs of trichiasis, cornea, and tarsal conjunctiva were taken, using a Nikon D90 digital SLR camera with 105 mm macro lens and R1C1 flash units.22

the degree of eyelid margin inward rotation.21 Corneal scarring was graded using a previously described detailed system.20 Three standardised high-resolution digital photographs of trichiasis, cornea, and tarsal conjunctiva were taken, using a Nikon D90 digital SLR camera with 105 mm macro lens and R1C1 flash units.22 Before recruitment, nine experienced trichiasis nurse-surgeons, already trained, certified, and regularly performing PLTR surgery were trained in BLTR surgery. We followed the procedures described in the WHO Trichiasis Surgery for Trachoma manual.4 After training, surgeons were carefully observed throughout five operations and certified as correctly doing the procedure following the standardisation checklist.4 Surgeons then returned to their usual workplace, and regularly performed BLTR for 6 months. They then returned for repeat standardisation, assessment, and certification on both PLTR and BLTR procedures by two assessors. Before commencing the trial, each surgeon had done about 100 BLTR procedures (median 117, range 94–137). The best six surgeons did the surgery in this trial: they were all certified as consistently performing all component steps of both operations correctly, using the WHO certification procedures.

res by two assessors. Before commencing the trial, each surgeon had done about 100 BLTR procedures (median 117, range 94–137). The best six surgeons did the surgery in this trial: they were all certified as consistently performing all component steps of both operations correctly, using the WHO certification procedures. The procedures are described in detail in the WHO manual.4 Briefly, in the PLTR the eyelid is everted, an incision is made through the tarsal conjunctiva and tarsal plate (posterior lamella), parallel to and 3 mm above the lid margin. The posterior lamella is separated from the anterior lamella (orbicularis muscle and skin). Three sutures are placed to externally rotate and fix the eyelid. In the BLTR the eyelid is fixed with a clamp (Waddell type), of an appropriate size.23 A full-thickness incision is made through the anterior and posterior lamellae, parallel to and 3 mm above the lid margin. Three sutures are placed to externally rotate and fix the eyelid. Surgery was done under subcutaneous local anaesthesia (2–3 mL of lidocaine 2%, with adrenaline). In both surgical procedures, 4/0 silk sutures with 3/8th circle, 19 mm cutting needles were used. Surgery duration was measured and complications documented. Postoperatively, operated eyes were padded for 1 day and tetracycline eye ointment 1% was self-administered twice daily for 2 weeks. Participants were not given perioperative oral azithromycin because it is not the operational practice to use it in this region.

urgery duration was measured and complications documented. Postoperatively, operated eyes were padded for 1 day and tetracycline eye ointment 1% was self-administered twice daily for 2 weeks. Participants were not given perioperative oral azithromycin because it is not the operational practice to use it in this region. Participants were examined at 10 days, 6 months, and 12 months after operation. At 10 days, data were collected on patient-reported outcomes (improvement in preoperative symptoms, postoperative pain, and functioning). Participants were examined for recurrence, degree of lid eversion, infection, granulomata, and eyelid contour abnormality (ECA) before suture removal.

and 12 months after operation. At 10 days, data were collected on patient-reported outcomes (improvement in preoperative symptoms, postoperative pain, and functioning). Participants were examined for recurrence, degree of lid eversion, infection, granulomata, and eyelid contour abnormality (ECA) before suture removal. At 6 months and 12 months participants were re-examined following the same procedures as for baseline (SA at 6 months and EH at 12 months). The examiners were standardised and had very strong agreement for the primary outcome in grading validation studies (κ=0·95). Based on severity, trichiasis cases were categorised into minor trichiasis with less than six lashes or evidence of epilation in less than one third of the lid margin, and major trichiasis with six or more lashes or evidence of epilation in one third or more of the lid margin. The degree of entropion correction was graded as follows: (grade 1) extra eversion: main lashes point superiorly, whole lid margin visible, and tarsal plate surface visible; (grade 2) lid margin eversion: lashes point superiorly, whole lid margin visible, and tarsal surface not visible; (grade 3) partial lid margin entropion: some parts of the lashes might point anteriorly and some part of the lid margin not visible; (grade 4) total lid margin entropion: lashes might point inferiorly or towards the globe and lid margin is not visible. We considered grade 1 over-correction, grade 2 normal correction, and grades 3 and 4 under-correction. Granulomata were defined as fleshy tissue growth of at least 2 mm on the tarsal conjunctiva or at the edge of the eyelid.12 Grading of ECAs was based on the PRET trial method: mild, vertical deviation from the natural contour less than 1 mm in height and affecting more than one third of horizontal eyelid length; moderate, vertical deviation from the natural contour 1–2 mm in height or affecting one third to two thirds of horizontal eyelid length; severe, vertical deviation from the natural contour more than 2 mm in height or a defect more than two thirds the horizontal eyelid length.24 These were regrouped as: clinically non-significant ECA, which included mild ECA; and clinically significant ECA, which included moderate-to-severe ECA. The clinically significant ECAs also included other ECAs such as divot, which is a scarred depression or tissue loss including lashes at the eyelid margin. Visual acuity and contrast sensitivity were measured at 12 months.

, which included mild ECA; and clinically significant ECA, which included moderate-to-severe ECA. The clinically significant ECAs also included other ECAs such as divot, which is a scarred depression or tissue loss including lashes at the eyelid margin. Visual acuity and contrast sensitivity were measured at 12 months. Data on patient-reported outcomes were collected at 12 months. Individuals with recurrent trichiasis during follow-up were offered repeat surgery. Participants with other ophthalmic pathology (eg, cataract) were referred. High-resolution digital photographs of upper eyelid, cornea, and tarsal conjunctiva were taken at 6 months and 12 months.22 To address potential concerns of bias which might arise from identifying procedure type from surgical scars, the upper eyelid photograph was taken after covering the incision area with a shaped occluder to prevent any unmasking of the independent photograph grader (an ophthalmologist with 15 years' experience of examining for trachomatous trichiasis). Images were viewed on a 15 inch high-resolution “retina” screen (Apple). Trachomatous trichiasis was considered to be present if there was one or more lashes touching the eye, identified by the lashes deviating over the globe and appearing to touch the eye.

5 years' experience of examining for trachomatous trichiasis). Images were viewed on a 15 inch high-resolution “retina” screen (Apple). Trachomatous trichiasis was considered to be present if there was one or more lashes touching the eye, identified by the lashes deviating over the globe and appearing to touch the eye. Outcomes The primary outcome was the cumulative proportion of individuals who developed recurrent trichiasis by 12 months. Recurrent trichiasis was defined as one or more lashes touching the eye or clinical evidence of epilation, or a history of repeat trichiasis surgery by 12 months. A-priori defined secondary outcome measures were: recurrent trichiasis at 6 months and 12 months; trichiasis recurrence difference by surgeon; trichiasis recurrence difference by baseline disease severity; number, type, and location of recurrent lashes at 12 months; corneal opacity, vision, and contrast sensitivity changes at 12 months; intraoperative, immediate, and late postoperative surgical complications (bleeding, infection, and granulomas); ECA at 12 months; and patient-reported outcomes.

baseline disease severity; number, type, and location of recurrent lashes at 12 months; corneal opacity, vision, and contrast sensitivity changes at 12 months; intraoperative, immediate, and late postoperative surgical complications (bleeding, infection, and granulomas); ECA at 12 months; and patient-reported outcomes. Statistical analysis In the STAR trial, the 1 year trichiasis recurrence rate using BLTR surgery was about 10% (tetracycline group).5 In our recent trials in Ethiopia involving patients with a similar severity of disease to the STAR trial, we found PLTR surgery had a 1 year recurrence rate of 18%.8 A sample of 836 participants was estimated to have 90% power and 95% confidence to detect a similar difference in recurrent trichiasis (18% vs 10%). Therefore, we aimed to recruit 1000 cases (500 in each group), to allow for about 15% loss to follow-up. Data were double-entered into Access 13 (Microsoft) and transferred to Stata 11 (StataCorp) for analysis. For participants who had bilateral surgery, we randomly designated one eye to be the study eye for the analysis. A modified intention-to-treat analysis was done, with primary outcome data analysed on all participants seen at either the 6 month or 12 month follow-up or both. Those not seen at either of these follow-up visits were excluded from the analysis.

gery, we randomly designated one eye to be the study eye for the analysis. A modified intention-to-treat analysis was done, with primary outcome data analysed on all participants seen at either the 6 month or 12 month follow-up or both. Those not seen at either of these follow-up visits were excluded from the analysis. The primary outcome and binary secondary outcomes were compared between the two surgical groups with logistic regression analyses to estimate the odds ratio (OR) and 95% CI. All comparisons between the two surgical procedures were controlled for surgeon as a fixed effect in the model to account for the stratified randomisation. The risk difference in the primary outcome (recurrent trichiasis by 12 months) between BLTR and PLTR procedures was estimated. The possibility of effect modification between group and a-priori defined factors such as surgeon, preoperative trichiasis severity, papillary inflammation, age, and sex was investigated by including interaction terms in the model and using a likelihood ratio test to assess statistical significance of the interaction term. Ordered categorical secondary outcomes (changes in visual acuity and corneal opacity, bleeding, and patient-reported outcomes) were compared between the two surgical interventions using ordinal logistic regression. Categorical secondary outcomes (type and location of recurrent lashes, ECAs, and entropion correction) were analysed using multinomial logistic regression to estimate relative risk ratio (RRR) and 95% CI. Negative binomial regression was used to analyse the difference in the number of recurrent lashes touching the eye between the two intervention groups. The signed-rank test was used to analyse visual acuity and contrast sensitivity changes between baseline and the 12 month follow-up. The risk of trichiasis recurrence difference by surgeon between the two surgical interventions was analysed using logistic regression adjusted for baseline disease severity such as entropion and trichiasis. To investigate the possibility of a learning curve effect during recruitment, the trichiasis recurrence rates for the first 50% of cases to be recruited versus the second 50% of cases recruited for each surgeon were compared using logistic regression adjusted for baseline disease severity such as entropion and trichiasis. The trial is registered at the Pan African Clinical Trials Registry (PACTR201401000743135).

currence rates for the first 50% of cases to be recruited versus the second 50% of cases recruited for each surgeon were compared using logistic regression adjusted for baseline disease severity such as entropion and trichiasis. The trial is registered at the Pan African Clinical Trials Registry (PACTR201401000743135). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between Feb 13, 2014, and May 31, 2014, 5168 people were examined for eligibility, of whom 1314 (25%) had trachomatous trichiasis (figure). The remaining 3854 had other ocular conditions. Of the 1314 trichiasis cases, 312 did not meet inclusion criteria, largely because they had previously received surgery for trichiasis. Of the 1002 eligible participants, two (<1%) declined surgery. Thus, 1000 trichiasis cases consented, were enrolled, and randomly assigned (501 in the BLTR group and 499 in the PLTR group).

e 1314 trichiasis cases, 312 did not meet inclusion criteria, largely because they had previously received surgery for trichiasis. Of the 1002 eligible participants, two (<1%) declined surgery. Thus, 1000 trichiasis cases consented, were enrolled, and randomly assigned (501 in the BLTR group and 499 in the PLTR group). Participants were reassessed at 10 days (range 7–14) for suture removal, 6 months, and 12 months after enrolment. Almost all (98%) participants were examined at each follow-up. At 10 days, two people had travelled to another region, and had sutures removed in their new locality. Eight (1%) participants were not seen at either 6 month or 12 month follow-up visits and were therefore excluded from the analysis: three from the PLTR group and five from the BLTR group. Hence, primary outcome data were available and analysed for 992 (99%): 496 in each group.

had sutures removed in their new locality. Eight (1%) participants were not seen at either 6 month or 12 month follow-up visits and were therefore excluded from the analysis: three from the PLTR group and five from the BLTR group. Hence, primary outcome data were available and analysed for 992 (99%): 496 in each group. Baseline demographic and clinical characteristics were balanced between the trial groups (table 1). The majority of the participants were female (77%) and their mean age was 47·3 years. The two groups were comparable for visual acuity and prevalence of corneal opacity, conjunctival inflammation, scarring, entropion, and trichiasis. There was evidence of epilation in 588 (59%) participants: 281 (56%) in the PLTR group and 307 (61%) in the BLTR group; among these 82 (8%) in both groups had successfully epilated, with no lashes touching. Major trichiasis was present in 145 (29%) of the PLTR group and 144 (29%) of the BLTR group. About 90% of the participants in both groups had corneal lashes (table 1). PLTR surgery took slightly less time than BLTR surgery (15 min 33 s vs 16 min 39 s; p<0·0001).

roups had successfully epilated, with no lashes touching. Major trichiasis was present in 145 (29%) of the PLTR group and 144 (29%) of the BLTR group. About 90% of the participants in both groups had corneal lashes (table 1). PLTR surgery took slightly less time than BLTR surgery (15 min 33 s vs 16 min 39 s; p<0·0001). By 12 months, the primary outcome, cumulative recurrent trichiasis, had developed in 173 (17%) of 992 study eyes. Cumulative recurrence was significantly more frequent in the BLTR group (110/496 [22%]) than in the PLTR group (63/496 [13%]); after adjusting for surgeon, the OR was 1·96 (95% CI 1·40–2·75; p=0·0001). The risk difference for recurrent trichiasis between BLTR and PLTR procedures was 9·50% (95% CI 4·79–14·16). There was no evidence of effect modification between group and a-priori defined other factors on the primary outcome, including surgeon. The primary outcome analysis using the photograph grading results was similar to the field grading. By 12 months, cumulative recurrent trichiasis was recorded for 250 (25%) of 992 study eyes. Recurrence was significantly more frequent in the BLTR group than in the PLTR group (32% vs 19%; OR 1·97 [95% CI 1·47–2·65]; p<0·0001). The risk difference for recurrent trichiasis between BLTR and PLTR procedures was 12·5% (95% CI 7·2–17·8).

cumulative recurrent trichiasis was recorded for 250 (25%) of 992 study eyes. Recurrence was significantly more frequent in the BLTR group than in the PLTR group (32% vs 19%; OR 1·97 [95% CI 1·47–2·65]; p<0·0001). The risk difference for recurrent trichiasis between BLTR and PLTR procedures was 12·5% (95% CI 7·2–17·8). At 10 days, recurrent trichiasis was present in three study eyes, one in the PLTR group and two in the BLTR group. At 6 months, recurrent trichiasis was present in 114 (12%) of 983 study eyes, and was significantly more frequent in the BLTR group than the PLTR group (71 [14%] vs 43 [9%]; OR 1·77 [95% CI 1·19–2·65]; p=0·0001). At 12 months, recurrent trichiasis was present in 131 (13%) of 981 study eyes and again remained significantly more frequent in the BLTR group than the PLTR group (85 [17%] vs 46 [9%]; OR 2·04 [95% CI 1·39–2·99]; p=0·0003). There was no evidence of a difference in the risk of trichiasis recurrence between surgeons by 12 months for either PLTR (p=0·80) or BLTR (p=0·44), or for a learning curve during the course of the trial for either procedure. For PLTR, recurrence risks during the first and second half of recruitment were 32 (13%) of 248 and 31 (12%) of 248, respectively (p=0·68). For BLTR, recurrence risks during the first and second half of recruitment were 55 (22%) of 247 and 55 (22%) of 249, respectively (p=0·93).

ourse of the trial for either procedure. For PLTR, recurrence risks during the first and second half of recruitment were 32 (13%) of 248 and 31 (12%) of 248, respectively (p=0·68). For BLTR, recurrence risks during the first and second half of recruitment were 55 (22%) of 247 and 55 (22%) of 249, respectively (p=0·93). The number, type, and location of recurrent lashes were comparable between the two groups (Table 1, Table 2). BLTR surgery had more frequent recurrence than PLTR surgery for major trichiasis cases and across all baseline entropion grades. There was no evidence of a difference in visual acuity, contrast sensitivity, corneal opacity, and entropion changes at 12 months between the two groups (table 2). However, compared with the baseline, at 12 months there was a statistically significant overall improvement in visual acuity (baseline median logMAR, 0·6 [IQR 0·3–0·8] vs 12 month median logMAR, 0·5 [0·2–0·7]; signed-rank test, p<0·0001) and contrast sensitivity (baseline median contrast sensitivity, 3% [2–5] vs 12 month contrast sensitivity, 2% [1–3]; signed-rank test, p<0·0001) in the entire combined study sample.

rovement in visual acuity (baseline median logMAR, 0·6 [IQR 0·3–0·8] vs 12 month median logMAR, 0·5 [0·2–0·7]; signed-rank test, p<0·0001) and contrast sensitivity (baseline median contrast sensitivity, 3% [2–5] vs 12 month contrast sensitivity, 2% [1–3]; signed-rank test, p<0·0001) in the entire combined study sample. After adjusting for surgeon, there was evidence of a difference in odds of intraoperative, immediate, or late postoperative complications between the two surgical interventions (table 3). There was more intraoperative and immediate postoperative bleeding in the BLTR surgery group than the PLTR surgery group (OR 2·76 [95% CI 1·27–6·00]; p=0·01), and also more postoperative infection in the BLTR surgery group than the PLTR surgery group (OR 4·44 [95% CI 2·11–9·33]; p=0·0001; table 3). Granulomata were less frequent in the BLTR group compared with the PLTR group (OR 0·41 [95% CI 0·20–0·83]; p=0·01; table 3).

TR surgery group (OR 2·76 [95% CI 1·27–6·00]; p=0·01), and also more postoperative infection in the BLTR surgery group than the PLTR surgery group (OR 4·44 [95% CI 2·11–9·33]; p=0·0001; table 3). Granulomata were less frequent in the BLTR group compared with the PLTR group (OR 0·41 [95% CI 0·20–0·83]; p=0·01; table 3). The frequency of clinically non-significant (mild) ECA at 12 months was lower in the BLTR surgery group than the PLTR surgery group (RRR 0·50 [95% CI 0·34–0·73]; p<0·0001; table 3). However, there was no evidence of a difference in the frequency of clinically significant (moderate-to-severe) ECA between the two groups (RRR 1·10 [95% CI 0·66–1·81]; p=0·72; table 3). A similar pattern in ECA was found by independent photograph grading. Clinically mild ECA at 12 months was less frequent in the BLTR group (27/484 [5%]) than in the PLTR group (58/489 [12%]; RRR 0·43 [95% CI 0·27–0·70]; p=0·001). However, again we found no evidence of a difference in moderate-to-severe ECA between the two groups (BLTR, 4% vs PLTR, 5%; RRR 0·76 [95% CI 0·41–1·44]; p=0·40). There was evidence of more under-correction at 12 months with BLTR surgery than PLTR surgery (table 3).

(58/489 [12%]; RRR 0·43 [95% CI 0·27–0·70]; p=0·001). However, again we found no evidence of a difference in moderate-to-severe ECA between the two groups (BLTR, 4% vs PLTR, 5%; RRR 0·76 [95% CI 0·41–1·44]; p=0·40). There was evidence of more under-correction at 12 months with BLTR surgery than PLTR surgery (table 3). There was no evidence of a difference between groups in the patient-reported pain experienced during surgery (p=0·84; table 4). However, participants in the BLTR group reported more pain and discomfort during the days between surgery and suture removal than the PLTR group (OR 1·46 [95% CI 1·12–1·89]; p=0·004; table 4). There was no evidence of a difference in patient satisfaction between the two groups for treatment of trichiasis (p=0·20) or the cosmetic appearance of the operated eyelid (p=0·64; table 4). Discussion Around 7 million people have trachomatous trichiasis and require high-quality surgical intervention.2 A major global effort exists to scale up surgical programmes. However, high postoperative trichiasis recurrence rates are undermining trachoma control.25 Identifying the surgical intervention with the lowest recurrence rate has been a research priority for many years.18 In this trial, we compared the relative effectiveness of the two most commonly used operations and found that PLTR surgery has a significantly lower trichiasis recurrence rate at 12 months than BLTR surgery, particularly for more severe cases.

cedures, to reduce the learning curve to achieve proficiency in the new procedure.26 During training and standardisation, before the commencement of the trial, each surgeon did about 100 BLTR operations and was confirmed by two assessors to be performing the procedure per protocol, using the WHO Certification process.4 There is clear evidence that during the trial the surgeons continued to do both operations consistently well and that the recorded difference in the primary outcome was not attributable to having learnt the BLTR procedure more recently. First, there were only three recurrent cases by 10 days, indicating that primary surgical failure was rare. If the recorded differences in recurrence were due to poor surgical technique, we would anticipate this to be more apparent by 10 days. This finding suggests that the subsequent difference in the primary outcome is attributable to fundamental differences in the surgical method that achieves a more stable and long-lasting correction in the case of PLTR surgery. Second, trichiasis recurrence rates between the first and second half of recruitment were very similar. If surgeons were still on a BLTR learning curve, a lower recurrence rate in the second half of recruitment would have been anticipated. Third, there was no significant difference in recurrence for either surgical procedure between surgeons. Finally, the recurrence rates for both procedures were generally similar to or lower than those reported in other trials, with the exception of the STAR trial which reported a lower BLTR recurrence rate.5, 6, 7, 8, 9, 10, 11, 12, 13, 14

ignificant difference in recurrence for either surgical procedure between surgeons. Finally, the recurrence rates for both procedures were generally similar to or lower than those reported in other trials, with the exception of the STAR trial which reported a lower BLTR recurrence rate.5, 6, 7, 8, 9, 10, 11, 12, 13, 14 The only other trial to compare BLTR and PLTR procedures was done in Ethiopia.13 This trial reported comparable outcomes for the two procedures: BLTR, 10·4%, and PLTR, 12·3%, recurrence at 3 months. However, this earlier study had a number of constraints. First, it was under-powered to detect a difference (153 patients, 256 eyes operated). Second, it was done at a tertiary teaching hospital by ophthalmologists. By contrast, most programmatic trichiasis surgery is done by non-physicians with limited training in remote, low-level, health facilities. Alternative techniques might give different results in more programmatic settings. Third, the 3 month follow-up period was too short to assess the relative performance because differences might take longer to become apparent.6, 7, 11, 27, 28

non-physicians with limited training in remote, low-level, health facilities. Alternative techniques might give different results in more programmatic settings. Third, the 3 month follow-up period was too short to assess the relative performance because differences might take longer to become apparent.6, 7, 11, 27, 28 The PLTR surgical procedure did better than BLTR for several secondary outcomes. A higher rate of postoperative infection occurred following BLTR, probably because of the skin incision. All infections were treated successfully with oral antibiotics. The skin and orbicularis incision probably also explains the greater intraoperative and postoperative bleeding and postoperative pain that occurred with the BLTR procedure because these structures have an extensive vascular and sensory supply. These are important considerations for improving surgical uptake, which might be reduced by patient reports of pain and bleeding. Participants reported very high levels of satisfaction with the cosmetic outcome and effect of surgery in alleviating the trichiasis in both groups at 12 months. There was no difference by group. However, some caution needs to be taken in drawing firm conclusions from such data; the questions were asked by members of the study team, and there could be some reticence in expressing dissatisfaction in this context. Of note, the Kenyan National Trachoma Control Programme recently switched from the BLTR to PLTR surgery because of reports of widespread patient dissatisfaction with the appearance of the full-thickness incision in BLTR surgery. Under-correction was more frequent with BLTR surgery at 12 months, suggesting it is less effective at correcting underlying entropion.

ol Programme recently switched from the BLTR to PLTR surgery because of reports of widespread patient dissatisfaction with the appearance of the full-thickness incision in BLTR surgery. Under-correction was more frequent with BLTR surgery at 12 months, suggesting it is less effective at correcting underlying entropion. We found that BLTR had a lower rate of mild ECAs. We consider this degree of ECA to be clinically and cosmetically non-significant because the vertical deviation from the lid contour is less than 1 mm. It is possible that this difference reflects consistently greater degrees of evertion with PLTR. There was no difference in moderate-to-severe ECA by group. Conjunctival granulomata developed more frequently after PLTR surgery.8 Granulomata are probably a vigorous healing response that occurs in a tissue defect.12 The additional rotation effected by the PLTR might create a larger posterior lamella defect and thereby a higher likelihood of granuloma formation. However, they are usually a minor complication that either self-resolve or need only a simple shave under topical anaesthesia. In the earlier comparison of BLTR and PLTR in Ethiopia, both eyelid notching and granulomata were significantly more common in the BLTR group than in the PLTR group (p=0·002).13

loma formation. However, they are usually a minor complication that either self-resolve or need only a simple shave under topical anaesthesia. In the earlier comparison of BLTR and PLTR in Ethiopia, both eyelid notching and granulomata were significantly more common in the BLTR group than in the PLTR group (p=0·002).13 We think that it is biologically plausible that the PLTR achieves a more effective and stable correction of the entropion and trichiasis due to a key difference in technique from the BLTR. In the PLTR procedure, the lower edge of the dissected upper portion of the tarsal plate is drawn down and tucked into the dissected space between the anterior lamella and the lower portion of the tarsal plate behind.4 Once healed, this provides a wedge of tissue that continues to rotate the distal end of the eyelid outwards, and stabilise the correction. This study has several strengths. It had a large sample size and very high follow-up rates. Demographic and clinical characteristics were balanced between groups. The surgeons were rigorously trained and standardised to ensure the procedures were done correctly.

We think that it is biologically plausible that the PLTR achieves a more effective and stable correction of the entropion and trichiasis due to a key difference in technique from the BLTR. In the PLTR procedure, the lower edge of the dissected upper portion of the tarsal plate is drawn down and tucked into the dissected space between the anterior lamella and the lower portion of the tarsal plate behind.4 Once healed, this provides a wedge of tissue that continues to rotate the distal end of the eyelid outwards, and stabilise the correction. This study has several strengths. It had a large sample size and very high follow-up rates. Demographic and clinical characteristics were balanced between groups. The surgeons were rigorously trained and standardised to ensure the procedures were done correctly. A potential design limitation in a trial of these two procedures is the risk of unmasking at the time of follow-up observations because some BLTR cases might very occasionally have a faint skin scar. The baseline, 6 month, and 12 month observers were masked to the randomisation. However, to independently assess the primary outcome for observer bias, photographs in which the upper lid skin was covered by a mask were graded. We found that the analysis of primary outcome using field and photograph grading were comparable, showing no systematic bias in the field grading. The observations of some of the secondary outcomes made during the operative procedure and at 10 days were impossible to mask. The use of surgeons who had previously been trained in PLTR and then provided with a second round of training in BLTR could be viewed as a potential limitation. However, we think that there was ample pre-trial training, practice, and assessment to bring the surgeons to a proficient standard, and that there is clear evidence that during the trial high standards were maintained, as discussed above. In this trial we used silk sutures, which were removed at 10 days. Although absorbable sutures such as polyglactan-910 (vicryl) offer the operational advantage of not needing to be removed, we have previously found in a randomised trial that silk and absorbable sutures have comparable outcomes, and therefore it is unlikely that the outcome of this present study would be modified by their use.8

absorbable sutures such as polyglactan-910 (vicryl) offer the operational advantage of not needing to be removed, we have previously found in a randomised trial that silk and absorbable sutures have comparable outcomes, and therefore it is unlikely that the outcome of this present study would be modified by their use.8 Overall, the PLTR procedure was superior to the BLTR in terms of lower trichiasis recurrence and fewer intraoperative and immediate postoperative complications. All other factors being equal, PLTR could be the preferred procedure for the programmatic management of trachomatous trichiasis. We suggest new surgical trainees in both established and new programmes should be trained in the PLTR procedure. Additionally, consideration could be given to further research to investigate whether individuals previously trained to do BLTR surgery need to be re-trained in PLTR surgery. Acknowledgments This study was supported by The Wellcome Trust (grant number 098481/Z/12/Z). We thank the trachoma control programme in Amhara National Regional State of Ethiopia, which is a collaboration between the Regional Health Bureau and the Lions-Carter Center SightFirst Initiative; the research study team; the study participants; West Gojam Zone Health Office, the Woreda Health Offices, and Mecha Woreda Administration Office.

ntrol programme in Amhara National Regional State of Ethiopia, which is a collaboration between the Regional Health Bureau and the Lions-Carter Center SightFirst Initiative; the research study team; the study participants; West Gojam Zone Health Office, the Woreda Health Offices, and Mecha Woreda Administration Office. Contributors EH and MJB did the literature search. EH, ABK, PME, RLB, DCWM, SNR, KC, HAW, and MJB were responsible for the study conception and design. EH, TW, SA, ZT, MZ, ZZ, ABK, and MJB collected the data. EH, HAW, and MJB did the statistical analysis. EH, HAW, and MJB were responsible for the interpretation of the data. EH and MJB drafted the manuscript. All authors critically revised the manuscript for important intellectual content. RLB, DCWM, and MJB obtained the funding. TW, SA, ZT, MZ, ZZ, ABK, CHR, PME, and KC provided administrative, technical, or material support. MJB was the study supervisor. Declaration of interests We declare no competing interests. Figure Trial profile Table 1 Baseline and 12 month characteristics of participants

Contributors EH and MJB did the literature search. EH, ABK, PME, RLB, DCWM, SNR, KC, HAW, and MJB were responsible for the study conception and design. EH, TW, SA, ZT, MZ, ZZ, ABK, and MJB collected the data. EH, HAW, and MJB did the statistical analysis. EH, HAW, and MJB were responsible for the interpretation of the data. EH and MJB drafted the manuscript. All authors critically revised the manuscript for important intellectual content. RLB, DCWM, and MJB obtained the funding. TW, SA, ZT, MZ, ZZ, ABK, CHR, PME, and KC provided administrative, technical, or material support. MJB was the study supervisor. Declaration of interests We declare no competing interests. Figure Trial profile Table 1 Baseline and 12 month characteristics of participants Baseline 12 months PLTR group (n=499) BLTR group (n=501) PLTR group (n=491) BLTR group (n=490) Sex (female) 388 (78%) 377 (75%) .. .. Age, years (mean, SD) 47·2 (15·0) 47·5 (14·9) .. .. Illiterate 441 (88%) 445 (89%) .. .. Best corrected logMAR visual acuity in study eye −0·1 to 0·3 141 (28%) 137 (27%) 175 (36%) 169 (34%) 0·3 to 0·7 190 (38%) 209 (42%) 186 (38%) 212 (43%) 0·7 to 1·1 107 (21%) 103 (21%) 90 (18%) 78 (16%) 1·1 to 2·0 18 (4%) 18 (4%) 10 (2%) 10 (2%) CF/HM/PL 37 (7%) 27 (5%) 25 (5%) 15 (3%) NPL 6 (1%) 7 (1%) 4 (1%) 5 (1%) Not possible to measure .. .. 1 (<1%) 1 (<1%) Entropion grade 0 11 (2%) 7 (1%) 467 (95%) 446 (91%) 1 93 (19%) 85 (17%) 17 (3%) 39 (8%) 2 315 (63%) 334 (67%) 6 (1%) 5 (1%) 3 71 (14%) 66 (13%) 1 (<1%) 0 4 9 (2%) 9 (2%) 0 0 Trichiasis (number of lashes) No trichiasis .. .. 445 (91%) 405 (83%) None (epilating) 38 (8%) 44 (9%) 7 (1%) 13 (3%) 1–5 316 (63%) 312 (62%) 37 (8%) 66 (13%) 6–9 87 (17%) 87 (17%) 1 (<1%) 5 (1%) 10–19 41 (8%) 46 (9%) 1 (<1%) 1 (<1%) 20+ 17 (3%) 12 (2%) 0 0 Mean (SD)* 5·6 (6·6) 5·4 (5·7) 2·7 (2·7) 2·6 (2·5) Lash location None (epilating) 38 (8%) 45 (9%) 7 (15%) 13 (15%) Corneal with or without peripheral 450 (90%) 451 (90%) 30 (65%) 59 (69%) Peripheral only 11 (2%) 5 (1%) 9 (20%) 13 (15%) Corneal opacity None (CC0) 121 (24%) 132 (26%) 155 (32%) 159 (32·5) Peripheral (CC1) 204 (41%) 201 (40%) 140 (29%) 157 (32·0) Off centre faint (CC2a) 94 (19%) 94 (19%) 98 (20%) 85 (17%) Off centre dense (CC2b) 19 (4%) 11 (2%) 7 (1%) 4 (1%) Central faint (CC2c) 48 (10%) 50 (10%) 77 (16%) 76 (16%) Central dense (CC2d) 7 (1%) 7 (1%) 10 (2%) 5 (1%) Total central dense (CC3) 4 (1%) 6 (1%) 2 (<1%) 4 (1%) Phthisis (CC4) 2 (<1%) 0 2 (<1%) 0 Tarsal conjunctiva inflammation None (P0) 6 (1%) 9 (2%) 9 (2%) 12 (2%) Mild (P1) 117 (23%) 131 (26%) 104 (21%) 98 (20%) Moderate (P2) 306 (61%) 297 (59%) 332 (68%) 321 (66%) Severe (P3) 70 (14%) 64 (13%) 46 (9%) 59 (12%) Tarsal conjunctival scarring None (C0) 0 0 .. .. Mild (C1) 51 (10%) 56 (11%) .. .. Moderate (C2) 373 (75%) 367 (73%) .. .. Severe (C3) 75 (15%) 78 (16%) .. .. Recurrent trichiasis by surgeon† 1 .. .. 8/89 (9%) 27/91 (30%) 2 .. .. 14/95 (15%) 17/93 (18%) 3 .. ..

) 321 (66%) Severe (P3) 70 (14%) 64 (13%) 46 (9%) 59 (12%) Tarsal conjunctival scarring None (C0) 0 0 .. .. Mild (C1) 51 (10%) 56 (11%) .. .. Moderate (C2) 373 (75%) 367 (73%) .. .. Severe (C3) 75 (15%) 78 (16%) .. .. Recurrent trichiasis by surgeon† 1 .. .. 8/89 (9%) 27/91 (30%) 2 .. .. 14/95 (15%) 17/93 (18%) 3 .. .. 12/84 (14%) 17/85 (20%) 4 .. .. 10/92 (11%) 17/91 (19%) 5 .. .. 6/47 (13%) 12/47 (26%) 6 .. .. 13/89 (15%) 20/89 (22%) Data are n (%) unless otherwise stated. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. CF=counting fingers. HM=hand movement. PL=perception of light. NPL=no perception of light. * Excluding those with no lashes touching the eyeball. † Data are n/N (%). Table 2 Secondary clinical outcomes and changes in clinical phenotype at 12 months

12/84 (14%) 17/85 (20%) 4 .. .. 10/92 (11%) 17/91 (19%) 5 .. .. 6/47 (13%) 12/47 (26%) 6 .. .. 13/89 (15%) 20/89 (22%) Data are n (%) unless otherwise stated. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. CF=counting fingers. HM=hand movement. PL=perception of light. NPL=no perception of light. * Excluding those with no lashes touching the eyeball. † Data are n/N (%). Table 2 Secondary clinical outcomes and changes in clinical phenotype at 12 months PLTR group BLTR group OR or RRR (95% CI) p value Cumulative recurrence by baseline trichiasis severity* Minor trachomatous trichiasis 26/266 (10%) 36/257 (14%) 1·47 (0·85–2·53) 0·16 Major trachomatous trichiasis 37/230 (16%) 74/239 (31%) 2·29 (1·46–3·59) 0·0003 Cumulative recurrence by baseline entropion severity* None or mild 11/102 (11%) 28/90 (31%) 3·98 (1·80–8·80) 0·0007 Moderate 37/314 (12%) 59/331 (18%) 1·59 (1·02–2·49) 0·04 Severe 15/80 (19%) 23/75 (31%) 2·04 (0·95–4·37) 0·07 Number of recurrent lashes (mean, SD)† 2·67 (2·72) 2·65 (2·45) 0·97‡ (0·71–1·32) 0·84 Types of recurrent lashes§ Entropic 4/46 (9%) 10/85 (12%) 1·79¶ (0·48–6·69) 0·38 Metaplastic (base outcome) 32/46 (70%) 55/85 (65%) 1 .. Misdirected 3/46 (6%) 7/85 (8%) 1·37¶ (0·32–5·99) 0·67 Epilating 7/46 (15%) 13/85 (15%) 1·26¶ (0·43–3·69) 0·68 Location of recurrent lashes§ Corneal or corneal and peripheral (base outcome) 30/46 (65%) 59/85 (69%) 1 .. Peripheral 9/46 (20%) 13/85 (15%) 0·82¶ (0·30–2·22) 0·69 Epilating 7/46 (15%) 13/85 (15%) 1·08¶ (0·37–3·12) 0·89 Visual acuity change|| Worse 123/490 (25%) 111/489 (23%) 0·97 (0·77–1·23) 0·81 Same 172/490 (35%) 198/489 (41%) Better 195/490 (40%) 180/489 (37%) Contrast sensitivity|| Worse 114/490 (23%) 100/489 (20%) 1·05 (0·83–1·32) 0·71 Same 149/490 (30%) 165/489 (34%) Better 227/490 (46%) 224/489 (46%) Corneal opacity change|| More opacity 84/491 (17%) 68/490 (14%) 1·19 (0·92–1·55) 0·19 No change 329/491 (67%) 338/490 (69%) Less opacity 78/491 (16%) 84/490 (17%) Entropion grade change|| <2 grade change or no change 119/491 (24%) 113/490 (23%) 1·07 (0·79–1·44) 0·67 ≥2 grade change 372/491 (76%) 377/490 (77%) Data are n/N (%), unless otherwise stated. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. OR=odds ratio. RRR=relative risk ratio.

%) 84/490 (17%) Entropion grade change|| <2 grade change or no change 119/491 (24%) 113/490 (23%) 1·07 (0·79–1·44) 0·67 ≥2 grade change 372/491 (76%) 377/490 (77%) Data are n/N (%), unless otherwise stated. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. OR=odds ratio. RRR=relative risk ratio. * Analysis done using logistic regression adjusted for surgeon to see the effect of the two surgical procedures on cumulative recurrence (by 12 months) across baseline trichiasis and entropion severity level. † Analysis done using negative binomial regression. ‡ Incidence rate ratio. § Multinomial logistic regression. ¶ Relative risk ratio. || Ordinal logistic regression. Table 3 Complications and eyelid contour abnormalities

* Analysis done using logistic regression adjusted for surgeon to see the effect of the two surgical procedures on cumulative recurrence (by 12 months) across baseline trichiasis and entropion severity level. † Analysis done using negative binomial regression. ‡ Incidence rate ratio. § Multinomial logistic regression. ¶ Relative risk ratio. || Ordinal logistic regression. Table 3 Complications and eyelid contour abnormalities PLTR group BLTR group OR or RRR (95% CI) p value Intraoperative or postoperative bleeding* Mild 490/499 (98%) 477/501 (95%) 2·76 (1·27–6·00) 0·01 Moderate 8/499 (2%) 18/501 (4%) Excessive 1/499 (<1%) 6/501 (1%) Sign of infection at 7–14 days†‡ 9/498 (2%) 37/500 (7%) 4·44 (2·11–9·33) 0·0001 Granuloma by 12 months† 26/496 (5%) 11/496 (2%) 0·41 (0·20–0·83) 0·01 Lagophthalmos (present) 3/491 (1%) 7/490 (1%) Eyelid contour abnormality at 12 months§ None (base outcome) 371/491 (76%) 404/490 (82%) 1 .. Clinically non-significant (mild) 89/491 (18%) 49/490 (10%) 0·50¶ (0·34–0·73) 0·000 Clinically significant (moderate-to- severe) 31/491 (6%) 37/490 (8%) 1·10¶ (0·66–1·81) 0·72 Central correction at 12 months§ Corrected (base outcome) 468/491 (95%) 454/490 (93%) 1 .. Over-corrected 12/491 (2%) 6/490 (1%) 0·52¶ (0·19–1·39) 0·19 Under-corrected 11/491 (2%) 30/490 (6%) 2·81¶ (1·39–5·68) 0·004 Medial correction at 12 months§ Corrected (base outcome) 469/491 (96%) 450/490 (92%) 1 .. Over-corrected 0 0 .. Under-corrected 22/491 (4%) 40/490 (8%) 1·90¶ (1·11–3·26) 0·02 Lateral correction at 12 months§ Corrected (base outcome) 486/491 (99%) 469/490 (96%) 1 .. Over-corrected 1/491 (<1%) 0 .. .. Under-corrected 4/491 (1%) 21/490 (4%) 5·44¶ (1·85–16·00) 0·002 Data are n/N (%), unless stated otherwise. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. OR=odds ratio. RRR=relative risk ratio.

onths§ Corrected (base outcome) 486/491 (99%) 469/490 (96%) 1 .. Over-corrected 1/491 (<1%) 0 .. .. Under-corrected 4/491 (1%) 21/490 (4%) 5·44¶ (1·85–16·00) 0·002 Data are n/N (%), unless stated otherwise. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. OR=odds ratio. RRR=relative risk ratio. * Ordinal logistic regression. † Logistic regression. ‡ Erythematous swelling and discharge. § Multinomial logistic regression. ¶ Relative risk ratio. Table 4 Patient-reported outcomes PLTR group BLTR group OR*(95% CI) p value Pain during surgery None 441/499 (88%) 441/501 (88%) 1·04 (0·71–1·53) 0·84 Mild 40/499 (8%) 39/501 (8%) Moderate 7/499 (1%) 11/501 (2%) Severe 11/499 (2%) 10/501 (2%) Pain between surgery and suture removal None 347/498 (70%) 309/500 (62%) 1·46 (1·12–1·89) 0·004 Mild 94/498 (19%) 107/500 (21%) Moderate 38/498 (8%) 56/500 (11%) Severe 19/498 (4%) 28/500 (6%) Satisfaction with the effect of surgery on the trichiasis at 12 months Satisfied 463/491 (94%) 452/490 (92%) 1·39 (0·84–2·31) 0·20 Neither satisfied nor dissatisfied 13/491 (3%) 16/490 (3%) Dissatisfied 15/491 (3%) 22/490 (4%) Satisfaction with the cosmetic appearance at 12 months Satisfied 465/491 (95%) 461/490 (94%) 1·14 (0·66–1·97) 0·64 Neither satisfied nor dissatisfied 10/491 (2%) 7/490 (1%) Dissatisfied 16/491 (3%) 22/490 (4%) Data are n/N (%), unless stated otherwise. BLTR=bilamellar tarsal rotation. PLTR=posterior lamellar tarsal rotation. OR=odds ratio. * Ordinal logistic regression.

The London Declaration on Neglected Tropical Diseases1 marked its fourth year on Jan 30, 2016. The declaration represents a coordinate effort to control or eliminate ten of the neglected diseases by 2020, and has already led to important coordination and partnership, and mobilised considerable resources. Although the progress made so far is to be celebrated, now is the time to count down to 2020 and start monitoring the progress and trends towards achieving the control and elimination targets. First, progress towards the 2020 targets should be measured through collation, presentation, and analysis of standardised core metrics that track country progress. Elimination and control targets in the WHO Roadmap could serve as initial standardised metrics.2 Efforts are being made to measure the progress and forecast future trends, but most efforts are disease-specific and occur at a global scale.3 Therefore, comprehensive measurements to track the goal of the declaration are needed. Country-specific case studies, considering different aspects of policy, programming, and intervention would provide the whole picture of the analysis needed for the countdown to 2020. Focus on assessment from a multidimensional aspect and pinpointing of bottlenecks will help to identify the gaps in implementation and point out areas that need improvement.

ering different aspects of policy, programming, and intervention would provide the whole picture of the analysis needed for the countdown to 2020. Focus on assessment from a multidimensional aspect and pinpointing of bottlenecks will help to identify the gaps in implementation and point out areas that need improvement. Second, development of standardised methods to measure the progress towards reaching the targets is crucial. Modelling methods, which use the available data to forecast future trends will be important for understanding how countries are progressing towards achieving the goals of the London declaration. These methods will enable comparison of the outputs of such an analysis, and several sources of data would be needed to provide continuous high-quality data for analysis. Collaboration with national surveys, such as the demographic and health surveys and several indicators surveys, would be crucial for achieving this goal. Third, establishment of a global independent monitoring group would be essential to assess country-specific reports and set up an accountability framework.4 The group would have an important role in the definition of the core methods to monitor progress and develop a framework for country assessment mechanisms. This approach will help to further improve the quality of country-level reports. As we go further down the road, understanding if we are on track to achieve the targets is necessary. Standardised metrics, continued monitoring, and comprehensive country-specific case studies will be crucial to advance the goal of the declaration.

Third, establishment of a global independent monitoring group would be essential to assess country-specific reports and set up an accountability framework.4 The group would have an important role in the definition of the core methods to monitor progress and develop a framework for country assessment mechanisms. This approach will help to further improve the quality of country-level reports. As we go further down the road, understanding if we are on track to achieve the targets is necessary. Standardised metrics, continued monitoring, and comprehensive country-specific case studies will be crucial to advance the goal of the declaration. I am supported by a Wellcome Trust fellowship in public health and tropical medicine (grant number 099876).

Introduction Ocular surface squamous neoplasia (OSSN) covers a range of conjunctival and corneal diseases, from intra-epithelial dysplasia to invasive squamous cell carcinoma.1 Risk factors for OSSN include ultraviolet radiation, HIV infection, and human papillomavirus infection.2 In temperate regions, OSSN is uncommon, usually growing slowly, and most often affects elderly men. By contrast, in sub-Saharan Africa, OSSN is more common and aggressive.3 It affects younger adults, predominately women (around two-thirds of cases), and is strongly associated with HIV infection (in about 70% of cases). OSSN has a wide range of clinical phenotypes and late presentation with invasive orbital disease is not uncommon (figure 1). Surgical excision is the mainstay of treatment, although primary chemotherapy has also been used (appendix). OSSN often recurs after surgery. The highest recorded recurrence is 67%.4 Recurrence is a particular problem in sub-Saharan Africa, where it typically occurs in 30–40% of patients.4, 5, 6, 7 In temperate climates, recurrence typically occurs in 5–25% of patients (appendix). Surgical outcomes seem to be affected by delays in diagnosis, tumour size, histological grade, ocular location, scleral invasion, excision margin involvement, prior excision, and coexisting xeroderma pigmentosum.5, 8, 9 Several adjunctive treatment regimens are used during or after surgery to reduce recurrence: cryotherapy, topical chemotherapy (mitomycin and fluorouracil), interferon alfa-2b, retinoic acid, and radiotherapy (appendix p 1, 3, 7, 11, 13, 14). Most data on adjuvant treatment are case series. There is one previous randomised trial, from Australia, which assessed the effectiveness of topical mitomycin and there are no trial data on interventions for people with HIV.10, 11 Radical surgery (enucleation or exenteration) is usually needed for advanced disease.12

. Most data on adjuvant treatment are case series. There is one previous randomised trial, from Australia, which assessed the effectiveness of topical mitomycin and there are no trial data on interventions for people with HIV.10, 11 Radical surgery (enucleation or exenteration) is usually needed for advanced disease.12 The antimetabolite fluorouracil is often used in ophthalmology, particularly for its anti-scarring properties during surgical procedures (trabeculectomy, pterygium excision, and lacrimal surgery).13 Eye drops containing 1% fluorouracil have also been used for several years to treat patients with OSSN after tumour excision (appendix p 11), on the basis of case series, which suggest that fluorouracil reduces tumour recurrence and is safe.14, 15, 16, 17, 18, 19, 20 However, there are no data from trials. Fluorouracil is widely available and relatively cheap in sub-Saharan Africa, therefore, if shown to be an effective adjuvant, it would be a deliverable intervention. Research in context Evidence before this study

The antimetabolite fluorouracil is often used in ophthalmology, particularly for its anti-scarring properties during surgical procedures (trabeculectomy, pterygium excision, and lacrimal surgery).13 Eye drops containing 1% fluorouracil have also been used for several years to treat patients with OSSN after tumour excision (appendix p 11), on the basis of case series, which suggest that fluorouracil reduces tumour recurrence and is safe.14, 15, 16, 17, 18, 19, 20 However, there are no data from trials. Fluorouracil is widely available and relatively cheap in sub-Saharan Africa, therefore, if shown to be an effective adjuvant, it would be a deliverable intervention. Research in context Evidence before this study Ocular surface squamous neoplasia (OSSN) is an eye cancer, common in people with HIV. A Cochrane systematic review from 2013 showed no evidence from trials for the effectiveness of interventions used in this population. We searched electronic databases (PubMed, Embase, The Cochrane Library), clinical trial registries (WHO International Clinical Trials Registry Platform and the US National Institutes of Health ClinicalTrials.gov), and international conference proceedings of HIV/AIDS and AIDS-related cancers from the AIDS Education Global Education System for studies published up to Aug 31, 2015, irrespective of language or publication status. We used the terms “randomized controlled trial”, “controlled clinical trial”, “randomized”, “placebo”, “drug therapy”, “randomly”, “trial”, “conjunctiva*”, “ocular surface”, “carcinoma”, “cancer”, “neoplasia”, “neoplasm”, “neoplastic”, “dysplasia”, “dysplastic”, “squamous”, and “squamous cell”. We found only one trial, on topical mitomycin in a non-HIV-infected population in Australia. We identified some case series and case reports (appendix). Only series that reported recurrence as an outcome were included.

”, “neoplasia”, “neoplasm”, “neoplastic”, “dysplasia”, “dysplastic”, “squamous”, and “squamous cell”. We found only one trial, on topical mitomycin in a non-HIV-infected population in Australia. We identified some case series and case reports (appendix). Only series that reported recurrence as an outcome were included. Added value of this study Our study provides the first evidence from a trial of the effectiveness of fluorouracil as adjunctive treatment for OSSN. Our results show that the simple and relatively inexpensive use of 4 weeks of fluorouracil 1% eye drops after surgical excision substantially reduced the relative risk of recurrence compared with placebo and was safe. There were transient side-effects, such as watery eye, discomfort when taking the drops, and eyelid skin inflammation but these were mostly mild and resolved in a few weeks after completion of 4 week treatment. Implications of all the available evidence Recurrence is a huge issue in the management of this common and aggressive eye disease. Fluorouracil does not need stringent storage conditions and cytotoxics have a low risk of contamination. It is on the WHO Essential Medicines List, and is a widely available and low-cost option, particularly in sub-Saharan Africa, which has the highest incidence of OSSN in the world. Translation of these trial results into clinical practice is therefore feasible. We assessed whether use of fluorouracil 1% eye drops could reduce recurrence of OSSN following surgical excision in Kenya.

Recurrence is a huge issue in the management of this common and aggressive eye disease. Fluorouracil does not need stringent storage conditions and cytotoxics have a low risk of contamination. It is on the WHO Essential Medicines List, and is a widely available and low-cost option, particularly in sub-Saharan Africa, which has the highest incidence of OSSN in the world. Translation of these trial results into clinical practice is therefore feasible. We assessed whether use of fluorouracil 1% eye drops could reduce recurrence of OSSN following surgical excision in Kenya. Methods Study design and participants We did a double-blind, parallel-group, randomised, placebo-controlled trial at four centres in Kenya: Kenyatta National Hospital Eye Clinic in Nairobi, PCEA Kikuyu Eye Unit in central Kenya, Sabatia Eye Hospital in western Kenya, and Kitale District Hospital in the north Rift Valley. We enrolled consecutive patients presenting with suspicious conjunctival lesions. Entry criteria were: histologically proven OSSN involving two or fewer quadrants; attendance for follow-up within the first 2 months after excision; healing of the excision site; and age at least 18 years. Exclusion criteria were: previous treatment with topical antimetabolite drugs such as fluorouracil or mitomycin to the same eye or systemic cytotoxic drugs; extensive disease requiring more radical surgery than a simple excision; and pregnant or breastfeeding mothers. Patients were not enrolled if they did not think that they could return for follow-up.

reatment with topical antimetabolite drugs such as fluorouracil or mitomycin to the same eye or systemic cytotoxic drugs; extensive disease requiring more radical surgery than a simple excision; and pregnant or breastfeeding mothers. Patients were not enrolled if they did not think that they could return for follow-up. All participants gave written informed consent before enrolment. Ethics approval was granted by the Kenyatta National Hospital/University of Nairobi ethics and research committee and the London School of Hygiene & Tropical Medicine ethics committee. Approval was also obtained from the Kenya Pharmacy and Poisons Board to produce and use the active intervention drops because they are not commercially available. An independent data and safety monitoring board oversaw the study, confirmed data integrity, and approved the results and report for release. Trial personnel received good clinical practice training and certification. This study adhered to the tenets of the Declaration of Helsinki. Randomisation and masking Participants were randomly assigned (1:1) to either fluorouracil 1% or placebo eye drops. The fluorouracil eye drops were prepared by dilution of fluorouracil 50 g/L solution for injection in hydroxypropyl methylcellulose 0·7% artificial tear eye drops. The placebo was the same hydroxypropyl methylcellulose 0·7% artificial tear eye drops.

ndomly assigned (1:1) to either fluorouracil 1% or placebo eye drops. The fluorouracil eye drops were prepared by dilution of fluorouracil 50 g/L solution for injection in hydroxypropyl methylcellulose 0·7% artificial tear eye drops. The placebo was the same hydroxypropyl methylcellulose 0·7% artificial tear eye drops. The randomisation sequence was generated by computer by the trial statistician using Stata (version 12). The permuted block size (known only to the statistician) varied randomly between two and four, and randomisation was stratified by surgeon. The allocation sequence was transferred to the manufacturing pharmacy, where an independent pharmacist applied labels with sequential code numbers to the appropriate eye drop bottles. Participants, clinicians, and study personnel were masked to the allocation: the bottles, liquid content, and packaging had identical appearances. The supervising clinician at each of the four study centres issued the trial drug to participants. The allocation followed the order of enrolment.

ropriate eye drop bottles. Participants, clinicians, and study personnel were masked to the allocation: the bottles, liquid content, and packaging had identical appearances. The supervising clinician at each of the four study centres issued the trial drug to participants. The allocation followed the order of enrolment. Procedures Lesions involving two or fewer quadrants of the conjunctiva were fully excised with a 4 mm clear margin by the no-touch technique, with use of an operating microscope and under local anaesthesia.21 Absolute alcohol was applied to any corneal component of the tumour to loosen it and facilitate dissection. The conjunctival component was dissected down to bare sclera. Cryotherapy was not applied, because it is not generally available in sub-Saharan Africa. Topical adrenaline, and where necessary mild diathermy, were used for haemostasis. The conjunctiva around the defect was undermined and mobilised for primary closure. Specimens were placed on suture-pack polystyrene, to keep the tissue flat and oriented for the pathologist, and fixed in 10% neutral buffered formalin. All histopathological tests were done centrally and reported by a single pathologist. Combined gentamicin 0·3% and prednisolone acetate 1% eye drops were applied four times per day for 3–4 weeks after surgery. Patients were reviewed after about 4 weeks to confirm wound healing and for recruitment into the trial.

malin. All histopathological tests were done centrally and reported by a single pathologist. Combined gentamicin 0·3% and prednisolone acetate 1% eye drops were applied four times per day for 3–4 weeks after surgery. Patients were reviewed after about 4 weeks to confirm wound healing and for recruitment into the trial. Participants were asked to self-administer one drop of their allocated medication four times a day to the affected eye for 4 weeks. Each participant was given a 28 day medication diary to monitor treatment. They were asked to record each dose taken or missed. The record card had a similar diary for adverse effects (pain or burning sensation, excessive tears, and redness).

their allocated medication four times a day to the affected eye for 4 weeks. Each participant was given a 28 day medication diary to monitor treatment. They were asked to record each dose taken or missed. The record card had a similar diary for adverse effects (pain or burning sensation, excessive tears, and redness). Participants underwent a detailed ophthalmic examination with a slit-lamp biomicroscope before surgery and at about 1 month after surgery. After enrolment, follow-up visits were scheduled for 1 month, 3 months, 6 months, and 12 months after randomisation. Participants were telephoned 1 week before their appointments to remind them. Individuals who missed follow-up visits were contacted by telephone. At each follow-up visit, a symptom history was taken and a detailed ophthalmic examination was done for evidence of recurrent disease. On each examination, high-resolution digital images of the surface of the eyes were taken. In addition, we assessed whether the lacrimal drainage system was blocked using the dye disappearance test. Fluorescein dye was applied in both eyes in the inferior conjunctival fornix and the tear film observed with the cobalt blue light of the slit lamp after 5 min for clearance of the dye. The presence of dye after this time was considered positive, indicating a functional or anatomical blockage.

e dye disappearance test. Fluorescein dye was applied in both eyes in the inferior conjunctival fornix and the tear film observed with the cobalt blue light of the slit lamp after 5 min for clearance of the dye. The presence of dye after this time was considered positive, indicating a functional or anatomical blockage. HIV status was initially tested by ELISA using Vironostika antigen/antibody kit (Biomerieux, France) then later changed to rapid tests with Alere Determine HIV-1/2 Ag/Ab (Alere, USA) and Unigold (Trinity Biotech, USA). CD4 count was measured with FacsCount (Becton Dickinson, USA). Serum retinol concentration was quantified by high-performance liquid chromatography (SHIMADZU Prominence HCT2010, Japan). When an obvious regrowth was found, re-excision for treatment and histopathology was advised. If a small potentially suspicious change was observed, it was initially photographed, the size measured, and the participant examined more frequently than the scheduled study visits. If on subsequent visits the lesion had progressed, the lesion was re-excised and sent for histopathology; the date of recurrence was recorded as the first time the possible regrowing lesion was noticed. Outcomes The primary outcome was clinical recurrence of the lesion at any time during the first year of follow-up, confirmed by histological assessment where available. The secondary outcomes were time to recurrence, cofactors of recurrence, and adverse events.

When an obvious regrowth was found, re-excision for treatment and histopathology was advised. If a small potentially suspicious change was observed, it was initially photographed, the size measured, and the participant examined more frequently than the scheduled study visits. If on subsequent visits the lesion had progressed, the lesion was re-excised and sent for histopathology; the date of recurrence was recorded as the first time the possible regrowing lesion was noticed. Outcomes The primary outcome was clinical recurrence of the lesion at any time during the first year of follow-up, confirmed by histological assessment where available. The secondary outcomes were time to recurrence, cofactors of recurrence, and adverse events. Primary outcome events were assessed and confirmed centrally by an ophthalmologist, who had either directly examined all patients at Kenyatta National Hospital and Kikuyu Eye Unit, or reviewed the clinical images from Sabatia Eye Hospital and Kitale District Hospital, supported by histopathological results. For cases where histopathology was not available, mostly because the participant did not return for the repeat surgery, the images of recurrent lesions were reviewed by two consultant ophthalmologists experienced in OSSN in east Africa to confirm clinical recurrence. Adverse effects were monitored by reviewing the medication diary with the participant and asking about discomfort and tearing.

cipant did not return for the repeat surgery, the images of recurrent lesions were reviewed by two consultant ophthalmologists experienced in OSSN in east Africa to confirm clinical recurrence. Adverse effects were monitored by reviewing the medication diary with the participant and asking about discomfort and tearing. Statistical analysis Case series of the use of surgical excision with or without adjuvant fluorouracil treatment have reported recurrences in 3·2–43% of patients.7, 19, 22, 23 Assuming a recurrence of 20% in the control group and 10% in the treatment group, power of 80%, and a two-sided α of 5%, the target sample size was initially calculated to be 219 participants in each group. 1 year into the study, we noted that recruitment was slow but recurrences were more common overall than anticipated, so after review by the trial steering committee in discussion with the data safety and monitoring board, a pragmatic decision was made to revise the sample size assuming that 30% of patients in the placebo group and 5% in the treatment group would have disease recurrence. As such, a sample size of 43 participants in each group would provide 80% power to detect an absolute difference in recurrence rates of 25%.

as older than ours (mean age 67 years), predominantly male (75%), and probably not infected with HIV (no data were provided). Although the lesions regressed clinically on treatment with mitomycin, more than half of patients had persistent OSSN on repeat histological assessment of the lesion site 1 year after treatment. Overall, the use of topical fluorouracil was associated with transient side-effects: watery eye, discomfort when taking the drops, and eyelid skin inflammation. However, these were mostly mild and resolved after the completion of treatment. A transiently positive dye disappearance test indicates temporary reversible obstruction of the nasolacrimal duct, a known complication of fluorouracil treatment.18 The most significant adverse effect was the eyelid skin inflammation. This was reliably prevented by protecting the skin with a tissue while applying the drops to catch any overflow. Epiphora was reported by 10% of participants at 1 month in the placebo group and 61% reported occasional discomfort at 1 month. We think that these effects were not caused by the placebo, which was a bland lubricant, but rather related to having recently had, often extensive, excision surgery to remove a tumour. It is quite common for excision of conjunctival lesions, OSSN or other pathology such as a pterygium, to result in a degree of ocular surface inflammation and irritation that can persist for several weeks. Such effects are especially common in young people of African origin, who are more likely to scar and have inflammation than are older white patients.25 Overall, we think that these side-effects can be partly mitigated, do not usually represent a problem after cessation of treatment, and are outweighed by the benefit of the reduced tumour recurrence.

ring board, a pragmatic decision was made to revise the sample size assuming that 30% of patients in the placebo group and 5% in the treatment group would have disease recurrence. As such, a sample size of 43 participants in each group would provide 80% power to detect an absolute difference in recurrence rates of 25%. The analysis was predefined. We compared the two groups for balance in terms of predefined factors that could have a bearing on aetiology or recurrence: age, sex, smoking history, outdoor occupation, HIV status, CD4 count, vitamin A concentration, tumour size, prior excision, and histological grade.3, 5, 8, 9 The primary analysis of the primary outcome and the safety analysis were done by intention to treat. Data were managed in Microsoft Access, cleaned, and transferred into Stata (version 12.1) for analysis. We calculated the numbers of events, person-months, and rate of recurrence in each group. We estimated the effect size as the odds ratio (OR) for recurrence, estimated by logistic regression, with 95% CIs. We adjusted the crude OR for the seven surgeons as a random effect and for additional baseline factors that were greater in one group than the other.24 We analysed the effect of the intervention on time to recurrence with Kaplan-Meier survival curves, and we used Cox regression to estimate hazard ratios and 95% CIs, adjusting for substantial baseline imbalances. To assess whether survival was the same by treatment group, we used the log rank test. We report the risk of any adverse effects at any follow-up in the treated eye by group.

th Kaplan-Meier survival curves, and we used Cox regression to estimate hazard ratios and 95% CIs, adjusting for substantial baseline imbalances. To assess whether survival was the same by treatment group, we used the log rank test. We report the risk of any adverse effects at any follow-up in the treated eye by group. The trial is registered with the Pan-African Clinical Trials Registry, number PACTR201207000396219. Role of the funding source The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between August, 2012, and July 2014, 496 patients with conjunctival lesions had surgical excision followed by histopathological tests. 187 of these patients had OSSN and 309 had other pathological disease types. For all participants with OSSN, only one eye was involved. 89 (48%) of 187 patients were ineligible. Therefore, we enrolled 98 patients (figure 2). 49 patients were assigned to receive fluorouracil eye drops and 49 were assigned to placebo. 58 of 98 patients were enrolled at Kikuyu Eye Unit, 20 at Sabatia Eye Hospital, 12 at Kitale District Hospital, and eight at Kenyatta National Hospital. The final follow-up visits were scheduled for 12 months, although we included data from up to 13 months for late participants. Follow-up was completed in July, 2015. Four individuals without a recorded recurrence (two from each group) did not complete a full year of follow-up.

, and eight at Kenyatta National Hospital. The final follow-up visits were scheduled for 12 months, although we included data from up to 13 months for late participants. Follow-up was completed in July, 2015. Four individuals without a recorded recurrence (two from each group) did not complete a full year of follow-up. The main reasons for exclusion were: inability to return for regular follow-up (n=41), extensive disease requiring radical surgery (n=24), not returning at all after surgery (n=16), or returning more than 2 months after surgery (n=3; figure 2). There were no significant differences between the enrolled and non-enrolled patients in terms of age, sex, smoking, HIV infection, or stage of OSSN (excluding larger tumours; data not shown). The mean age of all participants was 41·0 years (SD 11·3) and most were female. Baseline characteristics were reasonably balanced between the two groups with the exception of past passive cigarette smoking and use of antiretroviral therapy, which were more common in the placebo group than in the fluorouracil group, and therefore were adjusted for in the primary analysis (table 1).

ost were female. Baseline characteristics were reasonably balanced between the two groups with the exception of past passive cigarette smoking and use of antiretroviral therapy, which were more common in the placebo group than in the fluorouracil group, and therefore were adjusted for in the primary analysis (table 1). By the end of follow-up, recurrent lesions had developed in 22 eyes of 22 participants. The four participants who did not complete a full year of follow-up were excluded from the primary analysis of clinical recurrence of the lesion by 1 year. Lesion recurrence was significantly less common in the fluorouracil group (five [11%] of 47 patients) than in the placebo group (17 [36%] of 47 patients; crude OR 0·21, 95% CI 0·07–0·63; p=0·01). This effect remained significant when adjusted for use of antiretroviral therapy and passive cigarette smoking (OR 0·23, 95% CI 0·07–0·75; p=0·02). The relative risk of recurrence was reduced by 70·7% and the absolute difference was 25·6%. Treatment with fluorouracil after surgery for four patients would therefore prevent an estimated one recurrence (number needed to treat 3·9, 95% CI 2·4–11·8).

py and passive cigarette smoking (OR 0·23, 95% CI 0·07–0·75; p=0·02). The relative risk of recurrence was reduced by 70·7% and the absolute difference was 25·6%. Treatment with fluorouracil after surgery for four patients would therefore prevent an estimated one recurrence (number needed to treat 3·9, 95% CI 2·4–11·8). 16 of 22 recurrent lesions underwent repeated surgical excision and recurrent OSSN was confirmed by histopathology in all cases. Re-excision was not done for six recurrent cases: four participants in the placebo group did not return for repeat surgery or further follow-up after the recurrence was noted, and two participants in the placebo group died before re-excision could be done (one from a presumed myocardial infarction and one from HIV-related complications). The images of these cases were reviewed by two ophthalmologists experienced in OSSN in east Africa. All were judged to be recurrent OSSN disease on clinical grounds, in the context of previous histologically confirmed OSSN.

ould be done (one from a presumed myocardial infarction and one from HIV-related complications). The images of these cases were reviewed by two ophthalmologists experienced in OSSN in east Africa. All were judged to be recurrent OSSN disease on clinical grounds, in the context of previous histologically confirmed OSSN. There was a significant difference in the recurrence rate between the two groups over the follow-up period (hazard ratio [HR] 0·24, 95% CI 0·09–0·66; p=0·01; figure 3), which changed slightly after adjusting for smoking and use of antiretroviral therapy (HR 0·32, 95% CI 0·11–0·95; p=0·04). The test for proportional hazards assumption showed that the assumption of proportionality was appropriate (p=0·59). Median tumour-free survival was 7·3 months (IQR 2·3–13·5) in the fluorouracil group and 4·8 months (3·0–7·6) in the placebo group but the difference was not statistically significant (p=0·23). Sensitivity analysis, assuming the four participants who did not complete 1 year of follow-up did not have a recurrence in the first year, made little difference to the results (crude OR 0·21, 95% CI 0·07–0·64, p=0·01; adjusted OR 0·25, 95% CI 0·08–0·79, p=0·02; HR 0·25, 95% CI 0·07–0·67, p=0·01; adjusted HR 0·33, 95% CI 0·11–0·97, p=0·04) and the number needed to treat remained at four (4·1, 95% CI 2·5–12·5).

did not have a recurrence in the first year, made little difference to the results (crude OR 0·21, 95% CI 0·07–0·64, p=0·01; adjusted OR 0·25, 95% CI 0·08–0·79, p=0·02; HR 0·25, 95% CI 0·07–0·67, p=0·01; adjusted HR 0·33, 95% CI 0·11–0·97, p=0·04) and the number needed to treat remained at four (4·1, 95% CI 2·5–12·5). Tumour size at baseline was a significant cofactor of recurrence (crude OR 1·27, 95% CI 1·04–1·54, p=0·02). Participants who had a recurrence had significantly larger mean tumour diameter (7·3 mm, SE 0·26) than did patients who did not have a recurrence (5·8 mm, SE 0·66; p=0·01). There was no effect modification by tumour size or rate of tumour growth before surgery, defined as tumour diameter divided by duration between noticing the growth and time of surgery, assuming a linear rate (likelihood ratio test p=0·49). The mean growth rate was 1·6 mm per month (SE 0·61) in patients who had a recurrence and 1·7 mm per month (SE 0·22) in patients who did not. Surgical margin involvement was not a significant cofactor for recurrence (crude OR 1·28, 95% CI 0·49–3·33, p=0·62) and nor was having invasive carcinoma rather than carcinoma in situ at baseline (crude OR 1·09, 95% CI 0·42–2·81, p=0·87).

ents who had a recurrence and 1·7 mm per month (SE 0·22) in patients who did not. Surgical margin involvement was not a significant cofactor for recurrence (crude OR 1·28, 95% CI 0·49–3·33, p=0·62) and nor was having invasive carcinoma rather than carcinoma in situ at baseline (crude OR 1·09, 95% CI 0·42–2·81, p=0·87). Adverse effects were more common in the fluorouracil group than in the placebo group, as shown by analysis of the time to first adverse event (p=0·005; table 2). Epiphora (watery eye) was more common in the fluorouracil group than in the placebo group and the dye disappearance test became transiently positive in six patients in the fluorouracil group at 1 month. None of the participants with epiphora had a positive dye disappearance test at 1 year. Ocular discomfort was more common in the fluorouracil group than in the placebo group at 1 month (p=0·004). In one (2%) patient, discomfort was sufficient to discontinue treatment after 3 weeks. After the 1 month visit, no participants reported ocular discomfort. Seven participants, all in the fluorouracil group, developed inflammation or irritation of the eyelid skin after about 2–3 weeks of treatment. This was attributed to overflow spillage of the eye drop onto the skin. All skin changes fully resolved within 1 month. We advised all further enrollees to apply drops while holding a piece of tissue paper against the lid; subsequently, no episodes of eyelid skin irritation occurred.

after about 2–3 weeks of treatment. This was attributed to overflow spillage of the eye drop onto the skin. All skin changes fully resolved within 1 month. We advised all further enrollees to apply drops while holding a piece of tissue paper against the lid; subsequently, no episodes of eyelid skin irritation occurred. Discussion We showed that 4 weeks of treatment with topical fluorouracil 1% after surgical excision substantially reduced the 1 year recurrence of OSSN tumours. The study was done in a region with a relatively high incidence of OSSN, which is often associated with HIV infection, and where patients often present late with advanced disease. Tumour recurrence has been a major problem in managing this disease. Most of our participants were relatively young, and women outnumbered men, the typical demographic pattern in Africa.3 The whole range of OSSN disease was represented, enabling us to draw a general conclusion about the effectiveness of the intervention.

mour recurrence has been a major problem in managing this disease. Most of our participants were relatively young, and women outnumbered men, the typical demographic pattern in Africa.3 The whole range of OSSN disease was represented, enabling us to draw a general conclusion about the effectiveness of the intervention. This study was the first randomised controlled trial of topical fluorouracil as adjunctive treatment for OSSN. However, our results are consistent with those of non-randomised case series of adjuvant fluorouracil, which have reported similarly low proportions of recurrence.14, 15, 16, 17, 18, 19, 20 Some investigators have reported the effectiveness of fluorouracil as primary treatment in presumed OSSN lesions without surgical excision and histopathological assessment.14 We chose to test fluorouracil in this setting because it is cheap and readily available with a history of use and wide acceptance for other types of ophthalmic surgery. Fluorouracil is on WHO's list of essential drugs. It does not require stringent storage conditions such as refrigeration. Therefore, the translation of this result into clinical practice, given the resource limitations of the Kenyan health system and other similar settings, is realistic. Because surgical cryotherapy is not routinely available, topical fluorouracil is therefore an alternative strategy to prevent recurrence.

on in young people of African origin, who are more likely to scar and have inflammation than are older white patients.25 Overall, we think that these side-effects can be partly mitigated, do not usually represent a problem after cessation of treatment, and are outweighed by the benefit of the reduced tumour recurrence. Adjunctive fluorouracil probably works through its effect on residual OSSN cells that are left after surgical excision. It interferes with DNA and RNA processes through several active metabolites.26 Fluorodeoxyuridine monophosphate inhibits thymidylate synthase, blocking thymidylate production, and thus DNA replication. Rapidly dividing neoplastic cells are much more vulnerable to thymidine depletion than are normal cells. Fluorodeoxyuridine triphosphate is misincorporated into DNA and fluorouridine triphosphate is misincorporated into RNA. These different metabolites disrupt crucial cellular mechanisms, triggering apoptosis. We were able to follow up participants to 1 year. This was attributable to two factors. First, we excluded people who said that they would be unlikely to return for follow-up. Second, the study team were careful to build and maintain good relationships with study participants, and actively communicated with those who missed follow-up visits.

up participants to 1 year. This was attributable to two factors. First, we excluded people who said that they would be unlikely to return for follow-up. Second, the study team were careful to build and maintain good relationships with study participants, and actively communicated with those who missed follow-up visits. Our study has several limitations. First, recruitment was slower than anticipated, resulting in a smaller study than originally anticipated. The initial study size was based on previously reported recurrence rates from several case series, with heterogeneous inclusion criteria, treatment regimens, and follow-up.4, 7, 19, 22, 23, 27, 28 However, the higher than expected recurrence rate and high retention enabled us to have good power with a smaller sample size. In common with other studies from Africa, the recurrence rate in both groups was higher than that reported in many case series from temperate countries. This difference is possibly because OSSN in this population, with a high proportion of patients who are HIV positive, is more aggressive and patients probably present later with more advanced disease. Second, not all the cases of clinical recurrence had re-excision and histopathological tests done. However, all suspected recurrences for which histopathological results were available were confirmed as recurrent OSSN, which suggests that our clinical judgment in this situation is highly concordant with the pathology.

Our study has several limitations. First, recruitment was slower than anticipated, resulting in a smaller study than originally anticipated. The initial study size was based on previously reported recurrence rates from several case series, with heterogeneous inclusion criteria, treatment regimens, and follow-up.4, 7, 19, 22, 23, 27, 28 However, the higher than expected recurrence rate and high retention enabled us to have good power with a smaller sample size. In common with other studies from Africa, the recurrence rate in both groups was higher than that reported in many case series from temperate countries. This difference is possibly because OSSN in this population, with a high proportion of patients who are HIV positive, is more aggressive and patients probably present later with more advanced disease. Second, not all the cases of clinical recurrence had re-excision and histopathological tests done. However, all suspected recurrences for which histopathological results were available were confirmed as recurrent OSSN, which suggests that our clinical judgment in this situation is highly concordant with the pathology. Third, we excluded a high proportion of potential participants. The most common reason for exclusion was that the patient was unlikely to return for follow-up. Excluding those who did not think they could return helped us achieve good follow-up among those who were enrolled in the trial. However, there was no systematic difference between participants and excluded patients in terms of age, sex, HIV status, smoking status, or OSSN grade (when those with large lesions requiring alternative radical surgery were excluded). This finding suggests that our results can be generalised. The participants who were lost to follow-up attended at least the first visit after randomisation and were recurrence-free at that point. The challenge of ensuring high follow-up rates in Kenyans with HIV has been reported previously.29

surgery were excluded). This finding suggests that our results can be generalised. The participants who were lost to follow-up attended at least the first visit after randomisation and were recurrence-free at that point. The challenge of ensuring high follow-up rates in Kenyans with HIV has been reported previously.29 Fourth, we excluded individuals with very large tumours that required either enucleation or exenteration. This exclusion could reduce the generalisability of the findings. However, such patients are not suitable for less radical surgery and topical chemotherapy, as the tumour is already invading the deeper tissues of the orbit. Finally, there were some differences in adverse events by group, which could have led to unmasking. However, we think that this is unlikely: discomfort was common and similar in each group at 1 month, and eyelid inflammation, positive dye disappearance test, and epiphora after 1 month were uncommon. In conclusion, 4 weeks of topical fluorouracil 1% after surgical excision of OSSN substantially reduced the 1 year recurrence of tumours. The treatment is safe, generally well tolerated, and easy to use. Fluorouracil is widely available, affordable, and easy to formulate into eye drops. It is suitable in settings without cryotherapy. Fluorouracil eye drops are an effective and realistic intervention to improve outcomes for people with OSSN. Supplementary Material Supplementary appendix

In conclusion, 4 weeks of topical fluorouracil 1% after surgical excision of OSSN substantially reduced the 1 year recurrence of tumours. The treatment is safe, generally well tolerated, and easy to use. Fluorouracil is widely available, affordable, and easy to formulate into eye drops. It is suitable in settings without cryotherapy. Fluorouracil eye drops are an effective and realistic intervention to improve outcomes for people with OSSN. Supplementary Material Supplementary appendix Acknowledgments SG received funding from the British Council for Prevention of Blindness fellowship programme. MJB is supported by the Wellcome Trust (grant number 098481/Z/12/Z). Contributors SG, MSS, HAW, and MJB designed the study and interpreted the data. SG did the literature search. SG, EM, JK, AMZ, HR, EO, JW, RM, JM, and TO collected data. SG, HAW, and MJB analysed data and obtained funding. SG wrote the first draft of the Article, all authors revised it. JK, AMZ, HR, EO, JW, RM, and JM provided administrative, technical, and material support. HAW and MJB supervised the study. Declaration of interests We declare no competing interests. Figure 1 Ocular surface squamous neoplasia Moderately differentiated conjunctival squamous cell carcinoma, (A) moderate size, (B) large lesion involving the cornea, limbus, and extending to the fornix. Fornix involvement is often associated with orbital spread. Figure 2 Trial profile OSSN=ocular surface squamous neoplasia. Figure 3 Kaplan-Meier analysis of time to recurrence Table 1 Baseline characteristics

Moderately differentiated conjunctival squamous cell carcinoma, (A) moderate size, (B) large lesion involving the cornea, limbus, and extending to the fornix. Fornix involvement is often associated with orbital spread. Figure 2 Trial profile OSSN=ocular surface squamous neoplasia. Figure 3 Kaplan-Meier analysis of time to recurrence Table 1 Baseline characteristics Fluorouracil group (n=49) Placebo group (n=49) Sex Male 17 (35%) 14 (29%) Female 32 (65%) 35 (71%) Age (years) 39·1 (9·2) 42·9 (13·0) Marital status* Single 6 (12%) 11 (22%) Married 32 (65%) 28 (58%) Divorced or separated 4 (8%) 2 (4%) Widowed 7 (14%) 7 (14%) Formal education (years)* Tertiary (>12) 5 (10%) 4 (8%) Secondary completed (12) 18 (37%) 20 (41%) Some secondary (8–12) 4 (8%) 6 (12%) Primary completed (8) 13 (27%) 11 (22%) Some primary (<8) 6 (12%) 4 (8%) None 3 (6%) 3 (6%) Past cigarette smoking* No 41 (84%) 32 (67%) Yes 5 (10%) 4 (8%) Passive (spouse or partner smokes) 3 (6%) 12 (25%) Current cigarette smoking† No 44 (90%) 41 (87%) Yes 3 (6%) 0 (0%) Passive (spouse or partner smokes) 2 (4%) 6 (13%) Cigarettes smoked daily 9 (6) 13 (13) Years of cigarette smoking 10·7 (6·8) 16·2 (7·4) Location of current occupation* Indoors 20 (41%) 19 (40%) Outdoors 29 (59%) 29 (60%) Wears hat or cap outdoors† 8 (16%) 6 (13%) Wears sunglasses outdoors† 4 (8%) 4 (9%) HIV infection‡ 29 (63%) 31 (71%) Use of antiretroviral therapy§ 10 (22%) 19 (42%) CD4 count (cells per μL)¶ 444 (370) 460 (421) Serum retinol concentration (μg/L)‖ 489 (157) 529 (215) Vitamin A deficiency (serum retinol <300 μg/L)‖ 3 (6%) 3 (6%) Tumour diameter (mm) 5·9 (2·6) 6·3 (2·4) Prior excision 8 (16%) 9 (19%) Histological grading of tumours CIN 1 4 (8%) 4 (8%) CIN 2 13 (27%) 8 (16%) CIN 3 11 (22%) 14 (29%) Carcinoma in situ 0 (0%) 1 (2%) Poorly differentiated squamous cell carcinoma 1 (2%) 1 (2%) Moderately differentiated squamous cell carcinoma 17 (35%) 18 (37%) Well differentiated squamous cell carcinoma 3 (6%) 3 (6%) Surgical margin involvement 21 (43%) 19 (39%) Stage of OSSN** T1N0M0 15 (31%) 9 (18%) T2N0M0 10 (20%) 9 (18%) T3N0M0 23 (47%) 31 (63%) T3N1M0 1 (2%) 0 (0%) Data are n (%) or mean (SD).

%) 1 (2%) Moderately differentiated squamous cell carcinoma 17 (35%) 18 (37%) Well differentiated squamous cell carcinoma 3 (6%) 3 (6%) Surgical margin involvement 21 (43%) 19 (39%) Stage of OSSN** T1N0M0 15 (31%) 9 (18%) T2N0M0 10 (20%) 9 (18%) T3N0M0 23 (47%) 31 (63%) T3N1M0 1 (2%) 0 (0%) Data are n (%) or mean (SD). * Data missing for one participant in the placebo group. † Data missing for two participants in the placebo group. ‡ Data missing for three participants in the fluorouracil group and five in the placebo group. § Data missing for three participants in the fluorouracil group and four in the placebo group. ¶ Data missing for eight participants in the fluorouracil group and 11 in the placebo group. ‖ Data missing for ten participants in the fluorouracil group and 13 in the placebo group. ** As per the American Joint Committee on Cancer. CIN=cervical intra-epithelial neoplasia. OSSN=ocular surface squamous neoplasia. Table 2 Adverse events

¶ Data missing for eight participants in the fluorouracil group and 11 in the placebo group. ‖ Data missing for ten participants in the fluorouracil group and 13 in the placebo group. ** As per the American Joint Committee on Cancer. CIN=cervical intra-epithelial neoplasia. OSSN=ocular surface squamous neoplasia. Table 2 Adverse events Fluorouracil group (n=49) Placebo group (n=49) p value Epiphora .. .. .. 1 month 24 (49%) 5 (10%) <0·001 3 months* 3 (6%) 2 (4%) 0·66 6 months† 1 (2%) 0 (0%) 0·32 12 months† 4 (8%) 0 (0%) 0·04 Positive dye disappearance test .. .. .. At baseline 0 (0%) 1 (2%) 0·32 1 month 6 (12%) 1 (2%) 0·05 3 months* 1 (2%) 0 (0%) 0·32 6 months† 0 (0%) 0 (0%) - 12 months† 0 (0%) 0 (0%) - Discomfort in the treated eye at 1 month‡ .. .. 0·004 Occasional discomfort 21 (43%) 30 (61%) .. Discomfort for <5 min 12 (24%) 3 (6%) .. Discomfort for ≥5 min 4 (8%) 2 (4%) .. Discomfort making treatment difficult 6 (12%) 1 (2%) .. Eyelid inflammation at 1 month 7 (14%) 0 (0%) <0·001 Any adverse event .. .. 0·005§ 1 month 34 (69%) 19 (39%) .. 3 months 10 (20%) 9 (18%) .. 6 months 5 (10%) 1 (2%) .. 12 months 7 (14%) 6 (12%) .. * 48 patients in the fluorouracil group and 47 in the placebo group because of loss to follow-up. † 47 participants in each group because of loss to follow-up. ‡ No participant reported any discomfort at 3 months, 6 months, or 12 months. § Computed with the stratified log-rank test for equality of survivor functions.

Introduction The seasonal patterns, incidence, and severity of influenza virus infection are poorly defined in many tropical regions.1 In Bangladesh, over a decade of surveillance data show that influenza virus circulates during most months of the year and that infection is a frequent cause of febrile illness and lower respiratory tract infections in young children.2, 3, 4, 5, 6, 7 Furthermore, in urban Kamalapur, which has a background clinical pneumonia incidence of 500 cases per 1000 child-years, community-based surveillance identified that 10% of children younger than 5 years with clinical pneumonia are positive for influenza virus.4 Thus, prevention of influenza illness could have a substantial effect on childhood morbidity. Live attenuated influenza vaccines (LAIVs) are attractive for use in young children because they may be delivered intranasally and have good efficacy. In a WHO-sponsored technology transfer programme, several manufacturers are developing reassortant LAIVs based on A/Leningrad/17 and B/USSR/60 master donor viruses. This initiative could provide affordable supplies of influenza vaccine in Bangladesh and other low-resource countries. In 2012, we did a phase 2 study involving 300 children aged 24–59 months in urban Bangladesh that supported the safety of a Russian-backbone LAIV.8 We have now done a clinical efficacy trial to investigate the benefit of a single-dose Russian-backbone LAIV in children in urban and rural sites in Bangladesh.

countries. In 2012, we did a phase 2 study involving 300 children aged 24–59 months in urban Bangladesh that supported the safety of a Russian-backbone LAIV.8 We have now done a clinical efficacy trial to investigate the benefit of a single-dose Russian-backbone LAIV in children in urban and rural sites in Bangladesh. Methods Study design This was a two-site, randomised, double-blind, placebo-controlled, parallel-group clinical trial. The study was done at two demographic surveillance sites of the International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh: one in urban Kamalapur and one in rural Matlab. The study was approved by the ethics review committees of the International Centre for Diarrhoeal Disease Research, and by the Western Institutional Review Board, Puyallup, WA, USA. Participant safety was overseen by an independent international data safety monitoring board that was convened by the study's sponsor, PATH, Seattle, WA, USA, and a local data safety monitoring board convened by the International Centre for Diarrhoeal Disease Research. The study complied with the principles of the Declaration of Helsinki, and was done in compliance with Good Clinical Practice guidelines. Research in context Evidence before this study

Methods Study design This was a two-site, randomised, double-blind, placebo-controlled, parallel-group clinical trial. The study was done at two demographic surveillance sites of the International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh: one in urban Kamalapur and one in rural Matlab. The study was approved by the ethics review committees of the International Centre for Diarrhoeal Disease Research, and by the Western Institutional Review Board, Puyallup, WA, USA. Participant safety was overseen by an independent international data safety monitoring board that was convened by the study's sponsor, PATH, Seattle, WA, USA, and a local data safety monitoring board convened by the International Centre for Diarrhoeal Disease Research. The study complied with the principles of the Declaration of Helsinki, and was done in compliance with Good Clinical Practice guidelines. Research in context Evidence before this study Cold-adapted, temperature-sensitive live attenuated influenza vaccines (LAIVs) have been developed in the USA and Russia. These vaccines contain viruses produced by reassorting master donor viruses (A/Ann Arbor/6/60 and B/Ann Arbor/1/66 or A/Leningrad/134/17/57 and B/USSR/69/60, respectively) with viruses recommended by WHO or the US Public Health Service. Decisions are based on the expected circulation in the forthcoming northern or southern hemisphere influenza season. LAIVs based on Ann-Arbor-derived strains are approved for use in children ages 2 years and older, and those based on the Russian-derived strains are approved for single-dose administration to children aged 3 years and older. Through a WHO agreement, manufacturers in several developing countries have access to Russian-derived vaccine strains for production of their own LAIVs. The Serum Institute of India, Pune, India, is one such manufacturer. Its LAIV is approved for use in children aged 2 years and upwards. Policy decisions and sustainable use of LAIVs in developing countries will depend on the generation of data that show their efficacy in representative populations. We searched PubMed from Jan 1, 1980, to Jan 1, 2016, for efficacy trials assessing protection achieved with the use of LAIVs in children, using the search terms, “human influenza”, “vaccines, attenuated”, and “children”. LAIVs have been studied primarily in developed countries in Europe and Asia and in the USA. Use of the Ann-Arbor-derived LAIV in children younger than 72 months was studied, with culture confirmation of illness, in several trials from 1996 to 2005 in these regions plus one study in South Africa in children aged from 6 up to 36 months. These studies were done under Good Clinical Practice guidelines. In nearly all studies, protection was significant. Before those studies, five studies of Russian-derived LAIVs had been done in the Soviet Union and Cuba, primarily in 1986–91. These included nearly 28 000 children aged 3–6 years. These studies used serological confirmation of illness and were done before Good Clinical Practice standards were introduced for trials.

nt. Before those studies, five studies of Russian-derived LAIVs had been done in the Soviet Union and Cuba, primarily in 1986–91. These included nearly 28 000 children aged 3–6 years. These studies used serological confirmation of illness and were done before Good Clinical Practice standards were introduced for trials. Moreover, except for one study in Senegal, no efficacy trials of Russian-derived LAIVs had been done in developing countries. We found one case-control study assessing the clinical protection of a single-dose monovalent Serum Institute of India LAIV containing vaccine virus antigenically similar to pandemic A/H1N1 (2009). Added value of this study Lower respiratory tract disease caused by viral infections is common among young children in Bangladesh. LAIVs could potentially reduce some of this morbidity by preventing primary influenza disease, its complications, or both. The feasibility of influenza immunisation in Bangladesh would be increased if a single-dose LAIV provided protection. In our randomised trial of LAIV done exclusively among children in Bangladesh, we showed significant protection against PCR-confirmed influenza illness in children aged 2–4 years. Implications of all the available evidence

Lower respiratory tract disease caused by viral infections is common among young children in Bangladesh. LAIVs could potentially reduce some of this morbidity by preventing primary influenza disease, its complications, or both. The feasibility of influenza immunisation in Bangladesh would be increased if a single-dose LAIV provided protection. In our randomised trial of LAIV done exclusively among children in Bangladesh, we showed significant protection against PCR-confirmed influenza illness in children aged 2–4 years. Implications of all the available evidence Russian-backbone LAIV provided laboratory-confirmed protection in a low-resource setting, in rural and urban sites. This finding suggests that single-dose LAIVs could be the optimum choice for protection against influenza virus in paediatric populations in Asia. We are uncertain why the single-dose LAIV was protective in Bangladesh but not in Senegal, where no efficacy was seen, but this difference emphasises the variability in responses to influenza vaccine and the need for testing in multiple populations and, ideally, in multiple seasons.

s in paediatric populations in Asia. We are uncertain why the single-dose LAIV was protective in Bangladesh but not in Senegal, where no efficacy was seen, but this difference emphasises the variability in responses to influenza vaccine and the need for testing in multiple populations and, ideally, in multiple seasons. Participants Healthy children aged 2–4 years who lived in either of the two demographic surveillance areas were eligible for enrolment if a parent or legal guardian provided written informed consent and the family was not expecting to migrate out of the area during the study. In Kamalapur, we approached parents of eligible children during weekly field visits in households previously selected by cluster randomisation for active surveillance. In Matlab, we approached parents of eligible children as determined by our demographic surveillance system by visiting their households. Exclusion criteria were the same as in the previous phase 2 study,8 and included serious, active medical disorders and having previously received any influenza vaccine. The complete list of eligibility criteria is provided in the appendix.

determined by our demographic surveillance system by visiting their households. Exclusion criteria were the same as in the previous phase 2 study,8 and included serious, active medical disorders and having previously received any influenza vaccine. The complete list of eligibility criteria is provided in the appendix. Randomisation and masking The random allocation sequence was computer generated by PATH staff not involved with the trial, using a ratio for LAIV and placebo of 2:1 and block sizes of three. The sequence was delivered to the Serum Institute of India, Pune, India, where it was used to label the vaccine and placebo syringes, which were identical in appearance except for the allocation numbers. The labelled syringes of vaccine and placebo were shipped to Bangladesh for use. Study vaccine and placebo The syringes were used to fill spray devices with vaccine or placebo. Each dose was 0·5 mL, with half delivered into each nostril. The LAIV was 2012–13 Northern Hemisphere formulation (Nasovac-S, Serum Institute of India, lot 167E2002) and contained A/California/7/2009 (H1N1)-like, A/Victoria/361/2011 (H3N2)-like, and B/Wisconsin/1/2010 (Yamagata lineage)-like reassortants. The placebo was the vaccine vehicle without the virus components (Serum Institute of India, lot E9001PCB). The sprays were administered to each child according to the manufacturer's recommendations.9

fornia/7/2009 (H1N1)-like, A/Victoria/361/2011 (H3N2)-like, and B/Wisconsin/1/2010 (Yamagata lineage)-like reassortants. The placebo was the vaccine vehicle without the virus components (Serum Institute of India, lot E9001PCB). The sprays were administered to each child according to the manufacturer's recommendations.9 Procedures Participants were given one dose of study vaccine or placebo in the clinic and asked to remain for 30 min after administration, under the supervision of trained study nurses. Field workers did daily home visits up to day 4 after vaccination to monitor solicited events, unsolicited events, protocol-defined wheezing illness (PDWI), and serious adverse events (SAEs). If children presented at the clinic on days 4–7 with symptoms, these were included in solicited events. Thereafter children were monitored weekly at home by trained field workers. All children identified through home-visit surveillance as meeting protocol-defined criteria for physician assessment (signs of illness) were assessed by a study physician in the clinic according to standardised criteria. If children met protocol-defined criteria for specimen collection and presented within 7 days of illness onset, a nasopharyngeal wash specimen was collected for testing by real-time RT-PCR for evidence of influenza virus infection, according to the WHO laboratory protocol.10

e clinic according to standardised criteria. If children met protocol-defined criteria for specimen collection and presented within 7 days of illness onset, a nasopharyngeal wash specimen was collected for testing by real-time RT-PCR for evidence of influenza virus infection, according to the WHO laboratory protocol.10 The physician assessment criteria included the presence of at least one major or two minor signs. Major signs were fever (axillary temperature ≥38·0°C), tachypnoea (≥40 breaths per min), danger signs (chest wall indrawing, lethargy, cyanosis, inability to drink, convulsions), difficulty breathing, noisy breathing, ear pain, or ear discharge. Minor signs were subjective fever (feverishness), cough, rhinorrhoea, sore throat, myalgia or arthralgia, chills, headache, irritability or decreased activity, or vomiting. The criteria for specimen collection were two or more of axillary temperature 37·5°C or higher, cough, sore throat, and runny nose or nasal congestion present on the same day, or any of fever (axillary temperature ≥38·0°C), upper respiratory illness (axillary temperature ≥37·5°C with cough and rhinorrhoea), pneumonia, acute otitis media, meningitis, or sepsis, as diagnosed by the physician.4, 11

37·5°C or higher, cough, sore throat, and runny nose or nasal congestion present on the same day, or any of fever (axillary temperature ≥38·0°C), upper respiratory illness (axillary temperature ≥37·5°C with cough and rhinorrhoea), pneumonia, acute otitis media, meningitis, or sepsis, as diagnosed by the physician.4, 11 Assays were done at the International Centre for Diarrhoeal Disease Research, Bangladesh, as previously described.12, 13 Antigenic characterisation of positive samples was done at the virology laboratory at the International Centre for Diarrhoeal Disease Research.13 Genotyping was done to distinguish vaccine virus from wild-type virus at the Department of Virology, Institute of Experimental Medicine, Saint Petersburg, Russia, using specimens that were PCR positive within 14 days of vaccination.14 Outcomes The primary efficacy endpoint was symptomatic, laboratory-confirmed influenza virus infection with vaccine-matched strains up to December, 2013. The secondary efficacy endpoint was symptomatic, laboratory-confirmed influenza virus infection with any influenza virus strain.

Assays were done at the International Centre for Diarrhoeal Disease Research, Bangladesh, as previously described.12, 13 Antigenic characterisation of positive samples was done at the virology laboratory at the International Centre for Diarrhoeal Disease Research.13 Genotyping was done to distinguish vaccine virus from wild-type virus at the Department of Virology, Institute of Experimental Medicine, Saint Petersburg, Russia, using specimens that were PCR positive within 14 days of vaccination.14 Outcomes The primary efficacy endpoint was symptomatic, laboratory-confirmed influenza virus infection with vaccine-matched strains up to December, 2013. The secondary efficacy endpoint was symptomatic, laboratory-confirmed influenza virus infection with any influenza virus strain. Safety endpoints included immediate reactions occurring within 30 min of taking the vaccine, solicited reactions (nasal congestion, runny nose, ear pain, cough, sore throat, headache, fever, tachypnoea, muscle or joint pain, chills, irritability or decreased activity, and vomiting) and unsolicited adverse events occurring in the first 4 days after vaccination, as assessed by daily home visits, and up to day 7, as captured in the clinic. Severity of adverse events was graded as mild, moderate, severe, and life threatening. PDWI was defined as an illness meeting the physician assessment criteria and characterised by a long, high-pitched whistling or musical sound on expiration heard by auscultation over the lung fields. Severity of PDWI was graded by study physicians as mild (wheezing illness without other findings associated with moderate or greater severity disease), moderate (nasal flaring, chest in-drawing, or pulse oximetry ≥90% to <95%), severe (dyspnoea at rest causing inability to perform usual activities or pulse oximetry <90%), or life threatening. This definition was designed to be similar to the definitions of medically important wheezing used in previous trials8, 15 and to exclude incidental wheezing illness without other clinically important signs or symptoms.

at rest causing inability to perform usual activities or pulse oximetry <90%), or life threatening. This definition was designed to be similar to the definitions of medically important wheezing used in previous trials8, 15 and to exclude incidental wheezing illness without other clinically important signs or symptoms. Statistical analyses Assuming 60% efficacy, we calculated that 57 symptomatic, laboratory-confirmed influenza virus infections would be needed to test the hypothesis that LAIV efficacy was greater than 0 with a one-sided type I error of less than 2·5% and power of at least 90% (exact type I error and power 1·9% and 91·5%, respectively). On the basis of this number and assuming 6% influenza illness incidence in the placebo group and that 90% of children would be assessable, we estimated that we would need a total sample size of 1761 enrolled and vaccinated children.

and power of at least 90% (exact type I error and power 1·9% and 91·5%, respectively). On the basis of this number and assuming 6% influenza illness incidence in the placebo group and that 90% of children would be assessable, we estimated that we would need a total sample size of 1761 enrolled and vaccinated children. The primary objective was to estimate the efficacy of LAIV to reduce the incidence of symptomatic, laboratory-confirmed influenza virus infection with vaccine-matched strains, compared with placebo. Vaccine efficacy, expressed as a percentage, was defined as 1 minus the relative rate of influenza in the LAIV group compared with that in the placebo group. Efficacy with 95% CIs was computed with a binomial distribution of LAIV cases. We used Fisher's exact test to obtain two-sided p values for the test of the null hypothesis of zero vaccine efficacy. Primary efficacy analyses and summaries were done on a per-protocol basis. The per-protocol analysis set included all children who met the inclusion criteria, were randomised, and received one dose of study vaccine or placebo, and who remained in the study area for at least 8 days after vaccination. Analyses were based on the first case of influenza occurring from study day 8 onwards until Dec 31, 2013. Supportive analyses were done in the total vaccinated cohort (ie, all children who were randomised and received one dose of vaccine or placebo, irrespective of how long they stayed in the study area). All analyses excluded samples identified as containing vaccine virus.

ring from study day 8 onwards until Dec 31, 2013. Supportive analyses were done in the total vaccinated cohort (ie, all children who were randomised and received one dose of vaccine or placebo, irrespective of how long they stayed in the study area). All analyses excluded samples identified as containing vaccine virus. Safety was assessed in the total vaccinated cohort and was described as the proportion (95% CI) of children who had any reaction or adverse event. Differences between groups were calculated, including two-sided p values, with Fisher's exact test. Statistical analyses were done with SAS version 9.3. This study is registered with ClinicalTrials.gov, number NCT01797029. Role of the funding source The funder had no role in the study, data collection, data analysis, data interpretation, or the writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Study population Of 1811 children enrolled, 1761 received vaccine or placebo between Feb 27 and April 9, 2013 (total vaccinated analysis set), of whom 1637 were followed up for study outcomes until Dec 31, 2013 (figure 1). No participants were lost to follow-up before day 8. 1637 (92·9%) of study participants completed home follow-up visits to December, 2013, 1081 (92·0%) of those who received LAIV and 556 (94·7%) of those who received placebo (p=0·048).

et), of whom 1637 were followed up for study outcomes until Dec 31, 2013 (figure 1). No participants were lost to follow-up before day 8. 1637 (92·9%) of study participants completed home follow-up visits to December, 2013, 1081 (92·0%) of those who received LAIV and 556 (94·7%) of those who received placebo (p=0·048). Baseline demographic characteristics and medical history were similar in the two study groups (table 1), although differences were noted between the urban Kamalapur site and the rural Matlab site. Nearly twice as many children had severe stunting in Kamalapur as in Matlab, and previous medical treatment for asthma or wheezing illness was 31·3% in Kamalapur, including 6·5% who were admitted to hospital, compared with no children in Matlab.

ces were noted between the urban Kamalapur site and the rural Matlab site. Nearly twice as many children had severe stunting in Kamalapur as in Matlab, and previous medical treatment for asthma or wheezing illness was 31·3% in Kamalapur, including 6·5% who were admitted to hospital, compared with no children in Matlab. Influenza circulated in the study area from February to November, 2013 (figure 2). Symptomatic, laboratory-confirmed influenza illness due to vaccine-matched influenza viruses from 8 days after vaccination onwards was seen in 79 (6·7%) participants in the LAIV group and 93 (15·8%) participants in the placebo group, giving a vaccine efficacy of 57·5% (95% CI 43·6–68·0; table 2). In the strain-specific analysis in the per-protocol population (table 2), efficacy was 50–60% for the H3N2 and H1N1 strains. Only three children had vaccine-matched B/Yamagata strains isolated during the study. The attack rate for the mismatched B/Victoria lineage was statistically similar in the two study groups and, therefore, there was no vaccine efficacy against mismatched B lineage viruses. When all strains of influenza virus were analysed in the per-protocol population, including the mismatched B strains, vaccine efficacy was 41·0% (table 2). Results were similar in the total vaccinated cohort (appendix). When vaccine efficacy was assessed by study site, the rural Matlab site had higher influenza illness attack rates and higher vaccine efficacy than the urban Kamalapur site (table 2). In a post-hoc analysis, children with a history of asthma or wheezing illness in Kamalapur had a lower point estimate of efficacy than children without such a history (table 2).

udy site, the rural Matlab site had higher influenza illness attack rates and higher vaccine efficacy than the urban Kamalapur site (table 2). In a post-hoc analysis, children with a history of asthma or wheezing illness in Kamalapur had a lower point estimate of efficacy than children without such a history (table 2). No immediate reactions were seen after vaccine or placebo were administered. In the 7 days after receiving the study drug or placebo, the most common local and systemic events were runny nose, cough, and tachypnoea (table 3). Nearly all reactions were mild and similar proportions of children were affected in the vaccine and placebo groups. Among children with any history of asthma or wheezing illness, the frequency of PDWI was very similar in the LAIV and placebo groups, and among those without such a history, more in the placebo group than in the LAIV group had PDWI (table 3). No proportions of children with PDWI were significantly higher in the LAIV than in the placebo group when assessments were done by age group, site, site and age group, or history of asthma or wheezing illness at baseline. Two children died in the vaccine group during the study, both owing to drowning. Neither death was deemed to be related to vaccination.

were significantly higher in the LAIV than in the placebo group when assessments were done by age group, site, site and age group, or history of asthma or wheezing illness at baseline. Two children died in the vaccine group during the study, both owing to drowning. Neither death was deemed to be related to vaccination. Discussion This prospective trial of a single-dose Russian-backbone LAIV with intranasal delivery showed vaccine efficacy in children in Bangladesh with a good safety profile. These findings add important information to the data available on this vaccine because the original licensing studies done in Russia took place before PCR was used to confirm influenza virus infection, and licensing in India did not require laboratory-confirmed prospective efficacy studies. LAIV was efficacious against infection with circulating influenza H1N1pdm09 and H3N2 viruses, which are antigenically similar to the vaccine components. LAIV was not efficacious against infection with circulating B strains, as most were mismatched to the lineage in the vaccine.

y-confirmed prospective efficacy studies. LAIV was efficacious against infection with circulating influenza H1N1pdm09 and H3N2 viruses, which are antigenically similar to the vaccine components. LAIV was not efficacious against infection with circulating B strains, as most were mismatched to the lineage in the vaccine. When differences in study year, and thus vaccine and circulating strains, are taken into account, our primary efficacy against vaccine-matched strains of 57·5% (95% CI 43·6–68·0) is very close to the 59·9% (31·1–77·4) efficacy against vaccine-matched strains reported for a single-dose Ann-Arbor-based LAIV given to children aged 2–3 years in southeast Asia.16 Among children aged 2 years in South Africa and South America, an Ann-Arbor-backbone LAIV had 71·5% (95% CI 52·9–83·4) and 81·8% (66·8–90·8) efficacy against laboratory-confirmed influenza illness after one or two doses, respectively.17 As the vaccine used in our study is licensed for single-dose administration,10 we did not include a two-dose group. The increased cost and logistical challenges of a second dose would need to be weighed against any incremental benefit in low-resource settings. If the Russian-backbone vaccine were integrated into vaccine programmes in most low-income and middle-income countries, use of one dose would be likely to have substantial logistical and cost advantages over a two-dose regimen. Future studies could explore the benefit of a second dose of vaccine in children receiving influenza vaccine for the first time.

integrated into vaccine programmes in most low-income and middle-income countries, use of one dose would be likely to have substantial logistical and cost advantages over a two-dose regimen. Future studies could explore the benefit of a second dose of vaccine in children receiving influenza vaccine for the first time. The proportions of children in our study who received LAIV and had influenza illness were similar in Kamalapur (7·0%) and Matlab (6·1%), but the attack rate was higher in the Matlab placebo group than in the Kamalapur placebo group (21·9% vs 13·0%, table 2). Although the 95% CIs overlapped, vaccine efficacy for vaccine-matched strains was higher in Matlab than in Kamalapur in the per-protocol analysis set (72·0%, 95% CI 54·7–82·6 vs 46·2%, 23·0–62·4), possibly because of a lower force of infection in the rural setting than in the densely populated urban setting. Furthermore, the enrolled population in Kamalapur might have been less healthy than those in Matlab, since higher proportions of children had severe stunting and asthma or wheezing illness.

6·2%, 23·0–62·4), possibly because of a lower force of infection in the rural setting than in the densely populated urban setting. Furthermore, the enrolled population in Kamalapur might have been less healthy than those in Matlab, since higher proportions of children had severe stunting and asthma or wheezing illness. We chose our vaccination period to ensure children were vaccinated before the onset of peak influenza season, which is typically between April and September.2, 4 The efficacy we showed against the A(H1N1)pdm09 viruses is important because several observational studies of the Ann-Arbor-backbone LAIV in the USA in the 2013–14 and 2015–16 seasons and the Russian-backbone LAIV in Senegal showed no such efficacy.18, 19, 20 The same vaccine lot was used for this and the Senegal trial, and the storage conditions at both sites were well monitored and well maintained. Thus, a vaccine-specific cause for the difference in results seems unlikely. Influenza vaccine is not widely used in Bangladesh. Whether the lack of previous exposure to influenza vaccine, use of a different vaccine formulation than had been used previously, or both, affected influenza A(H1N1) efficacy in our Bangladesh population differently from that in the same period in the USA is unclear. It is likely, however, that excluding children who had previously been vaccinated ensured generalisability of our findings to the wider Bangladesh population. Unfortunately, the generalisability of our findings to other settings is uncertain.

sh population differently from that in the same period in the USA is unclear. It is likely, however, that excluding children who had previously been vaccinated ensured generalisability of our findings to the wider Bangladesh population. Unfortunately, the generalisability of our findings to other settings is uncertain. The major limitation of our study is that it focused on clinical efficacy and did not include immunogenicity measurements. At the time of this study, the preferential use of LAIVs was being considered in several countries due to superior performance in young children in randomised trials over inactivated vaccines. Our study was done before the studies in Senegal and the USA showed lack of efficacy for A(H1N1)pdm09 strains. In retrospect, information on the immunogenicity and replicative ability of the vaccine used in this study might have informed our interpretation of the variation in results between settings that subsequently emerged. However, the overall usefulness of immunogenicity data in this context is uncertain because of the difficulty in identifying protective immunological correlates for LAIVs in previous studies.21

d in this study might have informed our interpretation of the variation in results between settings that subsequently emerged. However, the overall usefulness of immunogenicity data in this context is uncertain because of the difficulty in identifying protective immunological correlates for LAIVs in previous studies.21 In a previous study of 300 children in Kamalapur, we prospectively assessed wheezing endpoints for the single-dose Russian-backbone LAIV.8 In this larger study, background wheezing was higher in urban Kamalapur (9·1%) than in rural Matlab (3·9%), but increases in or worsening of wheezing-related illness were not seen in either site. These results are reassuring, particularly in combination with a similar lack of wheezing signal in the Senegal study,18 and support careful assessment of the Russian-backbone LAIV in younger age groups. Previous data from Bangladesh show high attack rates and severe illness in young children3, 4, 5, 7 particularly those younger than 2 years.4, 6

, particularly in combination with a similar lack of wheezing signal in the Senegal study,18 and support careful assessment of the Russian-backbone LAIV in younger age groups. Previous data from Bangladesh show high attack rates and severe illness in young children3, 4, 5, 7 particularly those younger than 2 years.4, 6 This study corroborates data obtained over many years in urban and rural Bangladesh, which have shown sustained circulation of multiple influenza strains and associated high clinical attack rates.2, 3, 4, 5, 6, 7, 12, 22, 23, 24 Overall, 24·5% of children in our study had laboratory-confirmed influenza illness (any strain) during the course of this trial, including over a third (34·8%) of children at the rural Matlab site. Circulation of B viruses was common, which suggests that the use of a quadrivalent vaccine with efficacy against both B lineages might further increased the magnitude of the effect. Our results support the use of a single-dose LAIV to prevent medically attended lower respiratory tract illness in young children in Bangladesh. Large multisite trials of vaccines in children done over multiple years could improve measurement of the effects of influenza vaccines, establish the burden of severe influenza disease in young children, and inform policy and financing decisions.25 Improved understanding of population-based differences in influenza vaccine performance is crucial to designing effective public health programmes in low-resource settings. Supplementary Material Supplementary appendix

Our results support the use of a single-dose LAIV to prevent medically attended lower respiratory tract illness in young children in Bangladesh. Large multisite trials of vaccines in children done over multiple years could improve measurement of the effects of influenza vaccines, establish the burden of severe influenza disease in young children, and inform policy and financing decisions.25 Improved understanding of population-based differences in influenza vaccine performance is crucial to designing effective public health programmes in low-resource settings. Supplementary Material Supplementary appendix Acknowledgments We thank all the families who participated in this trial and the research team in Bangladesh, especially the clinical and field teams at both sites, without whose diligence and attention to detail during a period of social unrest this study would not have been possible. We thank Fatimah Dawood, US Centers for Disease Control and Prevention, Atlanta, GA, USA, for background work on wheezing illness and Bilkis Ara Anjali, icddr,b, Dhaka, Bangladesh, for assistance with data management. We thank the Serum Institute of India, Pune, India for donating the masked vaccine and placebo used in this study. We also thank Larisa Rudenko and Irina Isakova-Sivak, Institute of Experimental Medicine, St Petersburg, Russia, for the sequencing of specimens to identify vaccine-matched and wild-type viruses. Supporting PATH, Seattle, WA, USA, in fulfilling its sponsor obligations, we thank the teams at Clinogent, Hyderabad, India, for their diligence in site monitoring and at EMMES, Rockville, MD, USA for statistical support. At PATH we also thank Kristin Bedell for assistance with trial management and, at Johns Hopkins University, Baltimore, MD, USA, Larry Moulton for his helpful comments. This work was supported through funding from the Bill & Melinda Gates Foundation, which provides financial support to PATH's Influenza Vaccine Project (OPP48805). The findings and conclusions in this report are those of the authors and they do not necessarily represent the decisions or policies of their organisations or the study funder.

rted through funding from the Bill & Melinda Gates Foundation, which provides financial support to PATH's Influenza Vaccine Project (OPP48805). The findings and conclusions in this report are those of the authors and they do not necessarily represent the decisions or policies of their organisations or the study funder. Contributors WAB, KZ, KDCL, JRO, DG, and KMN conceived and designed the study. KZ, DG, ATS, KN, and MR acquired the data. All authors analysed and interpreted the data. WAB and KMN drafted the Article and all authors provided critical revisions to the content. Declaration of interests We declare no competing interests. International data safety monitoring board Margaret Rennels, Oxford, MD, USA; Edwin Asturias, Children's Hospital Colorado, Aurora, CO, USA; Susan Chiu, University of Hong Kong, Hong Kong Special Administrative Region, China; and John Tam, Hong Kong Polytechnic University, Hong Kong Special Administrative Region, China. Local data safety monitoring board Choudhury Ali Kawser, Bangabandhu Sheikh Mujib Medical University, Dhaka, Bangladesh; Jalal Udin Ashraful Haq, Ibrahim Medical College and BIRDEM, Shahbagh, Dhaka, Bangladesh; Nazam Haque, Ibrahim Medical College, Shahbagh, Dhaka, Bangladesh; Md Nurul Alam, icddr,b, Dhaka, Bangladesh; and Wasif Ali Khan, Centre for Vaccine Sciences, Dhaka, Bangladesh. Figure 1 Trial profile LAIV=live attenuated influenza vaccine. Figure 2 Influenza circulation in the study area overall and in Kamalapur and Matlab, by type and subtype or lineage, in weeks 9–52 of 2013

Local data safety monitoring board Choudhury Ali Kawser, Bangabandhu Sheikh Mujib Medical University, Dhaka, Bangladesh; Jalal Udin Ashraful Haq, Ibrahim Medical College and BIRDEM, Shahbagh, Dhaka, Bangladesh; Nazam Haque, Ibrahim Medical College, Shahbagh, Dhaka, Bangladesh; Md Nurul Alam, icddr,b, Dhaka, Bangladesh; and Wasif Ali Khan, Centre for Vaccine Sciences, Dhaka, Bangladesh. Figure 1 Trial profile LAIV=live attenuated influenza vaccine. Figure 2 Influenza circulation in the study area overall and in Kamalapur and Matlab, by type and subtype or lineage, in weeks 9–52 of 2013 (A) The whole study area. (B) Kamalapur. (C) Matlab. Circulating and vaccine B strains were mismatched and, therefore, did not contribute to vaccine efficacy Table 1 Baseline characteristics

Figure 2 Influenza circulation in the study area overall and in Kamalapur and Matlab, by type and subtype or lineage, in weeks 9–52 of 2013 (A) The whole study area. (B) Kamalapur. (C) Matlab. Circulating and vaccine B strains were mismatched and, therefore, did not contribute to vaccine efficacy Table 1 Baseline characteristics Kamalapur Matlab All LAIV (n=800) Placebo (n=400) LAIV (n=374) Placebo (n=187) LAIV (n=1174) Placebo (n=587) Total (n=1761) Mean (range) age (months) 42·7 (24–59) 42·3 (24–59) 42·2 (24–59) 41·8 (24–59) 42·5 (24–59) 42·1 (24–59) 42·4 (24–59) Age group (years) ≥2 to <3 248 (31·0%) 138 (34·5%) 105 (28·1%) 53 (28·3%) 353 (30·1%) 191 (32·5%) 544 (30·9%) ≥3 to <4 229 (28·6%) 105 (26·3%) 134 (35·8%) 69 (36·9%) 363 (30·9%) 174 (29·6%) 537 (30·5%) ≥4 to <5 323 (40·4%) 157 (39·3%) 135 (36·1%) 65 (34·8%) 458 (39·0%) 222 (37·8%) 680 (38·6%) Sex Male 410 (51·3%) 191 (47·8%) 175 (4·8%) 90 (48·1%) 585 (49·8%) 281 (47·9%) 866 (49·2%) Female 390 (48·8%) 209 (52·3%) 199 (53·2%) 97 (51·9%) 589 (50·2%) 306 (52·1%) 895 (50·8%) Underweight (weight for age malnutrition)* None 203 (25·4%) 103 (25·8%) 106 (28·3%) 48 (25·7%) 309 (26·3%) 151 (25·7%) 460 (26·1%) Mild 324 (40·5%) 173 (43·3%) 162 (43·3%) 84 (44·9%) 486 (41·4%) 257 (43·8%) 743 (42·2%) Moderate 219 (27·4%) 105 (26·3%) 91 (24·3%) 43 (23·0%) 310 (26·4%) 148 (25·2%) 458 (26·0%) Severe 54 (6·8%) 19 (4·8%) 15 (4·0%) 12 (6·4%) 69 (5·9%) 31 (5·3%) 100 (5·7%) Stunting (height for age malnutrition)* None 152 (19·0%) 84 (21·0%) 111 (29·7%) 47 (25·1%) 263 (22·4%) 131 (22·3%) 394 (22·4%) Mild 275 (34·4%) 144 (36·0%) 140 (37·4%) 81 (43·3%) 415 (35·3%) 225 (38·3%) 640 (36·3%) Moderate 257 (32·1%) 127 (31·8%) 95 (25·4%) 50 (26·7%) 352 (30·0%) 177 (30·2%) 529 (30·0%) Severe 116 (14·5%) 45 (11·3%) 28 (7·5%) 9 (4·8%) 144 (12·3%) 54 (9·2%) 198 (11·2%) Wasting (weight for height malnutrition)* None 495 (61·9%) 249 (62·3%) 209 (55·9%) 100 (53·5%) 704 (60·0%) 349 (59·5%) 1053 (59·8%) Mild 243 (30·4%) 125 (31·3%) 120 (32·1%) 65 (34·8%) 363 (30·9%) 190 (32·4%) 553 (31·4%) Moderate 60 (7·5%) 20 (5·0%) 42 (11·2%) 21 (11·2%) 102 (8·7%) 41 (7·0%) 143 (8·1%) Severe 2 (0·3%) 6 (1·5%) 3 (0·8%) 1 (0·5%) 5 (0·4%) 7 (1·2%) 12 (0·7%) Asthma or wheezing illness Previous hospital admission 58 (7·3%) 20 (5·0%) 0 0 58 (4·9%) 20 (3·4%) 78 (4·4%) Previous treatment 242 (30·3%) 134 (33·5%) 0 0 242 (30·3%) 134 (33·5%) 376 (21·4%) LAIV=live attenuated influenza vaccine.

(7·0%) 143 (8·1%) Severe 2 (0·3%) 6 (1·5%) 3 (0·8%) 1 (0·5%) 5 (0·4%) 7 (1·2%) 12 (0·7%) Asthma or wheezing illness Previous hospital admission 58 (7·3%) 20 (5·0%) 0 0 58 (4·9%) 20 (3·4%) 78 (4·4%) Previous treatment 242 (30·3%) 134 (33·5%) 0 0 242 (30·3%) 134 (33·5%) 376 (21·4%) LAIV=live attenuated influenza vaccine. * Z score, mild (−2 to <–1), moderate (−3 to <–2), or severe (<–3). Table 2 Vaccine efficacy in the per-protocol population* LAIV (n=1174) Placebo (n=587) Vaccine efficacy (95% CI) Number of infections Attack rate (%) Number of infections Attack rate (%) Whole study population (n=1761) All vaccine-matched strains 79 6·7% 93 15·8% 57·5% (43·6 to 68·0) All strains 170 14·5% 144 24·5% 41·0% (28·0 to 51·6) H1N1 21 1·8% 21 3·6% 50·0% (9·2 to 72·5) H3N2 57 4·9% 72 12·3% 60·4% (44·8 to 71·6) B/Yamagata (vaccine-matched) 2 0·2% 1 0·2% 0% (−1001·0 to 90·9) B/Victoria (unmatched) 58 4·9% 31 5·3% 6·5% (−43·0 to 38·8) Kamalapur (n=1200)† All vaccine-matched strains 56 7·0% 52 13·0% 46·2% (23·0 to 62·4) With history of asthma or wheeze‡ 22 8·9% 17 12·6% 29·0% (−29·0 to 60·9) Without history of asthma or wheeze 34 6·1% 35 13·2% 53·4% (27·2 to 70·3) All strains 103 12·9% 79 19·8% 34·8% (14·8 to 50·1) Matlab (n=561)§ All vaccine-matched strains 23 6·1% 41 21·9% 72·0% (54·7 to 82·6) All strains 67 17·9% 65 34·8% 48·5% (30·9 to 61·5) LAIV=live attenuated influenza vaccine. * Includes laboratory-confirmed influenza infections occurring from 8 days onwards after receiving vaccine or placebo. † n=800 in the LAIV group, n=400 in the placebo group.

LAIV (n=1174) Placebo (n=587) Vaccine efficacy (95% CI) Number of infections Attack rate (%) Number of infections Attack rate (%) Whole study population (n=1761) All vaccine-matched strains 79 6·7% 93 15·8% 57·5% (43·6 to 68·0) All strains 170 14·5% 144 24·5% 41·0% (28·0 to 51·6) H1N1 21 1·8% 21 3·6% 50·0% (9·2 to 72·5) H3N2 57 4·9% 72 12·3% 60·4% (44·8 to 71·6) B/Yamagata (vaccine-matched) 2 0·2% 1 0·2% 0% (−1001·0 to 90·9) B/Victoria (unmatched) 58 4·9% 31 5·3% 6·5% (−43·0 to 38·8) Kamalapur (n=1200)† All vaccine-matched strains 56 7·0% 52 13·0% 46·2% (23·0 to 62·4) With history of asthma or wheeze‡ 22 8·9% 17 12·6% 29·0% (−29·0 to 60·9) Without history of asthma or wheeze 34 6·1% 35 13·2% 53·4% (27·2 to 70·3) All strains 103 12·9% 79 19·8% 34·8% (14·8 to 50·1) Matlab (n=561)§ All vaccine-matched strains 23 6·1% 41 21·9% 72·0% (54·7 to 82·6) All strains 67 17·9% 65 34·8% 48·5% (30·9 to 61·5) LAIV=live attenuated influenza vaccine. * Includes laboratory-confirmed influenza infections occurring from 8 days onwards after receiving vaccine or placebo. † n=800 in the LAIV group, n=400 in the placebo group. ‡ Vaccine efficacy analyses including history of asthma or wheezing could only be interpreted for the Kamalapur study site where history was identified at baseline. No participants at the Matlab site indicated a history of asthma or wheeze. § n=347 in the LAIV group, n=187 in the placebo group. Table 3 Local and systemic reactions in the 7 days after vaccination and protocol-defined wheezing illness at any time

‡ Vaccine efficacy analyses including history of asthma or wheezing could only be interpreted for the Kamalapur study site where history was identified at baseline. No participants at the Matlab site indicated a history of asthma or wheeze. § n=347 in the LAIV group, n=187 in the placebo group. Table 3 Local and systemic reactions in the 7 days after vaccination and protocol-defined wheezing illness at any time LAIV (n=1174) Placebo (n=587) Whole population (n=1761) Mild Moderate Severe Mild Moderate Severe Mild Moderate Severe Local and systemic reactions Fever (≥38°C) 17 (1·4%) 20 (1·1%) 0 8 (1·4%) 8 (1·4%) 0 25 (1·4%) 28 (1·6%) 0 Nasal congestion 1 (0·1%) 0 0 1 (0·2%) 0 0 2 (0·1%) 0 0 Runny nose 68 (5·8%) 0 0 39 (6·6%) 0 0 107 (6·1%) 0 0 Cough 72 (6·1%) 1 (0·1%) 0 43 (7·3%) 0 0 115 (6·5%) 1 (0·1%) 0 Sore throat 4 (0·3%) 0 0 2 (0·3%) 0 0 6 (0·3%) 0 0 Ear pain 2 (0·2%) 0 0 3 (0·5%) 0 0 5 (0·3%) 0 0 Headache 0 0 0 1 (0·2%) 0 0 1 (0·1%) 0 0 Vomiting 4 (0·3%) 0 0 5 (0·9%) 1 (0·2%) 0 9 (0·5%) 1 (0·1%) 0 Chills 0 0 0 0 0 0 0 0 0 Irritability or decreased activity 1 (0·1%) 0 0 0 0 0 1 (0·1%) 0 0 Muscle/joint pain 3 (0·3%) 1 (0·1%) 0 0 0 0 3 (0·2%) 1 (0·1%) 0 Tachypnoea* 77 (6·6%) 9 (0·5%) 0 53 (9·0%) 1 (0·2%) 0 130 (7·4%) 10 (0·6%) 0 Protocol-defined wheezing illness Days 0–7 4 (0·3%) 0 0 1 (0·2%) 0 0 5 (0·3%) 0 0 Days 8–42 16 (1·4%) 0 0 9 (1·5%) 0 0 25 (1·4%) 0 0 Day 43 to 6 months 38 (3·2%) 1 (0·1%) 0 31 (5·3%) 1 (0·2%) 0 69 (3·9%) 2 (0·1%) 0 Day 0 to 6 months 53 (4·5%) 1 (0·1%) 0 37 (6·3%) 1 (0·2%) 0 90 (5·1%) 2 (0·1%) 0 Anytime† 78 (6·6%) 3 (0·3%) 0 46 (7·8%) 4 (0·7%) 0 124 (7·0%) 7 (0·4%) 0 LAIV=live attenuated influenza vaccine.

–42 16 (1·4%) 0 0 9 (1·5%) 0 0 25 (1·4%) 0 0 Day 43 to 6 months 38 (3·2%) 1 (0·1%) 0 31 (5·3%) 1 (0·2%) 0 69 (3·9%) 2 (0·1%) 0 Day 0 to 6 months 53 (4·5%) 1 (0·1%) 0 37 (6·3%) 1 (0·2%) 0 90 (5·1%) 2 (0·1%) 0 Anytime† 78 (6·6%) 3 (0·3%) 0 46 (7·8%) 4 (0·7%) 0 124 (7·0%) 7 (0·4%) 0 LAIV=live attenuated influenza vaccine. * Mild 31–40 breaths per min, moderate 41–50 breaths per min, and severe ≥51 breaths per min. † Including events occurring >6 months after vaccination.

Introduction Globally, lower respiratory infection remains the leading cause of death in children younger than 5 years,1, 2, 3 and influenza virus infections can be an important contributor to serious lower respiratory disease.4 Live attenuated influenza vaccines are an attractive option for young children in low-resource settings. In US trials done before this trial was initiated, live attenuated influenza vaccines had shown superior efficacy to inactivated vaccines among young children and might be cross-protective against drifted influenza viruses.5, 6 Live attenuated influenza vaccines are likewise attractive from a manufacturing perspective because higher yields, a simpler purification process, and quicker lot release should lower the cost of production of live attenuated influenza vaccines relative to inactivated vaccines.7 Finally, live attenuated influenza vaccines might be more feasible given their intranasal administration and potential protection with a single dose.8 In Senegal, there are no recommendations for routine influenza vaccination of children, and generally only trivalent inactivated influenza vaccine of WHO-recommended northern hemisphere formulation has been available in the country for use in vaccination of pilgrims to the Hajj. The National Influenza Center of Senegal has been undertaking influenza surveillance for more than 20 years and has accumulated data that influenza, especially type A, occurs in association with the rainy season9 (typically between July and September) and can circulate widely, especially among children.

ms to the Hajj. The National Influenza Center of Senegal has been undertaking influenza surveillance for more than 20 years and has accumulated data that influenza, especially type A, occurs in association with the rainy season9 (typically between July and September) and can circulate widely, especially among children. Under the Global Pandemic Influenza Action Plan to Increase Vaccine Supply, the WHO leads a programme to enhance regional access to influenza vaccines by supporting developing country vaccine manufacturers.10 Serum Institute of India, Ltd (Pune, India) participated in this initiative and developed a reassortant live attenuated seasonal influenza vaccine based on A/Leningrad/17 and B/USSR/60 master donor viruses. To understand how live attenuated influenza vaccines such as this might perform in young children in resource-limited tropical settings in Africa, we assessed the efficacy of Serum Institute of India's trivalent live attenuated influenza vaccine, which is licensed as a single dose for prevention of influenza, in Senegal. Research in context Evidence before this study

Under the Global Pandemic Influenza Action Plan to Increase Vaccine Supply, the WHO leads a programme to enhance regional access to influenza vaccines by supporting developing country vaccine manufacturers.10 Serum Institute of India, Ltd (Pune, India) participated in this initiative and developed a reassortant live attenuated seasonal influenza vaccine based on A/Leningrad/17 and B/USSR/60 master donor viruses. To understand how live attenuated influenza vaccines such as this might perform in young children in resource-limited tropical settings in Africa, we assessed the efficacy of Serum Institute of India's trivalent live attenuated influenza vaccine, which is licensed as a single dose for prevention of influenza, in Senegal. Research in context Evidence before this study We searched PubMed from Jan 1, 1980, to Jan 1, 2016, for efficacy trials assessing protection by live attenuated influenza vaccines among children using the search terms, “human influenza”, “vaccines, attenuated”, and “children”. Live attenuated influenza vaccines have been studied mainly in developed countries in Europe, Asia, and in the USA, but one study has included children aged 6 months to 36 months of age from sites in South Africa. In nearly all studies, A/Ann Arbor-based live attenuated influenza vaccines provided significant protection. Russian-derived live attenuated influenza vaccines used among young children had been studied primarily from 1986 to 1991 in five trials in the Soviet Union and Cuba. These studies used serological confirmation of illness and were done before introduction of Good Clinical Practice standards for the designing, conducting, recording, and reporting of trials. Moreover, except for the companion study in Bangladesh, no efficacy trials of Russian-derived live attenuated influenza vaccines have been done in developing country settings. The only study of clinical protection of Serum Institute of India's live attenuated influenza vaccines was an observational case-control study of the effectiveness of one dose of its monovalent live attenuated influenza vaccines containing vaccine virus antigenically similar to pandemic A/H1N1 (2009).

country settings. The only study of clinical protection of Serum Institute of India's live attenuated influenza vaccines was an observational case-control study of the effectiveness of one dose of its monovalent live attenuated influenza vaccines containing vaccine virus antigenically similar to pandemic A/H1N1 (2009). Added value of this study To the best of our knowledge, our trial is the first randomised trial of any live attenuated influenza vaccines in tropical developing Africa, and our results indicate that children in this population were not protected with one dose. Our results point to the need for more study of live attenuated influenza vaccines in impoverished populations in Africa to understand how well live attenuated influenza vaccines do and can best be used to provide protection to young children. Implications of all the available evidence The results of this study should be viewed in light of data from early studies in developed populations and the results of our companion study in Asia in which live attenuated influenza vaccines provided moderately high protection to this young child age group. Counter to this finding is the fact that observational effectiveness studies in the USA in 2013–14 found a similar absence of efficacy of live attenuated influenza vaccines against A/H1N1 (2009) in young children. Live attenuated influenza vaccines are easy to administer and might be the optimal choice for child populations in developing populations. However, further study in such populations, which might experience extended periods of influenza circulation, is warranted.

fluenza vaccines against A/H1N1 (2009) in young children. Live attenuated influenza vaccines are easy to administer and might be the optimal choice for child populations in developing populations. However, further study in such populations, which might experience extended periods of influenza circulation, is warranted. Methods Study design and participants This double-blind, placebo-controlled, parallel group, single-centre trial was done in the rural area comprising the Niakhar Demographic Surveillance System, about 110 km southeast of Dakar, Senegal. Generally healthy children aged 2–5 years were eligible if a parent was willing to provide written informed consent and was not expecting to migrate out of the area during the study period. Informed consent was obtained through a well established two-step process in accordance with established social structures and cultural norms in this population. First, the investigator held group meetings in communities of the study area to obtain village permission for study conduct. Second, a list of eligible children was constructed using demographic surveillance system data, and parents of eligible children were approached by trained study staff during the month prior to study initiation to provide them with information about the upcoming research study. Those interested in participating were given dates and locations of enrollment days at their locale. On days of enrolment, trained study staff reviewed study information with presenting parents and children, and children were enrolled consecutively as parents provided written informed consent and children were determined to be eligible to participate. Trained study physicians were present on all enrolment days. Exclusion criteria included serious active medical conditions. Acute malnutrition is common in Senegal, particularly during the typical influenza season from June through October before the harvest. Therefore, malnourished children were not specifically excluded. Eligibility criteria are listed in the appendix (p 1).

. Exclusion criteria included serious active medical conditions. Acute malnutrition is common in Senegal, particularly during the typical influenza season from June through October before the harvest. Therefore, malnourished children were not specifically excluded. Eligibility criteria are listed in the appendix (p 1). Ethics review was provided by the National Ethics Committee for Health Research (Senegal Ministry of Health and Social Welfare) and Western Institutional Review Board. The study was done in accordance with the principles of the Declaration of Helsinki (2008) and in compliance with Good Clinical Practice guidelines. Randomisation and masking Participants were randomly allocated (2:1) to receive the live attenuated influenza vaccine or placebo. The allocation sequence was computer-generated by PATH with block sizes of three. The sequence was delivered to Serum Institute of India where vaccine and placebo were labelled before shipping to Senegal for use in the field. Vaccine and placebo were identical in appearance and vials containing the study products were labelled only with a clinical trial label and unique allocation numbers to preserve blinding.

e was delivered to Serum Institute of India where vaccine and placebo were labelled before shipping to Senegal for use in the field. Vaccine and placebo were identical in appearance and vials containing the study products were labelled only with a clinical trial label and unique allocation numbers to preserve blinding. Procedures Study products were lyophilised live attenuated influenza vaccine of 2012–13 Northern Hemisphere formulation (Nasovac-S, Serum Institute of India, Pune, India; lot 167E2002) containing A/California/7/2009 (H1N1)-like, A/Victoria/361/2011 (H3N2)-like, and B/Wisconsin/1/2010 (Yamagata lineage)-like reassortants. Matched placebo was identical to vaccine in appearance and content but was missing viral components (Serum Institute of India; lot E9001PCB). A single 0·5 mL dose of either was administered intranasally (divided evenly per nostril) to each participant using Wolfe-Tory mucosal atomiser devices. Lyophilised vaccine was stored at 2–8°C until reconstitution with diluent of ambient temperature. Administration after reconstitution was immediate.

ia; lot E9001PCB). A single 0·5 mL dose of either was administered intranasally (divided evenly per nostril) to each participant using Wolfe-Tory mucosal atomiser devices. Lyophilised vaccine was stored at 2–8°C until reconstitution with diluent of ambient temperature. Administration after reconstitution was immediate. Participant follow-up was planned from June, 2013, until December, 2013, during which participants were monitored weekly through home visits by trained field workers. On day 7 postvaccination, participants were assessed for solicited reactions and unsolicited adverse events occurring since vaccination, and on day 28 for unsolicited adverse events. Adverse events were documented by field workers using standardised data collection forms after interviews with parents at each visit. Throughout study follow-up, standardised criteria were used to identify participants with signs and symptoms of influenza. Additional criteria were used to refer participants with medically important illness to a study physician for further evaluation. These physicians provided treatment as per the standard of care in Senegal and assessed participants for study outcomes of clinical influenza, protocol-defined wheezing illness, and serious adverse events. For all participants with signs or symptoms of influenza, both a nasal swab and a pharyngeal swab specimen were collected and pooled.

ovided treatment as per the standard of care in Senegal and assessed participants for study outcomes of clinical influenza, protocol-defined wheezing illness, and serious adverse events. For all participants with signs or symptoms of influenza, both a nasal swab and a pharyngeal swab specimen were collected and pooled. A subset of participants already enrolled in the main trial were additionally consented to participate in a vaccine infectivity and extended safety subset to assess the ability of cold-adapted vaccine viruses to replicate in the nasal passages of children in this population in which ambient daily temperatures regularly exceed 40°C during the period of trial vaccinations and to provide a more detailed description of vaccine safety during the 7 days postvaccination in this low-resource population. For this analysis, both a nasal swab and a pharyngeal swab specimen were collected prevaccination and on days 2 and 4 postvaccination from all subset participants. Detailed solicited reactions and unsolicited adverse events were also collected on standard data collection forms by field workers through interview with parents/guardians.

s, both a nasal swab and a pharyngeal swab specimen were collected prevaccination and on days 2 and 4 postvaccination from all subset participants. Detailed solicited reactions and unsolicited adverse events were also collected on standard data collection forms by field workers through interview with parents/guardians. Outcomes The primary efficacy endpoint was symptomatic laboratory-confirmed influenza caused by any virus type, including those not included in the vaccine, and occurring from 15 days post vaccination to the end of the study in December, 2013. Symptomatic influenza was defined as sudden onset of measured fever (>37·5°C axillary) or subjectively reported feverishness and a cough or sore throat. Laboratory confirmation (type and subtype or lineage) was defined as detection of seasonal influenza virus in a swab specimen. Specimens were tested at Senegal's National Influenza Centre Laboratory at the Institut Pasteur de Dakar for the presence of influenza virus by real-time RT-PCR using methods and reagents, including standard seasonal influenza oligonucleotide primers and probes, provided by the Influenza Division of the US Centers for Disease Control and Prevention (CDC). Additionally, although not protocol-specified, all influenza-positive specimens from symptomatic cases occurring within 1 month of vaccination were shipped to the US CDC and tested post-hoc by real-time RT-PCR using CDC's oligonucleotide primer-probe sets specific for live attenuated influenza virus A and live attenuated influenza virus B, as was done for specimens from the vaccine infectivity subset. Antigenic characterisation of about 20% of positive specimens was done at CDC. Secondary efficacy endpoints were symptomatic laboratory-confirmed influenza matched to vaccine and influenza strain-specific symptomatic influenza.

attenuated influenza virus B, as was done for specimens from the vaccine infectivity subset. Antigenic characterisation of about 20% of positive specimens was done at CDC. Secondary efficacy endpoints were symptomatic laboratory-confirmed influenza matched to vaccine and influenza strain-specific symptomatic influenza. Safety endpoints included solicited local and systemic reactions (nasal congestion, runny nose, stuffy nose, ear pain, cough, sore throat, headache, fever, tachypnoea, muscle or joint pain, chills, irritability or decreased activity, and vomiting), unsolicited adverse events and serious adverse events, and protocol-defined wheezing illness. Protocol-defined wheezing illness was defined as an illness meeting physician evaluation criteria (fever >38°C axillary, tachypnoea >40 breaths per min, ear pain, seizure or convulsions, or any other condition believed to warrant physician evaluation) and characterised by a long, high-pitched whistling or musical sound on expiration heard by auscultation over the lung fields. Severity of protocol-defined wheezing illness was graded by study physicians as mild (wheezing illness without other findings associated with disease of moderate or greater severity), moderate (nasal flaring, chest in-drawing, or pulse oximetry 90–95%), severe (dyspnoea at rest causing inability to perform usual social and functional activities or pulse oximetry <90%), or life threatening (per physician's medical opinion).

hout other findings associated with disease of moderate or greater severity), moderate (nasal flaring, chest in-drawing, or pulse oximetry 90–95%), severe (dyspnoea at rest causing inability to perform usual social and functional activities or pulse oximetry <90%), or life threatening (per physician's medical opinion). Swab specimens from the vaccine infectivity subset were tested at the US CDC by real-time RT-PCR first for the presence of influenza virus using standard seasonal influenza oligonucleotide primers and probes. Positive specimens in the vaccine group were further tested for the presence of live attenuated influenza vaccine by real-time RT-PCR using oligonucleotide primers and probes designed to detect internal virus genes specific to the A/Leningrad/134/17/57 (H2N2) and B/USSR/60/69 master donor viruses used to create the cold-adapted reassortants contained in the vaccine.

ther tested for the presence of live attenuated influenza vaccine by real-time RT-PCR using oligonucleotide primers and probes designed to detect internal virus genes specific to the A/Leningrad/134/17/57 (H2N2) and B/USSR/60/69 master donor viruses used to create the cold-adapted reassortants contained in the vaccine. Statistical analysis The primary objective was to estimate the efficacy of live attenuated influenza vaccine in reducing the rate of symptomatic laboratory-confirmed influenza regardless of vaccine match among children receiving live attenuated influenza vaccine. Vaccine efficacy was defined as one minus the relative rate (times 100%) of influenza in the live attenuated influenza vaccine group compared with that in the placebo group. A Cox proportional hazards model with censoring at time of first endpoint was fitted to estimate the relative rate and its 95% CI. The protocol specified that endpoints occurring within 2 weeks postvaccination were not counted to allow sufficient time for development of the immune response and to ensure that any vaccinees with non-influenza acute respiratory illness during this period would not be misclassified if vaccine virus was identified by real-time RT-PCR using seasonal influenza primer-probe sets. Secondary efficacy endpoints were similarly analysed. Primary efficacy analyses were done on a per-protocol basis and included all children who met eligibility criteria, were randomised, received one dose of study vaccine, and contributed at least one day of person-time. Supportive analyses were also done on the total vaccinated cohort of all randomised participants who received a dose of vaccine or placebo, counting endpoints occurring at any time postvaccination. Effect of previous receipt of trivalent inactivated vaccine on efficacy was explored in post-hoc analyses, as was the effect of malnutrition, with adjustment for grade of malnutrition. Grades for weight-for-age, height-for-age, and weight-for-height were based on calculated Z scores and categorised as mild (–2 to <–1), moderate (–3 to <–2), or severe (<–3).11, 12

ivated vaccine on efficacy was explored in post-hoc analyses, as was the effect of malnutrition, with adjustment for grade of malnutrition. Grades for weight-for-age, height-for-age, and weight-for-height were based on calculated Z scores and categorised as mild (–2 to <–1), moderate (–3 to <–2), or severe (<–3).11, 12 Safety in the total vaccinated cohort and the extended safety subset was described as the proportion of participants in each study group experiencing reactions or events of any severity (and by severity grade) with its corresponding exact 95% CI. Fisher's exact test was used to compare proportions of reactions of any severity grade between study groups. Participant safety was also overseen by an independent safety monitoring committee convened by PATH. Assuming 60% efficacy, 57 total primary endpoints were required to test the hypothesis that live attenuated influenza vaccine efficacy was greater than 0% with a one-sided type I error of less than 2·5% and power of at least 90%. Based on this number and assuming a 6% incidence rate and 90% evaluability, a total sample size of 1761 enrolled subjects was estimated. Data were analysed with SAS version 9.3. This study is registered with ClinicalTrials.gov, number NCT01854632.

Assuming 60% efficacy, 57 total primary endpoints were required to test the hypothesis that live attenuated influenza vaccine efficacy was greater than 0% with a one-sided type I error of less than 2·5% and power of at least 90%. Based on this number and assuming a 6% incidence rate and 90% evaluability, a total sample size of 1761 enrolled subjects was estimated. Data were analysed with SAS version 9.3. This study is registered with ClinicalTrials.gov, number NCT01854632. Role of the funding source This work was funded through a Cooperative Agreement to PATH from the US Centers for Disease Control and Prevention (U01IP000476). Supplementary funding from the Bill & Melinda Gates Foundation to PATH (OPP1017334) supported statistical analyses. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US CDC, Gates Foundation, or PATH. All authors had full access to all the data in the study and JCV had final responsibility for the decision to submit for publication. Results Between May 23, and July 1, 2013, 1761 children were enrolled and randomised; 1174 were allocated to receive study vaccine and 587 to receive placebo (figure 1). Nearly all (1757) contributed person-time to the primary (per-protocol) analysis of efficacy. Retention was very high, with 1722 (98%) being monitored to the end of the follow-up period, Dec 20, 2013 (table 1). Children who additionally gave consent into the vaccine infectivity and extended safety subset had their day 0 swabs taken between May 23 and June 5.

e to the primary (per-protocol) analysis of efficacy. Retention was very high, with 1722 (98%) being monitored to the end of the follow-up period, Dec 20, 2013 (table 1). Children who additionally gave consent into the vaccine infectivity and extended safety subset had their day 0 swabs taken between May 23 and June 5. Baseline characteristics were similar in the two study groups (table 1). A substantial proportion of participants were malnourished to some extent, with 46% underweight, 48% stunted, and 36% showing wasting. Otherwise, participants had few identifiable medical issues. Only one child, in the live attenuated influenza vaccine group, had a reported history of asthma. Type B influenza viruses of both lineages were already circulating in Senegal by the beginning of the trial (figure 2). At the trial site, the circulation of Victoria lineage B virus (unmatched to vaccine) intensified as the season progressed. Midway through the trial, H1N1pdm09 appeared and circulated extensively until the end of the study period. H3N2 circulation was negligible.

ing in Senegal by the beginning of the trial (figure 2). At the trial site, the circulation of Victoria lineage B virus (unmatched to vaccine) intensified as the season progressed. Midway through the trial, H1N1pdm09 appeared and circulated extensively until the end of the study period. H3N2 circulation was negligible. Influenza incidence among young children in this study was high. In the primary analysis, 210 endpoints of symptomatic laboratory-confirmed influenza were reached among vaccine recipients (18%) and 105 were reached among placebo recipients (18%), yielding an efficacy point estimate of 0·0% (95% CI −26·4 to 20·9%; table 2). Analyses of secondary endpoints caused by vaccine-matched strains and by specific type and subtype or lineage likewise revealed no statistical evidence for vaccine efficacy. 79 (7%) endpoints of H1N1pdm09 were noted in the vaccine group and 36 (6%) in the placebo group; 20 (2%) endpoints of vaccine-matched B in the vaccine group and 11 (2%) in the placebo group; and 115 (10%) endpoints of vaccine-mismatched B in the vaccine group and 62 (11%) in the placebo group. Only three (<1%) endpoints of H3N2, all in the vaccine group, were reached. In a post-hoc analysis of efficacy against all vaccine-matched strains using the per-protocol dataset, among children previously vaccinated with trivalent inactivated influenza vaccine (containing A/California/7/2009) in 2010 or 2011 (average age of this subgroup, 4·4 years), live attenuated influenza vaccine efficacy was −28·4% (95% CI −108·7 to 19·9) compared with 15·4% (–27·5 to 43·4) among children with no previous receipt of trivalent inactivated influenza vaccine (average age of this subgroup, 3·5 years). Additional post-hoc analyses in which we stratified primary and secondary analyses by age groups or measures of malnutrition also did not show any significant efficacy for any measure (data not shown).

hildren with no previous receipt of trivalent inactivated influenza vaccine (average age of this subgroup, 3·5 years). Additional post-hoc analyses in which we stratified primary and secondary analyses by age groups or measures of malnutrition also did not show any significant efficacy for any measure (data not shown). Detection of at least one strain of vaccine virus was confirmed in 55 (83%) of 66 live attenuated influenza vaccine recipients postvaccination (table 3). Virus detection was highest on day 2 postvaccination. Live attenuated influenza vaccine-A/H3N2 and live attenuated influenza vaccine-B viruses were detected postvaccination among 34 (52%) and 42 (66%) of recipients, respectively. Live attenuated influenza vaccine-A/H1N1pdm09 was detected among only 14 (22%) of recipients. One child in the live attenuated influenza vaccine group experienced mild epistaxis in the 30 min postvaccination. In the 7 days postvaccination, the most common solicited events reported overall were runny nose (296; 17%), cough (172; 10%), and nasal congestion (57; 3%). Nearly all reactions were mild, and there were no significant differences between the vaccine and placebo groups in terms of the proportions experiencing reactions of any severity (table 4) or unsolicited adverse events (appendix p 2). No significant differences were noted in occurrence of solicited reactions in the extended safety subset (appendix p 4).

there were no significant differences between the vaccine and placebo groups in terms of the proportions experiencing reactions of any severity (table 4) or unsolicited adverse events (appendix p 2). No significant differences were noted in occurrence of solicited reactions in the extended safety subset (appendix p 4). Nine (1%) participants experienced protocol-defined wheezing illness, with no significant differences in occurrence between vaccine groups (table 4). Serious adverse events occurred in 14 children; seven in each of the vaccine (1%) and placebo (1%) groups (appendix p 5). All serious adverse events were considered unrelated to vaccination. Two girls, both aged 2 years, died. Both had received live attenuated influenza vaccine, and causes of death were anasarca 12 days postvaccination and malnutrition 70 days postvaccination.

ch of the vaccine (1%) and placebo (1%) groups (appendix p 5). All serious adverse events were considered unrelated to vaccination. Two girls, both aged 2 years, died. Both had received live attenuated influenza vaccine, and causes of death were anasarca 12 days postvaccination and malnutrition 70 days postvaccination. Discussion Among this low-resource paediatric population in sub-Saharan Africa, symptomatic influenza infection was common, with an overall attack rate of about 20%. Although live attenuated influenza vaccine was well tolerated, a single dose did not provide protection against symptomatic laboratory-confirmed influenza for all strains or for the predominant vaccine-matched strain, H1N1pdm09. Efficacy was also not found for influenza B strains, although attack rates were low for the vaccine-matched B lineage. There were insufficient H3N2 cases to calculate vaccine strain-specific efficacy for that subtype. Although nearly half of participants had been previously vaccinated with trivalent inactivated influenza vaccine, randomisation ensured equal distribution of this factor to study groups in this trial, and thus it should not have biased efficacy results. The study was carefully monitored throughout, and after the results were known, we did an additional audit of the site and a validation of real-time RT-PCR testing by the Senegal National Influenza Center laboratory. These revealed no concerns which would call into question the findings of this trial.

efficacy results. The study was carefully monitored throughout, and after the results were known, we did an additional audit of the site and a validation of real-time RT-PCR testing by the Senegal National Influenza Center laboratory. These revealed no concerns which would call into question the findings of this trial. The absence of efficacy against strain-matched, laboratory-confirmed symptomatic influenza illness in this study contrasts with the findings of a similar study conducted in the same season with a single dose of Nasovac-S among children aged 2–4 years in Bangladesh.13 In Bangladesh, H1N1pdm09 and H3N2 were the predominant strains, with attack rates of 3·6% and 12·3% in the placebo group, respectively, and vaccine efficacy of 50·0% (95% CI 9·2–72·5) against H1N1pdm09 and 60·4% (44·8–71·6) against H3N2. The reasons for the discrepancy between efficacy estimates in these two studies are unclear. It is unlikely that vaccine potency accounted for the efficacy differences given that the same lot of lyophilised vaccine was used for both studies. In fact, potency of shelf stocks of vaccine had been assessed monthly by Serum Institute of India, and in June, 2013, 9 months after manufacture, the A/California H1N1pdm09 component had a 50% egg infectious dose (EID50) of 107·532. Cold chain, vaccine reconstitution, and administration were carefully monitored at both sites.

, potency of shelf stocks of vaccine had been assessed monthly by Serum Institute of India, and in June, 2013, 9 months after manufacture, the A/California H1N1pdm09 component had a 50% egg infectious dose (EID50) of 107·532. Cold chain, vaccine reconstitution, and administration were carefully monitored at both sites. Differences in previous history of influenza exposure could exist in the Senegal and Bangladesh populations. Although we do not have prevaccination serological data to test this hypothesis, in a concurrent trial of inactivated influenza vaccines that we did in the neighbouring village of Niakhar (NCT01819155), baseline serological data showed that less than 25% of children younger than 6 years had evidence of previous exposure to H1N1pdm09, as determined by haemagglutinin antibody levels of >1:10 to that strain. Most of that seropositivity was focused in the older age groups, consistent with Senegal's national surveillance data, showing extensive H1N1pdm09 circulation in early 2010, and low levels of circulation later in 2011 and in 2012, the years before this study. Thus, it is likely that most participants in this study were naive to the H1N1 strain and would not have had antibodies that could have interfered with replication of this live vaccine virus strain. Even stratification by previous vaccination with inactivated influenza vaccine 2–3 years before this study showed no significant efficacy.

y that most participants in this study were naive to the H1N1 strain and would not have had antibodies that could have interfered with replication of this live vaccine virus strain. Even stratification by previous vaccination with inactivated influenza vaccine 2–3 years before this study showed no significant efficacy. Ecology of the nasopharynx or nutritional deficiencies might differentially affect the ability of the vaccine virus to infect or the recipient to mount an immune response and, therefore, the performance of vaccine in these two populations. Although post-hoc analyses stratifying efficacy by measures of malnutrition in this study population showed no effect of nutritional status, other unmeasured micronutrient deficiencies might be important. Whether the history of recent oral polio vaccine receipt in participants in Senegal could have adversely affected the performance of the live attenuated influenza vaccine is also unknown. Previously, a study of oral polio vaccine given concomitantly with the Ann Arbor-based live attenuated influenza vaccine showed no effect on antibody responses as determined by haemagglutination inhibition assay.14 However, no studies have determined the effects on the immunogenicity of live attenuated influenza vaccine with oral polio vaccine use in the prior 30 days or using alternative measures of immune response.

ed influenza vaccine showed no effect on antibody responses as determined by haemagglutination inhibition assay.14 However, no studies have determined the effects on the immunogenicity of live attenuated influenza vaccine with oral polio vaccine use in the prior 30 days or using alternative measures of immune response. In this study, vaccine virus of at least one strain was detected in 80% of live attenuated influenza vaccine recipients 2–4 days after administration. Since the mucociliary clearance rate is minutes, not days,15 we believe that our results are evidence of viral replication and that vaccine was viable at the time of administration in most recipients. However, detection of H1N1pdm09 was much lower than the other two vaccine strains. Although this low detection rate in Senegal stands in contrast to studies in Russia16 and Bangladesh17 with monovalent and trivalent formulations, respectively, where no H1N1pdm09 could be detected in nasal swabs or washes postvaccination, one study in the USA with initial trivalent live attenuated influenza vaccine containing A/California (H1N1) did find a substantial proportion of children (9/13) shed H1N1pdm09 after receipt of one dose.18 As mentioned earlier, differences in exposure history might exist between the Senegal and Bangladesh populations, and national surveillance data indicate that H1N1pdm09 circulation has been limited since the first wave of the pandemic in Senegal in early 2010. Prior immunity to vaccine strains has been shown to reduce live attenuated influenza vaccine viral shedding.19 Previous receipt of influenza vaccine has also been shown to inhibit shedding,18 although in that study previous vaccination was only 1 month prior and previous receipt of live attenuated influenza vaccine reduced shedding more than previous receipt of inactivated vaccine. Additionally, the earlier Bangladesh safety study used an earlier formulation of trivalent vaccine, and vaccine strains can differently interfere with each other depending on the formulation.20 Nonetheless, we are unable to explain why we were able to detect H1N1pdm09 vaccine virus postvaccination in about a fifth of recipients and yet measure no efficacy.

afety study used an earlier formulation of trivalent vaccine, and vaccine strains can differently interfere with each other depending on the formulation.20 Nonetheless, we are unable to explain why we were able to detect H1N1pdm09 vaccine virus postvaccination in about a fifth of recipients and yet measure no efficacy. Although we assessed infectivity of vaccine virus, a major limitation of this study is its absence of immunological measurements. Standard immunogenicity measures following receipt of live attenuated influenza vaccine have lacked correlation with efficacy, and clinical efficacy studies are the standard for licensure and vaccine policy determination.21, 22 One study described cell-mediated immunity as determined by ELISPOT assays that measure γ interferon as correlating with protection following live attenuated influenza vaccine in children; however, the results have not been corroborated and this measurement is difficult in remote field settings and among children.23 A recent study of children aged 2–9 years further highlights the difficulty in identifying the immunological basis for protection by live attenuated influenza vaccine;24 the study could not identify such a mechanism in a live attenuated influenza vaccine challenge model. Regardless, had we done mucosal or serum immunologic assays, we might at least have been able to confirm that vaccinees were responding immunologically in some way to live attenuated influenza vaccine. Another limitation is that we did not assess the protection of two doses of this vaccine. Live attenuated influenza vaccines based on Ann Arbor master donor viruses (FluMist/Fluenz) are licensed beginning at 2 years of age, and two doses are recommended for children receiving influenza vaccine for the first time.25 The Russian-derived vaccines are licensed for single dose administration.

es of this vaccine. Live attenuated influenza vaccines based on Ann Arbor master donor viruses (FluMist/Fluenz) are licensed beginning at 2 years of age, and two doses are recommended for children receiving influenza vaccine for the first time.25 The Russian-derived vaccines are licensed for single dose administration. Although comparative studies of one versus two doses of live attenuated influenza vaccines based on Russian-derived master donor viruses have not been done, studies of the Ann Arbor-based vaccines suggest that efficacy can be achieved after one dose, and might be improved by a second dose.8, 26, 27 However, an observational study in a broadly aged population in India found that Serum Institute of India's monovalent H1N1pdm09 vaccine had high effectiveness with only one dose.28 Although only half of the young children in this study in Senegal were influenza vaccine naive, whether a second dose of Nasovac-S could have provided efficacy to those either naive or previously exposed to influenza vaccine is uncertain.

India's monovalent H1N1pdm09 vaccine had high effectiveness with only one dose.28 Although only half of the young children in this study in Senegal were influenza vaccine naive, whether a second dose of Nasovac-S could have provided efficacy to those either naive or previously exposed to influenza vaccine is uncertain. Several observational studies of FluMist in the USA in 2013–14 reported reduced effectiveness against the current H1N1pdm09 among young children aged 2–8 years, with effectiveness against B preserved.29 Unfortunately, in view of the absence of sufficient circulating H3N2 and matched B-lineage in Senegal, we cannot determine whether the absence of efficacy in this study was strain-specific or more generalised. One hypothesis for the US findings in 2013–14 is that the H1N1pdm09 strain used for vaccine manufacture might not have been optimal because it contained a glutamic acid at position 47 (E47) in the haemagglutinin stalk that made it less thermally stable and infectious in ferrets than A/H1N1 (2009) circulating globally today.30 In early 2015, the manufacturer of FluMist reported that it would replace the A/California (H1N1) strain with an antigenically similar strain with a more stable haemagglutinin.29 However, the US CDC's effectiveness study for 2015–16, a season when H1N1pdm09 predominated,31 again showed no effectiveness of live attenuated influenza vaccine in mid-year analyses, leading the US Advisory Committee on Immunization Practices to recommend that the live attenuated influenza vaccine not be used in the USA for 2016–17.32 Confusing the issue further is that findings in other studies, other countries, and in other years have shown variable levels of effectiveness of live attenuated influenza vaccine against H1N1pdm09.33 Serum Institute of India's live attenuated influenza vaccine used in Senegal and the parallel Bangladesh trial13 also contained the E47 residue, so it is unclear how this contributed to the finding of no efficacy against H1N1pdm09 in Senegal given that efficacy was demonstrated with this vaccine in the Bangladesh trial. Finally, although standard antigenic techniques indicated circulating H1N1pdm09 was matched to vaccine, we did not sequence isolates; however, a study in Canada showed that effectiveness was preserved for current sequence changes hypothesised to result in reduced match to vaccine.34

his vaccine in the Bangladesh trial. Finally, although standard antigenic techniques indicated circulating H1N1pdm09 was matched to vaccine, we did not sequence isolates; however, a study in Canada showed that effectiveness was preserved for current sequence changes hypothesised to result in reduced match to vaccine.34 The vaccine was safe and well tolerated, with fever and systemic symptoms being uncommon in the 7 days after vaccination. Local symptoms, including nasal congestion, runny nose, and cough, were common among both vaccine and placebo recipients, with no imbalances between groups. These safety results are similar to those reported in Bangladesh during both a phase 2 study and the aforementioned efficacy trial there.13, 35 In view of the association of wheezing and admissions to hospital noted among children younger than 2 years who were vaccinated with FluMist,36 we carefully monitored participants for wheezing illness using a standardised definition. Incidence of wheezing was overall low in this population, and we did not observe differences between the vaccine and placebo groups. These safety findings are reassuring, particularly because prelicensure trials in Russia pre-dated the FluMist finding and were not designed to identify wheezing illness as a solicited event.

dence of wheezing was overall low in this population, and we did not observe differences between the vaccine and placebo groups. These safety findings are reassuring, particularly because prelicensure trials in Russia pre-dated the FluMist finding and were not designed to identify wheezing illness as a solicited event. Our study corroborates several years of influenza surveillance in this rural community in Senegal, which show that influenza illness is common in young children.9, 37, 38, 39 Furthermore, in most years in this tropical setting, multiple strains circulate and for extended periods. During the period of this study, vaccine-mismatched influenza B was the predominant circulating strain, supporting the need to further investigate the use of quadrivalent vaccines containing both B lineages that are already available in many parts of the world. Likewise, two doses of live attenuated influenza vaccine might be necessary for young children in Senegal and elsewhere, recognising that a two-dose schedule would be costlier and less feasible in low-resource areas of the world.

vaccines containing both B lineages that are already available in many parts of the world. Likewise, two doses of live attenuated influenza vaccine might be necessary for young children in Senegal and elsewhere, recognising that a two-dose schedule would be costlier and less feasible in low-resource areas of the world. The Senegal trial of live attenuated influenza vaccine and its Bangladesh companion contribute essential data for the protection afforded to children by live attenuated influenza vaccines, especially in low-resource settings. Currently, active global discussion on the effectiveness of live attenuated influenza vaccines, especially for the current H1N1pdm09 strain, is ongoing and policy makers are basing decisions on differing results from an array of observational studies in the USA, Canada, Europe, and elsewhere. Although the Bangladesh efficacy trial provides confidence that this live attenuated influenza vaccine from India can provide protection with a single dose, our findings from Senegal are in line with recent US CDC studies in which live attenuated influenza vaccine failed to provide protection against H1N1pdm09. Further studies are needed to assess whether our findings are specific to the H1N1pdm09 strain and whether two doses might be efficacious, and to better understand the performance of live attenuated influenza vaccines in diverse paediatric populations for whom the burden of respiratory disease remains unacceptably high. For WHO Flunet data see http://www.who.int/influenza/gisrs_laboratory/flunet/charts/en/ Supplementary Material Supplementary appendix

The Senegal trial of live attenuated influenza vaccine and its Bangladesh companion contribute essential data for the protection afforded to children by live attenuated influenza vaccines, especially in low-resource settings. Currently, active global discussion on the effectiveness of live attenuated influenza vaccines, especially for the current H1N1pdm09 strain, is ongoing and policy makers are basing decisions on differing results from an array of observational studies in the USA, Canada, Europe, and elsewhere. Although the Bangladesh efficacy trial provides confidence that this live attenuated influenza vaccine from India can provide protection with a single dose, our findings from Senegal are in line with recent US CDC studies in which live attenuated influenza vaccine failed to provide protection against H1N1pdm09. Further studies are needed to assess whether our findings are specific to the H1N1pdm09 strain and whether two doses might be efficacious, and to better understand the performance of live attenuated influenza vaccines in diverse paediatric populations for whom the burden of respiratory disease remains unacceptably high. For WHO Flunet data see http://www.who.int/influenza/gisrs_laboratory/flunet/charts/en/ Supplementary Material Supplementary appendix Acknowledgments This work was funded through a Cooperative Agreement to PATH from the US Centers for Disease Control and Prevention (U01IP000476). Supplementary funding from the Bill & Melinda Gates Foundation to PATH (OPP1017334) supported statistical analyses. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US CDC, Gates Foundation, or PATH. Our sincere thanks to all the families who participated in this trial and to the full research teams at Institut de Recherche pour le Développement and Institut Pasteur de Dakar in Senegal. We are most grateful to Serum Institute of India Ltd for donating the masked vaccine and placebo used in this study. We thank Noelle Benzekri of the University of Washington (Seattle, WA, USA) for assisting in monitoring and training of study physicians and Xiyan Xu of CDC for antigenic characterisation of submitted influenza-positive specimens. Supporting PATH in fulfilling its sponsor obligations, the Agence Africaine De Recherche en Sante Humaine (AARSH) conducted site monitoring, and FHI360 conducted data management. Finally, we are indebted to the members of our Safety Monitoring Committee, Samba Sow, Margaret Rennels, and Kathryn Edwards, who generously volunteered their time to provide independent safety review and oversight for this trial.

en Sante Humaine (AARSH) conducted site monitoring, and FHI360 conducted data management. Finally, we are indebted to the members of our Safety Monitoring Committee, Samba Sow, Margaret Rennels, and Kathryn Edwards, who generously volunteered their time to provide independent safety review and oversight for this trial. Contributors KMN conceived the study and KMN, JRO, KDCL, JCV, AD, MNN, and M-AW designed the trial. KDCL, JRO, JF, JCV, KMN, AD, BD, MNN, M-AW, and KEL developed study methods or data collection instruments. AD and BD collected the data and biological specimens. JRO and KMN served as the medical monitors. KDCL, assisted by JF, directed the site monitors, and KDCL and JF assisted the investigator with site management. ND did the rRT-PCR testing in Senegal and SLE did the rRT-PCR testing in Atlanta, USA. KDCL and JF oversaw and coordinated data management, respectively. RH, KDCL, and JCV planned the statistical analyses. RH did the statistical analyses, and KDCL and JF verified their accuracy. KDCL managed compiling of the official clinical study report. AD served as the Principal Investigator in Senegal. MNN directed testing at the Senegal National Influenza Laboratory. JCV and KMN directed the team at PATH. All authors had full access to the data, took part in meetings to discuss and interpret the results, drafted or critically revised the report, and approved its final version. Declarations of interests We declare no competing interests. Figure 1 Trial profile LAIV=live attenuated influenza vaccine.

Contributors KMN conceived the study and KMN, JRO, KDCL, JCV, AD, MNN, and M-AW designed the trial. KDCL, JRO, JF, JCV, KMN, AD, BD, MNN, M-AW, and KEL developed study methods or data collection instruments. AD and BD collected the data and biological specimens. JRO and KMN served as the medical monitors. KDCL, assisted by JF, directed the site monitors, and KDCL and JF assisted the investigator with site management. ND did the rRT-PCR testing in Senegal and SLE did the rRT-PCR testing in Atlanta, USA. KDCL and JF oversaw and coordinated data management, respectively. RH, KDCL, and JCV planned the statistical analyses. RH did the statistical analyses, and KDCL and JF verified their accuracy. KDCL managed compiling of the official clinical study report. AD served as the Principal Investigator in Senegal. MNN directed testing at the Senegal National Influenza Laboratory. JCV and KMN directed the team at PATH. All authors had full access to the data, took part in meetings to discuss and interpret the results, drafted or critically revised the report, and approved its final version. Declarations of interests We declare no competing interests. Figure 1 Trial profile LAIV=live attenuated influenza vaccine. Figure 2 Influenza circulation in Niakhar, by type and subtype or lineage, from week 22 to week 51 during the period of the trial

Contributors KMN conceived the study and KMN, JRO, KDCL, JCV, AD, MNN, and M-AW designed the trial. KDCL, JRO, JF, JCV, KMN, AD, BD, MNN, M-AW, and KEL developed study methods or data collection instruments. AD and BD collected the data and biological specimens. JRO and KMN served as the medical monitors. KDCL, assisted by JF, directed the site monitors, and KDCL and JF assisted the investigator with site management. ND did the rRT-PCR testing in Senegal and SLE did the rRT-PCR testing in Atlanta, USA. KDCL and JF oversaw and coordinated data management, respectively. RH, KDCL, and JCV planned the statistical analyses. RH did the statistical analyses, and KDCL and JF verified their accuracy. KDCL managed compiling of the official clinical study report. AD served as the Principal Investigator in Senegal. MNN directed testing at the Senegal National Influenza Laboratory. JCV and KMN directed the team at PATH. All authors had full access to the data, took part in meetings to discuss and interpret the results, drafted or critically revised the report, and approved its final version. Declarations of interests We declare no competing interests. Figure 1 Trial profile LAIV=live attenuated influenza vaccine. Figure 2 Influenza circulation in Niakhar, by type and subtype or lineage, from week 22 to week 51 during the period of the trial The graph is a stacked column chart where numbers of positive detections for each strain each week are stacked in the graph and can be visually summed. Study vaccinations occurred from week 22 to week 28. Laboratory testing data from the trial indicated that B circulation was of mixed lineage from week 22 through 35 but became almost only Victoria lineage (unmatched to vaccine) thereafter. Inset show influenza circulation in Senegal, by type and subtype, during the entire year of the trial, 2013, as measured by the Senegal National Influenza Center. Note: determination of B lineage was not standard practice for national surveillance in 2013. Data from WHO Flunet.

ineage (unmatched to vaccine) thereafter. Inset show influenza circulation in Senegal, by type and subtype, during the entire year of the trial, 2013, as measured by the Senegal National Influenza Center. Note: determination of B lineage was not standard practice for national surveillance in 2013. Data from WHO Flunet. Table 1 Participant completion rates and baseline characteristics

ineage (unmatched to vaccine) thereafter. Inset show influenza circulation in Senegal, by type and subtype, during the entire year of the trial, 2013, as measured by the Senegal National Influenza Center. Note: determination of B lineage was not standard practice for national surveillance in 2013. Data from WHO Flunet. Table 1 Participant completion rates and baseline characteristics Live attenuated influenza vaccine (n=1174) Placebo (n=587) Study completion Followed up to Dec 20, 2013 1148 (98%) 574 (98%) Lost to follow-up 10 (1%) 7 (1%) Temporary migration 14 (1%) 6 (1%) Died 2 (<1%) 0 Age, months 47·7 (13·2) 47·3 (13·5) Age group (years) 2–<3 278 (24%) 150 (26%) 3–<4 310 (26%) 157 (27%) 4–<5 323 (28%) 145 (25%) 5–<6 263 (22%) 135 (23%) Sex Male 610 (52%) 297 (51%) Female 564 (48%) 290 (49%) Height and weight Height (cm) 98·2 (9·5) 98·0 (9·4) Weight (kg) 14·4 (2·7) 14·3 (2·8) Underweight: weight-for-age malnutrition grade* None 641 (55%) 319 (54%) Mild 333 (28%) 166 (28%) Moderate 162 (14%) 75 (13%) Severe 38 (3%) 27 (5%) Stunting: height-for-age malnutrition grade* None 608 (52%) 311 (53%) Mild 360 (31%) 176 (30%) Moderate 153 (13%) 72 (12%) Severe 53 (5%) 28 (5%) Wasting: weight-for-height malnutrition grade* None 604 (51%) 289 (49%) Mild 192 (16%) 98 (17%) Moderate 98 (8%) 54 (9%) Severe 41 (4%) 21 (4%) Unknown† 239 (20%) 125 (21%) History of chronic illness, including asthma or wheezing illness‡ 1 (<1%) 0 Receipt of trivalent inactivated influenza vaccine in previous field trials at site Never 580 (49%) 285 (49%) In mid-2009 (containing A/Brisbane/59/2007 [H1N1]-like virus) 65 (6%) 24 (4%) In mid-2010 or mid-2011 (containing A/California/7/2009 [H1N1]-like virus) 512 (44%) 272 (46%) Previous receipt could not be determined 17 (1%) 6 (1%) Receipt of oral polio vaccine in past 30 days Yes 817 (70%) 413 (70%) No 357 (30%) 174 (30%) Data are n (%) or mean (SD).

bane/59/2007 [H1N1]-like virus) 65 (6%) 24 (4%) In mid-2010 or mid-2011 (containing A/California/7/2009 [H1N1]-like virus) 512 (44%) 272 (46%) Previous receipt could not be determined 17 (1%) 6 (1%) Receipt of oral polio vaccine in past 30 days Yes 817 (70%) 413 (70%) No 357 (30%) 174 (30%) Data are n (%) or mean (SD). * Based on Z score: mild (−2 to <–1), moderate (−3 to <–2), severe (<–3). † Weight for height Z score cannot be calculated for children 60 months or older and so is categorised as unknown. ‡ One child in the live attenuated influenza vaccine group reported a history of asthma. Table 2 Symptomatic laboratory-confirmed influenza outcomes among children aged 2–5 years in Senegal receiving live-attenuated influenza vaccine or placebo and estimated vaccine efficacy

† Weight for height Z score cannot be calculated for children 60 months or older and so is categorised as unknown. ‡ One child in the live attenuated influenza vaccine group reported a history of asthma. Table 2 Symptomatic laboratory-confirmed influenza outcomes among children aged 2–5 years in Senegal receiving live-attenuated influenza vaccine or placebo and estimated vaccine efficacy Per-protocol population Total vaccinated cohort Live attenuated influenza vaccine (n=1173) Placebo (n=584) Vaccine efficacy (95% CI)* Live attenuated influenza vaccine (n=1174) Placebo (n=587) Vaccine efficacy (95% CI)* Primary virological endpoint All strains* 210 (18%) 105 (18%) 0·0% (−26·4 to 20·9) 225 (19%) 107 (18%) −6·5% (−34·1 to 15·4) Secondary virological endpoint All vaccine-matched strains† 100 (9%) 47 (8%) −6·1% (−50·0 to 25·0) 114 (10%) 47 (8%) −22·6% (−72·2 to 12·7) H1N1 79 (7%) 36 (6%) −9·7% (−62·6 to 26·1) 79 (7%) 36 (6%) −9·9% (−62·9 to 25·9) H3N2‡ 3 (<1%) 0 .. 10 (1%) 0 .. B (Yamagata lineage, matched to vaccine) 20 (2%) 11 (2%) 9·5% (−88·9 to 56·6) 28 (2%) 11 (2%) −27·7% (−156·5 to 36·4) B (Victoria lineage, unmatched to vaccine) 115 (10%) 62 (11%) 7·3% (−26·3 to 31·9) 118 (10%) 64 (11%) 7·7% (−25·1 to 31·9) Data are n (%). * 1 minus the relative rate × 100%; relative rate was estimated using a Cox proportional hazards model. † Only the first laboratory-confirmed influenza-like illness occurring more than 14 days postvaccination was counted in the all strains and all vaccine-matched strains analyses.

Per-protocol population Total vaccinated cohort Live attenuated influenza vaccine (n=1173) Placebo (n=584) Vaccine efficacy (95% CI)* Live attenuated influenza vaccine (n=1174) Placebo (n=587) Vaccine efficacy (95% CI)* Primary virological endpoint All strains* 210 (18%) 105 (18%) 0·0% (−26·4 to 20·9) 225 (19%) 107 (18%) −6·5% (−34·1 to 15·4) Secondary virological endpoint All vaccine-matched strains† 100 (9%) 47 (8%) −6·1% (−50·0 to 25·0) 114 (10%) 47 (8%) −22·6% (−72·2 to 12·7) H1N1 79 (7%) 36 (6%) −9·7% (−62·6 to 26·1) 79 (7%) 36 (6%) −9·9% (−62·9 to 25·9) H3N2‡ 3 (<1%) 0 .. 10 (1%) 0 .. B (Yamagata lineage, matched to vaccine) 20 (2%) 11 (2%) 9·5% (−88·9 to 56·6) 28 (2%) 11 (2%) −27·7% (−156·5 to 36·4) B (Victoria lineage, unmatched to vaccine) 115 (10%) 62 (11%) 7·3% (−26·3 to 31·9) 118 (10%) 64 (11%) 7·7% (−25·1 to 31·9) Data are n (%). * 1 minus the relative rate × 100%; relative rate was estimated using a Cox proportional hazards model. † Only the first laboratory-confirmed influenza-like illness occurring more than 14 days postvaccination was counted in the all strains and all vaccine-matched strains analyses. ‡ H3N2: none of the three live attenuated influenza vaccine cases have been tested for live attenuated influenza vaccine vs wild-type virus. One case occurred 15 days postvaccination and the other two cases, one of which could not be confirmed by the WHO Collaborating Center at the US CDC as H3N2, occurred more than 100 days postvaccination. The seven additional H3N2 cases added in the total vaccinated cohort occurred during the 14-day postvaccination period; six of these were confirmed to be live attenuated influenza vaccine type influenza A.

t be confirmed by the WHO Collaborating Center at the US CDC as H3N2, occurred more than 100 days postvaccination. The seven additional H3N2 cases added in the total vaccinated cohort occurred during the 14-day postvaccination period; six of these were confirmed to be live attenuated influenza vaccine type influenza A. Table 3 Detection of vaccine virus on day 2 and day 4 postvaccination among vaccine recipients in the vaccine infectivity subset Day 2 Day 4* Either day LAIV-A/H1N1 12/65 (19%) 3/66 (5%) 14/65 (22%) LAIV-A/H3N2 31/65 (48%) 18/66 (27%) 34/65 (52%) LAIV-B 34/65† (52%) 28/65 (43%) 42/64 (66%) Any 48/65 (74%) 39/66 (59%) 55/66 (83%) Data are n/total n (%). Specimens test-inconclusive for a particular strain were removed from the denominator of the respective table cell, except for the Any row where the specimen is counted in the denominator if it was test-positive for one of the other three vaccine strains. LAIV=live attenuated influenza vaccine. * On day 4, three specimens had LAIV-A detected but subtype was undetermined. † On day 2, four specimens were positive using the seasonal influenza B primers but negative using the LAIV-type B primers. On day 4, one of these four was then negative using both seasonal and LAIV-B primers, and the other three were then positive using both seasonal and LAIV-B primers. Only one participant in the placebo group (n=34) was ever positive for influenza, with only that participant's day 0 specimen testing positive for seasonal influenza B only.

hese four was then negative using both seasonal and LAIV-B primers, and the other three were then positive using both seasonal and LAIV-B primers. Only one participant in the placebo group (n=34) was ever positive for influenza, with only that participant's day 0 specimen testing positive for seasonal influenza B only. Table 4 Local and systemic reactions in the first 7 days after vaccination and protocol-defined wheezing illness occurring throughout the trial, total vaccinated cohort

hese four was then negative using both seasonal and LAIV-B primers, and the other three were then positive using both seasonal and LAIV-B primers. Only one participant in the placebo group (n=34) was ever positive for influenza, with only that participant's day 0 specimen testing positive for seasonal influenza B only. Table 4 Local and systemic reactions in the first 7 days after vaccination and protocol-defined wheezing illness occurring throughout the trial, total vaccinated cohort Live attenuated influenza vaccine (n=1171) Placebo (n=587) Mild Moderate Severe All* Mild Moderate Severe All* Local and systemic reactions in first 7 days following vaccination Fever (measured ≥38°C) 2 (0·2%) 2 (0·2%) 1 (0·1%) 5 (0·4%) 1 (0·2%) 3 (0·5%) 0 (0·0%) 4 (0·7%) Nasal congestion 41 (3·5%) 0 (0·0%) 0 (0·0%) 41 (3·5%) 16 (2·7%) 0 (0·0%) 0 (0·0%) 16 (2·7%) Runny nose 199 (17·0%) 2 (0·2%) 0 (0·0%) 201 (17·1%) 95 (16·2%) 0 (0·0%) 0 (0·0%) 95 (16·2%) Stuffy nose 18 (1·5%) 0 (0·0%) 0 (0·0%) 18 (1·5%) 9 (1·5%) 0 (0·0%) 0 (0·0%) 9 (1·5%) Cough 110 (9·4%) 4 (0·3%) 0 (0·0%) 114 (9·7%) 58 (9·9%) 0 (0·0%) 0 (0·0%) 58 (9·9%) Sore throat 4 (0·3%) 0 (0·0%) 0 (0·0%) 4 (0·3%) 4 (0·7%) 1 (0·2%) 0 (0·0%) 5 (0·9%) Ear pain 6 (0·5%) 1 (0·1%) 0 (0·0%) 7 (0·6%) 2 (0·3%) 0 (0·0%) 0 (0·0%) 2 (0·3%) Headache 14 (1·2%) 0 (0·0%) 0 (0·0%) 14 (1·2%) 7 (1·2%) 0 (0·0%) 0 (0·0%) 7 (1·2%) Vomiting 11 (0·9%) 1 (0·1%) 0 (0·0%) 12 (1·0%) 6 (1·0%) 0 (0·0%) 0 (0·0%) 6 (1·0%) Chills 3 (0·3%) 0 (0·0%) 0 (0·0%) 3 (0·3%) 1 (0·2%) 0 (0·0%) 0 (0·0%) 1 (0·2%) Irritability/decreased activity 14 (1·2%) 2 (0·2%) 0 (0·0%) 16 (1·4%) 7 (1·2%) 1 (0·2%) 0 (0·0%) 8 (1·4%) Muscle/joint pain 3 (0·3%) 0 (0·0%) 0 (0·0%) 3 (0·3%) 1 (0·2%) 0 (0·0%) 0 (0·0%) 1 (0·2%) Tachypnoea (≥40 breaths per min) 1 (0·1%) 0 (0·0%) 0 (0·0%) 1 (0·1%) 0 (0·0%) 1 (0·2%) 0 (0·0%) 1 (0·2%) Protocol-defined wheezing illness by study period Days 0–7 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) Days 8–42 0 (0·0%) 0 (0·0%) 0 (0·0%) 0 (0·0%) 1 (0·2%) 0 (0·0%) 0 (0·0%) 1 (0·2%) Day 43–6+ months 3 (0·3%) 1 (0·1%) 0 (0·0%) 4 (0·3%) 2 (0·3%) 2 (0·3%) 0 (0·0%) 4 (0·7%) Anytime 3 (0·3%) 1 (0·1%) 0 (0·0%) 4 (0·3%) 3 (0·5%) 2 (0·3%) 0 (0·0%) 5 (0·9%) * No significant differences using Fisher's exact test (two-sided p value was never <0·05) between live attenuated influenza vaccine and placebo for events of any severity.

Introduction 2015 marked the end of the Millennium Development Goal (MDG) era and the beginning of the implementation of the Sustainable Development Goals (SDGs). China represents a unique example of success in achieving the MDG 4 target of reducing under-5 mortality rate by two-thirds between 1990 and 2015. Based on estimates published by the UN Interagency Group on Child Mortality Estimation (UN IGME), China's under-5 mortality rate declined from 53·8 to 10·7 per 1000 livebirths at an average annual rate of reduction (ARR) of 6·5% in 1990–2015, faster than the ARR needed to achieve the MDG 4 target of 4·4%.1 In fact, China achieved the MDG 4 target in 2009, 6 years ahead of time. China is also one of about only a dozen countries that achieved a faster decline of mortality in neonates than in children aged 1–59 months since 1990.1, 2 In this Article, we estimated the national and subnational levels and causes of child mortality in China annually in 1996–2015, and attempted to draw implications for achievement of the SDGs for China and other low-income and middle-income countries (LMICs). We updated our approach to estimating causes of child deaths in China from the previous single-cause models based on published data3 to use of adjusted empirical data from the Maternal and Child Health Surveillance System (MCHSS). Research in context Evidence before this study

In this Article, we estimated the national and subnational levels and causes of child mortality in China annually in 1996–2015, and attempted to draw implications for achievement of the SDGs for China and other low-income and middle-income countries (LMICs). We updated our approach to estimating causes of child deaths in China from the previous single-cause models based on published data3 to use of adjusted empirical data from the Maternal and Child Health Surveillance System (MCHSS). Research in context Evidence before this study We searched multiple databases such as PubMed, Embase, ISIS Web of Knowledge, Popline, and LILACS for studies that reported causes of deaths in children younger than 5 years in China published up to Feb 12, 2015. There were no language restrictions. Twelve eligible studies that reported causes of death in China for children younger than 5 years were identified. Two studies were nationally representative, reporting causes of deaths in children younger than 5 years for some years in 1996–2010. Ten studies reported data only for selected years and subnational geographic areas. In 2015, we published the county level under-5 mortality estimates for China for the years 1996–2012. We used data from the Maternal and Child Health Surveillance System (MCHSS), censuses, surveys, surveillance sites, and disease surveillance points. We also published in 2010 causes of child mortality for China in 2008, in which input data for the estimation were based on systematic review of peer-reviewed publications. Added value of this study

We searched multiple databases such as PubMed, Embase, ISIS Web of Knowledge, Popline, and LILACS for studies that reported causes of deaths in children younger than 5 years in China published up to Feb 12, 2015. There were no language restrictions. Twelve eligible studies that reported causes of death in China for children younger than 5 years were identified. Two studies were nationally representative, reporting causes of deaths in children younger than 5 years for some years in 1996–2010. Ten studies reported data only for selected years and subnational geographic areas. In 2015, we published the county level under-5 mortality estimates for China for the years 1996–2012. We used data from the Maternal and Child Health Surveillance System (MCHSS), censuses, surveys, surveillance sites, and disease surveillance points. We also published in 2010 causes of child mortality for China in 2008, in which input data for the estimation were based on systematic review of peer-reviewed publications. Added value of this study In this study, we updated and improved our estimates from the previous two papers on levels and causes of child mortality, specifically to 1996–2015, to present China's progress in improving child survival in the era of the Millennium Development Goals. We introduce in detail the MCHSS, which is regarded as the highest quality data source among all sources used previously. We adjusted the empirical estimates from MCHSS, with the adjustment and estimation methods being largely independent of the previous estimation approaches. We found that estimates presented in this Article agree in general with our previous estimates for the same years. Additionally, we highlighted areas where more research is needed to understand the key factors that contributed to China's progress in most of the MDG period. Our research has implications for China and other low-income and middle-income countries (LMICs) to achieve the Sustainable Development Goals.

or the same years. Additionally, we highlighted areas where more research is needed to understand the key factors that contributed to China's progress in most of the MDG period. Our research has implications for China and other low-income and middle-income countries (LMICs) to achieve the Sustainable Development Goals. Implications of all the available evidence China has made impressive and widespread progress in improving child survival in 1996–2015, and the establishment and evolution of the MCHSS have been important in documenting this progress. Establishment of the sample registration system and its improvement over time could be useful for other LMICs that do not have quality vital and sample registration systems. However, more rigorous research is needed to fully understand the effects of China's child-survival policy and programmes implemented during this period, which could be adopted by other LMICs. Methods Study design and participants MCHSS is a sample registration system collecting vital statistics on levels and causes of maternal and child mortality and data on congenital abnormalities. The system is designed to be nationally and regionally representative. A child mortality surveillance network was first established in 1991 by the Ministry of Health in China. In 1996, child mortality surveillance was combined with the maternal mortality and congenital abnormality surveillance networks to form MCHSS. Since then, MCHSS has been generating estimates on levels and causes of child mortality at the national and regional level.4, 5

ished in 1991 by the Ministry of Health in China. In 1996, child mortality surveillance was combined with the maternal mortality and congenital abnormality surveillance networks to form MCHSS. Since then, MCHSS has been generating estimates on levels and causes of child mortality at the national and regional level.4, 5 In 1991, the child mortality surveillance network covered a population of 8·5 million in 81 cities or counties. When MCHSS was established in 1996, the system was expanded to 116 surveillance sites covering 37 urban districts and 79 rural counties in 31 provinces, autonomous regions, and municipalities, monitoring a total population of 12·7 million in mainland China.6 A major expansion of MCHSS was initiated in 2006 because of rapidly declining rates of maternal and child mortality. Training and quality assessments were done in new sites in 2006–08. Data from new sites have been incorporated since 2009. After the expansion, MCHSS came to its current structure in 2013, covering 334 sites representing 124 urban districts and 210 rural counties with a surveillance population of 47·1 million (figure 1). Additional details of the MCHSS evolution are in the appendix (p 3).

a from new sites have been incorporated since 2009. After the expansion, MCHSS came to its current structure in 2013, covering 334 sites representing 124 urban districts and 210 rural counties with a surveillance population of 47·1 million (figure 1). Additional details of the MCHSS evolution are in the appendix (p 3). MCHSS has a multistage, stratified, clustered sampling design; stratification changed over time.6, 7 The sampling design and sample size calculation before 2006 are in the appendix (pp 4–5). When MCHSS expanded in 2006, the sample size was calculated based on infant mortality from MCHSS in 2005, crude birth rate from the National Bureau of Statistics in 2005, total population from the 2000 national census, and a design effect of 2·0 to adjust for correlation of events within the sampled areas (see appendix p 6 for more details).

2006, the sample size was calculated based on infant mortality from MCHSS in 2005, crude birth rate from the National Bureau of Statistics in 2005, total population from the 2000 national census, and a design effect of 2·0 to adjust for correlation of events within the sampled areas (see appendix p 6 for more details). The sampling was also updated in 2006. The country was divided into eastern, central, and western regions, and each region was further stratified into urban and rural areas (appendix p 7). The urban or rural classification was derived from criteria used in the 1993 National Health Services Survey and the 2006 administrative division codes published by the National Bureau of Statistics.8, 9 MCHSS is thus regarded to be representative of the six region-residency strata: eastern urban, eastern rural, central urban, central rural, western urban, and western rural areas, and of the aggregates of the six strata. The original surveillance sites remained in the sample. New sites added in 2006 were randomly sampled from urban neighbourhoods and rural townships covering 10% of the total population of the newly sampled urban districts and rural counties.

stern urban, and western rural areas, and of the aggregates of the six strata. The original surveillance sites remained in the sample. New sites added in 2006 were randomly sampled from urban neighbourhoods and rural townships covering 10% of the total population of the newly sampled urban districts and rural counties. All children under 5 years of age, living in the surveillance sites and all livebirths of mothers who are either permanent residents of the sites or have lived in the sites for at least 1 year are included in MCHSS. There were no exclusion criteria. A livebirth was defined as a fetus of at least 28 weeks' gestation (or with birthweight of more than 1000 g if the gestational age was unknown), and showing any of the following signs of life after separation from his or her mother: heartbeat, breathing, umbilical cord pulsation, or voluntary muscle contraction.6 Livebirths were recorded by community health workers in urban areas or village doctors in rural areas and reported to neighbourhood (urban) or township (rural) Maternal and Child Health Centers (MCHCs) on a monthly basis. The information was compiled by district (urban) or county (rural) MCHCs and reported to higher (than district or county) level (city, prefecture, and province) MCHCs on a quarterly basis.

rural areas and reported to neighbourhood (urban) or township (rural) Maternal and Child Health Centers (MCHCs) on a monthly basis. The information was compiled by district (urban) or county (rural) MCHCs and reported to higher (than district or county) level (city, prefecture, and province) MCHCs on a quarterly basis. Outcomes The recording and reporting of deaths and causes of deaths were implemented in both communities and health facilities. Details of the reporting and cause ascertainment have been published elsewhere.10 Briefly, deaths occurring in communities were recorded and reported to township MCHCs within 10 days of the events by village doctors. Deaths and causes of deaths were then investigated, ascertained, and reported using the Child Death Registration Card (appendix pp 8–9) by township MCH physicians within the next 7 days. Cause ascertainment was based on death certificates of children who died in health facilities, the last clinical diagnosis if children were discharged from health facilities within 1 month of death and died on their way home or at home, or verbal autopsy if there was no contact with the health-care system (the distribution of deaths by cause ascertaining method by age and year are in the appendix [p 10]). Verbal autopsy was done using non-standardised instruments before 2012, and the 2012 WHO verbal autopsy instrument and physician review from 2012 onwards.11 Deaths occurring in health facilities were recorded on the Child Death Registration Card on site and reported quarterly to district or county MCHCs. Information on deaths and causes of deaths was then compiled by district or county MCHCs and crosschecked between community and facility reports for duplicates or missing records on a quarterly basis. Health-facility reports not confirmed by community reports were further investigated to ensure they were eligible MCHSS cases. Once crosschecked, deaths and causes of deaths were reported to higher level MCHCs on a quarterly basis. Since 2007, internet-based reporting has been implemented in parallel with paper-based reporting and both systems are currently in use.6, 10, 12

ts were further investigated to ensure they were eligible MCHSS cases. Once crosschecked, deaths and causes of deaths were reported to higher level MCHCs on a quarterly basis. Since 2007, internet-based reporting has been implemented in parallel with paper-based reporting and both systems are currently in use.6, 10, 12 Quality control Coverage, completeness, and validity of reported livebirths, deaths, and causes of deaths were reviewed at the neighbourhood or township level quarterly, and at the district or county and higher levels annually. An annual quality control study is done by the MCHSS national office in two selected provinces.10 Provinces rotate through the quality control study over time. Three surveillance sites, including one urban and two rural sites from each province, are randomly selected. All child deaths occurring in those sites are reviewed by the MCHSS National Office. Data triangulation is done at all levels of MCHSS through cross-validation of reported births and deaths through multiple sources to identify missing events. These sources include local health facilities, family planning offices, Centers of Disease Control, and public security bureaus.6, 10, 12 Since 2010, the neonatal death audit has been implemented in MCHSS (see appendix p 11 for details).

of reported births and deaths through multiple sources to identify missing events. These sources include local health facilities, family planning offices, Centers of Disease Control, and public security bureaus.6, 10, 12 Since 2010, the neonatal death audit has been implemented in MCHSS (see appendix p 11 for details). Rates of under-reporting for livebirths and child deaths were calculated as the fraction of missed cases identified in the annual quality control study out of total cases captured (appendix p 11). The rate of under-reporting of livebirths was less than 1·5% in urban areas and less than 3% in rural areas in 1996–2015. Rates of under-reporting were less than 1% in the eastern region, less than 4% in the central region, and less than 3·5% in the western region. The rates of under-reporting of under-5 deaths were 10–30% at the national level, 10–25% in the eastern region, 12–35% in the central region, and 16–33% in the western region. Data analysis Before 2014, information on causes of deaths was collected and grouped into 35 categories in MCHSS (appendix p 9). These categories were mapped to aggregated codes of the International Classification of Diseases 10th version (ICD-10) during the analyses (appendix pp 12–14). Since 2014, causes of deaths have been directly coded using ICD-10 in MCHSS. Causes were further mapped and aggregated to the Child Health Epidemiology Reference Group categories in this study.13, 14

ted codes of the International Classification of Diseases 10th version (ICD-10) during the analyses (appendix pp 12–14). Since 2014, causes of deaths have been directly coded using ICD-10 in MCHSS. Causes were further mapped and aggregated to the Child Health Epidemiology Reference Group categories in this study.13, 14 In this study, MCHSS data are available annually for the period 1996–2015. We adjusted MCHSS empirical data to generate nationally and subnationally representative estimates of the numbers of deaths and mortality rates of deaths by cause. Specifically, we first adjusted the annual number of deaths by cause using a 3 year moving average of under-reporting rates by age group (0–6 days, 7–27 days, 28 days–5 months, 6–11 months, 12–23 months, and 24–59 months), type of residency (urban or rural), and region (eastern, central, or western). We calculated sampling probabilities based on best available estimate of the total population under surveillance and the region-residency census population (appendix p 15). For 1996, we based the total population on the 1990 national census, whereas the population under surveillance was based on the 1996 MCHSS. For 2000 and 2010, we based the total population on national censuses and the population under surveillance on MCHSS data in the corresponding years. The sampling probability ranged from 0·4% in rural central regions in 1996 to 6·6% in urban western regions in 2010. The sampling probabilities increased consistently in 2010, and more so in rural strata. We applied the sampling probabilities to the number of livebirths in MCHSS to derive livebirth estimates for each region-residency stratum such that the 1996 probabilities were applied to livebirths estimates for 1996–99, the 2000 probabilities were applied to 2000–09, and the 2010 probabilities were applied to 2010 onwards. We then normalised stratum-specific livebirths to the UN's estimates of livebirths for China15 at the national level proportionally to ensure global and regional comparability and smoothed stratum-specific proportions of livebirths using a 3 year moving average to obtain stable trends.

ies were applied to 2010 onwards. We then normalised stratum-specific livebirths to the UN's estimates of livebirths for China15 at the national level proportionally to ensure global and regional comparability and smoothed stratum-specific proportions of livebirths using a 3 year moving average to obtain stable trends. We calculated the total numbers of all-cause deaths by age-region-residency strata using region-residency-specific livebirths defined above, and a 3 year moving average of age-region-residency-specific all-cause mortality rates. We normalised the total numbers of stratified all-cause deaths to fit the total number of deaths of neonates and 1–59-month-old children generated by UN IGME for 1996–2015.1 Within each age-region-residency stratum, we smoothed the cause-specific mortality fractions using a 7 year moving average, with the highest weight given to the fourth year and the lowest and equal weights given to the first and seventh year. We then calculated the numbers of deaths by cause based on the smoothed cause-specific mortality fractions and age-region-residency-specific all-cause deaths.

y fractions using a 7 year moving average, with the highest weight given to the fourth year and the lowest and equal weights given to the first and seventh year. We then calculated the numbers of deaths by cause based on the smoothed cause-specific mortality fractions and age-region-residency-specific all-cause deaths. Deaths due to infrequent causes, including HIV or AIDS, pertussis, malaria, and measles, were challenging to capture in MCHSS. We used WHO programme estimates for these causes instead.2 We derived subnational estimates for these causes assuming the number of cause-specific deaths were distributed similarly to all-cause deaths across strata.2 Although the assumption could be problematic for localised infectious epidemic such as malaria, the total numbers of deaths due to these conditions were rather small (eg, 3–8 malaria deaths were estimated in 1996–2015). Therefore, biases associated with this assumption are likely to be small. The numbers and fractions of all the causes were normalised to all-cause deaths within each age-region-residency stratum. Subnational estimates were aggregated to derive regional and national estimates.

alaria deaths were estimated in 1996–2015). Therefore, biases associated with this assumption are likely to be small. The numbers and fractions of all the causes were normalised to all-cause deaths within each age-region-residency stratum. Subnational estimates were aggregated to derive regional and national estimates. Comparison with other estimates, uncertainty estimation, and transparency We compared our estimates with those produced by systematic analysis16 for the period 2009–15. The analysis16 was based on a systematic review of Chinese scientific literature that included all published studies meeting prespecified inclusion criteria on causes of child mortality,16 and was an update of our previous estimates applying single-cause models to derive cause-specific mortality fractions for the period 2000–08.3, 14 These estimates are based on nearly 300 studies published from a subset of MCHSS sites.1 Because input data are not exactly the same and the estimation strategies of the two approaches have limited overlap, the two sets of estimates are partially independent. We used accuracy of cause-specific mortality fractions17 to measure the population-level agreement between the two sets of estimates on cause distribution. Cause-specific mortality fraction accuracy is scaled between 0 and 1 (1 indicating perfect agreement).

overlap, the two sets of estimates are partially independent. We used accuracy of cause-specific mortality fractions17 to measure the population-level agreement between the two sets of estimates on cause distribution. Cause-specific mortality fraction accuracy is scaled between 0 and 1 (1 indicating perfect agreement). The uncertainty of estimated deaths by cause for each year, and the uncertainty in the time trend was estimated by incorporating resampling methods of the bootstrap.18 The stratified total number of surveillance sites was selected at random with replacement in each region over 1000 independent instances, building a distribution for possible observed causes of deaths in each stratum. The 2·5th and 97·5th percentiles of this bootstrapped distribution were used as the uncertainty bounds for the numbers of deaths by cause. This study conforms to the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) statement,19 which promotes transparency and replicability of global health estimates (see appendix p 16 for checklist). Additional details of the input data, estimation methodology including statistical codes, and estimates are online and publicly available through the Maternal and Child Epidemiology Estimations website. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All coauthors involved in the analyses had access to the data in the study and the corresponding authors had final responsibility for the decision to submit for publication.

funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All coauthors involved in the analyses had access to the data in the study and the corresponding authors had final responsibility for the decision to submit for publication. Results In 2015, 181 600 children died before the age of 5 in China, with 93 400 (51·5%, 95% uncertainty range [UR] 44·0–60·0) occurring in the neonatal period. At the national level, under-5 mortality rate fell by nearly 80% in the past two decades, from 50·8 deaths per 1000 livebirths in 1996 to 10·7 per 1000 livebirths in 2015 (figure 2A), resulting in an ARR of 8·2% per year (table 1). Neonatal mortality rate also declined substantially, from 25·7 deaths per 1000 livebirths in 1996 to 5·5 per 1000 livebirths in 2015. The rate of decline was faster in the first decade than in the second decade for under-5 mortality rate, but steadier for neonatal mortality rate (figure 2A).

of 8·2% per year (table 1). Neonatal mortality rate also declined substantially, from 25·7 deaths per 1000 livebirths in 1996 to 5·5 per 1000 livebirths in 2015. The rate of decline was faster in the first decade than in the second decade for under-5 mortality rate, but steadier for neonatal mortality rate (figure 2A). Child mortality rates showed large variation by region in 1996–2015 (figure 2B). The western region had the highest under-5 mortality rate and neonatal mortality rate in 2015, with an estimated under-5 mortality rate of 18·5 (95% UR 12·6–25·2) deaths per 1000 livebirths and neonatal mortality rate of 9·5 (6·8–13·7) per 1000 livebirths. The under-5 mortality rate in the eastern region was 5·8 deaths per 1000 livebirths and neonatal mortality rate in the eastern region was 3·1 per 1000 livebirths, similar to the rates in the USA and Canada. The geographical inequalities, represented by the ratio of under-5 mortality rate and neonatal mortality rate between western and eastern regions, fluctuated but generally increased during this period. For example, the ratio of under-5 mortality rate increased from 2·9:1 in 1996 to 4·4:1 in 2004, dropped to 2·3:1 in 2005–2010, then increased again to 3·2:1 in 2015. Since 1996, under-5 mortality rate in all three regions have been decreasing with an ARR of 7·5% or greater; the central region had the fastest ARR of 8·5% (95% UR 6·8–10·8; table 1).

er-5 mortality rate increased from 2·9:1 in 1996 to 4·4:1 in 2004, dropped to 2·3:1 in 2005–2010, then increased again to 3·2:1 in 2015. Since 1996, under-5 mortality rate in all three regions have been decreasing with an ARR of 7·5% or greater; the central region had the fastest ARR of 8·5% (95% UR 6·8–10·8; table 1). Inequity in child mortality between rural and urban areas was also apparent in China in 1996–2015 (figure 2C). In 2015, 81·1% (95% UR 80·9–81·4) of under-5 deaths occurred in rural China. This proportion reduced from 96·2% (93·5–96·5) in 1996, which is probably reflective of rapid urbanisation. As a result, despite a continued reduction in under-5 mortality rate in urban areas (by 56·5% [39·5–68·2] between 1996 and 2015), the actual number of under-5 deaths occurring in urban areas increased by 27·2% (9·7–40·6). In rural areas, mortality rates fell by more than 75% for neonates (77·5% [68·1–82·4]) and children aged under-5 years (77·8% [67·9–80·2]). The leading causes of under-5 mortality in 2015 in China were congenital abnormalities (19·7% [95% UR 18·1–21·5]), preterm birth complications (17·0% [14·9–19·7]), and injuries (14·6% [13·4–15·8]) (figure 3; table 2). The leading causes in neonates were preterm birth complications (16·0% [13·7–18·9] of under-5 deaths), intrapartum-related events (14·1% [12·1–16·6]), and congenital abnormalities (9·2% [8·0–11·1]). Among children aged 1–59 months, the leading causes were injuries (12·1% [10·6–13·5]), congenital abnormalities (10·4% [8·0–12·6]), and pneumonia (9·3% [6·9–11·0]).

birth complications (16·0% [13·7–18·9] of under-5 deaths), intrapartum-related events (14·1% [12·1–16·6]), and congenital abnormalities (9·2% [8·0–11·1]). Among children aged 1–59 months, the leading causes were injuries (12·1% [10·6–13·5]), congenital abnormalities (10·4% [8·0–12·6]), and pneumonia (9·3% [6·9–11·0]). The contribution of major infectious causes, such as pneumonia and diarrhoea, to all-cause under-5 deaths decreased from 1996 to 2015 (figure 4). Pneumonia was the leading cause of death in 1996, accounting for 22·6% (95% UR 20·8–25·3) of all-cause under-5 deaths (table 2). The contribution of pneumonia dropped to 12·2% (10·1–13·8), making it the 6th leading cause in 2015. The proportion of deaths due to diarrhoea decreased from 6·9% (5·2–7·8) in 1996 to 2·9% (2·1–3·5) in 2015. Conversely, the proportions of under-5 deaths due to congenital abnormalities increased from 9% (8·1–11·0) in 1996 to 20% (18·1–21·5) in 2015. Congenital abnormalities have become the leading cause of under-5 deaths since 2009. The contribution of preterm birth complications and intrapartum-related events were more stable in this period, ranging between 12·7% (12·4–15·7) and 19·2% (17·4–21·0).

s increased from 9% (8·1–11·0) in 1996 to 20% (18·1–21·5) in 2015. Congenital abnormalities have become the leading cause of under-5 deaths since 2009. The contribution of preterm birth complications and intrapartum-related events were more stable in this period, ranging between 12·7% (12·4–15·7) and 19·2% (17·4–21·0). In 1996–2015, cause-specific mortality rates decreased with striking, albeit varying ARRs (table 1). The causes of death with the most rapid decline in children younger than 5 years were infectious diseases, including neonatal tetanus, diarrhoea, and pneumonia, with ARRs of greater than 10% at the national level and across regions. All other causes decreased at ARRs greater than that needed to achieve MDG 4 (4·4%), with the exception of congenital abnormalities. The reduction of mortality due to congenital abnormalities was particularly slow in the central region (ARR 3·5% [95% UR 1·4–6·2]) and the western region (2·6% [0·0–6·2]).

across regions. All other causes decreased at ARRs greater than that needed to achieve MDG 4 (4·4%), with the exception of congenital abnormalities. The reduction of mortality due to congenital abnormalities was particularly slow in the central region (ARR 3·5% [95% UR 1·4–6·2]) and the western region (2·6% [0·0–6·2]). The cause-of-death composition also varied across age groups. Distributions of age-specific causes-of-death are shown in the appendix (p 18). Intrapartum-related events, preterm birth complications, and congenital abnormalities were the leading causes in the early neonatal period (0–6 days). Collectively, they accounted for over three-quarters of all early neonatal deaths. For deaths occurring in the late neonatal period (7–27 days), pneumonia replaced intrapartum-related events and ranked among the top three leading causes. Congenital abnormalities, pneumonia, and injuries were the leading causes of death in children who died at the ages of 28 days–5 months and 6–11 months. After the first year of life, injuries became the leading cause, contributing 41·1% (95% UR 38·8–42·9) of deaths in children aged 12–23 months and 53·8% (51·4–57·5) of deaths in children aged 24–59 months. The relative importance of preterm birth complications dropped sharply across the six age groups once children had survived the neonatal period. The contribution of congenital abnormalities to mortality remained high through infancy, after which it declined. Injuries started to gain prominence after the neonatal period and became dominant after the first year of life (appendix p 18).

rply across the six age groups once children had survived the neonatal period. The contribution of congenital abnormalities to mortality remained high through infancy, after which it declined. Injuries started to gain prominence after the neonatal period and became dominant after the first year of life (appendix p 18). The cause distribution was also somewhat different between rural and urban areas (appendix p 19). In rural areas, congenital abnormalities and preterm births were followed by injuries as the leading causes, whereas in urban areas intrapartum-related events were ranked third. The distribution of causes of under-5 deaths also differed by region (appendix p 19). Whereas congenital abnormalities, preterm birth complications, and injuries were the leading causes in the eastern and central regions, pneumonia caused more deaths than did injuries in the western region.

ed events were ranked third. The distribution of causes of under-5 deaths also differed by region (appendix p 19). Whereas congenital abnormalities, preterm birth complications, and injuries were the leading causes in the eastern and central regions, pneumonia caused more deaths than did injuries in the western region. The western rural area is the least developed area of the six region-residency strata. Under-5 mortality rate in the western rural area was estimated to be 22·4 (95% UR 18·1–28·1) per 1000 livebirths in 2015, at least twice that of under-5 mortality rate in other region-residency strata (figure 5). The western rural area was also the only stratum with an infectious disease (pneumonia) as the leading cause, whereas congenital abnormalities were the leading cause in all other strata. Injuries contributed to higher mortality rates in rural areas than urban areas across all regions. The gap in mortality rates between western rural and eastern urban areas widened for major non-infectious causes in 1996–2015. For example, the ratio of mortality rates due to congenital abnormalities between the two areas increased from 1·2 in 1996 to 2·8 in 2015, and the ratio of intrapartum-related events increased from 4·1 in 1996 to 5·9 in 2015.

en western rural and eastern urban areas widened for major non-infectious causes in 1996–2015. For example, the ratio of mortality rates due to congenital abnormalities between the two areas increased from 1·2 in 1996 to 2·8 in 2015, and the ratio of intrapartum-related events increased from 4·1 in 1996 to 5·9 in 2015. Our MCHSS based cause-specific mortality fractions estimates for 2015 in children younger than 5 years were similar to those derived by Song and colleagues16 (appendix p 20). The accuracy of the cause-specific mortality fractions was estimated to be 0·92. One exception was the fraction of congenital abnormalities, which was estimated to be 20% in this study compared with 14% by Song and colleagues.16 More discrepancies are noted in the neonatal period; despite overall good agreement between the two sets of estimates (cause-specific mortality fractions accuracy of 0·96), the fraction of congenital abnormalities was greater in our study than that in Song and colleagues' study (18% vs 11%). In children aged 1–59 months, the agreement was good (cause-specific mortality fractions accuracy is 0·94), with similar fractions of major causes, including injuries, congenital abnormalities, pneumonia, and diarrhoea.

genital abnormalities was greater in our study than that in Song and colleagues' study (18% vs 11%). In children aged 1–59 months, the agreement was good (cause-specific mortality fractions accuracy is 0·94), with similar fractions of major causes, including injuries, congenital abnormalities, pneumonia, and diarrhoea. Discussion China has made remarkable strides in improving child survival in 1996–2015. It has achieved an impressive average annual rate of decline of under-5 mortality rate of 8·2% compared with the global figure of 3·7% in this period. Based on our estimates, sufficient progress (ie, at an ARR of at least 4·4% per year to achieve the MDG 4 target) has been achieved across regions, residency, age groups, and cause categories. The progress coincided with China's widespread socioeconomic development in the past two decades.20, 21 This development includes, for example, the improvement in girls' education through achievement of 9 year compulsory education for all and narrowing the gender gap in years of education from 1·3 years in 2000 to 0·8 years in 2014.20, 21 National commitment to the improvement of maternal and child survival and health could be another important contributor, manifested by strengthened strategic legislation (eg, the Law on Maternal and Infant Health Care) and institutionalised policy frameworks (eg, the National Action Plan for the Development of Children).20, 21 Rigorous quantitative and qualitative research is still needed to understand the mechanisms through which these national development and action plans have been successful.

aw on Maternal and Infant Health Care) and institutionalised policy frameworks (eg, the National Action Plan for the Development of Children).20, 21 Rigorous quantitative and qualitative research is still needed to understand the mechanisms through which these national development and action plans have been successful. China officially eliminated neonatal tetanus in 2012.22 Despite not having a national programme on maternal tetanus toxoid immunisation, this success was credited to increased institutional delivery through the Safe Motherhood Initiative and the national Program to Reduce Maternal Mortality and Eliminate Neonatal Tetanus to improve local obstetric infrastructure, establish a fast-channel referral mechanism for pregnant women at labour, subsidise institutional delivery, and offer community-based health education.20, 23 More research is needed to establish the contribution of economic development in addition to these national programmes.24 China outperformed most of the 74 Countdown to 2015 countries (countries with high under-5 mortality rate, large numbers of under-5 deaths, or both) in its progress on major infectious causes, such as pneumonia and diarrhoea.25 Improved socioeconomic status, water, sanitation, nutrition, and access to integrated management of childhood illness could be contributors to success.10, 26, 27 Major causes of neonatal mortality, such as preterm birth complications and intrapartum-related events, have also been steadily declining, which might be associated with increased quality of institutional delivery and neonatal resuscitation.28, 29 Although existing research has implied causal relationships between some of these interventions and cause-specific mortality rates, more rigorous evaluation studies that can adequately establish causal relationships are needed.

ssociated with increased quality of institutional delivery and neonatal resuscitation.28, 29 Although existing research has implied causal relationships between some of these interventions and cause-specific mortality rates, more rigorous evaluation studies that can adequately establish causal relationships are needed. Despite the progress, 181 600 children died before their fifth birthday in China in 2015. Major causes of child deaths included congenital abnormalities, preterm birth complications, and injuries. The list of leading causes is similar to that in high-income countries, such as the USA.30 Congenital abnormalities are not only the leading cause, but also one that becomes increasingly important. Previous studies suggest that integrated preventive and treatment strategies combining political commitment, folic acid supplementation, and improved insurance coverage of surgeries addressing congenital heart diseases could be effective.31, 32, 33 Staple food fortification with folic acid could also be considered.20, 32 Primary and secondary prevention of preterm birth complications and injuries should receive more attention from the medical community, the government, and the general public. Effective interventions could include antenatal corticosteroids and kangaroo mother care for preterm births, and seat-belts and helmets to prevent road traffic injuries.34, 35 More research is needed to further understand the effectiveness of these interventions in the Chinese context and, if effective, how to implement these interventions at scale.2

clude antenatal corticosteroids and kangaroo mother care for preterm births, and seat-belts and helmets to prevent road traffic injuries.34, 35 More research is needed to further understand the effectiveness of these interventions in the Chinese context and, if effective, how to implement these interventions at scale.2 Despite a similar pace of decline across regions, almost half of all deaths in children younger than 5 years occurred in the least developed western region in 2015. The mortality rate in the western region was more than three times that of the most developed eastern region, and this gap increased by 10% since 1996. Overall, inadequate health service infrastructures and low health expenditure due to economic underdevelopment are thought to have resulted in limited access and use of health services in mothers and children, which led to inequity in child survival status in the western region.36, 37 Although the quantity of MCH workers did not seem to be an issue in this region, their education level was worse than that in the other two regions.38 This might have affected the quality of child health services provided. Investment in strategies to encourage the flow of better-educated and well trained MCH staff to the western region could be a policy option. MCH leadership and champions are also needed to improve child survival in the less developed areas. Clearly, the regional disparities in child mortality need to be systematically addressed in the SDG era. Child survival policy, programmes, and resources targeting the western region should receive more attention.

ption. MCH leadership and champions are also needed to improve child survival in the less developed areas. Clearly, the regional disparities in child mortality need to be systematically addressed in the SDG era. Child survival policy, programmes, and resources targeting the western region should receive more attention. With 80% of under-5 deaths in 2015, rural areas should be another focus for child survival in the SDG era. Several rural-focused national policies and programmes, such as the Program to Reduce Maternal Mortality and Eliminate Neonatal Tetanus and the Safe Motherhood Initiative, have been in place since the early 2000s through enhanced subsidies of institutional delivery in rural women. These programmes might have contributed to the steady decline of child mortality in rural China and more evidence is needed to support this claim.20 Ensuring continued success of these programmes is crucial, especially in hard-to-reach areas where child mortality is probably the highest. Rapid urbanisation has been shifting the burden of deaths in children younger than 5 years from rural to urban areas, yet the survival, health, and wellbeing of migrant children or children of migrant parents who are left behind are poorly understood.39 Injuries and pneumonia should receive more attention from the central and local government, the medical and public health community, and the general public in rural areas whereas congenital abnormalities and preterm birth complications are the priorities in urban areas.

ents who are left behind are poorly understood.39 Injuries and pneumonia should receive more attention from the central and local government, the medical and public health community, and the general public in rural areas whereas congenital abnormalities and preterm birth complications are the priorities in urban areas. More than half of under-5 deaths in China in 2015 were of neonates. Based on the estimated age-specific cause-of-death distributions, interventions targeting intrapartum-related events should focus on the early neonatal period. Programmes for preterm birth complications need to target the entire first month of life. Interventions addressing congenital abnormalities are important throughout the first year of life. After the age of 1 year, child survival policy and programmes should clearly concentrate on preventing and treating injuries. Research to obtain similar understanding of the age-specific cause composition is needed in other LMICs to help policy makers and programme managers to better prioritise and allocate scarce resources by age.

of 1 year, child survival policy and programmes should clearly concentrate on preventing and treating injuries. Research to obtain similar understanding of the age-specific cause composition is needed in other LMICs to help policy makers and programme managers to better prioritise and allocate scarce resources by age. Our data source, MCHSS, has been growing and evolving since its establishment. Reporting completeness has improved over time and the proportion of causes ascertained by medical certification has increased, aided in part by increased use of hospitals.40, 41 Most deaths ascertained by verbal autopsy were due to injuries, which are relatively accurately identified by this method.42 Increasing completeness suggests that surveillance capability and data quality might have improved. However, the pace of improvement might not have been even. For example, neonatal mortality rate in the western region increased in 2000–03 before resuming the downward trend in 2004 (figure 2B). This small peak in the trend line might not be real, given the rapid decline in neonatal mortality rate in other regions and in the national under-5 mortality rate. Instead, the increase in mortality rate observed during that period could reflect improved surveillance capacity in detecting neonatal deaths after 2000 in the western region.

he trend line might not be real, given the rapid decline in neonatal mortality rate in other regions and in the national under-5 mortality rate. Instead, the increase in mortality rate observed during that period could reflect improved surveillance capacity in detecting neonatal deaths after 2000 in the western region. Our study has some limitations. First, MCHSS considered local capacity when sampling prefectural-level urban cities and rural counties, which could have introduced biases. Second, the sampling probabilities were calculated using total population under surveillance and region-residence-stratum-specific census population estimates that are the best available. Because a census is only done every 10 years and China has been experiencing rapid urbanisation, the sampling probabilities are approximations. Third, the definition of livebirths was restricted to those born after 28 weeks of gestation, which could underestimate neonatal and under-5 mortality rates and the contribution of major neonatal causes, especially preterm birth complications. Given the relatively small burden of those born alive before 28 weeks, this bias is likely to be small. We did a sensitivity analysis comparing estimates based on the narrow definition of livebirth and that of all livebirths in selected sites where information on both was collected and found only small differences (data not shown). A plan is being developed to update the livebirth definition in MCHSS. This update should be easier to implement now because China has officially adopted the two-child policy. Lastly, given the nature of sample surveillance systems in which coverage is overall low at the national level, trends based on raw data are not smooth (appendix p 21). We have chosen to smooth the raw data and fit to the UN IGME estimates. The process might have inadvertently masked real short-term trends.

ild policy. Lastly, given the nature of sample surveillance systems in which coverage is overall low at the national level, trends based on raw data are not smooth (appendix p 21). We have chosen to smooth the raw data and fit to the UN IGME estimates. The process might have inadvertently masked real short-term trends. Despite the additional adjustments performed, however, MCHSS raw estimates fall within the uncertainty of UN IGME estimates for most years (appendix p 21). We also published in 2014 the national trends of under-5 mortality rate and neonatal mortality rate for China for the period 2000–12.43 The under-5 mortality rate estimates agree well with those produced by UN IGME, but the neonatal mortality rate estimates are significantly lower than those of UN IGME in 1996–2005 and start to converge to the UN IGME estimates after 2006. In the 2014 paper, input data from MCHSS and other sources were used, including vital registration and maternal and child annual reporting. Estimation methods were also different from those used in the UN IGME. That the two sets of relatively independent estimates agree reasonably well in 10 years is encouraging. Similarly, our estimates of cause-of-death distributions generally agree with estimates produced by Song and colleagues.16

hild annual reporting. Estimation methods were also different from those used in the UN IGME. That the two sets of relatively independent estimates agree reasonably well in 10 years is encouraging. Similarly, our estimates of cause-of-death distributions generally agree with estimates produced by Song and colleagues.16 China's Western Development is arguably one of the biggest development initiatives in the world. With a combination of socioeconomic development and regional MCH programmes, such as the Program to Reduce Maternal Mortality and Eliminate Neonatal Tetanus which was first piloted regionally in 2000,20 the western region has had great progress in improving child survival. Lessons learnt in this process, particularly those assessed by rigorous evaluations, are probably the most relevant to other LMICs. However, not all of China's experiences are readily transferrable to other countries and context. For example, institutional delivery rate has increased by about 30% in urban areas (76·5% in 1996 to 99·7% in 2012) and 91% in rural areas (51·7% in 1996 to 98·8% in 2012).44 These increases might have contributed to China's advances in child, and particularly neonatal, survival. But such a rapid increase in institutional delivery rate could be challenging to replicate in other LMICs. In addition to coverage, quality of institutional delivery rate is also a key consideration. For low-income settings not ready for the scale-up of institutional delivery rate, community-based neonatal interventions, such as kangaroo mother care and chlorhexidine cord cleansing,45 could be suitable options.

other LMICs. In addition to coverage, quality of institutional delivery rate is also a key consideration. For low-income settings not ready for the scale-up of institutional delivery rate, community-based neonatal interventions, such as kangaroo mother care and chlorhexidine cord cleansing,45 could be suitable options. The experiences of establishing and improving MCHSS could also prove informative to other LMICs where adequate vital or sample registration system is unavailable or of low quality. Where access to health care is high, a dual system of collecting vital events through community and health facilities can be attempted, as was done in MCHSS, so that crosschecking can be routinely implemented during quality control. Other forms of data triangulation across sectors and routine quality control mechanisms are also possible lessons from which other LMICs could learn.

ng vital events through community and health facilities can be attempted, as was done in MCHSS, so that crosschecking can be routinely implemented during quality control. Other forms of data triangulation across sectors and routine quality control mechanisms are also possible lessons from which other LMICs could learn. We consider our transition from model-based estimates to estimates based on empirical data for China an important step forward. This approach not only helps improves the validity of our estimates, but also increases the use of sample registration system data for national policy making and international exchange. Our study shows that strengthened civil registration and vital statistics could support and inform health policy and planning and help global health debates. We hope that China's experiences of developing MCHSS and achieving rapid improvement of child survival can contribute to the discussion on strengthening civil registration and vital statistics and on continued investment in child survival in LMICs in the SDG era. For the Maternal and Child Epidemiology website see http://tinyurl.com/Hopkins-MNCH-Chinacod-openacce Supplementary Material Supplementary appendix

We consider our transition from model-based estimates to estimates based on empirical data for China an important step forward. This approach not only helps improves the validity of our estimates, but also increases the use of sample registration system data for national policy making and international exchange. Our study shows that strengthened civil registration and vital statistics could support and inform health policy and planning and help global health debates. We hope that China's experiences of developing MCHSS and achieving rapid improvement of child survival can contribute to the discussion on strengthening civil registration and vital statistics and on continued investment in child survival in LMICs in the SDG era. For the Maternal and Child Epidemiology website see http://tinyurl.com/Hopkins-MNCH-Chinacod-openacce Supplementary Material Supplementary appendix Acknowledgments This study was funded by the National Twelfth Five-Year Plan for Science and Technology Support (2014BAI06B01), National Health and Family Planning Commission of The People's Republic of China (grants QT2003–009 and 05wsb-02), the National Science and Technology Basic Work Project of the Ministry of Science and Technology of China (grant 2014FY110700), UNICEF, and Bill & Melinda Gates Foundation. We thank the countless health workers who have contributed to data collection in the MCHSS; Danan Gu at UN Population Division for useful discussion on adjusting MCHSS estimates; and Yujiang Chen for her assistance with preparing webappendix 8 (appendix).

rant 2014FY110700), UNICEF, and Bill & Melinda Gates Foundation. We thank the countless health workers who have contributed to data collection in the MCHSS; Danan Gu at UN Population Division for useful discussion on adjusting MCHSS estimates; and Yujiang Chen for her assistance with preparing webappendix 8 (appendix). Contributors LL, REB, YW, and JZ conceived the idea. CH, YW, and JZ collected the data. LL and YC wrote the first draft with important contributions from CH. All authors reviewed and commented on subsequent drafts of the manuscript. Declaration of interests We declare no competing interests. Figure 1 China Maternal and Child Health Surveillance System site map Figure 2 Age-specific mortality rates in children younger than 5 years in China, 1996–2015 (A) National. (B) By region. (C) By residency. UR=uncertainty range. Figure 3 Causes of deaths in children aged 1–59 months and neonates in China, 2015 Causes of death that caused less than 1% of all deaths are not labelled. Figure 4 Cause of mortality in children younger than 5 years in China, 1996–2015 Figure 5 Cause-specific mortality rate in children younger than 5 years by region-residency strata in China, 2015 Causes with mortality rate of less than 0·5 per 1000 livebirths are not labelled. Table 1 National and regional average annual rate of reduction and uncertainty range of all-cause mortality rates and cause-specific mortality rates in children younger than 5 years in China, 1996–2015

Figure 5 Cause-specific mortality rate in children younger than 5 years by region-residency strata in China, 2015 Causes with mortality rate of less than 0·5 per 1000 livebirths are not labelled. Table 1 National and regional average annual rate of reduction and uncertainty range of all-cause mortality rates and cause-specific mortality rates in children younger than 5 years in China, 1996–2015 National Eastern region Central region Western region All cause Under-5 mortality rate 8·2% (6·4–9·7) 8·1% (6·4–9·6) 8·5% (6·8–10·8) 7·5% (5·8–9·0) Under-5 mortality rate in urban areas 4·4% (2·8–6·0) 6·0% (4·4–7·6) 4·1% (2·6–5·8) 4·5% (2·8–6·1) Under-5 mortality rate in rural areas 7·9% (6·2–9·4) 7·9% (6·2–9·4) 8·6% (6·8–10·0) 6·7% (5·0–8·2) Early neonatal mortality rate 8·4% (5·0–9·1) 8·6% (6·9–11·0) 9·4% (7·7–11·8) 6·7% (5·0–9·1) Late neonatal mortality rate 7·1% (4·9–9·0) 7·9% (6·2–10·3) 7·0% (5·3–9·4) 6·6% (4·9–9·0) Neonatal mortality rate 8·1% (6·4–10·5) 8·5% (5·9–10·8) 8·9% (7·2–11·3) 6·6% (5·0–9·0) Infant mortality rate 7·2% (5·3–8·6) 7·1% (5·9–9·2) 7·7% (6·5–9·8) 6·5% (5·3–8·6) By cause Neonatal tetanus 26·4% (22·2–39·3) 23·7% (8·6–29·8) 34·0% (9·4–42·7) 20·8% (16·4–30·5) Diarrhoea 12·8% (10·1–15·7) 12·9% (9·3–19·1) 15·5% (11·7–19·9) 11·7% (8·8–14·9) Pneumonia 11·4% (9·7–13·7) 12·5% (9·4–16·7) 12·4% (10·5–15·0) 10·3% (8·3–12·8) Measles 9·4% (6·9–14·1) 8·7% (5·8–13·1) 9·2% (6·7–13·9) 9·4% (8·0–15·3) Injury 9·3% (5·4–9·6) 9·2% (5·8–10·4) 9·3% (5·2–9·6) 8·5% (3·7–10·2) Intrapartum-related conditions 8·6% (6·3–10·5) 8·5% (5·6–10·2) 8·6% (7·3–11·5) 8·1% (4·3–9·1) Sepsis or meningitis 8·0% (5·9–10·6) 7·6% (5·0–11·9) 9·1% (5·5–11·0) 6·4% (4·4–11·0) Preterm birth complications 6·6% (5·0–9·1) 6·9% (5·2–9·5) 8·3% (6·6–10·9) 4·5% (2·6–7·2) Congenital abnormalities 4·1% (2·3–6·1) 6·5% (4·6–9·1) 3·5% (1·4–6·2) 2·6% (0·0–6·2) Other conditions 7·0% (5·9–9·7) 6·8% (4·6–8·7) 6·7% (5·8–9·3) 7·0% (5·5–9·9) Data are annual rate of reduction (95% uncertainty range).

(4·4–11·0) Preterm birth complications 6·6% (5·0–9·1) 6·9% (5·2–9·5) 8·3% (6·6–10·9) 4·5% (2·6–7·2) Congenital abnormalities 4·1% (2·3–6·1) 6·5% (4·6–9·1) 3·5% (1·4–6·2) 2·6% (0·0–6·2) Other conditions 7·0% (5·9–9·7) 6·8% (4·6–8·7) 6·7% (5·8–9·3) 7·0% (5·5–9·9) Data are annual rate of reduction (95% uncertainty range). Table 2 Estimated numbers of child deaths and mortality rates by cause in China in 1996 and 2015

(4·4–11·0) Preterm birth complications 6·6% (5·0–9·1) 6·9% (5·2–9·5) 8·3% (6·6–10·9) 4·5% (2·6–7·2) Congenital abnormalities 4·1% (2·3–6·1) 6·5% (4·6–9·1) 3·5% (1·4–6·2) 2·6% (0·0–6·2) Other conditions 7·0% (5·9–9·7) 6·8% (4·6–8·7) 6·7% (5·8–9·3) 7·0% (5·5–9·9) Data are annual rate of reduction (95% uncertainty range). Table 2 Estimated numbers of child deaths and mortality rates by cause in China in 1996 and 2015 1996 2015 Estimated number of deaths Cause-specific mortality rate (per 1000 livebirths) Estimated number of deaths Cause-specific mortality rate (per 1000 livebirths) Children aged 0–59 months Congenital abnormalities 62 700 (50 200–69 800) 4·5 (3·6–5·1) 35 700 (28 400–45 200) 2·1 (1·7–2·7) Preterm birth complications 88 800 (76 500–101 200) 6·4 (5·5–7·3) 30 900 (24 200–40 800) 1·8 (1·4–2·4) Injury 126 100 (83 400–146 700) 9·1 (6·0–10·6) 26 600 (21 000–33 400) 1·6 (1·3–2·0) Intrapartum-related events 97 600 (84 600–111 400) 7·1 (6·1–8·1) 26 100 (20 500–34 600) 1·5 (1·2–2·0) Pneumonia 158 200 (129 800–163 000) 11·5 (9·4–11·8) 22 200 (15 900–28 000) 1·3 (0·9–1·6) Diarrhoea 48 700 (32 300–49 400) 3·5 (2·3–3·6) 5300 (3400–6900) 0·3 (0·2–0·4) Sepsis or meningitis 20 600 (14 200–20 800) 1·5 (1·0–1·5) 5000 (3800–6500) 0·3 (0·2–0·4) Other conditions 97 900 (70 400–111 400) 7·1 (5·1–8·1) 29 700 (24 200–40 800) 1·7 (1·4–2·3) Neonates aged 0–27 days Preterm birth complications 85 800 (73 700–97 900) 6·2 (5·3–7·1) 29 000 (22 600–38 800) 1·7 (1·3–2·3) Intrapartum-related events 97 500 (84 400–111 200) 7·1 (6·1–8·1) 25 700 (20 000–34 100) 1·5 (1·2–2·0) Congenital abnormalities 36 200 (29 900–42 800) 2·6 (2·2–3·1) 16 800 (13 100–22 700) 1 (0·8–1·4) Pneumonia 56 600 (48 100–66 000) 4·1 (3·5–4·8) 5400 (4100–7200) 0·3 (0·2–0·4) Injury 32 500 (26 200–40 100) 2·4 (1·9–2·9) 4600 (3600–6200) 0·3 (0·2–0·4) Sepsis or meningitis 5600 (4000–7400) 0·4 (0·3–0·5) 2500 (1800–3500) 0·1 (0·1–0·2) Diarrhoea 2700 (2100–3400) 0·2 (0·1–0·2) 500 (300–600) 0 (0–0) Tetanus 6500 (4100–9100) 0·5 (0·3–0·7) .. ..

0–66 000) 4·1 (3·5–4·8) 5400 (4100–7200) 0·3 (0·2–0·4) Injury 32 500 (26 200–40 100) 2·4 (1·9–2·9) 4600 (3600–6200) 0·3 (0·2–0·4) Sepsis or meningitis 5600 (4000–7400) 0·4 (0·3–0·5) 2500 (1800–3500) 0·1 (0·1–0·2) Diarrhoea 2700 (2100–3400) 0·2 (0·1–0·2) 500 (300–600) 0 (0–0) Tetanus 6500 (4100–9100) 0·5 (0·3–0·7) .. .. Other conditions 31 200 (25 800–36 600) 2·3 (1·9–2·7) 9000 (6900–12 200) 0·5 (0·4–0·7) Children aged 1–59 months Injuries 93 600 (73 500–113 600) 6·8 (5·3–8·2) 22 000 (16 700–27 800) 1·3 (1·0–1·6) Congenital abnormalities 26 600 (18 300–29 300) 1·9 (1·3–2·1) 19 000 (13 000–25 100) 1·1 (0·8–1·5) Pneumonia 101 600 (75 100–104 000) 7·4 (5·4–7·5) 16 800 (11 100–21 700) 1·0 (0·7–1·3) Diarrhoea 45 900 (30 000–46 600) 3·3 (2·2–3·4) 4800 (3100–6300) 0·3 (0·2–0·4) Meningitis 14 600 (8 500–15 100) 1·1 (0·6–1·1) 2100 (1500–2800) 0·1 (0·1–0·2) Preterm birth complications 3000 (2200–3800) 0·2 (0·2–0·3) 1900 (1300–2500) 0·1 (0·1–0·1) Pertussis 2200 (0–10 200) 0·2 (0·0–0·7) 1400 (1300–1400) 0·1 (0·1–0·1) Measles 2400 (1500–3100) 0·2 (0·1–0·2) 500 (300–900) 0·0 (0·0–0·1) Intrapartum-related events 100 (100–200) 0·0 (0·0–0·0) 400 (300–600) 0·0 (0·0–0·0) HIV/AIDS 300 (200–600) 0·0 (0·0–0·0) 400 (200–800) 0·0 (0·0–0·0) Malaria .. .. .. .. Other conditions 55 800 (36 300–64 700) 4·0 (2·6–4·7) 18 800 (13 700–24 400) 1·1 (0·8–1·4) Data are n (95% UR) or rate (95% UR). Other conditions in children aged 1–59 months included causes originating during the perinatal period, cancer, severe malnutrition, and other specified causes. Intrapartum-related events were previously referred to as birth asphyxia. UR=uncertainty range. Blank cells indicate that less than 100 deaths were estimated for the condition.

Introduction Despite improvements in child survival over recent decades, progress in newborn survival remains slow, with 44% of all child deaths occurring in the first month of life. Of these neonatal deaths, 23–30% are due to infections.1 WHO recommends hospital referral and 7 days of injectable penicillin and gentamicin for neonates and young infants (aged 0–59 days) with suspected sepsis.2 However, up to three-quarters of families of sick young infants in Karachi, Pakistan, refuse hospital referrals, despite free transport and treatment, because of the substantial opportunity costs to very poor families of prolonged admissions at locations far from their place of residence.3 Stated reasons for refusal are financial constraints, cultural beliefs, and concern about poor quality of care at hospitals.3, 4 Similar constraints to optimum care of sick newborn babies in high-mortality settings have also been noted from other low-resource settings.5

at locations far from their place of residence.3 Stated reasons for refusal are financial constraints, cultural beliefs, and concern about poor quality of care at hospitals.3, 4 Similar constraints to optimum care of sick newborn babies in high-mortality settings have also been noted from other low-resource settings.5 An expert consultation reviewed the issue of low adherence to referral advice for sick young infants and recommended that clinical trials were needed to evaluate simplified antibiotic regimens to manage severe infections in young infants when referral was not possible, to improve access to care and newborn survival.6 Thus, randomised controlled trials assessing simplified antibiotic regimens—ie, fewer injections, addition of high-dose oral amoxicillin in lieu of penicillin—for outpatient management of young infants with clinical severe infection were undertaken in several countries (Democratic Republic of Congo, Bangladesh, Kenya, Nigeria, and Pakistan) to ensure wide generalisability.7, 8, 9, 10 These trials were not designed to show that the simpler regimens were better than the standard regimen but rather that they had similar efficacy to the reference regimen—namely, an equivalence or non-inferiority design.11 Thus, the trials were designed to produce narrow confidence intervals when examining the difference in risk of treatment failure between the simplified regimens and the reference treatment and, hence, provide a high degree of confidence that any differences between efficacy of the treatments were small. The choice of treatment regimens followed a systematic review of pathogens causing neonatal sepsis,12 their antimicrobial resistance patterns,13 antibiotic pharmacodynamics in neonates,14 and existing evidence on treatment success with various oral and injectable antibiotics in young infants.15, 16

eatments were small. The choice of treatment regimens followed a systematic review of pathogens causing neonatal sepsis,12 their antimicrobial resistance patterns,13 antibiotic pharmacodynamics in neonates,14 and existing evidence on treatment success with various oral and injectable antibiotics in young infants.15, 16 Research in context Evidence before the study We searched PubMed between January, 1990, and October, 2015, with the terms “young infant”, “clinical severe infection”, and “simplified antibiotic regimens” to identify peer-reviewed publications in the English language about simplified antibiotic regimens for severe infections in neonates and young infants in the primary-care setting. We identified five reports (three protocol papers), of which two reported findings of similar trials undertaken in Africa (Kenya, Nigeria, and Democratic Republic of Congo) and Bangladesh, comparing the WHO-recommended regimen of parenteral penicillin and gentamicin with simpler antibiotic regimens. Added value of this study Compared with previous studies in Africa and Bangladesh, our trial from Pakistan had a higher representation of very young infants (those in the first week of life) and was enriched by the availability of bacterial aetiological data and antimicrobial susceptibility data. A pooled analysis can now be done of data from all three related trials, to support policy recommendations for this very young group of patients. Implications of all the available evidence

Compared with previous studies in Africa and Bangladesh, our trial from Pakistan had a higher representation of very young infants (those in the first week of life) and was enriched by the availability of bacterial aetiological data and antimicrobial susceptibility data. A pooled analysis can now be done of data from all three related trials, to support policy recommendations for this very young group of patients. Implications of all the available evidence The findings of our trial are consistent with those published previously, showing that simplified antibiotic regimens are as efficacious as the WHO-recommended regimen of parenteral penicillin and gentamicin for young infants with severe infection. Our study has contributed to development of new WHO guidelines for treatment of severe infection in young infants where referral is not feasible. We targeted young infants (aged 0–59 days) with clinical severe infections because increased susceptibility to infection persists into the second month of life.17 Additionally, the signs and management of sepsis in young infants (aged 29–59 days) are similar to those in neonates (aged 0–28 days). WHO and UNICEF's Integrated Management of Childhood Illness (IMCI) strategy addresses children aged 0–59 days as a separate group (young infants) from children aged 2–59 months.18

fe.17 Additionally, the signs and management of sepsis in young infants (aged 29–59 days) are similar to those in neonates (aged 0–28 days). WHO and UNICEF's Integrated Management of Childhood Illness (IMCI) strategy addresses children aged 0–59 days as a separate group (young infants) from children aged 2–59 months.18 Here, we present data from the Simplified Antibiotic Therapy Trial (SATT) undertaken in Karachi, Pakistan,8 to compare the risk of treatment failure in young infants with a diagnosis of clinical severe infection between a reference treatment and two simplified antibiotic regimens, comprising fewer injections and a high dose of an oral antibiotic in lieu of parenteral administration. Methods Study design and participants SATT Pakistan is a randomised open-label trial designed to assess equivalence of three outpatient-based antibiotic regimens. We undertook the trial in five low-income settlements in coastal Karachi, Pakistan (Rehri Goth, Ibrahim Hyderi, Ali Akbar Shah Goth, Bhains colony, and Bilal colony), which are roughly 1 h drive from the main campus of the Aga Khan University in Karachi. These settlements are served by five primary health-care clinics, have an annual birth cohort of 8000, and have ongoing household surveillance of young infants aged 0–59 days (appendix p 1).

h Goth, Bhains colony, and Bilal colony), which are roughly 1 h drive from the main campus of the Aga Khan University in Karachi. These settlements are served by five primary health-care clinics, have an annual birth cohort of 8000, and have ongoing household surveillance of young infants aged 0–59 days (appendix p 1). Infants from the catchment area were either referred to a study clinic by community health workers during routine household surveillance or presented with their family at one of the five primary health-care clinics, at which study clinicians screened them for eligibility to participate in the trial.8 Inclusion criteria were age 0–59 days, living in the catchment area, refusal by family to be admitted to hospital, and one or more signs of clinical severe infection (panel). Infants were excluded from the study if their family agreed to admission, weight at presentation was less than 1500 g, major congenital malformations or suspected chromosomal abnormalities were present, surgical conditions needed hospital referral, they had been admitted for illness in the past 2 weeks, they had been included previously in the study, or they had one or more signs of critical illness (panel). If the infant had signs of clinical severe infection, the study clinician first recommended hospital referral. If the family refused admission and the infant had no signs of critical illness, the study clinician offered trial enrolment with facility-based and home-based treatment. If the infant had signs of critical illness, parents were counselled again on the importance of hospital referral. If parents still refused admission, they were offered the reference treatment (ie, procaine benzylpenicillin and gentamicin) and were not enrolled or randomly allocated. Study methods have been described in detail.8

ant had signs of critical illness, parents were counselled again on the importance of hospital referral. If parents still refused admission, they were offered the reference treatment (ie, procaine benzylpenicillin and gentamicin) and were not enrolled or randomly allocated. Study methods have been described in detail.8 We obtained written informed consent for infants to participate in the trial from parents or guardians. Consent included documentation of the refusal for hospital referral and acceptance of enrolment. A third party (community member) witnessed the consent procedure, and this individual also signed the consent form. We read the consent form to illiterate participants and took a thumbprint in lieu of a signature, which a third party witness countersigned. The study was approved by the ethics review committee of the Aga Khan University, ethics review committee of WHO, and the ethics committee at the London School of Hygiene & Tropical Medicine.

consent form to illiterate participants and took a thumbprint in lieu of a signature, which a third party witness countersigned. The study was approved by the ethics review committee of the Aga Khan University, ethics review committee of WHO, and the ethics committee at the London School of Hygiene & Tropical Medicine. Randomisation and masking After obtaining consent, we randomly allocated infants to one of three antibiotic regimens: 7 days of procaine benzylpenicillin and gentamicin, administered intramuscularly (reference treatment); intramuscular gentamicin once a day and oral amoxicillin twice daily for 7 days; or procaine benzylpenicillin and gentamicin administered intramuscularly once a day for 2 days followed by oral amoxicillin twice daily for 5 days. We used a site-specific and age-specific (<7 days and 7–59 days) randomisation sequence list generated by the London School of Hygiene & Tropical Medicine.8 The allocation sequence for every site and age group was placed in serially numbered, sealed, opaque envelopes by the Data Management Unit at Aga Khan University and delivered to every site. Study clinicians selected the next envelope and the treatment corresponding to the allocation code printed within was assigned to the infant. Study participants' families and study clinicians were not blinded to treatment allocation because giving placebo injections to sick young infants was judged unethical.

ery site. Study clinicians selected the next envelope and the treatment corresponding to the allocation code printed within was assigned to the infant. Study participants' families and study clinicians were not blinded to treatment allocation because giving placebo injections to sick young infants was judged unethical. Procedures Study drugs were provided by the Aga Khan University pharmacy, stored at room temperature away from direct sunlight at study clinics, and administered using study-specific dosage charts present at every public health-care clinic (appendix p 2). We administered intramuscular procaine benzylpenicillin (40 000–60 000 units per kg) once a day, intramuscular gentamicin (4·0–5·0 mg/kg in early neonates aged 0–6 days; 5·0–6·5 mg/kg in infants aged 7–59 days) once a day, and oral amoxicillin (75–100 mg/kg per day) twice daily in divided doses. Paramedics or study clinicians administered intramuscular injections at study clinics; study personnel gave the morning dose of oral amoxicillin at the clinic, and a community health worker visiting the child's household administered the evening dose. If the baby vomited within 30 min, the oral drug was re-administered. We followed up every participant daily at the public health-care clinics, from enrolment to day 8, then on day 11 and day 14 for vital signs (respiratory rate, temperature, and heart rate), danger signs of critical illness (panel), improvement or deterioration in clinical status (defined as resolution or presence of one danger sign of critical illness),8 and adverse events (ie, relapse, death, or treatment failure). We referred young infants who failed treatment to hospital. If families still refused admission, we treated the infant with intramuscular ceftriaxone once a day for 1 week.

status (defined as resolution or presence of one danger sign of critical illness),8 and adverse events (ie, relapse, death, or treatment failure). We referred young infants who failed treatment to hospital. If families still refused admission, we treated the infant with intramuscular ceftriaxone once a day for 1 week. We collected blood samples at enrolment using appropriate aseptic precautions. We captured blood sample collection procedures on video for about 10% of participants as a quality assurance measure. We injected about 2–3 mL of blood (median 2·1 mL [IQR 1·4–2·8]) into a bottle (BACTEC Peds Plus; Becton Dickinson, Franklin Lakes, NJ, USA) and transported the sample to the Infectious Disease Research Laboratory at Aga Khan University within 3 h of collection; we incubated the sample in a continuous monitoring system (BACTEC 9240; Becton Dickinson) for 5 days. If bottles were flagged positive by the automated system, we Gram-stained and subcultured blood culture broth on appropriate media—eg, 5% sheep blood agar and chocolate agars incubated in 5% CO2 at 35°C; Maconkey agar incubated in air at 35°C; or 5% sheep blood agar incubated anaerobically at 35°C. We did all identification and susceptibility tests in accordance with American Society for Microbiology procedures and Clinical Laboratory Standards Institute guidelines,19 when applicable. For a few blood cultures that were smear-positive for campylobacter-like organisms but that did not yield bacterial growth on conventional culture, we also did PCR of the 23S RNA-conserved region of campylobacter, helicobacter, and arcobacter complex. Briefly, we extracted DNA (MagNA Pure extraction kit; Roche Diagnostics, Burgess Hill, UK) then did conventional PCR (Eppendorf MasterCycler Gradient; Marshall Scientific, Hampton, NH, USA)20 and gel electrophoresis.

we also did PCR of the 23S RNA-conserved region of campylobacter, helicobacter, and arcobacter complex. Briefly, we extracted DNA (MagNA Pure extraction kit; Roche Diagnostics, Burgess Hill, UK) then did conventional PCR (Eppendorf MasterCycler Gradient; Marshall Scientific, Hampton, NH, USA)20 and gel electrophoresis. We categorised blood culture results as no growth, contamination, or bacteraemia (known or probable pathogen). We classed blood cultures with growth of common skin flora as contamination. We used bacteraemia to describe blood cultures positive for known pathogens, whether isolated singly or mixed with other pathogens or contaminants, or pathogens less commonly associated with neonatal sepsis, whether isolated singly or mixed with other pathogens or contaminants (appendix p 3). We checked for adverse events at every follow-up visit. Community health workers reported adverse events to study doctors at every study clinic, who graded them as serious or non-serious. Serious adverse events were those possibly related to study drugs, including: severe diarrhoea with dehydration requiring facility management; Stevens-Johnson syndrome; anaphylaxis; and acute renal failure. A study supervisor (trained paediatrician) verified the grading. Quarterly adverse event reports were reviewed by the study's data safety and monitoring board and technical scientific committee.

ere diarrhoea with dehydration requiring facility management; Stevens-Johnson syndrome; anaphylaxis; and acute renal failure. A study supervisor (trained paediatrician) verified the grading. Quarterly adverse event reports were reviewed by the study's data safety and monitoring board and technical scientific committee. Outcomes The primary outcome of our trial was treatment failure within 7 days of enrolment, which we defined as either: death; admission; clinical deterioration;8 change in antibiotic regimen because of infectious comorbidity (to intramuscular ceftriaxone); serious adverse event; occurrence of a new sign of clinical severe infection (panel) on or after day 3; persistence of presenting signs at day 4; or recurrence of initial signs of sepsis on or after day 5. Among young infants who had treatment failure, secondary outcomes were: death within 7 days of enrolment; death at any time before the day 14–15 assessment; and admission for any reason at any time within 7 days of enrolment. Among children who did not have treatment failure, secondary outcomes were: admission at any time between the day 8 and day 14–15 visits; death at any time between the day 8 and day 14–15 visits; and non-fatal relapse at any time between the day 8 and day 14–15 visits (defined as admission, development of any sign of critical illness, or development of any sign of suspected sepsis).

omes were: admission at any time between the day 8 and day 14–15 visits; death at any time between the day 8 and day 14–15 visits; and non-fatal relapse at any time between the day 8 and day 14–15 visits (defined as admission, development of any sign of critical illness, or development of any sign of suspected sepsis). Statistical analysis We postulated (from our previous experience) that the simplified antibiotic regimens would have equivalent efficacy to the reference treatment and that the risk of treatment failure would be 10% in all groups. For every comparison with the reference treatment, we planned to estimate the difference in the risk of failure between the two treatment groups and to use a two-sided 95% CI to assess the equivalence of the two simplified antibiotic regimens. We judged simplified antibiotic regimens as efficacious as the reference if the upper bound of the 95% CI for the difference in treatment failure was less than 5. Based on this criterion, we estimated that a sample size of 750 assessable children per treatment group would provide at least 90% power to show that the simplified antibiotic regimens were as efficacious as the reference treatment.

the upper bound of the 95% CI for the difference in treatment failure was less than 5. Based on this criterion, we estimated that a sample size of 750 assessable children per treatment group would provide at least 90% power to show that the simplified antibiotic regimens were as efficacious as the reference treatment. Since the aim of our trial was to show the equivalence of different antibiotic regimens, rather than superiority of one regimen over another, we analysed the primary outcome per protocol8 rather than by intention to treat, which would tend to reduce any differences between treatment regimens. We defined per-protocol infants as those who had completed clinical follow-up fully (eight visits on 8 days) or partly (three visits days 2–4, at least one visit days 5–8, and known vital status at day 8) and who had adhered to treatment fully or partly (appendix p 6). We defined infants who were fully adherent to treatment as those who received all doses of scheduled antibiotics for 7 days (or by the time of treatment failure, if failure occurred) and who had not received any other antibiotic from a study or non-study clinician. We defined infants who were partly adherent to treatment as those who had received all scheduled antibiotics on the first 3 days of treatment (or by the time of treatment failure) and at least 50% of all scheduled doses of each antibiotic on days 4–7 (or by the time of treatment failure), and who did not receive any non-study injectable antibiotic before the day 8 visit (unless given because of treatment failure) or any non-study oral antibiotic on days 1–3.

(or by the time of treatment failure) and at least 50% of all scheduled doses of each antibiotic on days 4–7 (or by the time of treatment failure), and who did not receive any non-study injectable antibiotic before the day 8 visit (unless given because of treatment failure) or any non-study oral antibiotic on days 1–3. We did statistical analyses with Stata version 13. This study is registered with ClinicalTrials.gov, number NCT01027429. Role of the funding source This trial was funded by the Saving Newborn Lives initiative of Save the Children, with support from the Bill & Melinda Gates Foundation, and by WHO and USAID. The funders had a role in study design but played no part in data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and was responsible for the decision to submit for publication.

elinda Gates Foundation, and by WHO and USAID. The funders had a role in study design but played no part in data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and was responsible for the decision to submit for publication. Results Between Jan 1, 2010, and Dec 26, 2013, 41 230 young infants were screened for trial eligibility and 2780 (7%) were eligible for enrolment (figure). Of these, 63 (2%) families agreed to admission and 264 (9%) refused participation. Thus, 2453 (88%) of 2780 young infants were randomly allocated one of the three study treatments: 820 were assigned procaine benzylpenicillin and gentamicin, 816 were allocated amoxicillin and gentamicin, and 817 were assigned procaine benzylpenicillin, gentamicin, and amoxicillin. Table 1 shows baseline characteristics of all young infants who were randomly allocated. Median age at presentation was 11 days (IQR 2–36). 1083 (44%) infants were early neonates (aged 0–6 days), 1309 (53%) were boys, and 940 (38%) had a low weight-for-age (Z score <–2). 2141 (87%) infants presented with one sign at enrolment. Fever was the most common presenting sign (1015/2453 [41%]) and was the only presenting sign for more than a third of infants (905/2453 [37%]; table 1). The next most frequent sign was severe chest indrawing (818/2453 [33%]), with around a third having this sign in isolation (717/2453 [29%]). 358 (15%) of 2453 infants had local infection, of whom 283 (79%) had an umbilical infection and 86 (24%) had a skin infection. 1223 (50%) of 2453 young infants were born in a health facility.

t frequent sign was severe chest indrawing (818/2453 [33%]), with around a third having this sign in isolation (717/2453 [29%]). 358 (15%) of 2453 infants had local infection, of whom 283 (79%) had an umbilical infection and 86 (24%) had a skin infection. 1223 (50%) of 2453 young infants were born in a health facility. 2251 (92%) of 2453 infants who were randomly allocated met per-protocol criteria for clinical follow-up and treatment adherence (figure). 747 received procaine benzylpenicillin and gentamicin, 751 were treated with amoxicillin and gentamicin, and 753 were given procaine benzylpenicillin, gentamicin, and amoxicillin. Per-protocol infants had similar characteristics at baseline to the intention-to-treat population (data not shown). Table 2 presents primary and secondary outcome data in the per-protocol population. Treatment failure was recorded within 7 days of enrolment in 90 (12%) of 747 infants who received procaine benzylpenicillin and gentamicin (reference), 76 (10%) of 751 who were given amoxicillin and gentamicin (risk difference with reference, −1·9, 95% CI −5·1 to 1·3), and 99 (13%) of 753 treated with procaine benzylpenicillin, gentamicin, and amoxicillin (risk difference with reference, 1·1, −2·3 to 4·5); the upper bound of the 95% CI for both comparisons was within the prespecified margin for equivalence. The most common causes of treatment failure across the three study groups were persistence of presenting signs at day 4 (n=62), admission (n=51), and clinical deterioration (n=43). In analyses of all randomly allocated infants (appendix p 4), treatment failure was recorded in 97 (12%) of 820 assigned procaine benzylpenicillin and gentamicin (reference), 81 (10%) of 816 allocated amoxicillin and gentamicin (risk difference with reference, −1·9, 95% CI −4·9 to 1·1), and 111 (14%) of 817 assigned procaine benzylpenicillin, gentamicin, and amoxicillin (risk difference with reference, 1·8, −1·5 to 5·0).

of 820 assigned procaine benzylpenicillin and gentamicin (reference), 81 (10%) of 816 allocated amoxicillin and gentamicin (risk difference with reference, −1·9, 95% CI −4·9 to 1·1), and 111 (14%) of 817 assigned procaine benzylpenicillin, gentamicin, and amoxicillin (risk difference with reference, 1·8, −1·5 to 5·0). 28 (1%) of 2251 infants in the per-protocol analysis died within 7 days of enrolment, due to clinical severe illness, and the risk of death was similar across the three treatment groups (table 2): 11 (1%) of 747 infants died who received procaine benzylpenicillin and gentamicin (reference), seven (1%) of 751 died who were treated with amoxicillin and gentamicin (risk difference with reference −0·5, 95% CI −1·6 to 0·6), and ten (1%) of 753 died who were given procaine benzylpenicillin, gentamicin, and amoxicillin (risk difference −0·1, −1·3 to 1·0). No deaths were attributable to study procedures. A further six children died between day 8 and day 15: two who were treated with procaine benzylpenicillin and gentamicin (both neonatal sepsis); two who received amoxicillin and gentamicin (diarrhoea, and neonatal sepsis); and two who were given procaine benzylpenicillin, gentamicin, and amoxicillin (neonatal tetanus, and neonatal sepsis). None of these six deaths was judged attributable to study procedures by the principal investigator or the data safety and monitoring board. Among all randomly allocated infants, 35 (1%) of 2453 children had died within 7 days of enrolment (appendix p 4), 12 (1%) of 820 children assigned procaine benzylpenicillin and gentamicin (reference), ten (1%) of 816 allocated amoxicillin and gentamicin (risk difference with reference, −0·2, 95% CI −1·4 to 0·9), and 13 (2%) of 817 assigned procaine benzylpenicillin, gentamicin, and amoxicillin (risk difference with reference, 0·1,–1·1 to 1·3). By day 15, a further three infants had died who were allocated procaine benzylpenicillin and gentamicin, two children assigned amoxicillin and gentamicin had died, and three children died who were allocated procaine benzylpenicillin, gentamicin, and amoxicillin.

cillin (risk difference with reference, 0·1,–1·1 to 1·3). By day 15, a further three infants had died who were allocated procaine benzylpenicillin and gentamicin, two children assigned amoxicillin and gentamicin had died, and three children died who were allocated procaine benzylpenicillin, gentamicin, and amoxicillin. Three non-fatal serious adverse events were reported among the 2453 randomly allocated infants, one in a child assigned procaine benzylpenicillin and gentamicin (diarrhoea with severe dehydration) and two in children allocated amoxicillin and gentamicin (diarrhoea with severe dehydration, and generalised rash). Two of these three events contributed to the initial reason for treatment failure at day 2. All three infants had recovered by the day 15 visit and were included in the per-protocol analysis. Among the 2251 children in the per-protocol analysis, 193 (9%) non-serious adverse events occurred, 79 (11%) in 747 infants treated with procaine benzylpenicillin and gentamicin, 55 (7%) in 751 who received amoxicillin and gentamicin, and 59 (8%) in 753 who were given procaine benzylpenicillin, gentamicin, and amoxicillin. The most frequent events were injection-site swelling (n=88) and mild diarrhoea (n=81).

curred, 79 (11%) in 747 infants treated with procaine benzylpenicillin and gentamicin, 55 (7%) in 751 who received amoxicillin and gentamicin, and 59 (8%) in 753 who were given procaine benzylpenicillin, gentamicin, and amoxicillin. The most frequent events were injection-site swelling (n=88) and mild diarrhoea (n=81). Blood cultures were obtained from 2067 (84%) of 2453 randomly allocated infants, of which 1713 (83%) were negative after 5 days of incubation and 273 (13%) were contaminated. 81 (4%) cultures were positive for various pathogens (appendix p 5); 79 grew a single organism whereas two were polymicrobial. Campylobacteraceae were the commonest group of pathogens cultured (n=18), followed by pseudomonads (n=13), enteric Gram-negative bacteria (n=12), and Streptococcus pyogenes (n=8). Of 23 Gram-positive organisms, 19 (83%) were susceptible to penicillin. Of the Gram-negative bacteria, gentamicin susceptibility results were available for 14 organisms, and all but one Klebsiella pneumoniae isolate were gentamicin susceptible. Ampicillin-susceptibility test results were available for 18 Gram-negative isolates, and nine (50%) were ampicillin susceptible. Overall, 32 (86%) of 37 microbes available for antimicrobial susceptibility were sensitive to a regimen including penicillin or amoxicillin and gentamicin. Ten (13%) of 75 children with bacteraemia and 227 (12%) of 1618 without bacteraemia had treatment failure; thus, bacteraemia did not predict treatment failure in per-protocol infants (risk difference 1·03, 95% CI −6·8 to 8·9).

usceptibility were sensitive to a regimen including penicillin or amoxicillin and gentamicin. Ten (13%) of 75 children with bacteraemia and 227 (12%) of 1618 without bacteraemia had treatment failure; thus, bacteraemia did not predict treatment failure in per-protocol infants (risk difference 1·03, 95% CI −6·8 to 8·9). Discussion Our results show that, in Pakistan, simplified antibiotic regimens are as efficacious as the reference treatment for young infants with clinical severe infection whose families refuse referral to hospital.2 This finding is consistent with those of two similar trials from Africa,9 and Bangladesh.10 Results from these trials have contributed to development of new WHO guidelines for management of young infants with possible serious bacterial infection where referral is not feasible.21

e families refuse referral to hospital.2 This finding is consistent with those of two similar trials from Africa,9 and Bangladesh.10 Results from these trials have contributed to development of new WHO guidelines for management of young infants with possible serious bacterial infection where referral is not feasible.21 A strength of our Pakistan study is that we included a much higher representation of infants aged 0–6 days with clinical severe infection (44% of all enrolled children) compared with the other randomised equivalence trials of the simplified antibiotic regimens from Bangladesh10 and Africa (Kenya, Nigeria, and Democratic Republic of Congo).9 Additional strengths of our study are the availability of bacterial aetiological data and antimicrobial susceptibility data. Enrolment of few infants younger than 7 days with clinical severe infection was judged a limitation of the Bangladesh trial,22 in terms of drawing conclusions for this important subgroup at high risk of vertically acquired infections and mortality. A pooled analysis of data from our trial, the Bangladesh trial, and the AFRINEST trial in Kenya, Nigeria, and Democratic Republic of Congo is underway and will have an adequate number of infants aged 0–6 days to support policy recommendations for this age group. These findings hold great promise in increasing access to treatment for sick young infants and improving their clinical outcomes.

AFRINEST trial in Kenya, Nigeria, and Democratic Republic of Congo is underway and will have an adequate number of infants aged 0–6 days to support policy recommendations for this age group. These findings hold great promise in increasing access to treatment for sick young infants and improving their clinical outcomes. Data for common bacterial pathogens among young infants at the community level are scarce.23 We gathered samples for blood culture at the first-level facility, from young infants recruited from the community. The proportion of infants with clinical sepsis and who grew a pathogen from the blood sample was low (4%) but comparable with yields reported elsewhere.24, 25 Biomarkers need to be developed to improve our ability to distinguish between infants with and without bacterial infection in first-level settings.23 Bacteraemia did not predict treatment failure in per-protocol infants in our study; however, in view of the fairly small number of culture-positive cases, we cannot exclude a modest overall increase in risk of treatment failure.

distinguish between infants with and without bacterial infection in first-level settings.23 Bacteraemia did not predict treatment failure in per-protocol infants in our study; however, in view of the fairly small number of culture-positive cases, we cannot exclude a modest overall increase in risk of treatment failure. Positive blood culture results showed diversity in causative agents of clinical severe infection in young infants. Campylobacteraceae were the commonest pathogens causing bacteraemia and bacteraemic treatment failure. Campylobacter jejuni and Campylobacter coli are common causes of diarrhoea in infants in developing countries, including the area from which these infants were recruited;26 however, none of the young infants in this study who grew campylobacteraceae from their blood had a history of diarrhoea. A related species, Campylobacter fetus, has a well known association with adverse birth outcomes in cattle.27 It is possible that the fastidious nature and special growth requirements of campylobacteraceae have resulted in scant recognition of its role as a pathogen of newborn bloodstream infections.28 One concern was not borne out—namely, that high rates of antimicrobial resistance among newborn pathogens reported from hospital-based series from low-income settings would be a problem for management of clinical severe infection in young infants in the community. Susceptibility testing showed that a regimen based on penicillin or amoxicillin plus gentamicin would cover more than 80% of pathogens encountered.

orn pathogens reported from hospital-based series from low-income settings would be a problem for management of clinical severe infection in young infants in the community. Susceptibility testing showed that a regimen based on penicillin or amoxicillin plus gentamicin would cover more than 80% of pathogens encountered. The mix in severity of clinical severe infections in our study was affected by the proportion of infants identified at a very early stage by household surveillance. However, we made this trade-off to ensure adequate enrolment of infants aged 0–6 days who otherwise do not present to facilities. Another consideration is whether the milder spectrum of illness, potentially including many children without a bacterial infection, could have biased the results of our trial towards equivalence. Although some enrolled children will undoubtedly not have had a bacterial infection, in a pilot study3 using the same inclusion criteria as this study, a significantly higher rate of treatment failure was reported among children treated with co-trimoxazole and gentamicin compared with those treated with procaine benzylpenicillin and gentamicin.

ldren will undoubtedly not have had a bacterial infection, in a pilot study3 using the same inclusion criteria as this study, a significantly higher rate of treatment failure was reported among children treated with co-trimoxazole and gentamicin compared with those treated with procaine benzylpenicillin and gentamicin. Our study shares some limitations with the trials in Africa and Bangladesh.22, 23 First, we did not mask clinicians or participants for ethical reasons;22 placebo injections were not judged justifiable in sick young infants. Second, families of enrolled infants most probably refused admission because their child was not perceived to be very sick.22, 23 Critical illness was a chosen exclusion criterion in the design phase because we believed treatment of very sick young infants with simplified antibiotic regimens would have been unethical. Nevertheless, 35 (1%) of 2453 trial participants died despite antimicrobial treatment and 265 (12%) of 2251 per-protocol infants did not respond to treatment, indicating a substantial level of severe illness among enrolled children. Third, concern has been raised about observer bias in assessing the soft study endpoints.23 Quality of clinical assessment was ensured through repeat training and supervision, and all treatment failures were confirmed independently by another study clinician, as described elsewhere.8, 29 Fourth, most infants were diagnosed clinically without availability of laboratory tests to indicate the presence or absence of infection. Since the main aim of this study was to find pragmatic solutions to scant access to antibiotic treatment for most young infants with newborn infections in low-resource settings, in environments where laboratory testing will not be available in the near future, we chose to use clinical definitions. Nevertheless, it is noteworthy that because of the absence of reliable laboratory tests to diagnose newborn infections, and the high risk of adverse outcomes, it is common practice to diagnose neonatal sepsis clinically and treat empirically with antibiotics pending laboratory results, even in high-resource hospital settings. Finally, although low case-fatality was recorded with the study treatments under stringent trial conditions and close monitoring, application of these findings in programmatic settings might not yield such impressive results when adherence with treatment might be lower. The programmatic effect of implementation of these guidelines must be assessed carefully.

recorded with the study treatments under stringent trial conditions and close monitoring, application of these findings in programmatic settings might not yield such impressive results when adherence with treatment might be lower. The programmatic effect of implementation of these guidelines must be assessed carefully. Our data show that simplified antibiotic regimens with fewer injections administered closer to home by a trained health provider were as efficacious as a reference strategy comprising more penicillin-gentamicin injections for treatment of young infants with clinical severe infection whose families declined admission. Overall, this evidence supports easier to administer regimens and can inform national and international policy on the treatment of sick young infants in situations when hospital admission is not accepted. Supplementary Material Supplementary appendix

Our data show that simplified antibiotic regimens with fewer injections administered closer to home by a trained health provider were as efficacious as a reference strategy comprising more penicillin-gentamicin injections for treatment of young infants with clinical severe infection whose families declined admission. Overall, this evidence supports easier to administer regimens and can inform national and international policy on the treatment of sick young infants in situations when hospital admission is not accepted. Supplementary Material Supplementary appendix Acknowledgments The study was funded by the Saving Newborn Lives initiative of Save the Children through support from the Bill & Melinda Gates Foundation; and by WHO and USAID. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of Save the Children. FM, IN, FJ, and SS received research training support from the National Institute of Health's Fogarty International Center (1 D43 TW007585-01). We thank Shahida Qureshi, Tayyab un Nisa, Aneeta Hotwani, and Furqan Kabir (Infectious Disease Research Laboratory, Aga Khan University) for processing of biological samples; and Najeeb ur Rehman and Zaid Bhatti (Data Management Unit, Aga Khan University) for data processing.

y International Center (1 D43 TW007585-01). We thank Shahida Qureshi, Tayyab un Nisa, Aneeta Hotwani, and Furqan Kabir (Infectious Disease Research Laboratory, Aga Khan University) for processing of biological samples; and Najeeb ur Rehman and Zaid Bhatti (Data Management Unit, Aga Khan University) for data processing. Contributors FM, IN, SST, and AKMZ contributed to the writing of the report. SST also oversaw study implementation and data collection. SS supervised processing of microbiology specimens and analysed and interpreted blood culture data. BB and FJ contributed to study monitoring and protocol adherence. SC contributed to the statistical analysis plan and data analysis. IA contributed to data entry, data processing, and the preliminary data analysis. AKMZ was the principal investigator for the study and oversaw study design and implementation and data analysis. All authors reviewed and approved the final report. Declaration of interests We declare no competing interests. Figure Trial flow diagram

Contributors FM, IN, SST, and AKMZ contributed to the writing of the report. SST also oversaw study implementation and data collection. SS supervised processing of microbiology specimens and analysed and interpreted blood culture data. BB and FJ contributed to study monitoring and protocol adherence. SC contributed to the statistical analysis plan and data analysis. IA contributed to data entry, data processing, and the preliminary data analysis. AKMZ was the principal investigator for the study and oversaw study design and implementation and data analysis. All authors reviewed and approved the final report. Declaration of interests We declare no competing interests. Figure Trial flow diagram *428 children were excluded because of the presence of at least one danger sign of critical illness; 125 had low weight; 44 had a congenital malformation; 25 needed admission for a surgical reason; 87 had a history of admission in the past 2 weeks; 80 had a history of previous enrolment in the study; 55 were out of catchment area; and 106 had other comorbid conditions that needed admission. The total excluded is more than 757 because some infants fulfilled more than one criterion. †Children were included in the per-protocol analysis if they had complete or adequate clinical follow-up and complete or adequate treatment adherence (appendix p 6). Table 1 Baseline characteristics

*428 children were excluded because of the presence of at least one danger sign of critical illness; 125 had low weight; 44 had a congenital malformation; 25 needed admission for a surgical reason; 87 had a history of admission in the past 2 weeks; 80 had a history of previous enrolment in the study; 55 were out of catchment area; and 106 had other comorbid conditions that needed admission. The total excluded is more than 757 because some infants fulfilled more than one criterion. †Children were included in the per-protocol analysis if they had complete or adequate clinical follow-up and complete or adequate treatment adherence (appendix p 6). Table 1 Baseline characteristics Procaine benzylpenicillin and gentamicin (n=820) Amoxicillin and gentamicin (n=816) Procaine benzylpenicillin, gentamicin, and amoxicillin (n=817) Age at enrolment (days) 0–6 361 (44%) 360 (44%) 362 (44%) 7–59 459 (56%) 456 (56%) 455 (56%) Sex Male 465 (57%) 419 (51%) 425 (52%) Female 355 (43%) 397 (49%) 392 (48%) Weight at enrolment (g) <2000 62 (8%) 67 (8%) 81 (10%) 2000–2499 145 (18%) 144 (18%) 128 (16%) ≥28 613 (75%) 605 (74%) 608 (74%) Weight-for-age (Z score) <–2 303 (37%) 322 (39%) 315 (39%) ≥–2 517 (63%) 494 (61%) 502 (61%) Number of signs present One 720 (88%) 717 (88%) 704 (86%) More than one 100 (12%) 99 (12%) 113 (14%) Fever 330 (40%) 337 (41%) 348 (43%) In isolation 296 (36%) 303 (37%) 306 (37%) Hypothermia 71 (9%) 71 (9%) 91 (11%) In isolation 43 (5%) 46 (6%) 55 (7%) Movement only when stimulated 45 (5%) 38 (5%) 39 (5%) In isolation 10 (1%) 5 (1%) 4 (<1%) Severe chest indrawing 277 (34%) 272 (33%) 269 (33%) In isolation 239 (29%) 238 (29%) 240 (29%) Poor feeding or suck 210 (26%) 205 (25%) 191 (23%) In isolation 132 (16%) 125 (15%) 99 (12%) Local infection 122 (15%) 110 (13%) 126 (15%) Facility delivery* 412 (50%) 388 (48%) 423 (52%) Maternal age (years)* 25·9 (5·5) 25·8 (5·7) 25·9 (5·5) Maternal education (years)* 0 (0–4) 0 (0–5) 0 (0–5) Data are number of children (%), mean (SD), or median (IQR). Signs in isolation were the only clinical sign present at enrolment.

tion 122 (15%) 110 (13%) 126 (15%) Facility delivery* 412 (50%) 388 (48%) 423 (52%) Maternal age (years)* 25·9 (5·5) 25·8 (5·7) 25·9 (5·5) Maternal education (years)* 0 (0–4) 0 (0–5) 0 (0–5) Data are number of children (%), mean (SD), or median (IQR). Signs in isolation were the only clinical sign present at enrolment. * Data missing for place of delivery (n=1), maternal age (n=297), and maternal education (n=194). Table 2 Primary and secondary treatment outcomes (per-protocol population)*

tion 122 (15%) 110 (13%) 126 (15%) Facility delivery* 412 (50%) 388 (48%) 423 (52%) Maternal age (years)* 25·9 (5·5) 25·8 (5·7) 25·9 (5·5) Maternal education (years)* 0 (0–4) 0 (0–5) 0 (0–5) Data are number of children (%), mean (SD), or median (IQR). Signs in isolation were the only clinical sign present at enrolment. * Data missing for place of delivery (n=1), maternal age (n=297), and maternal education (n=194). Table 2 Primary and secondary treatment outcomes (per-protocol population)* Procaine benzylpenicillin and gentamicin (n=747) Amoxicillin and gentamicin (n=751) Procaine benzylpenicillin, gentamicin, and amoxicillin (n=753) Risk difference†(95% CI) Risk difference‡(95% CI) Primary outcome Treatment failure within 7 days of enrolment 90 (12%) 76 (10%) 99 (13%) −1·9 (−5·1 to 1·3) 1·1 (−2·3 to 4·5) Initial reason for treatment failure Death 6 4 6 .. .. Admission 17 13 21 .. .. Clinical deterioration 14 12 17 .. .. New sign on or after day 3 8 11 3 .. .. Persistence of signs at day 4 22 12 28 .. .. Recurrence of signs on or after day 5 13 15 16 .. .. Persistence of signs at day 8 0 0 0 .. .. Serious adverse event 1 1 0 .. .. Antibiotic change because of infectious comorbidity 9 8 8 .. .. Secondary outcomes Admission within 7 days of enrolment 26 (3%) 20 (3%) 30 (4%) −0·8 (−2·6 to 0·9) 0·5 (−1·4 to 2·4) Died within 7 days of enrolment 11 (1%) 7 (1%) 10 (1%) −0·5 (−1·6 to 0·6) −0·1 (−1·3 to 1·0) Died at any time before day 15 visit 13 (2%) 9 (1%) 12 (2%) −0·5 (−1·8 to 0·7) −0·1 (−1·4 to 1·1) Not classified as treatment failure with follow-up on day 11 or day 15 642 (86%) 661 (88%) 643 (85%) .. .. Admission at any time between day 8 and day 15 visits§ 6 (1%) 2 (<1%) 1 (<1%) .. .. Died any time between day 8 and day 15 visits§ 0 1 (<1%) 2 (<1%) .. .. Non-fatal relapse at any time between day 8 and day 15 visits§ 20 (3%) 9 (1%) 6 (1%) −1·8 (−3·4 to −0·1) −2·2 (−3·7 to −0·6) Data are number of children (%), unless otherwise stated.

time between day 8 and day 15 visits§ 6 (1%) 2 (<1%) 1 (<1%) .. .. Died any time between day 8 and day 15 visits§ 0 1 (<1%) 2 (<1%) .. .. Non-fatal relapse at any time between day 8 and day 15 visits§ 20 (3%) 9 (1%) 6 (1%) −1·8 (−3·4 to −0·1) −2·2 (−3·7 to −0·6) Data are number of children (%), unless otherwise stated. * Secondary outcomes for all randomly allocated infants are presented in the appendix (p 4). † Difference between amoxicillin and gentamicin, and procaine benzylpenicillin and gentamicin. ‡ Difference between procaine benzylpenicillin, gentamicin, and amoxicillin, and procaine benzylpenicillin and gentamicin. § Denominator was children not classified as treatment failures with follow-up to day 11 or day 15. Panel Case definitions Clinical severe infection Signs of clinical severe infection were defined as: • Movement only when stimulated • Not feeding well on observation • Temperature ≥38°C or <35·5°C • Severe chest indrawing (inward chest movement with every breath in 1 min) Critical illness Signs of critical illness were defined as: • Unconsciousness • Convulsions • Inability to feed • Apnoea • Inability to cry • Cyanosis • Bulging fontanelle • Major congenital malformations inhibiting oral antibiotic intake • Active bleeding needing transfusion • Surgical conditions needing hospital referral • Persistent vomiting (ie, vomiting after three attempts to feed the infant within 30 min, with the infant vomiting after each attempt)

Introduction Salmonella infections contribute substantially to global morbidity and mortality.1, 2 The best described invasive salmonella serovars are Salmonella enterica serotype Typhi (S Typhi), causing typhoid fever, and S enterica serotype Paratyphi A, B, and C (S Paratyphi A, B, and C), which cause paratyphoid fever. Other non-typhoidal salmonella (NTS) serovars that typically cause self-limiting diarrhoea can also cause systemic infections, refered to as invasive NTS (iNTS) disease.3 Globally, typhoid fever is estimated to cause 21·7 million illnesses and 217 000 fatalities annually, and iNTS disease is estimated to cause 3·4 million illnesses and 681 000 fatalities annually.1, 2

typically cause self-limiting diarrhoea can also cause systemic infections, refered to as invasive NTS (iNTS) disease.3 Globally, typhoid fever is estimated to cause 21·7 million illnesses and 217 000 fatalities annually, and iNTS disease is estimated to cause 3·4 million illnesses and 681 000 fatalities annually.1, 2 Substantial knowledge gaps exist regarding the distribution of typhoid fever and iNTS disease in Africa. The few existing studies,4, 5, 6, 7, 8 reported over differing time periods and using various protocols, have been extrapolated and contribute to existing typhoid fever estimates, which limits international generalisability. The scarcity of data in sub-Saharan Africa prompted WHO, in 2008, to request more epidemiological information to reliably estimate the incidence of typhoid fever and iNTS disease and the antimicrobial susceptibilities of the corresponding organisms.9 Consequently, between 2010, and 2014, we established 13 surveillance sites across sub-Saharan Africa in locations where typhoid fever had been previously reported. This network formed the Typhoid Fever Surveillance in Africa Program (TSAP) and served as a platform to implement standardised surveillance of febrile illness and cross-sectional studies to investigate the health-care-seeking behaviour of the surveyed populations.10, 11, 12 Here, we present the adjusted incidence estimates of typhoid fever and iNTS disease and the antimicrobial susceptibility profiles of the causative agents at the 13 selected surveillance sites. Research in context Evidence before this study

Substantial knowledge gaps exist regarding the distribution of typhoid fever and iNTS disease in Africa. The few existing studies,4, 5, 6, 7, 8 reported over differing time periods and using various protocols, have been extrapolated and contribute to existing typhoid fever estimates, which limits international generalisability. The scarcity of data in sub-Saharan Africa prompted WHO, in 2008, to request more epidemiological information to reliably estimate the incidence of typhoid fever and iNTS disease and the antimicrobial susceptibilities of the corresponding organisms.9 Consequently, between 2010, and 2014, we established 13 surveillance sites across sub-Saharan Africa in locations where typhoid fever had been previously reported. This network formed the Typhoid Fever Surveillance in Africa Program (TSAP) and served as a platform to implement standardised surveillance of febrile illness and cross-sectional studies to investigate the health-care-seeking behaviour of the surveyed populations.10, 11, 12 Here, we present the adjusted incidence estimates of typhoid fever and iNTS disease and the antimicrobial susceptibility profiles of the causative agents at the 13 selected surveillance sites. Research in context Evidence before this study We did a literature search using PubMed with the following search terms: (“typhoid” OR “typhoid fever” OR “Salmonella Typhi” OR “S Typhi” OR “salmonella infection” OR “enteric fever” OR “non-typhoidal salmonella” OR “NTS”) AND (“incidence”OR “rate” OR “frequency” OR “prevalence” OR “morbidity” OR “burden” OR “surveillance” OR “epidemiology”). We restricted publication dates from Dec 31, 1995, to July 30, 2016, and no language restrictions were applied. The date of our last search was July 30, 2016.

OR “non-typhoidal salmonella” OR “NTS”) AND (“incidence”OR “rate” OR “frequency” OR “prevalence” OR “morbidity” OR “burden” OR “surveillance” OR “epidemiology”). We restricted publication dates from Dec 31, 1995, to July 30, 2016, and no language restrictions were applied. The date of our last search was July 30, 2016. Salmonella infections are a major cause of global morbidity and mortality; however, substantial knowledge gaps exist with regards to the distribution and incidence of disease caused by Salmonella enterica serotype Typhi and invasive non-typhoidal salmonella (iNTS) disease in sub-Saharan Africa. Before the Typhoid Fever Surveillance in Africa Program (TSAP), estimates of typhoid fever incidence data from Africa were available from four vaccine trials and one population-based study in Kenya. Other estimates of invasive salmonella infections originated from different descriptions of bacteraemia in febrile patients in The Gambia, Malawi, Mozambique, and Kenya. These few, unstandardised, published data are not sufficient for understanding the burden of the disease in sub-Saharan Africa.

-based study in Kenya. Other estimates of invasive salmonella infections originated from different descriptions of bacteraemia in febrile patients in The Gambia, Malawi, Mozambique, and Kenya. These few, unstandardised, published data are not sufficient for understanding the burden of the disease in sub-Saharan Africa. In 2008, WHO expressed the necessity for more epidemiological information to estimate the incidence and antimicrobial susceptibility of invasive salmonella disease. Consequently, in January, 2009, the International Vaccine Institute (Seoul, South Korea) and the Kenya Medical Research Institute (Kilifi, Kenya) co-hosted a meeting with five other international institutions and 28 investigators from 14 research sites across sub-Saharan Africa. The purpose of the meeting was to review existing data on invasive salmonella infections in sub-Saharan Africa and surveillance infrastructure from sites, and to discuss the way forward to investigate invasive salmonella in the African region. These 28 investigators and the five international institutions presented their data on invasive bacterial disease, focusing on invasive salmonellosis.

lla infections in sub-Saharan Africa and surveillance infrastructure from sites, and to discuss the way forward to investigate invasive salmonella in the African region. These 28 investigators and the five international institutions presented their data on invasive bacterial disease, focusing on invasive salmonellosis. The data indicated the presence of typhoid fever and iNTS disease; however, the studies were not standardised in design, data collection, and laboratory techniques. The meeting concluded that unless standardised methods of data collection and diagnostic procedure were used across countries, and patterns of health-care utilisation were understood and accounted for, the real disease burden of invasive salmonella infections in the region would remain unclear. As a result, a consortium was established and members agreed to form a network of surveillance sites in sub-Saharan Africa in areas with previous reports of cases of typhoid fever. The TSAP was created to address the knowledge gaps on the incidence and antimicrobial resistance patterns of invasive salmonella infections at different countries with previous reports of typhoid fever cases in sub-Saharan Africa. TSAP created a network of 13 surveillance sites across ten countries, and implemented cross-sectional studies to investigate the health-care-seeking behaviour of the populations under surveillance. Added value of this study

The TSAP was created to address the knowledge gaps on the incidence and antimicrobial resistance patterns of invasive salmonella infections at different countries with previous reports of typhoid fever cases in sub-Saharan Africa. TSAP created a network of 13 surveillance sites across ten countries, and implemented cross-sectional studies to investigate the health-care-seeking behaviour of the populations under surveillance. Added value of this study Original data collected in TSAP represent the most comprehensive standardised analysis done in sub-Saharan Africa of the incidence and antimicrobial resistance patterns of invasive salmonella infections. The results describe the incidence estimated, adjusted by health-care-seeking behaviour, and antimicrobial susceptibility of typhoid fever and iNTS diseases from 13 sites in ten sub-Saharan Africa countries. For typhoid fever disease, we estimate that the overall incidence is two to three times higher than a previous estimate (10–100 cases per 100 000 people), and is in some settings similar to data from Asia, where the burden is known to be very high. The data also revealed that children aged 2–14 years bear the greatest burden of the disease. For iNTS disease, the data also reflect a high incidence, especially in young children, and in specific sites (Ghana) the incidence could be more than five times that previously estimated. Implications of all the available evidence

Original data collected in TSAP represent the most comprehensive standardised analysis done in sub-Saharan Africa of the incidence and antimicrobial resistance patterns of invasive salmonella infections. The results describe the incidence estimated, adjusted by health-care-seeking behaviour, and antimicrobial susceptibility of typhoid fever and iNTS diseases from 13 sites in ten sub-Saharan Africa countries. For typhoid fever disease, we estimate that the overall incidence is two to three times higher than a previous estimate (10–100 cases per 100 000 people), and is in some settings similar to data from Asia, where the burden is known to be very high. The data also revealed that children aged 2–14 years bear the greatest burden of the disease. For iNTS disease, the data also reflect a high incidence, especially in young children, and in specific sites (Ghana) the incidence could be more than five times that previously estimated. Implications of all the available evidence The results of this study underscore the need for preventive measures, including vaccines, improved sanitation and hygiene, malaria control, antiretroviral therapy programmes, and improved nutrition. The results also emphasise that the implementation of effective antimicrobials might be impaired by the presence and potential increase of drug-resistance salmonella strains in the region. The advent of typhoid conjugate vaccines might provide more powerful tools to control typhoid fever; the first vaccine, which was manufactured in India, has already been submitted to WHO for prequalification. Data from this study will be included in the GAVI Alliance review of potential subsidies for typhoid fever vaccines in 2017; their recommendation will be crucial for the deployment of these vaccines. Hence, an urgent need exists to understand the pragmatic aspects of vaccine targeting and delivery, particularly given the burden of disease in children, the associated risk factors, and the focal nature of the disease. Further assessment of the incidence in infants (0–5 months vs 6–11 months) and data on severe typhoid fever or iNTS, including mortality, is crucial to determine the potential effect of future vaccines. Our follow-on study—Severe Typhoid in Africa (SETA)—which investigates severe typhoid burden, is underway.

e disease. Further assessment of the incidence in infants (0–5 months vs 6–11 months) and data on severe typhoid fever or iNTS, including mortality, is crucial to determine the potential effect of future vaccines. Our follow-on study—Severe Typhoid in Africa (SETA)—which investigates severe typhoid burden, is underway. Methods Study design, site selection, and participants We used a multicentre, population-based, prospective surveillance study design. Selection of the surveillance sites in sub-Saharan Africa was not random; locations were eligible if they had evidence of previous typhoid fever, a laboratory infrastructure suitable for blood culture, an onsite health-care facility, and staff experienced in microbiological laboratory research.10 13 sites in ten countries were selected (figure 1), four of which already had established surveillance systems: Pietermaritzburg, South Africa; Asante Akim North, Ghana; Moshi Urban District and Moshi Rural District, Tanzania; and Kibera, Kenya. Four sites were part of the International Network for the Demographic Evaluation of Populations and Their Health (INDEPTH): Polesgo and Nioko II, Burkina Faso; Butajira, Ethiopia; and Bandim, Guinea-Bissau. These sites had functional Health and Demographic Surveillance Systems (HDSS) in place.13 Additional surveillance sites were Isotry and Imerintsiatosika, Madagascar; Pikine, Senegal; and East Wad Medani, Sudan. The surveillance system in Kibera was established before TSAP with an active, population-based surveillance component. Home visits were done once every 2 weeks to screen for febrile patients and encourage visits to the affiliated health-care facility. Active surveillance in Kibera was continued throughout TSAP. All other sites implemented passive surveillance.10 The ethics committees of all collaborating institutions and the International Vaccine Institute (Seoul, South Korea) approved the study protocol.

and encourage visits to the affiliated health-care facility. Active surveillance in Kibera was continued throughout TSAP. All other sites implemented passive surveillance.10 The ethics committees of all collaborating institutions and the International Vaccine Institute (Seoul, South Korea) approved the study protocol. The catchment area for each site was determined through health-care facility records and through accessible administrative and demographic data.11 We determined the population of each catchment area using the latest census or the INDEPTH database. We categorised sites as urban, rural, or other using setting classifications at each site. Surveillance was implemented in each study location for a period of at least 12 months and recruitment occurred at primary, secondary, and tertiary health-care facilities. Recruitment was open to outpatients and inpatients who visited any of the health-care facilities participating in TSAP, who resided within the catchment area and presented with tympanic (≥38·0°C) or axillary temperature (≥37·5°C). Inpatients with a reported history of fever for 72 h or longer were excluded, as were patients with residence outside of the catchment area. Asante Akim North recruited children younger than age 15 years only; other sites recruited patients of all ages. Written informed consent preceded recruitment and clinical appraisal forms were completed for all participants.

or 72 h or longer were excluded, as were patients with residence outside of the catchment area. Asante Akim North recruited children younger than age 15 years only; other sites recruited patients of all ages. Written informed consent preceded recruitment and clinical appraisal forms were completed for all participants. Laboratory procedures We standardised laboratory, quality control, and blood sample collection procedures across sites.10 Blood (5–10 mL for adults; 1–3 mL for children) was inoculated into aerobic blood culture bottles and incubated in an automated blood culture system (BD BACTEC, Becton-Dickinson, USA, or BacT/ALERT, BioMérieux, France), with the exception of Sudan, where manual culturing with daily subculturing for up to 5 days was instituted. Gram staining and bacterial identification were done with standard microbiological techniques.14 Quality control of preanalytical processes included time and temperature control measures, during which every blood culture bottle was collected, transported, and placed into the incubator. Quality control of analytical processes included sterility and function control of culture media, controls of biochemical reactions, and antimicrobial susceptibility testing. For the quality control of manual culturing in Sudan, additionally, blood culture bottles were inoculated weekly with a suspension containing Escherichia coli or Staphylococcus aureus references. Inoculated blood culture bottles were incubated overnight and verified for growth by subculture.

ial susceptibility testing. For the quality control of manual culturing in Sudan, additionally, blood culture bottles were inoculated weekly with a suspension containing Escherichia coli or Staphylococcus aureus references. Inoculated blood culture bottles were incubated overnight and verified for growth by subculture. Contaminants were defined as organisms not typically associated with bloodstream infections; these included non-pathogens and those more commonly associated with commensal skin microbiota, including coagulase-negative Staphylococci, Bacillus spp, and Micrococcus spp. Antimicrobial susceptibility testing was done by disc diffusion according to Clinical and Laboratory Standards Institute15 standards for ampicillin, amoxicillin-clavulanic acid, chloramphenicol, co-trimoxazole, ceftriaxone, and ciprofloxacin. Multidrug resistance was defined as resistance to ampicillin or amoxicillin-clavulanic acid, chloramphenicol, and co-trimoxazole. Isolates with intermediate susceptibility were classified as resistant. Malaria blood smears were routinely done, except in South Africa. In Ethiopia, rapid diagnostic tests (SD BIOLINE Malaria Ag Pf/Pv, SD Standard Diagnostics, Yongin, South Korea) were used in addition to routine malaria blood smears.

co-trimoxazole. Isolates with intermediate susceptibility were classified as resistant. Malaria blood smears were routinely done, except in South Africa. In Ethiopia, rapid diagnostic tests (SD BIOLINE Malaria Ag Pf/Pv, SD Standard Diagnostics, Yongin, South Korea) were used in addition to routine malaria blood smears. Health-care utilisation survey and person-years of observation calculation The health-care-seeking behaviour of the populations under surveillance was investigated with the assumption that access to the TSAP health-care facility was non-uniform throughout the population.16, 17 A standardised and pretested health-care utilisation survey was implemented in a representative sample of households randomly selected from each study area.11 We investigated health-care-seeking behaviour in cases of self-reported fever lasting less than 3 days. The first choice of health-care facility in cases of fever was categorised by age-stratified groups and used to calculate the proportion of individuals from the catchment population who visited this TSAP health-care facility. This proportion constituted an adjustment factor to correct incidences. The time at risk in person-years of observation (PYO) stratified by age was calculated using the adjusted population. In HDSS sites, each resident contributed to PYO for the time present in the study area during the recruitment period. In non-HDSS sites, we calculated PYO by projecting the catchment population from the start to the end of the study recruitment period, and multiplied the calculated average population by the number of years of surveillance duration.

ontributed to PYO for the time present in the study area during the recruitment period. In non-HDSS sites, we calculated PYO by projecting the catchment population from the start to the end of the study recruitment period, and multiplied the calculated average population by the number of years of surveillance duration. Statistical analysis We established a multicountry database using FoxPro software. We excluded patients from the analysis who were recruited during pilot testing, failed to meet inclusion criteria, or had incomplete laboratory results. We estimated incidences per 100 000 PYO. Confirmed invasive salmonella cases, stratified by age group (0–1 years, 2–4 years, 5–14 years, and ≥15 years), were adjusted by the specific age-group recruitment proportion. We calculated this proportion by dividing the number of patients with complete data (numerator) by the total number of patients in the study area who had been diagnosed with a febrile illness at a recruitment facility during the surveillance period (denominator). We used health-care facility records, reviewed at the end of the surveillance activities, to estimate the number of patients diagnosed with a febrile illness. The catchment population in PYO, adjusted by health-care-seeking behaviour, was used as the denominator in crude and adjusted incidence rates (AIR).

enominator). We used health-care facility records, reviewed at the end of the surveillance activities, to estimate the number of patients diagnosed with a febrile illness. The catchment population in PYO, adjusted by health-care-seeking behaviour, was used as the denominator in crude and adjusted incidence rates (AIR). The 95% CI for AIR was derived on the log-scale and exponentiated. We used the error factor (exp[1·96/√adjusted cases]) to calculate the lower (adjusted rate/error factor) and upper (adjusted rate × error factor) 95% CIs. At the sites in Senegal, Ethiopia, and South Africa, incomplete health-care facility records did not allow for the estimation of the recruitment proportion and calculation of AIRs; for these sites we present crude rates. AIRs for typhoid fever and iNTS were assessed for all other sites. Differences in proportions of blood cultures positive for a pathogen between study years were assessed with the χ2 test (SAS, version 9.3). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

The 95% CI for AIR was derived on the log-scale and exponentiated. We used the error factor (exp[1·96/√adjusted cases]) to calculate the lower (adjusted rate/error factor) and upper (adjusted rate × error factor) 95% CIs. At the sites in Senegal, Ethiopia, and South Africa, incomplete health-care facility records did not allow for the estimation of the recruitment proportion and calculation of AIRs; for these sites we present crude rates. AIRs for typhoid fever and iNTS were assessed for all other sites. Differences in proportions of blood cultures positive for a pathogen between study years were assessed with the χ2 test (SAS, version 9.3). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between March 1, 2010, and Jan 31, 2014, we recruited 13 558 patients from 13 sites who met the inclusion criteria and resided in the catchment areas (Figure 1, Figure 2). We excluded data from 127 (1%) patients because of incomplete laboratory results; data from 13 431 patients were analysed, and 8582 patients (64%) were younger than 15 years (table 1). All patients had one blood culture sample analysed at recruitment and 11 421 (85%) were screened for malaria parasites (table 1). The proportion of contaminated blood cultures ranged from less than 1% in Imerintsiatosika to 24% in Nioko II. The proportion of blood cultures that yielded non-contaminant bacteria varied between sites, ranging from 1% in Imerintsiatosika to 9% in Kibera (table 1). In total, 568 non-contaminant bacteria were isolated from blood samples of febrile patients. The most frequent non-contaminant bacteria isolated were S Typhi (135 [24%]), NTS (94 [17%]), S aureus (70 [12%]), E coli (47 [8%]), and Streptococcus pneumoniae (43 [8%]). Of the sites with at least 2 years of surveillance (Asante Akim North, Kibera, and Pietermaritzburg), the proportion of blood cultures that were pathogen positive differed significantly between study years in Kibera only (12% at year 1 and 5% at year 2; p<0·0001; χ2 test).

[8%]), and Streptococcus pneumoniae (43 [8%]). Of the sites with at least 2 years of surveillance (Asante Akim North, Kibera, and Pietermaritzburg), the proportion of blood cultures that were pathogen positive differed significantly between study years in Kibera only (12% at year 1 and 5% at year 2; p<0·0001; χ2 test). With the exception of East Wad Medani, Salmonella spp were isolated from the blood of febrile patients at all sites (135 S Typhi and 94 iNTS isolates), which accounted for 33% or more of all isolated bacteria in all but four sites (East Wad Medani, Pietermaritzburg, Butajira, and Isotry). Seasonal variation was not observed at any site (data not shown). The most common iNTS serovars were S enterica serotype Typhimurium (38 [40%] of 94), S enterica serotype Enteriditis (11 [12%] of 94), and S enterica serotype Dublin (10 [11%] of 94). The highest AIRs for typhoid fever in the 15 years or younger age group were observed in Polesgo, Kibera, and Asante Akim North (table 2). S Paratyphi A (three isolates) was isolated in Senegal only. Among age groups of children younger than 15 years, the highest AIR for typhoid fever was observed in children aged 2–4 years from Polesgo, Asante Akim North, Moshi Urban District, and Kibera, and in children aged 5–14 years from Kibera and Polesgo (table 2). The AIR for typhoid fever in adults (aged ≥15 years) was less than 70 per 100 000 PYO at all sites except Moshi Urban District, Kibera, and Polesgo (table 2).

in children aged 2–4 years from Polesgo, Asante Akim North, Moshi Urban District, and Kibera, and in children aged 5–14 years from Kibera and Polesgo (table 2). The AIR for typhoid fever in adults (aged ≥15 years) was less than 70 per 100 000 PYO at all sites except Moshi Urban District, Kibera, and Polesgo (table 2). iNTS organisms were more frequently isolated from infants (0–1 years) or children aged 2–4 years than from adults (table 2), except for the sites in Pikine, Moshi Rural District, and Kibera. The AIR for iNTS among children aged 2–4 years was highest in Nioko II, Polesgo, and Asante Akim North. The AIR for iNTS in children younger than 15 years was less than 100 per 100 000 PYO in Kibera, Imerintsiatosika, and in both sites in Tanzania. No iNTS was isolated from sites in Sudan, South Africa, Ethiopia, and Isotry.

R for iNTS among children aged 2–4 years was highest in Nioko II, Polesgo, and Asante Akim North. The AIR for iNTS in children younger than 15 years was less than 100 per 100 000 PYO in Kibera, Imerintsiatosika, and in both sites in Tanzania. No iNTS was isolated from sites in Sudan, South Africa, Ethiopia, and Isotry. The antimicrobial susceptibility profiles of S Typhi and iNTS isolates differed between sites (table 3). Overall, 47% of S Typhi isolates and 48% of iNTS isolates were multidrug resistant. Most multidrug-resistant S Typhi isolates were obtained at the sites in Kenya, Ghana, and Tanzania (both sites combined). Multidrug-resistant iNTS isolates were isolated at the sites in Burkina Faso (both combined), Ghana, Guinea-Bissau, and Kenya (table 3). S Typhi isolates that had reduced ciprofloxacin susceptibility were cultured in Kenya and South Africa, only; one ciprofloxacin-resistant S Paratyphi A organism was isolated in Senegal. Ciprofloxacin-resistant iNTS was similarly uncommon, isolated only in Burkino Faso (once at the Nioko II site) and in Ghana. One iNTS isolate in Kenya was resistant to ceftriaxone (table 3). Discussion This study identified Salmonella as a major cause of invasive bacterial febrile illness across sub-Saharan Africa, affecting children aged 2–14 years rather than adults, and arising in both high-population and low-population density settings. Other major causes of invasive bacterial febrile illnesses varied by country; E coli and S aureus were the most frequent non-Salmonella pathogens isolated from blood.

cross sub-Saharan Africa, affecting children aged 2–14 years rather than adults, and arising in both high-population and low-population density settings. Other major causes of invasive bacterial febrile illnesses varied by country; E coli and S aureus were the most frequent non-Salmonella pathogens isolated from blood. Results from previous studies18, 19 suggest that typhoid fever in some sub-Saharan Africa settings occurs predominately in urban settlements with high-population densities, and that disease incidence could have been overestimated by the use of the Widal test. Our study, done using a standardised protocol in both urban and rural settings, indicated high incidences of typhoid fever and iNTS in areas with high-population and low-population densities. Separate analyses done at the Ghana site confirmed this observation and revealed a higher disease incidence in children living in rural areas than in those living in urban areas.20 Furthermore, we observed variable incidences of typhoid fever and iNTS among neighbouring populations in Burkina Faso, and in the same populations in Kenya and Ghana in consecutive years, indicating a focal nature and a fluctuating burden of iNTS disease.

ren living in rural areas than in those living in urban areas.20 Furthermore, we observed variable incidences of typhoid fever and iNTS among neighbouring populations in Burkina Faso, and in the same populations in Kenya and Ghana in consecutive years, indicating a focal nature and a fluctuating burden of iNTS disease. A previous global estimate of the burden of typhoid fever indicated that south-central and east-central Asia had the highest incidences of typhoid fever with more than 100 cases per 100 000 people annually; Africa was estimated to have a medium incidence (10–100 cases per 100 000).1 The AIR for typhoid fever estimated in our study reveals a higher burden than previously estimated.1 Four sites had an overall AIR for typhoid fever of more than 100 per 100 000 PYO, five sites had an AIR for typhoid fever of more than 100 per 100 000 PYO in children younger than 15 years, and six sites had an AIR for typhoid fever of more than 100 per 100 000 PYO in at least one age group. Similar to the Diseases of the Most Impoverished programme done in Asia,21 our results show that children aged 2–14 years bear the greatest burden of typhoid fever. Notably, our data indicate that the AIR for typhoid fever at TSAP sites was equal to or even greater than incidences reported in five Asian countries in the early 2000s.21, 22

ses of the Most Impoverished programme done in Asia,21 our results show that children aged 2–14 years bear the greatest burden of typhoid fever. Notably, our data indicate that the AIR for typhoid fever at TSAP sites was equal to or even greater than incidences reported in five Asian countries in the early 2000s.21, 22 For iNTS disease, we observed an AIR equal or higher than previously estimated and a bimodal age distribution with very young children and adults being the key age group for symptomatic infection.2 This age distribution differed from that observed for typhoid fever, in which children aged 2–14 years were the most affected, and emphasises substantial differences in the epidemiology of typhoid fever and iNTS disease. Malaria, malnutrition, and HIV infections have been reported to be associated with iNTS disease in Africa.23 At TSAP sites, a higher AIR for iNTS was observed in children with a malaria positivity rate of 30% or more than in those with a lower positivity rate; this observation was confirmed in a separate analysis.24

e. Malaria, malnutrition, and HIV infections have been reported to be associated with iNTS disease in Africa.23 At TSAP sites, a higher AIR for iNTS was observed in children with a malaria positivity rate of 30% or more than in those with a lower positivity rate; this observation was confirmed in a separate analysis.24 Results of our study identified a high prevalence of resistance against first-line antimicrobials in both S Typhi and iNTS infections. Reduced susceptibility to ciprofloxacin was identified in S Typhi from Kibera and Pietermaritzburg. Multidrug-resistant iNTS isolates were isolated at several sites and have been isolated in sub-Saharan Africa previously.18, 25, 26 Furthermore, a single iNTS isolate from Kibera showed resistance to ceftriaxone. Genomic analyses27 have described the spread of S Typhi haplotype H58 into Africa, a multidrug-resistant strain associated with reduced ciprofloxacin susceptibility. The susceptibility patterns observed in our study are concerning, particularly because some antimicrobial-resistant S Typhi can have a selective fitness advantage.28 Concerted measures are needed to monitor the emergence of fluoroquinolone-resistant Salmonella.29, 30, 31, 32

d with reduced ciprofloxacin susceptibility. The susceptibility patterns observed in our study are concerning, particularly because some antimicrobial-resistant S Typhi can have a selective fitness advantage.28 Concerted measures are needed to monitor the emergence of fluoroquinolone-resistant Salmonella.29, 30, 31, 32 We made all efforts to minimise bias; however, our study has some limitations. First, we did not adjust the disease incidences for blood culture sensitivity, which is approximately 40–60% of bone marrow culture.33, 34, 35, 36, 37 This correction factor is inconsistently applied in studies and, if applied here, the incidences presented would double. The restricted sensitivity of blood culture to detect Salmonella pathogens applies to other bacterial pathogens as well—ie, S pneumoniae and Haemophilus influenzae type b—however, those are universally recognised as important infections for which vaccines are cost-effective, and vaccination programmes have been established. Second, our results represent incidence in sites selected because of their previous reports on typhoid fever. The site selection strategy limits the generalisability of the AIR to other locations and might result in the reduced detection of iNTS disease. Third, given the vast number of patients (and restricted diagnostics capacity), not every patient with a history of fever was enrolled—eg, at sites where inpatients were recruited, patients with a fever for 72 h or longer were excluded to minimise the inclusion of patients pretreated with antimicrobials and to maximise blood culture yield. Fourth, the proportion of the catchment population using the TSAP health-care facilities for febrile illness was low in some sites, and antimicrobial treatment before blood collection and its potential effect on blood culture sensitivity were not assessed. Fifth, the classification of the settings as either urban, rural, semi-urban, or urban-slum reflects the classification commonly used at each site and does not refer to a standard definition; instead, the population density of each site is presented to make setting comparisons.

od culture sensitivity were not assessed. Fifth, the classification of the settings as either urban, rural, semi-urban, or urban-slum reflects the classification commonly used at each site and does not refer to a standard definition; instead, the population density of each site is presented to make setting comparisons. Sixth, sites with no previous experience of blood collection for blood culture assessment had a higher incidence of contamination than sites with previous experience of blood collection (South Africa, Ghana, Tanzania, and Kenya); these incidences might have led to errors in clinical interpretation and uncertainty to distinguish between clinically significant bacteraemia and contamination. Available isolates and blood samples collected from participants were PCR tested at the reference lab to minimise misclassification of isolated organisms. Seventh, the site in Ghana recruited only children younger than 15 years and the proportion of recruited inpatients varied greatly across all sites. Finally, data on disease severity, complications, mortality, and HIV status were not assessed because these were not primary study objectives. Despite these limitations, this multisite study, the largest study of typhoid fever and iNTS done across sub-Saharan Africa to date, provides the most current and accurate incidence figures for these major infectious diseases across the continent and has substantial implications for their control.

ary study objectives. Despite these limitations, this multisite study, the largest study of typhoid fever and iNTS done across sub-Saharan Africa to date, provides the most current and accurate incidence figures for these major infectious diseases across the continent and has substantial implications for their control. We surmise that the incidence of invasive salmonella infections among children in sub-Saharan Africa is much higher than previously estimated, underscoring the need for preventive measures. Therefore, until access to safe drinking water and improved sanitation is greatly expanded, the prevention of typhoid fever will require immunisation and effective treatment options.38 The advent of new typhoid fever conjugate vaccines might provide more powerful tools for disease control; the first typhoid fever conjugate vaccine (Bharat Biotech, Hyderabad) has been submitted to WHO for prequalification. Data from TSAP will be incorporated into the GAVI Alliances' review of potential subsidies for typhoid fever vaccines in 2017; their recommendation will be crucial for deployment of these vaccines. Hence, the need to understand the pragmatic aspects of vaccine targeting and delivery is pressing, particularly given the burden of disease in children, the associated risk factors, and the focal and unpredictable nature of the disease. Similarly, in the absence of vaccines targeting iNTS disease, prevention will require a major investment in infrastructure for diagnosis and effective treatment of iNTS disease. When appropriate diagnosis and treatment are available, the use of effective antimicrobials might be impaired by the presence and potential increase of multidrug-resistant salmonella. Further assessment of incidences in infants (0–5 months vs 6–11 months) and data on severe typhoid fever or iNTS, including mortality, is crucial to determine the potential effect of future vaccines. We are currently undertaking a follow-on study—Severe Typhoid in Africa (SETA)—which investigates severe typhoid burden.

Further assessment of incidences in infants (0–5 months vs 6–11 months) and data on severe typhoid fever or iNTS, including mortality, is crucial to determine the potential effect of future vaccines. We are currently undertaking a follow-on study—Severe Typhoid in Africa (SETA)—which investigates severe typhoid burden. We conclude that typhoid fever and iNTS disease are major agents of invasive bloodstream infections in urban and rural locations, affecting children more commonly than adults across sub-Saharan Africa. Immunisation of high-risk age groups with existing and new vaccines should be a priority. The next generation of epidemiological studies in sub-Saharan Africa needs to provide better data regarding the severity and mortality of typhoid fever and iNTS to guide the introduction of new typhoid and iNTS vaccines. Lastly, the accelerated development and introduction of iNTS vaccines needs to become a fundamental goal on the global health agenda. For the study protocol see http://www.ivi.int/?page_id=12479&uid=922&mod=document

We conclude that typhoid fever and iNTS disease are major agents of invasive bloodstream infections in urban and rural locations, affecting children more commonly than adults across sub-Saharan Africa. Immunisation of high-risk age groups with existing and new vaccines should be a priority. The next generation of epidemiological studies in sub-Saharan Africa needs to provide better data regarding the severity and mortality of typhoid fever and iNTS to guide the introduction of new typhoid and iNTS vaccines. Lastly, the accelerated development and introduction of iNTS vaccines needs to become a fundamental goal on the global health agenda. For the study protocol see http://www.ivi.int/?page_id=12479&uid=922&mod=document Acknowledgments This study was supported by the Bill & Melinda Gates Foundation (OPPGH5231). The findings and conclusions contained within are our own and do not necessarily reflect positions or policies of the Bill & Melinda Gates Foundation or the US Centers for Disease Control and Prevention. International Vaccine Institute acknowledges its donors, including the South Korea and the Swedish International Development Cooperation Agency (Sida). Research infrastructure at the Moshi site was supported by the US National Institutes of Health (R01TW009237; U01 AI062563; R24 TW007988; D43 PA-03–018; U01 AI069484; U01 AI067854; P30 AI064518), and by the UK Biotechnology and Biological Sciences Research Council (BB/J010367). SB is a Sir Henry Dale Fellow, jointly funded by the Wellcome Trust and the Royal Society (100087/Z/12/Z). We are grateful to Sooyoung Kwon for her invaluable administrative support of the project. We also thank all patients who consented to participate and hospital and clinic staff for their support. We especially acknowledge those who personally contributed to the implementation and execution of the study, additional to routine clinical work. Without the efforts of dedicated field staff this research would not have been possible.

atients who consented to participate and hospital and clinic staff for their support. We especially acknowledge those who personally contributed to the implementation and execution of the study, additional to routine clinical work. Without the efforts of dedicated field staff this research would not have been possible. Declaration of interests FM, JAC, TFW, and RFB report grants from Bill & Melinda Gates Foundation during the conduct of the study. All other authors declare no competing interests. Contributors FM and TFW contributed to study conception and design, analysis of data, interpretation of results, and drafting and editing of the paper. FK, JM, UP, VvK, EDM, and JDC contributed to study conception and design, data interpretation, and editing of the paper. MA, GDP, LMCE, VvK, and JKP contributed to data analysis. KT and BL contributed to study conception and design, data acquisition in the field, interpretation of the results, and editing of the paper. VvK, LMCE, SEP, CGM, CN, and JI drafted the manuscript and contributed to interpretation of results and editing of the paper. RFB, MA, FK, JM, UP, TFW, VvK, PA, YA-S, AA, MB-A, JAC, LMCE, JFD, NG, JTH, JI, HJJ, KHK, JMM, RK, RR, AGS, SEP, HJS, AS, MT, MRW, BY, MAET, HMB, LC, AJ, SVL, TMR, NS, and AT contributed to data acquisition in the field, interpretation of results, and editing of the paper. SB, JIC, UP, DMD, BSF, LPK, AAN, NVMH, BO, HR, TJLR, ES, HS-G, and AS contributed to laboratory work, interpretation of results, and editing of the paper. All authors read and approved the final draft.

VL, TMR, NS, and AT contributed to data acquisition in the field, interpretation of results, and editing of the paper. SB, JIC, UP, DMD, BSF, LPK, AAN, NVMH, BO, HR, TJLR, ES, HS-G, and AS contributed to laboratory work, interpretation of results, and editing of the paper. All authors read and approved the final draft. Figure 1 Sites participating in the Typhoid Fever Surveillance in Africa Program Figure 2 Visits to health-care facilities and recruitment of patients during surveillance period at each site NA=not available. *Data on health facility visits were collected retrospectively, after completion of surveillance period. Diagnosis of febrile illnesses was used at sites when temperature of patients was not recorded. †Number estimated by the proportion of the population under demographic surveillance at each respective site. ‡In Tanzania, before Nov 11, 2011, every fifth eligible patient was recruited; from Nov 11, 2011, every second eligible patient was recruited. This recruitment pattern was applied to this number. Table 1 Demographics and laboratory results of the sites in the Typhoid Fever Surveillance in Africa Program

‡‡ Population data for Tanzania were provided by the National Bureau of Statistics and correspond to the 2012 population and housing census. §§ Patients who met inclusion criteria, consented to take part in the study, and had a blood culture taken and a documented blood culture result. ¶¶ Recruitment health-care facility providing outpatient services only. ‖‖ Positive for non-contaminant isolates. *** Denominator differs from all blood cultures analysed because of missing values. Malaria results are based on blood smears, except for the site in Butajira (52% of patients positive for malaria were diagnosed with malaria rapid tests). Table 2 Invasive salmonella infections across sites in the Typhoid Fever Surveillance in Africa Program

NA=not available. *Data on health facility visits were collected retrospectively, after completion of surveillance period. Diagnosis of febrile illnesses was used at sites when temperature of patients was not recorded. †Number estimated by the proportion of the population under demographic surveillance at each respective site. ‡In Tanzania, before Nov 11, 2011, every fifth eligible patient was recruited; from Nov 11, 2011, every second eligible patient was recruited. This recruitment pattern was applied to this number. Table 1 Demographics and laboratory results of the sites in the Typhoid Fever Surveillance in Africa Program Nioko II, Burkina Faso Polesgo, Burkina Faso Bandim, Guinea-Bissau Pikine, Senegal Asante Akim North, Ghana East Wad Medani, Sudan Butajira, Ethiopia Imerintsiatosika, Madagascar Isotry, Madagascar Pietermaritzburg, South Africa Moshi Urban District, Tanzania Moshi Rural District, Tanzania Kibera, Kenya* Surveillance sites Type of health-care facility (IPD, OPD) 1 hospital (IPD, OPD) 1 health-care centre (OPD) 1 hospital, 1 health-care centre (IPD, OPD) 1 hospital, 3 health-care centres (IPD, OPD) 1 hospital (IPD) 3 health-care centres (OPD) 1 hospital, 3 health-care centres (IPD, OPD) 1 health-care centre (OPD) 1 health-care centre (OPD) 1 hospital (IPD) 1 hospital (IPD, OPD) 1 hospital (IPD, OPD) 1 health-care centre (OPD) Setting† Semi-urban Semi-urban Urban Urban and urban slum Urban and rural Urban Semi-urban and rural Rural Urban Urban Urban Rural Urban slum Population density, people per km2 2204 5163 17 078 16 695 121 7209 6545 225 29 301 1191 3069 332 77 000 Surveillance period (months)‡ April, 2012, to September, 2013 (18) April, 2012, to September, 2013 (18) December, 2011, to April, 2013 (17) December, 2011, to April, 2013 (17) March, 2010, to May, 2012 (27) July, 2012, to July, 2013 (13) May, 2012, to January, 2014 (21) November, 2011, to June, 2013 (20) February, 2012, to May, 2013 (16) February, 2012, to January, 2014 (24) September, 2011, to May, 2013 (21) September, 2011, to May, 2013 (21) January, 2012, to December, 2013 (24) Source of catchment population HDSS 2011§ HDSS 2011§ HDSS 2011§ Ministry of Health 2012¶ Census 2010‖ Census 2008** HDSS 2012§ Ministry of Health 2010¶ Ministry of Health 2010¶ Census 2010†† Census 2012‡‡ Census 2012f KEMRI/CDC 2012g Collaborating research institution UoO UoO BHP IPD§ KCCR/BNITM UoG AHRI UoA UoA NICD KCMC/Duke KCMC/Duke KEMRI/US-CDC Patient demographics Patients analysed, N§§ 918 756 1021 1058 2651 644 847 976 1501 1128 406 274 1251 Median age, years (IQR) 4 (1–12) 7 (3–21) 3 (1–7) 22 (14–32) 2 (0–5) 15 (9–32) 11 (5–25) 20 (9–32) 26 (17–40) 3 (1–29) 7 (1–29) 19 (2–39) 7 (4–14) 0–1 years, n (% of N) 247 (27%) 117 (15%) 369 (36%) 9 (1%) 1114 (42%) 2 (<1%) 74 (9%) 66 (7%) 12 (1%) 427 (38%) 114 (28%) 67 (24%)

8 2651 644 847 976 1501 1128 406 274 1251 Median age, years (IQR) 4 (1–12) 7 (3–21) 3 (1–7) 22 (14–32) 2 (0–5) 15 (9–32) 11 (5–25) 20 (9–32) 26 (17–40) 3 (1–29) 7 (1–29) 19 (2–39) 7 (4–14) 0–1 years, n (% of N) 247 (27%) 117 (15%) 369 (36%) 9 (1%) 1114 (42%) 2 (<1%) 74 (9%) 66 (7%) 12 (1%) 427 (38%) 114 (28%) 67 (24%) 99 (8%) 2–4 years, n (% of N) 235 (26%) 148 (20%) 271 (27%) 23 (2%) 841 (32%) 41 (6%) 124 (15%) 87 (9%) 58 (4%) 209 (19%) 62 (15%) 37 (14%) 312 (25%) 5–14 years, n (% of N) 228 (25%) 252 (33%) 274 (27%) 255 (24%) 696 (26%) 275 (43%) 303 (36%) 184 (19%) 234 (16%) 95 (8%) 56 (14%) 26 (9%) 539 (43%) ≥15 years, n (% of N) 208 (23%) 239 (32%) 107 (10%) 771 (73%) NA 326 (51%) 346 (41%) 639 (65%) 1197 (80%) 397 (35%) 174 (43%) 144 (53%) 301 (24%) Female patients, n (% of N) 467 (51%) 404 (53%) 487 (48%) 468 (44%) 1204 (45%) 348 (54%) 433 (51%) 570 (58%) 997 (66%) 586 (52%) 211 (52%) 149 (54%) 622 (50%) Inpatients, n (% of N) 66 (7%) NA¶¶ 224 (22%) 241 (23%) 2651 (100%) NA¶¶ 31 (4%) NA¶¶ NA¶¶ 1128 (100%) 220 (54%) 156 (57%) NA¶¶ Laboratory results Total blood culture, N 918 756 1021 1058 2651 644 847 976 1501 1128 406 274 1251 Total contaminated blood cultures, n (% of N) 220 (24%) 145 (19) 125 (12%) 96 (9%) 182 (7%) 54 (8%) 90 (11%) 6 (1%) 49 (3%) 192 (17%) 8 (2%) 13 (5%) 16 (1%) Total positive blood cultures, n (% of N)‖‖ 29 (3%) 31 (4) 30 (3%) 31 (3%) 175 (7%) 16 (2%) 26 (3%) 11 (1%) 30 (2%) 51 (5%) 17 (4%) 11 (4%) 110 (9%) Positive for malaria, n (% of all patients tested)*** 430/908 (47%) 444/744(60%) 206/525(39%) 297/1058(28%) 1139/2651(43%) 254/632(40%) 110/822(13%) 19/955(2%) 2/274(1%) 0 4/406(1%) 2/274 (1%) 226/956(24%) UoO=University of Ouagadougou, Ouagadougou.

(7%) 16 (2%) 26 (3%) 11 (1%) 30 (2%) 51 (5%) 17 (4%) 11 (4%) 110 (9%) Positive for malaria, n (% of all patients tested)*** 430/908 (47%) 444/744(60%) 206/525(39%) 297/1058(28%) 1139/2651(43%) 254/632(40%) 110/822(13%) 19/955(2%) 2/274(1%) 0 4/406(1%) 2/274 (1%) 226/956(24%) UoO=University of Ouagadougou, Ouagadougou. BHP=Bandim Health Project, Bissau. IPD=Institute Pasteur de Dakar, Dakar. KCCR/BNITM=Kumasi Centre for Collaborative Research in Tropical Medicine, Kumasi/Bernhard Nocht Institute for Tropical Medicine, Hamburg, Germany. UoG=University of Gezira, Wad Medani. AHRI=Armauer Hansen Research Institute, Addis Ababa. UoA=University of Antananarivo, Antananarivo. NICD=National Institute for Communicable Diseases, Johannesburg. KCMC/Duke=Kilimanjaro Christian Medical Center, Moshi/Duke University Medical Center, Durham, NC, USA. KEMRI/US-CDC=Kenya Medical Research Institute/US Centers for Disease Control and Prevention, Nairobi. IPD=inpatient department. OPD=outpatient department. HDSS=Health and Demographic Surveillance System. KEMRI=Kenya Medical Research Institute. NA=not available. * In Kibera, active population mobilisation was done in addition to passive surveillance. † Setting reflects the classification commonly used at each site and does not refer to a standard definition. ‡ Surveillance activities were scheduled for 12 months in Burkina Faso, Guinea-Bissau, Senegal, Sudan, Ethiopia, and Madagascar and for 24 months in Ghana, Kenya, South Africa, and Tanzania. If funds allowed, the scheduled period was extended. § Population data were provided from the HDSS country office.

† Setting reflects the classification commonly used at each site and does not refer to a standard definition. ‡ Surveillance activities were scheduled for 12 months in Burkina Faso, Guinea-Bissau, Senegal, Sudan, Ethiopia, and Madagascar and for 24 months in Ghana, Kenya, South Africa, and Tanzania. If funds allowed, the scheduled period was extended. § Population data were provided from the HDSS country office. ¶ Population data for Senegal and Madagascar were provided by Ministry of Health. Population data correspond to the 2012 population census and 2010 estimated population for the area, respectively. ‖ Population data for Ghana were obtained from the Ghana Statistical Service, 2010 population, and housing census. It includes 53 towns distributed in what is now Asante Akim North and Central. ** Population data for Sudan were provided by the Statistics Department, Population Center, University of Gezira, Sudan, and correspond to year 2008. †† Population data for South Africa were provided by the Statistics Department in South Africa and corresponds to the 2011 census. ‡‡ Population data for Tanzania were provided by the National Bureau of Statistics and correspond to the 2012 population and housing census. §§ Patients who met inclusion criteria, consented to take part in the study, and had a blood culture taken and a documented blood culture result. ¶¶ Recruitment health-care facility providing outpatient services only. ‖‖ Positive for non-contaminant isolates.

*** Denominator differs from all blood cultures analysed because of missing values. Malaria results are based on blood smears, except for the site in Butajira (52% of patients positive for malaria were diagnosed with malaria rapid tests). Table 2 Invasive salmonella infections across sites in the Typhoid Fever Surveillance in Africa Program Proportion of individuals from study population visiting recruitment facility in case of fever (95% CI) PYO estimation Recruitment proportion Salmonella Typhi iNTS Study population Study population adjusted by health-seeking behaviour PYO Crude cases Crude incidence per 100 000 PYO Cases adjusted for recruitment Adjusted incidence per 100 000 PYO (95% CI) Crude cases Crude incidence per 100 000 PYO Cases adjusted for recruitment Adjusted incidence per 100 000 PYO (95% CI) Nioko II, Burkina Faso 0–1 years 81% (74–88) 2208 1788 2097 247/1297 (19%) 0 0 0·0 0 (0–0) 3 143 15·8 753 (460–1233) 2–4 years 81% (75–86) 1823 1477 2097 235/1259 (19%) 1 48 5·3 251 (107–590) 3 143 16·0 753 (460–1233) 5–14 years 81% (78–84) 4295 3479 4889 228/889 (26%) 4 82 15·4 315 (191–519) 3 61 12·0 236 (133–420) <15 years NA 8326 6744 9083 NA 5 55 20·6 227 (148–350) 9 99 43·1 475 (352–640) ≥15 years 81% (79–83) 9428 7637 10 676 208/759 (27%) 0 0 0·0 0 (0–0) 1 9 4·0 35 (13–96) All ..

48 5·3 251 (107–590) 3 143 16·0 753 (460–1233) 5–14 years 81% (78–84) 4295 3479 4889 228/889 (26%) 4 82 15·4 315 (191–519) 3 61 12·0 236 (133–420) <15 years NA 8326 6744 9083 NA 5 55 20·6 227 (148–350) 9 99 43·1 475 (352–640) ≥15 years 81% (79–83) 9428 7637 10 676 208/759 (27%) 0 0 0·0 0 (0–0) 1 9 4·0 35 (13–96) All .. 17 754 14 381 19 759 NA 5 25 20·6 104 (68–161) 10 51 46·8 237 (178–316) Polesgo, Burkina Faso 0–1 years 92% (86–99) 896 824 929 117/475 (25% 0 0 0·0 0 (0–0) 1 108 4·0 431 (162–1147) 2–4 years 83% (76–89) 856 710 992 148/466 (32%) 6 605 18·8 1890 (1202–2972) 2 202 6·0 630 (288–1380) 5–14 years 87% (83–91) 1734 1509 2104 252/510 (49%) 5 238 10·2 485 (263–896) 0 0 0·0 0 (0–0) <15 years NA 3486 3043 4025 NA 11 273 29·0 719 (500–1035) 3 75 10·3 255 (138–470) ≥15 years 87% (84–89) 4088 3557 4917 239/629 (38%) 2 41 5·3 107 (46–252) 1 20 3·0 54 (16–179) All NA 7574 6600 8942 NA 13 145 34·2 383 (274–535) 4 45 12·9 144 (83–249) Bandim, Guinea-Bissau 0–1 years 46% (39–54) 10 852 4992 5198 206/631 (33%) 0 0 0·0 0 (0–0) 5 96 15·2 291 (176–482) 2–4 years 43% (37–48) 7307 3142 3866 175/359 (49%) 1 26 2·0 53 (13–208) 1 26 2·0 53 (13–208) 5–14 years 42% (41–48) 19 905 8360 11 101 187/380 (49%) 1 9 2·0 18 (5–72) 2 18 4·0 53 (14–97) <15 years NA 38 064 16 494 20 165 NA 2 10 4·1 20 (8–53) 8 40 21·3 116 (69–161) ≥15 years 45% (43–47) 62 694 28 212 37 109 105/163 (64%) 1 3 1·6 4 (1–20) 0 0 0·0 0 (0–0) All NA 100 758 44 706 57 274 NA 3 5 5·6 10 (4–22) 8 14 21·3 37 (24–57) Asante Akim North, Ghana 0–1 years 16% (14–18) 11 222 1760 4080 41%* 2 49 4·9 120 (49–290) 29 711 70·7 1733 (1373–2188) 2–4 years 16% (13–18) 8086 1268 2940 41%* 13 442 31·7 1079 (762–1528) 23 782 56·1 1908 (1469–2479) 5–14 years 16% (15–17) 34 439 5415 12 554 623/1657 (38%) 15 119 39·5 314 (230–430) 7 56 18·4 147 (93–232) <15 years NA 53 747 8443 19 574 NA 30 153 76·1 389 (310–486) 59 301 145·3 742 (631–873) ≥15 years NA NA† NA NA NA NA† NA NA NA NA† NA NA NA All NA NA† NA NA NA NA† NA NA NA NA† NA NA NA Pikine, Senegal‡§ 0–1 years 39% (32–46) 20 120 7837 11 194 NA 0 0 NA NA 0 0 NA NA 2–4 years 37% (33–41) 30 180 11 097 15 851 NA 0 0 NA NA 0 0 .. NA 5–14 years 31% (28–34) 96 152 29 807 42 577 NA 3 7 NA NA 1 5 .. NA <15 years NA 146 452 48 741 69 623 NA 3 4 NA NA 0 0 .. NA ≥15 years 30% (28–31) 195 726 58 718 83 874 NA 4 5 NA NA 3 6 .. NA All NA 342 178 107 459 153 496 NA 7 5 NA NA 4 5 ..

A 0 0 NA NA 2–4 years 37% (33–41) 30 180 11 097 15 851 NA 0 0 NA NA 0 0 .. NA 5–14 years 31% (28–34) 96 152 29 807 42 577 NA 3 7 NA NA 1 5 .. NA <15 years NA 146 452 48 741 69 623 NA 3 4 NA NA 0 0 .. NA ≥15 years 30% (28–31) 195 726 58 718 83 874 NA 4 5 NA NA 3 6 .. NA All NA 342 178 107 459 153 496 NA 7 5 NA NA 4 5 .. NA East Wad Medani, Sudan§ 0–1 years 23% (14–32) 2377 537 589 2/85 (2%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 2–4 years 22% (15–29) 3566 781 857 29/108 (27%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 5–14 years 25% (21–28) 11 071 2735 2999 160/234 (68%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) <15 years NA 17 014 4053 4445 NA 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) ≥15 years 29% (27–31) 29 843 8684 9525 131/147 (89%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) All NA 46 857 12 737 13 970 NA 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) Butajira, Ethiopia§ 0–1 years 69% (59–78) 2266 1563 2798 NA 0 0 NA NA 0 0 NA NA 2–4 years 62% (55–69) 3398 2107 3771 NA 0 0 NA NA 0 0 NA NA 5–14 years 65% (61–69) 14 015 9110 16 305 NA 1 6 NA NA 0 0 NA NA <15 years NA 19 679 12 780 22 874 NA 1 4 NA NA 0 0 NA NA ≥15 years 65% (62–68) 42 545 28 080 50 257 NA 2 4 NA NA 0 0 NA NA All NA 62 224 40 860 73 131 NA 3 4 NA NA 0 0 NA NA Moshi Rural District, Tanzania 0–1 years 4% (0–11)¶ 24 289 390 693 79%* 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 2–4 years 2% (0–4)‖ 25 281 406 721 79%* 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 5–14 years 13% (10–16) 118 219 15 487 27 508 79%* 2 (4)** 15 5·1 18 (8–44) 0 0 0·0 0 (0–0) <15 years NA 167 789 16 283 28 922 NA 2 (4)** 14 5·1 18 (7–42) 0 0 0·0 0 (0–0) ≥15 years 2% (1–2) 298 948 5172 9186 79%* 1 (2)** 22 2·5 28 (8–95) 1 (2)** 22 2·5 28 (8–95) All NA 466 737 21 454 38 108 NA 3 (6)** 16 7·6 20 (10–41) 1 (2)** 5 2·5 7 (2–23) Moshi Urban District, Tanzania 0–1 years 7% (0–19)¶ 10 406 335 595 79%* 0 0 0·0 0 (0–0) 1 (2)** 336 2·5 427 (125–1461) 2–4 years 2% (0–6)‖ 10 831 348 618 79%* 1 (5)** 809 6·4 1028 (472–2237) 0 0 0·0 0 (0–0) 5–14 years 13% (8–19) 37 309 4850 8615 79%* 2 (7)** 81 8·9 103 (54–199) 0 0 0·0 0 (0–0) <15 years NA 58 546 5533 9828 NA 3 (12)** 122 15·2 155 (94–256) 1 (2)** 20 2·5 26 (8–88) ≥15 years 2% (0–3) 125  746 2138 3796 79%* 3 (6)** 158 7·6 201 (99–408) 0 0 0·0 0 (0–0) All NA 184 292 7671 13 626 NA 6 (18)** 132 22·9 168 (111–253) 1 (2)** 15 2·5 19 (5–64) Kibera, Kenya†† 0–1 years 42% (38–47) 3467 1456 2031 99/99 (100%) 3 148 3·0 148 (48–458) 1 49 1·0 49 (7–350) 2–4 years 39% (36–43) 3070 1197 2039 312/312 (100%) 10 490 10·0 490 (264–912) 1 49 1·0 49 (7–348) 5–14 years 43% (39–47) 7514 3231 5722 539/539 (100%)

NA 6 (18)** 132 22·9 168 (111–253) 1 (2)** 15 2·5 19 (5–64) Kibera, Kenya†† 0–1 years 42% (38–47) 3467 1456 2031 99/99 (100%) 3 148 3·0 148 (48–458) 1 49 1·0 49 (7–350) 2–4 years 39% (36–43) 3070 1197 2039 312/312 (100%) 10 490 10·0 490 (264–912) 1 49 1·0 49 (7–348) 5–14 years 43% (39–47) 7514 3231 5722 539/539 (100%) 28 489 28·0 489 (338–709) 1 17 1·0 17 (2–124) <15 years NA 14 051 5884 9792 NA 41 419 41·0 419 (308–569) 3 31 3·0 31 (10–95) ≥15 years 35% (32–38) 15 263 5342 9228 301/301 (100%) 13 141 13·0 141 (82–243) 3 33 3·0 33 (10–101) All NA 29 314 11 227 19 020 NA 54 284 54·0 284 (217–371) 6 32 6·0 32 (14–70) Imerintsiatosika, Madagascar 0–1 years 28% (20–37) 3424 753 1287 66/85(78%) 0 0 0·0 0 (0–0) 1 78 1·3 100 (18–562) 2–4 years 19% (14–25) 5136 1130 1932 87/101 (86%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 5–14 years 18% (15–20) 13 188 2374 4057 184/256 (72%) 5 123 6·9 171 (81–360) 0 0 0·0 0 (0–0) <15 years NA 21 748 4257 7276 NA 5 69 6·9 95 (45–201) 1 14 1·3 18 (3–99) ≥15 years 17% (15–19) 24 632 4187 7153 639/919 (70%) 1 14 1·4 20 (4–103) 0 0 0·0 0 (0–0) All NA 46 380 8444 14 429 NA 6 42 8·4 58 (29–114) 1 7 1·3 9 (2–50) Isotry, Madagasar 0–1 years 6% (1–12) 3204 192 261 12/14 (86%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 2–4 years 10% (5–14) 4805 481 653 58/65 (89%) 0 0 0·0 0 (0–0) 0 0 0·0 0 (0–0) 5–14 years 9% (7–11) 16 386 1475 2005 234/288 (81%) 1 50 1·2 62 (11–359) 0 0 0·0 0 (0–0) <15 years NA 24 395 2147 2919 NA 1 34 1·2 42 (7–247) 0 0 0·0 0 (0–0) ≥15 years 9% (7–11) 45 928 4134 5621 1197/1421 (84%) 2 36 2·4 42 (12–151) 0 0 0·0 0 (0–0) All NA 70 323 6281 8540 NA 3 35 3·6 42 (15–119) 0 0 0·0 0 (0–0) Pietermaritzburg, South Africa§ 0–1 years 11% (5–17) 13 990 1511 3055 NA 0 0 NA NA 0 0 NA NA 2–4 years 7% (3–12) 20 985 1490 3013 NA 0 0 NA NA 0 0 NA NA 5–14 years 16% (13–19) 62 313 10 157 20 537 NA 0 0 NA NA 0 0 NA NA <15 years NA 97 288 13 158 26 605 NA 0 0 NA NA 0 0 NA NA ≥15 years 15% (13–17) 294 542 43 887 88 739 NA 2 2 NA NA 0 0 NA NA All NA 391 830 57 045 115 344 NA 2 2 NA NA 0 0 NA NA Study population was adjusted for health-seeking behaviour and crude cases were adjusted for recruitment proportion (number of patients analysed divided by number of patients with febrile illness from study area who visited a recruitment health facility, multiplied by 100).

391 830 57 045 115 344 NA 2 2 NA NA 0 0 NA NA Study population was adjusted for health-seeking behaviour and crude cases were adjusted for recruitment proportion (number of patients analysed divided by number of patients with febrile illness from study area who visited a recruitment health facility, multiplied by 100). iNTS=invasive non-typhoidal salmonella. NA=not available. PYO=person-years of observation. * *Recruitment portion was not available for each age strata. Broader values were applied to each stratum. † Target population for surveillance activities in Ghana included patients younger than 15 years of age; patients aged 15 years or older were not recruited. ‡ Three Salmonella Paratyphi A were identified at this site, but are not included in this table. § No salmonella was isolated in Sudan. Missing data on recruitment patterns in Senegal, Ethiopia, and South Africa did not allow calculation of adjusted incidences. Crude rates are presented. ¶ This proportion applies to age group <1 year, and it was used to adjust the study population by health-seeking behaviour. ‖ This proportion applies to age group 1–4 years, and it was used to adjust the study population by health-seeking behaviour. The adjusted populations in age groups <1 year and 1–4 years were added to estimate the total adjusted population age group 0–4 years. Subsequently, the percentage of children <2 years reported by the 2012 national census was applied to derive age groups 0–1 years and 2–4 years.

population by health-seeking behaviour. The adjusted populations in age groups <1 year and 1–4 years were added to estimate the total adjusted population age group 0–4 years. Subsequently, the percentage of children <2 years reported by the 2012 national census was applied to derive age groups 0–1 years and 2–4 years. ** Crude cases have been adjusted for recruitment pattern unique to the site in Tanzania: before Nov 11, 2011, every fifth eligible patient was recruited; from Nov 11, 2011, every second eligible patient was recruited. Adjusted cases (presented inside parentheses) were used to calculate crude rate. †† Active population mobilisation was done, in addition to passive surveillance. Table 3 Antimicrobial resistance patterns of Salmonella enterica serotype Typhi and iNTS isolates across sites

** Crude cases have been adjusted for recruitment pattern unique to the site in Tanzania: before Nov 11, 2011, every fifth eligible patient was recruited; from Nov 11, 2011, every second eligible patient was recruited. Adjusted cases (presented inside parentheses) were used to calculate crude rate. †† Active population mobilisation was done, in addition to passive surveillance. Table 3 Antimicrobial resistance patterns of Salmonella enterica serotype Typhi and iNTS isolates across sites Burkina Faso Guinea-Bissau Senegal* Ghana Ethiopia Madagascar South Africa Tanzania Kenya All Total S Typhi isolates, N 18 3 7 30 3 9 2 9 54 135 Isolate with antimicrobial resistance, n (%)† Ampicillin 0 NR NR 20 (67%) 2 (67%) NR 0 8 (89%) 41 (76%) 71 (53%) Amoxicillin-clavulanic acid 0 NR NR 3 (10%) 0 NR 0 4 (44%) 24 (44%) 31 (23%) Chloramphenicol 2 (11%) NR NR 23 (77%) 0 NR 0 5 (56%) 43 (80%) 73 (54%) Co-trimoxazole 2 (11%) NR NR 24 (80%) 0 NR 0 8 (89%) 43 (80%) 77 (57%) Ceftriaxone 0 NR NR 0 0 NR 0 0 0 0 Ciprofloxacin 0 NR NR 0 0 NR 1 (50%) 0 11 (20%) 12 (9%) Multidrug resistance‡ 0 NR NR 19 (63%) 0 NR 0 5 (56%) 40 (74%) 64 (47%) Total iNTS isolates, N 14 8 4 59 0 1 0 2 6 94 Isolate with antimicrobial resistance, n (%)† Ampicillin 10 (71%) 1 (13%) NR 38 (64%) NR NR NR 0 2 (33%) 51 (54%) Amoxicillin-clavulanic acid 3 (21%) 0 NR 9 (15%) NR NR NR 0 2 (33%) 14 (15%) Chloramphenicol 12 (86%) 1 (13%) NR 34 (58%) NR NR NR 0 1 (17%) 48 (51%) Co-trimoxazole 13 (93%) 1 (13%) NR 34 (58%) NR NR NR 0 2 (33%) 50 (53%) Ceftriaxone 0 0 NR 0 NR NR NR 0 1 (17%) 1 (1%) Ciprofloxacin 1 (7%) 0 NR 2 (3%) NR NR NR 0 0 3 (3%) Multidrug resistance‡ 10 (71%) 1 (13%) NR 33 (56%) NR NR NR 0 1 (17%) 45 (48%) Resistant isolates are reported per country, rather than per site. No Salmonella enterica serotype Typhi (S Typhi) or iNTS isolates were cultured in Sudan. iNTS=invasive non-typhoidal salmonella. NR=no resistant isolates identified.

0 3 (3%) Multidrug resistance‡ 10 (71%) 1 (13%) NR 33 (56%) NR NR NR 0 1 (17%) 45 (48%) Resistant isolates are reported per country, rather than per site. No Salmonella enterica serotype Typhi (S Typhi) or iNTS isolates were cultured in Sudan. iNTS=invasive non-typhoidal salmonella. NR=no resistant isolates identified. * Seven S Typhi, four iNTS, and three S enterica serotype Paratyphi (S Paratyphi) isolates. One of the S Paratyphi isolates was resistant to ciprofloxacin. † Includes isolates fully and intermediately resistant against the respective drug, as defined by the Clinical Laboratory and Standards Institute guidelines 2013.15 ‡ Defined as resistance against ampicillin or amoxicillin AND chloramphenicol AND co-trimoxazole.

Introduction More than 377 million people live in India's 7933 urban areas,1, 2 of which 53 house more than 1 million people each. Three mega-cities, Mumbai, Delhi, and Kolkata, house more than 10 million people each. Two-thirds of census towns include informal settlements (slums)3 that are characterised by overcrowding, insubstantial housing, insufficient water and sanitation, lack of tenure, and hazardous locations.4, 5 There will be around 105 million people living in informal settlements by 2017.2 India's National Urban Health Mission aims to facilitate equitable access to quality health care through an improved public health system, partnerships, and community-based mechanisms. Three tiers of provision are envisaged: secondary and tertiary institutions, urban health centres, and community outreach to informal settlements and other vulnerable groups.6 This structure frames a context of pluralistic health care. For example, Mumbai's population of 12·4 million7 is served by a pyramid of municipal tertiary hospitals, peripheral hospitals, maternity homes, and health posts. Additionally, there are charitable institutions and a wealth of private care providers (from specialist hospitals to unqualified practitioners), the latter of which are responsible for around 75% of outpatient consultations.8

amid of municipal tertiary hospitals, peripheral hospitals, maternity homes, and health posts. Additionally, there are charitable institutions and a wealth of private care providers (from specialist hospitals to unqualified practitioners), the latter of which are responsible for around 75% of outpatient consultations.8 Non-governmental organisations are important to public–private partnerships,9 in which they contribute services traditionally provided by the public sector,10 alone or in collaboration,11 and develop models for adoption by the public sector.6 The Society for Nutrition, Education and Health Action (SNEHA) is a non-governmental organisation whose programmes address priority issues that have emerged from 16 years of work with women and children in informal settlements: maternal and neonatal health, sexual and reproductive health, childhood nutrition, and prevention of violence against women and children. We wanted to integrate these activities in a model that could be useful to the National Urban Health Mission of India and other city governments in achieving a commitment to health in informal settlements.12 After a large trial focused on neonatal survival,13 we believed that integration of the programme in the community was appropriate because of the multiple health issues faced by women and children, and that communities were more likely to respond to an intervention with a physical presence and service delivery. Research in context Evidence before this study

• Number of consultations for violence against women or children, including physical (slapping, arm twisting, pushing, punching, kicking, choking, or use of implements or weapons), emotional (jealousy, accusations of infidelity, prevention of association with friends or family, aggressive monitoring of whereabouts, lack of trust with money, humiliation in front of others, threats of violence, or insults), or sexual (forced sex, coercion into undesired sexual acts) • Proportion of home births in the preceding year • Proportion of pregnancies in the preceding 2 years in women younger than 20 years • Proportion of children younger than 5 years with anthropometric stunting • Proportion of children younger than 5 years with anthropometric underweight • Proportion of children born in the preceding 2 years who received services from Government of India Integrated Child Development Services (food supplements, health checks, early childhood development intervention, or measurement of weight)

Non-governmental organisations are important to public–private partnerships,9 in which they contribute services traditionally provided by the public sector,10 alone or in collaboration,11 and develop models for adoption by the public sector.6 The Society for Nutrition, Education and Health Action (SNEHA) is a non-governmental organisation whose programmes address priority issues that have emerged from 16 years of work with women and children in informal settlements: maternal and neonatal health, sexual and reproductive health, childhood nutrition, and prevention of violence against women and children. We wanted to integrate these activities in a model that could be useful to the National Urban Health Mission of India and other city governments in achieving a commitment to health in informal settlements.12 After a large trial focused on neonatal survival,13 we believed that integration of the programme in the community was appropriate because of the multiple health issues faced by women and children, and that communities were more likely to respond to an intervention with a physical presence and service delivery. Research in context Evidence before this study We searched PubMed for articles published up to Oct 1, 2016, addressing health-care interventions in urban slums worldwide. We used English search terms, but placed no restriction on the language of retrieved articles. We used the combined search expression “(slum OR “informal settlement”) AND (healthcare OR “health care”) AND (provision OR delivery OR program$ OR project)”. We screened 1481 article titles, including 389 limited to Asia and 175 limited to India, from which we identified 48 relevant abstracts. We found no completed or published trial of a model of provision of integrated health care for informal settlement populations, although models with some similarities are operational in Delhi and Chennai in India and in Bangladesh.

imited to Asia and 175 limited to India, from which we identified 48 relevant abstracts. We found no completed or published trial of a model of provision of integrated health care for informal settlement populations, although models with some similarities are operational in Delhi and Chennai in India and in Bangladesh. Added value of this study We showed that a community resource centre model for women's and children's health was feasible and potentially replicable and incurred low cost in informal settlements. The intervention could be implemented by a non-governmental organisation in collaboration with public sector and civil society institutions. It was possible to measure population health outcomes, with effects seen after only 2 years of operation. Implications of all the available evidence This clearly defined model for integrated community-based health intervention in informal settlements merits adaptation and assessment in other contexts, particularly in Asia and Africa.

We showed that a community resource centre model for women's and children's health was feasible and potentially replicable and incurred low cost in informal settlements. The intervention could be implemented by a non-governmental organisation in collaboration with public sector and civil society institutions. It was possible to measure population health outcomes, with effects seen after only 2 years of operation. Implications of all the available evidence This clearly defined model for integrated community-based health intervention in informal settlements merits adaptation and assessment in other contexts, particularly in Asia and Africa. We conceived a model that included service provision, outreach, and community mobilisation activities, with a visible presence in SNEHA centres. The evidence base for this type of approach is limited. Data synthesis identified 17 reviews of interventions to improve health in informal settlements, including physical upgrading of the built environment, improvements in water and sanitation, infectious disease control, prevention of burns, and cash transfers.14 Ten randomised, controlled trials of health-promotion interventions included interventions by community health workers to improve handwashing and nutrition and reduce the risks of burns, poisoning, injuries, and HIV infection.14 Relevant Cochrane reviews included the effects on health of strategies to upgrade slums15 and a planned review of childhood nutritional interventions.16

ntions included interventions by community health workers to improve handwashing and nutrition and reduce the risks of burns, poisoning, injuries, and HIV infection.14 Relevant Cochrane reviews included the effects on health of strategies to upgrade slums15 and a planned review of childhood nutritional interventions.16 We found no completed trial of health-care provision to people living in informal settlements, but we identified three regional initiatives that informed our model. In Delhi, the non-governmental Asha Community Health and Development Society has, since 1988, developed a programme of land rights advocacy, education, savings and loans, health care and diagnosis, community health volunteers and groups, and campaigns on these issues.17 In Bangladesh, the non-governmental BRAC MANOSHI programme has provided interventions for maternal, neonatal, and child health in urban informal settlements since 2007. It involves community health workers, birth attendants working in local delivery centres, and settlement committees.18, 19 In an informal settlement in Chennai, India, the Sahishnatha Trust delivered an integrated intervention that addressed water and sanitation and provided female link workers, weekly health clinics, self-help groups, and community campaigns.20 In this study, we implemented the SNEHA centre model and assessed the effects across a range of outcomes representative of women's and children's health. The study protocol has been published.21

We found no completed trial of health-care provision to people living in informal settlements, but we identified three regional initiatives that informed our model. In Delhi, the non-governmental Asha Community Health and Development Society has, since 1988, developed a programme of land rights advocacy, education, savings and loans, health care and diagnosis, community health volunteers and groups, and campaigns on these issues.17 In Bangladesh, the non-governmental BRAC MANOSHI programme has provided interventions for maternal, neonatal, and child health in urban informal settlements since 2007. It involves community health workers, birth attendants working in local delivery centres, and settlement committees.18, 19 In an informal settlement in Chennai, India, the Sahishnatha Trust delivered an integrated intervention that addressed water and sanitation and provided female link workers, weekly health clinics, self-help groups, and community campaigns.20 In this study, we implemented the SNEHA centre model and assessed the effects across a range of outcomes representative of women's and children's health. The study protocol has been published.21 Methods Study design and setting We used a parallel-group, phased,22 cluster-randomised, controlled trial design, and did censuses before and after the intervention. About 41% of Mumbai's households are in informal settlements.3 We did this trial in two of 24 municipal wards, each with a population of around 700 000. The wards were chosen because they had the lowest Human Development Indices (M East ward 0·05 and L ward 0·29).23 The numbers of informal settlements in these wards have grown over the past 20 years. Most have surfaced roads, electricity supplies, and schools. They are low-lying and susceptible to flooding, and some adjoin the city's largest solid-waste dump. About 65% of the settlement populations are made up of migrants from Uttar Pradesh, 10% from Bihar, and 15% from Maharashtra, the state in which Mumbai is located. Work includes unskilled labour (40%), skilled artisanal work (27%), and transport (14%). At the start of the study, the study area was served by nine municipal health posts, one urban health centre, one maternity home that provided antenatal care, immunisations, and family planning, but was underused, and a tertiary public hospital that took about 20 min to reach from the study area by public transport. Additionally, 42 Anganwadi centres, run by the Indian Government Integrated Child Development Services (ICDS) programme, provided maternal and child health and nutrition services. Many private practitioners, with a range of qualifications, tailor their services to the local economy. We identified 35 private providers in the intervention clusters alone.

, run by the Indian Government Integrated Child Development Services (ICDS) programme, provided maternal and child health and nutrition services. Many private practitioners, with a range of qualifications, tailor their services to the local economy. We identified 35 private providers in the intervention clusters alone. Before the trial, data for Mumbai slums were available from the National Family Health Survey (NFHS) in 2005–06,8 which showed that met need for family planning was 55% (6% for spacing and 49% for limiting of pregnancies). Full immunisation with BCG, diphtheria, pertussis, and tetanus ([DPT] three doses), polio, and measles vaccines was 69% in children aged 12–23 months. Wasting was seen in 16% of children younger than 5 years, and stunting in 47%.8 Violence against women is common in India.24 Estimates of lifetime prevalence are 29% (range 2–99) for domestic physical violence, 12% (0–75) for sexual violence, and 30% (4–56) for multiple forms of violence.25 However, violence is under-reported, and the incidence has been cited as 54·8 assaults per 100 000 women in the general population.26

common in India.24 Estimates of lifetime prevalence are 29% (range 2–99) for domestic physical violence, 12% (0–75) for sexual violence, and 30% (4–56) for multiple forms of violence.25 However, violence is under-reported, and the incidence has been cited as 54·8 assaults per 100 000 women in the general population.26 We estimated that about 60 sizeable non-governmental organisations were working in informal settlements in Mumbai. Some, including the Society for Promotion of Area Resource Centres, Akanksha Foundation, Apnalaya, DoorstepSchool, and Pratham, had run community resource centres with varied purposes: education, vocational training, recreational activities, centres for people with disabilities, family counselling, collective savings and loan disbursement, physical space for community interaction, and health clinics. Some organisations, including Apnalaya, Stree Hitkarni, the Committed Communities Development Trust, Alert, and Navjeevan, had focused on community health. Their resource centres were staffed by a mix of volunteers and salaried teams. The trial was approved by the Multi-Institutional Ethics Committee of the Anusandhan Trust, Mumbai, India, in sequential reviews: formative research (February, 2011), cluster vulnerability (May, 2011), the preintervention census (August, 2011), and the intervention and assessments (January, 2012). It was also approved by the University College London Research Ethics Committee, London, UK, in January, 2012 (reference 3546/001).

sequential reviews: formative research (February, 2011), cluster vulnerability (May, 2011), the preintervention census (August, 2011), and the intervention and assessments (January, 2012). It was also approved by the University College London Research Ethics Committee, London, UK, in January, 2012 (reference 3546/001). Participants The resource centres targeted women of reproductive age and children younger than 5 years, but any other residents living in an allocated cluster were free to participate in activities and access services. Two censuses were done, one before and one 2 years after implementation of the intervention (figure 1). Interviewees were ever-married women aged 15–49 years. We obtained consent at the cluster and individual levels27 and no monetary compensation was given. Cluster consent was provided by cluster gatekeepers, who were identified with a standard protocol from among participants in community engagement activities who others in the area judged appropriate to speak for them on health issues. Receipt of the intervention was at individual discretion. The right to withdraw at any time was implicit. Respondents in both censuses were given standard information about the trial and the procedures for anonymising data.

agement activities who others in the area judged appropriate to speak for them on health issues. Receipt of the intervention was at individual discretion. The right to withdraw at any time was implicit. Respondents in both censuses were given standard information about the trial and the procedures for anonymising data. Ethical considerations We identified no specific risks of harm to individuals or the community associated with community resource centres themselves. If, however, women and children were identified by data collectors as being malnourished, concerned about family planning or birth in an institution, or being survivors of domestic violence, interviewers had a duty of care to act within personal and organisational abilities. We followed guidelines on the reporting of violence, which included obtaining consent from the survivor and explaining the available interventions. Campaigns and group sessions were used to explain to survivors that they could disclose abuse and access support at any time they felt ready. For individual or family problems, community organisers followed organisational support procedures and protocols, provided information packs on local sources of help, and aided in arranging consultations. The study ethics committee mandated protocols for training, information, and action across a range of issues. As the intervention was part of our service delivery programme, we judged it to carry minimum risk over 2 years and did not specify stopping rules.

n local sources of help, and aided in arranging consultations. The study ethics committee mandated protocols for training, information, and action across a range of issues. As the intervention was part of our service delivery programme, we judged it to carry minimum risk over 2 years and did not specify stopping rules. Randomisation and masking The sample frame included 159 clusters of around 600 households. We identified informal settlements through local knowledge and information from the Municipal Corporation of Greater Mumbai, the Tata Institute of Social Sciences, and non-governmental organisations. We excluded areas that had been involved in a previous trial13 and divided large settlements into clusters along obvious physical boundaries, using geographical distribution to minimise contamination. After assessment of health vulnerability in all clusters in the sample frame with a rapid assessment tool,28 we included the 40 with the lowest scores in the study (figure 2). On July 25, 2011, SD and DO used an online randomisation generator to randomly assign these clusters, in blocks of 12, 12, and 16, to the intervention group or the control group, with intervals of 6 months between the allocation of each block to create three implementation phases (figure 1). Project staff were unaware of allocation when they collected consent and did the preintervention census. Because of the nature of the intervention, the implementation team and the field investigators who did the postintervention census were aware of allocation. However, to keep familiarity with residents to a minimum, a new team was recruited to do the postintervention census.

d consent and did the preintervention census. Because of the nature of the intervention, the implementation team and the field investigators who did the postintervention census were aware of allocation. However, to keep familiarity with residents to a minimum, a new team was recruited to do the postintervention census. Procedures A SNEHA centre was set up in each intervention cluster in rented premises. Each centre employed three full-time, salaried community organisers who were educated to at least the ninth grade and had similar socioeconomic backgrounds to potential beneficiaries. Each organiser was responsible for around a third of the households in the cluster. We engaged the community organisers to integrate our themes of reproductive, maternal, and neonatal health, child health and nutrition, and prevention of violence against women and children into the community services. They were equipped with technical knowledge in these areas and with communication and negotiation skills through 1 month of training followed by regular supervision and follow-up visits by SNEHA staff and invited experts. They made home visits, organised group meetings, day care for malnourished children, and community events, provided services, and liaised with existing systems (panel 1). All activities were logged via a smartphone-based reporting system created from open-source software (CommCare, Dimagi, Cambridge, MA, USA), including information about the families with whom they worked.

re for malnourished children, and community events, provided services, and liaised with existing systems (panel 1). All activities were logged via a smartphone-based reporting system created from open-source software (CommCare, Dimagi, Cambridge, MA, USA), including information about the families with whom they worked. The censuses were done by two teams of six interviewers and a supervisor who clarified cluster boundaries and mapped and numbered households. Interviews with eligible women were used to obtain information on numbers of household members, duration of residence, assets and amenities, housing fabric, faith, maternity history, and family planning. Women who had been pregnant in the preceding 2 years were asked about antenatal care, birth location, outcome, and infant feeding. Information on immunisations and use of ICDS was collected for children younger than 5 years. Additionally, anthropometric characteristics were recorded for children younger than 5 years on designated days at the end of each cluster census. Length of children younger than 2 years was measured with a Rollameter accurate to 1 mm, with an assistant holding the child's head. Height of children aged 2 years and older was measured with a Leicester stadiometer accurate to 1 mm, at the end of expiration, with feet together against the backboard, back straight, and head in the Frankfort plane. Weight was measured with Seca 385 electronic scales accurate to 1 g. Training for data collectors was repeated on four occasions, for which the indicative technical errors of measurement for height were 0·14%, 0·38%, 0·6%, and 0·5%.

th feet together against the backboard, back straight, and head in the Frankfort plane. Weight was measured with Seca 385 electronic scales accurate to 1 g. Training for data collectors was repeated on four occasions, for which the indicative technical errors of measurement for height were 0·14%, 0·38%, 0·6%, and 0·5%. Interview data were collected on smartphones with an open-source tool from Open Data Kit (Seattle, WA, USA) running in Google Android (versions 3.0–4.4 [Honeycomb to Kitkat]). 5% of interviews selected at random were observed by a supervisor. The interview system included automatic skips and validation constraints. Encrypted electronic data were transferred to a secure Open Data Kit Aggregate cloud repository on a password-protected Google Appspot (Google App Engine). Data were checked after download for errors in key fields, and monitoring summaries were produced through do-files written in Stata version 12. Each week, after all interviews from that week were numbered, 50 records (20–25% of interviews) were extracted at random, printed on spreadsheets, and rechecked in the field by a supervisor. After the interviewees' names had been removed, access to data was restricted to the data manager and data analysts.

version 12. Each week, after all interviews from that week were numbered, 50 records (20–25% of interviews) were extracted at random, printed on spreadsheets, and rechecked in the field by a supervisor. After the interviewees' names had been removed, access to data was restricted to the data manager and data analysts. Outcomes Outcomes were assessed on the basis of the two censuses done before and after 2 years of the intervention being implemented (panel 2). We assessed three primary outcomes: met need for family planning in women aged 15–49 years; the proportion of children aged 12–23 months who were fully immunised (BCG, DPT [three doses], polio, hepatitis B virus, and measles); and the proportion of children younger than 5 years who had anthropometric wasting. We also assessed seven secondary outcomes: number of consultations for violence against women or children; the proportion of home births in the preceding year; the proportion of pregnancies in the preceding 2 years in women younger than 20 years; the proportion of children younger than 5 years with anthropometric stunting; the proportion of children younger than 5 years with anthropometric underweight; the proportion of children born in the preceding 2 years who received ICDS; and the proportion of children achieving WHO Infant and Young Child Feeding core indicators (panel 2).30

hildren younger than 5 years with anthropometric stunting; the proportion of children younger than 5 years with anthropometric underweight; the proportion of children born in the preceding 2 years who received ICDS; and the proportion of children achieving WHO Infant and Young Child Feeding core indicators (panel 2).30 Statistical analysis For our sample size calculations, we assumed that we would be able to achieve two treatment groups consisting of unmatched clusters of roughly equal size and with similar k values (coefficient of variation of true proportions between clusters).31 On the basis of the data from the preintervention census, when around 400 women were interviewed per cluster, and the estimates that per cluster around 80 children would have been born from around 80 pregnancies in the previous 2 years and that there would be roughly 120 children aged 2–5 years, we calculated that interviewing 350 women and measuring the weights and heights of 150 children younger than 5 years per cluster after the intervention would provide 80% power to detect a 5% increase in met need for family planning, a 13% increase in full immunisation, and a 4% reduction in anthropometric wasting with a 5% significance threshold. Wealth was described by asset scores. We used a principal components analysis to derive weights for scores.32, 33

Statistical analysis For our sample size calculations, we assumed that we would be able to achieve two treatment groups consisting of unmatched clusters of roughly equal size and with similar k values (coefficient of variation of true proportions between clusters).31 On the basis of the data from the preintervention census, when around 400 women were interviewed per cluster, and the estimates that per cluster around 80 children would have been born from around 80 pregnancies in the previous 2 years and that there would be roughly 120 children aged 2–5 years, we calculated that interviewing 350 women and measuring the weights and heights of 150 children younger than 5 years per cluster after the intervention would provide 80% power to detect a 5% increase in met need for family planning, a 13% increase in full immunisation, and a 4% reduction in anthropometric wasting with a 5% significance threshold. Wealth was described by asset scores. We used a principal components analysis to derive weights for scores.32, 33 We generated anthropometric Z scores from the 2006 WHO growth standards and the ZSCORE06 module in Stata/IC (version 13.1).34 Outliers were removed so that the Z scores ranged from −6 to 6 for height for age, from −5 to 5 for weight for length or height, and from −6 to 5 for weight for age.35 We derived binary variables describing wasting, stunting, and underweight, with the threshold for all set at 2 SD below the median WHO value.

n 13.1).34 Outliers were removed so that the Z scores ranged from −6 to 6 for height for age, from −5 to 5 for weight for length or height, and from −6 to 5 for weight for age.35 We derived binary variables describing wasting, stunting, and underweight, with the threshold for all set at 2 SD below the median WHO value. We compared frequencies and proportions of demographic, socioeconomic, and environmental descriptors and of primary and secondary outcomes before the intervention in the two allocation groups, and report odds ratios (ORs) with 95% CIs. The primary intention-to-treat analysis involved a series of logistic regression models, including a variable for the outcome of interest, a dummy variable for allocation, and a random effect for cluster (quadrature assumptions were met). The allocation groups were generally balanced and we did not introduce additional covariates. Likelihood ratio tests showed no evidence of effect modification by implementation phase and, therefore, models did not include an interaction term.36

ocation, and a random effect for cluster (quadrature assumptions were met). The allocation groups were generally balanced and we did not introduce additional covariates. Likelihood ratio tests showed no evidence of effect modification by implementation phase and, therefore, models did not include an interaction term.36 We did two additional analyses. First, because migration rates were high, we did a per-protocol analysis of women who had participated in both censuses (ie, those who had potentially been exposed to the intervention for the full 2 years). Second, because weight for length or height differed between groups before the intervention, we did a cluster-level analysis of anthropometric changes in mean Z scores and the proportions of children with wasting between censuses. We did the same for uptake of family planning and proportions fully immunised at 12–23 months, and applied t tests to the normally distributed changes. Data to develop a classification of met need were unavailable in the baseline census and, therefore, we defined use of modern contraception as female or male terminal methods, oral contraceptive pill, intrauterine device, hormone implant or injection, condom, or diaphragm.

d t tests to the normally distributed changes. Data to develop a classification of met need were unavailable in the baseline census and, therefore, we defined use of modern contraception as female or male terminal methods, oral contraceptive pill, intrauterine device, hormone implant or injection, condom, or diaphragm. We used the RE-AIM framework to describe delivery and uptake of the intervention.37 34 components were reported, classified under five general criteria:37, 38 reach (the proportion of the target population who participated in the intervention, according to the census data), efficacy (based on trial endpoint findings), adoption (use of the intervention in allocated clusters), implementation (the degree to which the intervention was delivered as intended, assessed by programme monitoring), and maintenance of the intervention after the implementation period. Scores allocated were within the potential range of 0·0–1·0. The overall score was the product of the five criteria scores. We estimated the cost of the intervention from organisational finance records. Cost codes covered salaries, communication, and conveyance for human resources at all levels up to programme director, set-up and running costs for centres, equipment and consumables, costs of training, meetings, events and campaigns, internal monitoring costs, and administrative overheads.

anisational finance records. Cost codes covered salaries, communication, and conveyance for human resources at all levels up to programme director, set-up and running costs for centres, equipment and consumables, costs of training, meetings, events and campaigns, internal monitoring costs, and administrative overheads. A data monitoring committee met twice during the trial and once after the second census (figure 1). Following DAMOCLES guidelines,39 at the first meeting in May, 2012, the committee considered the protocol, sealed the analysis plan, and reviewed the findings of the preintervention census from the 12 clusters in phase 1 of implementation. In the second meeting, in December, 2013, data from all three implementation phases were assessed; no changes to the protocol were recommended. In the third meeting, in January, 2016, the final data were reviewed and ancillary analyses were recommended, leading to an addendum to the published protocol. After the first meeting, we removed receipt of the Janani Suraksha Yojana birth incentive as a secondary indicator because we were unable to affect its use at the institutional level. At the time of protocol development, we planned to use the London Measure of Unplanned Pregnancy40 as a measure of family planning. In 2012, revised guidelines for estimating unmet need for family planning were released,41 which we used instead. This study is registered with ISRCTN, number ISRCTN56183183, and Clinical Trials Registry of India, number CTRI/2012/09/003004.

e the London Measure of Unplanned Pregnancy40 as a measure of family planning. In 2012, revised guidelines for estimating unmet need for family planning were released,41 which we used instead. This study is registered with ISRCTN, number ISRCTN56183183, and Clinical Trials Registry of India, number CTRI/2012/09/003004. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results The preintervention census started in August, 2011, and was completed in January, 2013, and the postintervention census began in February, 2014, and was completed in September, 2015 (figure 1). The preintervention census identified 24 853 households and the postintervention census identified 24 939 (figure 3). Of the postintervention households, 15 907 (63·8%) were home to 17 568 eligible women, of whom 16 236 (92·4%) were interviewed. Information was provided for 10 551 children younger than 5 years.

ure 1). The preintervention census identified 24 853 households and the postintervention census identified 24 939 (figure 3). Of the postintervention households, 15 907 (63·8%) were home to 17 568 eligible women, of whom 16 236 (92·4%) were interviewed. Information was provided for 10 551 children younger than 5 years. Characteristics of respondents were similar in the intervention and control groups in the preintervention census (table 1). We saw substantial environmental improvements in the postintervention census versus the preintervention census. The number of home owners was unchanged, but increases were seen for robust housing fabric (10 908 [75%] of 14 474 houses vs 8399 [59%] of 14 293), households with a private tap providing drinking water (3483 [24%] vs 2533 [18%]), access to a community tapstand for drinking water (9372 [65%] vs 2553 [18%]), and homes with private toilets (2604 [18%] vs 1612 [11%]), and a decrease was seen in the number of households buying drinking water from tankers (1619 [11%] vs 9207 [64%]).

h a private tap providing drinking water (3483 [24%] vs 2533 [18%]), access to a community tapstand for drinking water (9372 [65%] vs 2553 [18%]), and homes with private toilets (2604 [18%] vs 1612 [11%]), and a decrease was seen in the number of households buying drinking water from tankers (1619 [11%] vs 9207 [64%]). Before the intervention, about 30% of women said that they were using modern methods of contraception, of which female terminal methods were the most common, 64% of children aged 12–23 months were fully immunised, and around 16% of children younger than 5 years showed anthropometric wasting and 47% showed stunting (table 2). The mean Z scores for weight for length or height were −0·92 (SD 1·15) in the control group and −1·06 (1·12) in the intervention group, and for height for age were −1·82 (1·65) and −1·70 (1·67), respectively. Only 5% of children aged 6–23 months met the requirements for minimum acceptable diet (table 2). In the intention-to-treat analysis after the intervention, met need for family planning was greater in the intervention than the control group for both spacing and limiting of pregnancies (table 3). Adjustment for maternal age and parity increased the likelihood of overall met need (1·35, 1·14–1·60). Values for full immunisation and wasting did not differ between allocation groups, although children aged 12–23 months in the intervention group were more likely to have immunisation cards than those in the control group (table 3). For children younger than 5 years, the mean Z scores for weight for length or height were −0·90 (SD 1·00) in the control group and −0·88 (1·02) in the intervention group.

although children aged 12–23 months in the intervention group were more likely to have immunisation cards than those in the control group (table 3). For children younger than 5 years, the mean Z scores for weight for length or height were −0·90 (SD 1·00) in the control group and −0·88 (1·02) in the intervention group. The per-protocol analysis of the primary endpoint included 5838 households, 5830 women, and 3529 children younger than 5 years (2647 households, 2645 women, and 1560 children in the control group, and 3191 households, 3185 women, and 1969 children in the intervention group). Thus, 40% of all households and 36% of all women participated in both censuses. After the intervention, use of modern family planning methods had increased by a mean of 7·0% in the control group and 13·1% in the intervention group, leading to a significant difference between groups (figure 4). Full immunisation in children aged 12–23 months changed by a mean of −1·9% in the control group and 4·2% in the intervention group but the difference between groups was not significant (figure 4). Weight for height Z score increased by a mean of 0·03 in the control group and by a mean of 0·19 in the intervention group (p=0·013). The proportion of children with anthropometric wasting decreased by a mean of 2·5% and 6·4%, respectively, leading to a significant difference between groups (figure 4).

cant (figure 4). Weight for height Z score increased by a mean of 0·03 in the control group and by a mean of 0·19 in the intervention group (p=0·013). The proportion of children with anthropometric wasting decreased by a mean of 2·5% and 6·4%, respectively, leading to a significant difference between groups (figure 4). Services for survivors of violence reported 314 consultations in intervention clusters. The proportions of births at home, childhood stunting and underweight, and uptake of ICDS for children aged 0–23 months did not differ between allocation groups. Decreases in Z scores for height for age were seen in both groups (mean −1·84 [SD 1·41] in the control group and −1·86 [1·37] in the intervention group) and weight for age (–1·70 [1·13] and −1·68 [1·13]). Feeding exclusively with breastmilk up to age 6 months and achieving minimum dietary diversity in children aged 6–23 months were increased in the intervention group compared with in the control group after the intervention (OR 1·54, 95% CI 1·02–2·33 and 1·48, 1·01–2·17). In the intervention group, 84% of women knew of the SNEHA centre in their cluster, 79% recalled monthly visits from a community organiser, 39% had participated in SNEHA centre activities, and 88% of women with children younger than 5 years reported had received a service (table 4). In the control group, fewer than 1% of residents reported awareness of or participating in any similar activities. Uptake of municipal or non-governmental services did not differ between groups.

ed in SNEHA centre activities, and 88% of women with children younger than 5 years reported had received a service (table 4). In the control group, fewer than 1% of residents reported awareness of or participating in any similar activities. Uptake of municipal or non-governmental services did not differ between groups. In the summary of the intervention process done with the adapted RE-AIM framework (appendix), we estimated scores of 0·8 for reach, 0·7 for effectiveness, 1·0 for adoption, 0·8 of implementation, and 0·8 for maintenance. The overall product score was 0·36. We estimated that the cost of the intervention was INR694 000 per SNEHA centre per year (£7300, US$10 340) or INR231 350 (£2435, $3450) per 1000 general population. Discussion Our integrated intervention, which combined home visits, group work, some service provision, and liaison, was delivered in some of the poorest of Mumbai's informal settlements by a non-governmental organisation. We saw clear improvements in various indicators of women's and children's health, including met need for family planning and full immunisation of children. The effect on immunisation was, however, seen only when women had been exposed to the intervention for 2 years (the per-protocol population). Effects on childhood malnutrition were not evident from the analysis of anthropometric wasting, but could be inferred from additional analyses (figure 4).

munisation of children. The effect on immunisation was, however, seen only when women had been exposed to the intervention for 2 years (the per-protocol population). Effects on childhood malnutrition were not evident from the analysis of anthropometric wasting, but could be inferred from additional analyses (figure 4). In the NFHS of 2015–16 (NFHS-4),42 unmet need for family planning was estimated to be 14% for Mumbai. In our control group after the intervention, however, unmet need was 22%, which suggests a difference between informal settlements and the city as a whole. Intervention might, therefore, be particularly important in informal settlements. The NFHS-4 findings showed that 46% of children aged 12–23 months in Mumbai were fully immunised.42 The comparable proportion in this study in the postintervention census (excluding hepatitis B and measles) was higher at 69%, as were the values for individual vaccinations (BCG 95% in our intervention group vs 88% in the NFHS-4; DPT 82% vs 51%; and hepatitis B 79% vs 46%). Anthropometric wasting in children was lower in our intervention group than in the NFHS-4 (12% vs 26%), although the proportion for stunting was greater (47% vs 23%). The values for neonates breastfed within 1 h of birth were similar (47% after our intervention vs 50% in NFHS-4) and more children in our intervention group were receiving an adequate diet at age 6–23 months (13% vs 5%).

oup than in the NFHS-4 (12% vs 26%), although the proportion for stunting was greater (47% vs 23%). The values for neonates breastfed within 1 h of birth were similar (47% after our intervention vs 50% in NFHS-4) and more children in our intervention group were receiving an adequate diet at age 6–23 months (13% vs 5%). Our intervention and control groups were generally similar, with high coverage and fidelity to planned activities and negligible contamination. Some of our outcomes were proxies for longer-term effects. For example, the most widely used forms of contraception were female terminal methods (44% of women in the control group and 35% in the intervention group) and condoms (21% and 30%). The increase in met need for family planning was largely due to escalating condom use. Changes in spacing and limiting of pregnancies will only be possible to assess in the longer term, particularly given the possibility of best behaviour bias in the intervention clusters. Similarly, although an end itself, an aim of immunisation is to reduce cause-specific morbidity and mortality, but the effects will only be possible to assess over the long term.

s will only be possible to assess in the longer term, particularly given the possibility of best behaviour bias in the intervention clusters. Similarly, although an end itself, an aim of immunisation is to reduce cause-specific morbidity and mortality, but the effects will only be possible to assess over the long term. The effects of the intervention on full immunisation and anthropometric wasting in children were limited, for which we offer several possible explanations. First, exposure to the intervention at the individual level might have been insufficient. Although coverage was high, population turnover was around 30% annually, and 2 years was a short time in which to assess the effects of the intervention and to consolidate community involvement. We will continue to assess the effects over the coming years. Young children moving into the clusters from elsewhere without primary immunisation also meant that achieving full immunisation in all children was unlikely, despite the efforts of community organisers. Second, although we found no evidence of contamination of control clusters by the intervention, government schemes and the activities of municipal and non-governmental providers might have improved health in control clusters. Nevertheless, we saw no evidence of increased use of other providers in the control clusters (table 4). Additionally, births in institutions rather than at home have become the norm. We had hoped that the intervention would increase the use of municipal health care and ICDS and concurrently strengthen ICDS,43 but we saw no indication of these effects. A third possibility is secular change. Environmental indicators improved substantially during the period of the intervention. The proportion of homes made with robust fabric increased by 16% and that of households with private toilets by 7%. Purchase of drinking water from tankers fell by 53%. Although anthropometric wasting in children was reduced after the intervention in the intervention group (which monitoring data from day-care centres suggest was causal), improvement from low preintervention levels was also seen in control areas. We tested several hypotheses to explain improvements in the control group (appendix p 3), but saw no differential changes between control and intervention areas in housing quality, water supply, economic poverty, schooling, migration numbers, or state of origin.

ow preintervention levels was also seen in control areas. We tested several hypotheses to explain improvements in the control group (appendix p 3), but saw no differential changes between control and intervention areas in housing quality, water supply, economic poverty, schooling, migration numbers, or state of origin. Informal settlements have different cultural, structural, and legal statuses from formal settlements, but we think that our findings are generalisable to established informal settlements that have some amenities and high annual turnover. The range of issues that community organisers had to address led to rapport with beneficiaries, but challenged their ability to focus on our primary indicators. Training needs were extensive, daily work involved multiple home visits and group facilitation, and response to requests from beneficiaries was time-consuming. For example, a whole day might be spent in accompanying one woman to support her health-care needs or in responding to a domestic violence incident. A possible option is to prioritise activities that address risk over general activities, for instance by decreasing the frequency of growth monitoring for children who are doing well.

a whole day might be spent in accompanying one woman to support her health-care needs or in responding to a domestic violence incident. A possible option is to prioritise activities that address risk over general activities, for instance by decreasing the frequency of growth monitoring for children who are doing well. In pursuit of the 11th United Nations Sustainable Development Goal, city governments in India and elsewhere are seeking guidance on the use of resources to improve health in informal settlements. We believe the evidence from this trial suggests effectiveness with a community resource centre model. Certainly, activities may be protocolised, making the model feasible and replicable, and we are currently expanding the catchment area to achieve economy of scale. Supplementary Material Supplementary appendix

In pursuit of the 11th United Nations Sustainable Development Goal, city governments in India and elsewhere are seeking guidance on the use of resources to improve health in informal settlements. We believe the evidence from this trial suggests effectiveness with a community resource centre model. Certainly, activities may be protocolised, making the model feasible and replicable, and we are currently expanding the catchment area to achieve economy of scale. Supplementary Material Supplementary appendix Acknowledgments The Wellcome Trust supported the trial (091561/Z/10/Z). Child Rights and You supported the child nutrition intervention activities. Cipla provided medications for distribution by clinicians at the Society for Nutrition, Education and Health Action (SNEHA) centres. We thank the women, men, and children who worked with us in the SNEHA centres programme and who responded patiently to our questions. The intervention project officers were Dilip Bhane, Digambar Gaikwad, Jyoti Gaikwad, Akshay Kamble, Akhtar Khan, Sachin Kulkarni, Narendra Lolge, Rupesh Parab, Pradeep Pawar, Shubangi Sadakal, Arjun Thakur, and Kiran Thorat. The intervention coordinators were Mahesh Rajguru and Anagha Waingankar. The census officers were Rekha Bagul, Renuka Bhoir, Pratibha Doiphode, and Ashwini Jadhav. Clinical activities for the intervention were provided by Asmita Kukade, Neha Naik, Sangeeta Pawar, Ninad Salunke, Dipali Sud, and Vijayalaxmi Vasdani. Latika Chordekar, and Dhanlaxmi Solanki managed the data. We thank the data monitoring committee for their support and advice: Ashok Dyalchand, Nerges Mistry, Ramesh Potdar, and Sulabha Parasuraman. Archit Bhise, Shreesh Naik, Sheila Chanani, and Jeremy Wacksman (Dimagi) helped us develop methods for electronic data collection. Pouruchisti Wadia, Gauri Ambavkar, Gayatri Mehat, Pratibha Sakat, and Archana Wankhede provided counselling support for survivors of violence. We thank our partners in the Integrated Child Nutrition Services, the Municipal Corporation of Greater Mumbai, Mumbai Mobile Creches, and the Family Planning Association of India. We thank Alka Jadhav, Department of Pediatrics, Lokmanya Tilak Municipal General Hospital, Mumbai, India, and Padmaja Keskar and Sucheta Bhandary, Municipal Corporation of Greater Mumbai, Mumbai, India. Finally, thanks to all our colleagues at SNEHA, led by Vanessa D'Souza and the SNEHA Trustees, including Archana Bagra and Usha Ahuja for financial management, Aparna Patil for human resources management, Jaya Nuty for capacity management, Ajay Devrukhkar and Sachin Jadhav for administrative support, and Gazi Sayed for information technology support.

agues at SNEHA, led by Vanessa D'Souza and the SNEHA Trustees, including Archana Bagra and Usha Ahuja for financial management, Aparna Patil for human resources management, Jaya Nuty for capacity management, Ajay Devrukhkar and Sachin Jadhav for administrative support, and Gazi Sayed for information technology support. Anonymised census preintervention and postintervention datasets are available from DO, in accordance with SNEHA data sharing protocols. Contributors NSM. SD, GA, and DO conceived the study and designed the methods, and NSM and DO acquired funding. UB, SM, VK, RS, SS, ND, and SP supervised the study. NSM, SD, SM, VK, RS, and DO were responsible for project administration. SD collected the data and SD and DO curated and formally analysed the data. RS and DO wrote the first draft of the paper. All authors critically reviewed, commented on, and revised the drafts and GA and DO were responsible for preparation of the paper, figures, and tables. Declaration of interests We declare no competing interests. Figure 1 Trial design Figure 2 Locations of clusters in the M East and L wards of Mumbai included in randomisation Figure 3 Trial profile Figure 4 Changes after intervention in use of modern contraception, full immunisation of children aged 12–23 months, and anthropometric wasting in children younger than 5 years in the per-protocol analysis

Figure 1 Trial design Figure 2 Locations of clusters in the M East and L wards of Mumbai included in randomisation Figure 3 Trial profile Figure 4 Changes after intervention in use of modern contraception, full immunisation of children aged 12–23 months, and anthropometric wasting in children younger than 5 years in the per-protocol analysis (A) Use of modern contraception, defined as female or male terminal methods, oral contraceptive pill, intrauterine device, hormone implant or injection, condoms, or diaphragm. (B) Full immunisation, defined as BCG, diphtheria, pertussis, and tetanus (three doses), polio, hepatitis B virus (three doses), and measles. (C) Anthropometric wasting, defined as values more than 2 SD below the median WHO value for weight for length or height for age and sex. Table 1 Characteristics of households and respondents in the preintervention census

(A) Use of modern contraception, defined as female or male terminal methods, oral contraceptive pill, intrauterine device, hormone implant or injection, condoms, or diaphragm. (B) Full immunisation, defined as BCG, diphtheria, pertussis, and tetanus (three doses), polio, hepatitis B virus (three doses), and measles. (C) Anthropometric wasting, defined as values more than 2 SD below the median WHO value for weight for length or height for age and sex. Table 1 Characteristics of households and respondents in the preintervention census Control group Intervention group Households All 12 614 (100%) 12 239 (100%) Median (IQR) households per cluster 622 (609–649) 625 (584–637) Median (IQR) household size 5 (5–6) 5 (5–5) Number of homes 7317 (100%) 6976 (100%) Home owned 4359 (60%) 4189 (60%) Family had ration card 4660 (64%) 4453 (64%) Housing fabric Robust (pucca) 4071 (56%) 4328 (62%) Partly robust (semi-pucca) 1922 (26%) 1807 (26%) Temporary (kaccha) 1324 (18%) 841 (12%) Electricity supply None 8 (<1%) 8 (<1%) Metered (family pay bill) 4720 (64%) 4673 (67%) Family pay landlord 568 (8%) 674 (10%) Other 2021 (28%) 1621 (23%) Drinking water source Private tap 1091 (15%) 1442 (21%) Community tap stand 1241 (17%) 1312 (19%) Purchased from tanker or in containers 4985 (68%) 4222 (60%) Toilet Private 807 (11%) 805 (12%) Public 6382 (87%) 6170 (88%) No toilet facility 128 (2%) 1 (<1%) Flooring Dirt, sand, mud 481 (7%) 447 (6%) Concrete, brick, tiles 6836 (93%) 6529 (94%) Asset index quintile 1 (poorest) 1798 (25%) 1518 (22%) 2 1247 (17%) 1157 (17%) 3 1462 (20%) 1414 (20%) 4 1446 (20%) 1440 (21%) 5 (least poor) 1364 (19%) 1447 (21%) Women respondents Number of respondents 8227 (100%) 7947 (100%) Age (years) 15–19 244 (3%) 241 (3%) 20–29 3425 (42%) 3300 (41%) 30–39 2874 (35%) 2760 (35%) 40–49 1684 (20%) 1646 (21%) Religion Muslim 6231 (76%) 6591 (83%) Hindu 1967 (24%) 1291 (16%) Other 29 (<1%) 65 (<1%) Education None 3292 (40%) 2943 (37%) Primary 644 (8%) 565 (7%) Secondary 3846 (47%) 3928 (49%) Higher 441 (5%) 504 (6%) Missing 4 (<1%) 7 (<1%) Length of residence in Mumbai (years) <1 459 (6%) 461 (6%) 1–5 1189 (14%) 1080 (14%) 6–10 875 (11%) 870 (11%) >10 2096 (25%) 1977 (25%) Lifelong 2700 (33%) 2588 (33%) Missing 908 (11%) 971 (12%) Always lived in current home 3344 (41%) 3309 (42%) Parity 0 836 (10%) 817 (10%) 1 1218 (15%) 1199 (15%) 2 1637 (20%) 1588 (20%) 3 1577 (19%) 1518 (19%) ≥4 2958 (36%) 2824 (36%) Missing 1 (<1%) 1 (<1%) Pregnancy in previous 2 years 2225 (27%) 2117 (27%) Table 2 Outcome indicators in the preintervention census

971 (12%) Always lived in current home 3344 (41%) 3309 (42%) Parity 0 836 (10%) 817 (10%) 1 1218 (15%) 1199 (15%) 2 1637 (20%) 1588 (20%) 3 1577 (19%) 1518 (19%) ≥4 2958 (36%) 2824 (36%) Missing 1 (<1%) 1 (<1%) Pregnancy in previous 2 years 2225 (27%) 2117 (27%) Table 2 Outcome indicators in the preintervention census Control group Intervention group Primary outcomes Number of women respondents 8227 (100%) 7947 (100%) Number using family planning* 2414 (29%) 2294 (29%) Family planning method Female terminal 1333 (55%) 1207 (53%) Oral contraceptive pill 466 (19%) 419 (18%) Condom 293 (12%) 337 (15%) Intrauterine contraceptive device 202 (8%) 214 (9%) Other 120 (5%) 117 (5%) Immunisation in children aged 12–23 months† Number of children 1014 (100%) 945 (100%) Fully immunised 637 (63%) 613 (65%) BCG 935 (92%) 880 (93%) DPT and polio Dose 1 881 (87%) 839 (89%) Dose 2 832 (82%) 798 (84%) Dose 3 769 (76%) 757 (80%) Hepatitis B virus 731 (72%) 679 (72%) Measles 674 (66%) 672 (71%) Anthropometric wasting in children <5 years‡ 586 (15%) of 3881§ 640 (18%) of 3550§ Secondary outcomes Deliveries to women in the previous 2 years All 2022 (100%) 1905 (100%) Adolescent pregnancy 286 (14%) 264 (14%) Home delivery 297 (15%) 287 (15%) Institutional delivery 1719 (85%) 1588 (83%) Public institutional delivery 1102 (64%) 1017 (64%) Received public institutional delivery incentive 555 (50%) 550 (54%) Unknown 6 (<1%) 30 (2%) Anthropometric stunting and underweight in children aged 0–59 months Stunting 1851 (48%) of 3861§ 1632 (46%) of 3541§ Underweight 1541 (39%) of 3902§ 1472 (41%) of 3576§ Infant and young child feeding indicators Early initiation of breastfeeding 999 (48%) 841 (44%) Exclusive breastfeeding (<6 months) 357 (62%) 309 (62%) Continued breastfeeding at 1 year (12–15 months) 270 (73%) 245 (72%) Introduction of solid, semisolid, or soft foods (6–8 months) 90 (35%) 120 (47%) Minimum dietary diversity (6–23 months) 192 (13%) 186 (13%) Minimum meal frequency (6–23 months) 646 (43%) 630 (43%) Minimum acceptable diet (6–23 months) 67 (4%) 71 (5%) Consumption of iron-rich foods (6–23 months) 205 (14%) 238 (16%) Use of ICDS services Children <5 years eligible 5057 (100%) 4767 (100%) Used ICDS 466 (9%) 508 (11%) Food supplements almost daily 373 (7%) 391 (8%) Health check-ups at least once per month 163 (3%) 179 (4%) Regular early childhood development intervention 306 (6%) 340 (7%) Weight measured at least once per 3 months 333 (7%) 349 (7%) DPT=diphtheria, pertussis, and tetanus.

00%) 4767 (100%) Used ICDS 466 (9%) 508 (11%) Food supplements almost daily 373 (7%) 391 (8%) Health check-ups at least once per month 163 (3%) 179 (4%) Regular early childhood development intervention 306 (6%) 340 (7%) Weight measured at least once per 3 months 333 (7%) 349 (7%) DPT=diphtheria, pertussis, and tetanus. ICDS=Government of India Integrated Child Development Services. * Intracluster correlation coefficient 0·011. † Intracluster correlation coefficient 0·066. ‡ Intracluster correlation coefficient 0·008. § Height or weight measurements were not taken for some children. Table 3 Outcomes in the postintervention census

00%) 4767 (100%) Used ICDS 466 (9%) 508 (11%) Food supplements almost daily 373 (7%) 391 (8%) Health check-ups at least once per month 163 (3%) 179 (4%) Regular early childhood development intervention 306 (6%) 340 (7%) Weight measured at least once per 3 months 333 (7%) 349 (7%) DPT=diphtheria, pertussis, and tetanus. ICDS=Government of India Integrated Child Development Services. * Intracluster correlation coefficient 0·011. † Intracluster correlation coefficient 0·066. ‡ Intracluster correlation coefficient 0·008. § Height or weight measurements were not taken for some children. Table 3 Outcomes in the postintervention census Control group Intervention group Intention-to-treat OR (95% CI) Per-protocol OR (95% CI) Primary outcomes Met need for family planning among women aged 15–49 years 3134 (78%) of 4028 3439 (82%) of 4184 1·31 (1·11–1·53) 1·37 (1·07–1·75) Full immunisation among children aged 12–23 months 708 (62%) of 1143 751 (68%) of 1108 1·30 (0·84–2·01) 1·73 (1·05–2·86) Anthropometric wasting in children <5 years 580 (13%) of 4608 530 (12%) of 4570 0·92 (0·75–1·12) 0·88 (0·66–1·18) Secondary outcomes Home births in previous 2 years 276 (12%) of 2266 270 (12%) of 2202 1·14 (0·69–1·86) 1·25 (0·65–2·41) Adolescent pregnancies in previous 2 years in women <20 years 210 (9%) of 2266 199 (9%) of 2202 0·96 (0·69–1·33) 0·95 (0·55–1·65) Anthropometric stunting in children <5 years 2105 (46%) of 4610 2146 (47%) of 4573 1·03 (0·85–1·25) 1·08 (0·84–1·40) Anthropometric underweight in children <5 years 1766 (38%) of 4610 1791 (39%) of 4573 1·03 (0·88–1·21) 1·06 (0·88–1·29) Use of ICDS by children aged 0–23 months 25 (1%) of 2243 18 (<1%) of 2177 1·16 (0·72–1·89) 0·89 (0·53–1·51) Infant and young child feeding indicators Early initiation of breastfeeding 975 (44%) of 2198 1005 (47%) of 2146 1·10 (0·58–2·07) 1·11 (0·55–2·25) Exclusive breastfeeding (<6 months) 310 (56%) of 554 329 (66%) of 504 1·54 (1·02–2·33) 1·95 (1·02–3·76) Continued breastfeeding at 1 year (12–15 months) 318 (77%) of 411 288 (78%) of 368 1·05 (0·75–1·48) 1·58 (0·89–2·82) Introduction of solid, semisolid, or soft foods (6–8 months) 154 (52%) of 297 180 (58%) of 308 1·27 (0·75–2·15) 0·91 (0·44–1·88) Minimum dietary diversity (6–23 months) 274 (16%) of 1751 374 (22%) of 1718 1·48 (1·01–2·17) 1·54 (0·99–2·39) Minimum meal frequency (6–23 months) 1048 (60%) of 1751 1142 (66%) of 1718 1·26 (0·91–1·75) 1·21 (0·83–1·78) Minimum acceptable diet (6–23 months) 160 (9%) of 1751 218 (13%) of 1718 1·39 (0·89–2·17) 1·58 (0·94–2·65) Consumption of iron-rich foods (6–23 months) 282 (16%) of 1751 291 (17%) of 1718 1·05 (0·76–1·45) 1·25 (0·86–1·80) Further information on primary and secondary outcomes Family planning Use of modern contraception to space pregnancies 442 (14%) of 3134 598 (17%) of 3439 1·29 (1·06–1·58) 1·23 (0·95–1·60) Use of

·94–2·65) Consumption of iron-rich foods (6–23 months) 282 (16%) of 1751 291 (17%) of 1718 1·05 (0·76–1·45) 1·25 (0·86–1·80) Further information on primary and secondary outcomes Family planning Use of modern contraception to space pregnancies 442 (14%) of 3134 598 (17%) of 3439 1·29 (1·06–1·58) 1·23 (0·95–1·60) Use of modern contraception to limit pregnancies 1036 (33%) of 3134 1433 (42%) of 3439 1·44 (1·21–1·71) 1·15 (0·91–1·45) Planned pregnancy in previous 2 years (London Measure of Unplanned Pregnancy) 1678 (65%) of 2575 1799 (71%) of 2532 1·20 (0·84–1·73) 1·33 (0.86–2·04) Reported intimate partner violence in previous 1 year 1052 (14%) of 7705 556 (7%) of 7484 0·85 (0·44–1·64) 0·90 (0·46–1·72) Immunisation for children aged 12–23 months Immunisation card 551 (48%) of 1143 645 (58%) of 1108 1·52 (1·14–2·02) 1·58 (1·04–2·40) Fully immunised on immunisation card 395 (72%) of 1143 497 (77%) of 1108 1·30 (0·84–2·01) 1·73 (1·05–2·01) BCG 1077 (94%) of 1143 1054 (95%) of 1108 1·06 (0·61–1·87) 0·84 (0·24–3·01) DPT and polio Dose 1 991 (87%) of 1143 1000 (90%) of 1108 1·41 (0·78–2·55) 1·98 (0·87–4·49) Dose 2 905 (79%) of 1143 947 (85%) of 1108 1·52 (0·93–2·50) 2·80 (1·49–5·24) Dose 3 846 (74%) of 1143 905 (82%) of 1108 1·56 (1·03–2·38) 2·44 (1·42–4·20) Hepatitis B virus Dose 1 964 (84%) of 1143 973 (88%) of 1108 1·42 (0·80–2·54) 2·34 (1·03–5·29) Dose 2 883 (77%) of 1143 923 (83%) of 1108 1·49 (0·91–2·43) 2·36 (1·29–4·35) Dose 3 829 (73%) of 1143 874 (79%) of 1108 1·44 (0·93–2·25) 2·06 (1·24–3·43) Measles 735 (64%) of 1143 786 (71%) of 1108 1·34 (0·87–2·06) 1·99 (1·19–3·34) Immunisation for children aged 24–59 months Immunisation card 1077 (35%) of 3060 1713 (58%) of 2947 1·38 (0·91–2·10) 1·15 (0·76–1·74) BCG 2882 (94%) of 3060 2824 (96%) of 2947 1·25 (0·74–2·10) 1·36 (0·51–3·66) DPT and polio Dose 1 2684 (88%) of 3060 2685 (91%) of 2947 1·33 (0·86–2·05) 1·54 (0·87–2·75) Dose 2 2550 (83%) of 3060 2588 (88%) of 2947 1·32 (0·88–1·98) 1·72 (1·00–2·96) Dose 3 2475 (81%) of 3060 2517 (85%) of 2947 1·27 (0·87–1·86) 1·57 (0·95–2·60) Hepatitis B virus Dose 1 2634 (86%) of 3060 2610 (89%) of 2947 1·16 (0·76–1·77) 1·23 (0·69–2·19) Dose 2 2510 (82%) of 3060 2515 (85%) of 2947 1·18 (0·79–1·76) 1·36 (0·81–2·27) Dose 3 2436 (80%) of 3060 2442 (83%) of 2947 1·12 (0·77–1·63) 1·21 (0·73–2·01) Measles 2245 (73%) of 3060 2323 (79%) of 2947 1·25 (0·83–1·88) 1·39 (0·83–2·33) Anthropometric malnutrition in children aged 0–59 months Severe acute 93 (2%) of 4608 68 (1%) of 4570 0·73 (0·48–1·13) 0·54 (0·

1·18 (0·79–1·76) 1·36 (0·81–2·27) Dose 3 2436 (80%) of 3060 2442 (83%) of 2947 1·12 (0·77–1·63) 1·21 (0·73–2·01) Measles 2245 (73%) of 3060 2323 (79%) of 2947 1·25 (0·83–1·88) 1·39 (0·83–2·33) Anthropometric malnutrition in children aged 0–59 months Severe acute 93 (2%) of 4608 68 (1%) of 4570 0·73 (0·48–1·13) 0·54 (0· 27–1·09) Moderate acute 487 (11%) of 4608 462 (10%) of 4570 0·95 (0·78–1·17) 0·96 (0·71–1·29) Use of government ICDS by children aged 0–24 months Food supplements almost daily 13 (<1%) of 2243 10 (<1%) of 2177 1·16 (0·55–2·45) 1·73 (0·71–4·24) Health check-ups at least once per month 8 (<1%) of 2243 9 (<1%) of 2177 0·92 (0·49–1·73) 1·31 (0·64–2·68) Regular early childhood development intervention 3 (<1%) of 2243 5 (<1%) of 2177 1·40 (0·51–3·83) 1·71 (0·53–5·56) Weight measured at least once per 3 months 12 (<1) of 2243 8 (<1%) of 2177 1·73 (0·88–3·39) 2·39 (1·10–5·20) OR=odds ratio. ICDS=Government of India Integrated Child Development Services. DPT=diphtheria, pertussis, and tetanus. Table 4 Intervention coverage, contamination, and substitution in the postintervention census

27–1·09) Moderate acute 487 (11%) of 4608 462 (10%) of 4570 0·95 (0·78–1·17) 0·96 (0·71–1·29) Use of government ICDS by children aged 0–24 months Food supplements almost daily 13 (<1%) of 2243 10 (<1%) of 2177 1·16 (0·55–2·45) 1·73 (0·71–4·24) Health check-ups at least once per month 8 (<1%) of 2243 9 (<1%) of 2177 0·92 (0·49–1·73) 1·31 (0·64–2·68) Regular early childhood development intervention 3 (<1%) of 2243 5 (<1%) of 2177 1·40 (0·51–3·83) 1·71 (0·53–5·56) Weight measured at least once per 3 months 12 (<1) of 2243 8 (<1%) of 2177 1·73 (0·88–3·39) 2·39 (1·10–5·20) OR=odds ratio. ICDS=Government of India Integrated Child Development Services. DPT=diphtheria, pertussis, and tetanus. Table 4 Intervention coverage, contamination, and substitution in the postintervention census Control group Intervention group p value Women aged 15–49 years Number of women 8271(100%) 7965(100%) N/A Aware of local SNEHA centre 75 (<1%) 6661 (84%) <0·0001 Aware of services offered by SNEHA centre 41 (<1%) 6299 (79%) <0·0001 Growth monitoring 38 (<1%) 6170 (77%) N/A Immunisation 15 (<1%) 2437 (31%) N/A Child health checks 32 (<1%) 4802 (60%) N/A Nutrition education 18 (<1%) 3688 (46%) N/A Family planning 24 (<1%) 2896 (36%) N/A Counselling for violence against women and girls 12 (<1%) 1947 (24%) N/A Visited by community organiser 14 (<1%) 6981 (88%) <0·0001 Visited at least monthly 12 (<1%) 6291 (79%) N/A Participated in SNEHA centre activities 6 (<1%) 3108 (39%) <0·0001 Group meetings 5 (<1%) 2958 (37%) N/A Parents' meetings 1 (<1%) 704 (9%) N/A Recipe workshops 2 (<1%) 848 (11%) N/A Received municipal services in previous year 2711 (33%) 2760 (35%) 0·708 Antenatal care 717 (9%) 737 (9%) N/A Delivery care 759 (9%) 675 (8%) N/A Family planning 247 (3%) 156 (2%) N/A Immunisation 1209 (15%) 1231 (15%) N/A Health camp 1582 (19%) 1626 (20%) N/A Received other NGO services in previous year 259 (3%) 206 (3%) 0·493 Growth monitoring 717 (26%) 737 (27%) N/A Delivery care 53 (<1%) 3 (<1%) N/A Immunisation 18 (<1%) 1 (<1%) N/A Child health check 29 (<1%) 7 (<1%) N/A Family planning 103 (1%) 121 (2%) N/A Women with children <5 years Number of women 3800(100%) 3777(100%) N/A Received SNEHA centre service 16 (<1%) 3332 (88%) <0·0001 Growth monitoring 36 (<1%) 3299 (87%) N/A Immunisation 2 (<1%) 869 (23%) N/A Child health checks 9 (<1%) 2357 (62%) N/A Nutrition education 7 (<1%) 1506 (40%) N/A Family planning 2 (<1%) 714 (19%) N/A Counselling for violence against women and girls 0 69 (2%) N/A Aware of day-care centre for children 12 (<1%) 1803 (48%) <0·0001 Aware of services offered by day-care centre 9 (<1%) 1574 (42%) 0·256 Day care 7 (<1%) 983 (26%) N/A Growth monitoring 9 (<1%) 1458 (39%) N/A Immunisation 5 (<1%) 780 (21%) N/A Supplementary food 7 (<1%) 1321 (35%) N/A Child health checks 8 (<1%) 1162 (31%) N/A Play and learning 4 (<1%) 741 (20%) N/A Have used day-care centre 4 (<1%) 608 (16%) 0·894 Child admitted 1 (<1%) 443 (12%) N/A Growth monitoring 4 (<1%) 579

A Growth monitoring 9 (<1%) 1458 (39%) N/A Immunisation 5 (<1%) 780 (21%) N/A Supplementary food 7 (<1%) 1321 (35%) N/A Child health checks 8 (<1%) 1162 (31%) N/A Play and learning 4 (<1%) 741 (20%) N/A Have used day-care centre 4 (<1%) 608 (16%) 0·894 Child admitted 1 (<1%) 443 (12%) N/A Growth monitoring 4 (<1%) 579 (15%) N/A Immunisation 0 237 (6%) N/A Supplementary food 2 (<1%) 528 (14%) N/A Child health check 3 (<1%) 466 (12%) N/A Play and learning 1 (<1%) 331 (9%) N/A N/A=not applicable. SNEHA=Society for Nutrition, Education and Health Action. NGO=non-governmental organisation. Panel 1 Study intervention Inception microplanning We facilitated a series of participatory learning and action and resource-mapping exercises.19 5-day cycles involved community volunteers building rapport with each other and the cluster community, and facilitating understanding of community resources, patterns of health care, and local aspirations. Activities involved community members, front-line workers from other organisations, and resource agencies, and the findings were disseminated within communities and allied systems. Programme staff also established contact with public sector providers and local non-governmental organisations and held stakeholder workshops to streamline referrals. Communication emphasis

We facilitated a series of participatory learning and action and resource-mapping exercises.19 5-day cycles involved community volunteers building rapport with each other and the cluster community, and facilitating understanding of community resources, patterns of health care, and local aspirations. Activities involved community members, front-line workers from other organisations, and resource agencies, and the findings were disseminated within communities and allied systems. Programme staff also established contact with public sector providers and local non-governmental organisations and held stakeholder workshops to streamline referrals. Communication emphasis • Maternal and neonatal health: registration of pregnancy, antenatal care, referral for women at risk of pregnancy-related disorders, nutrition counselling, institutional delivery, essential neonatal care (eg, resuscitation, warmth, early and frequent breastfeeding, keeping with mother, cleanliness, and prompt identification of illness), and referral for obstetric and neonatal danger signs. • Child health and nutrition: feeding of infants and young children, immunisation, treatment of severe acute malnutrition, and seeking care for illness. • Sexual and reproductive health: family planning, adolescent sexual and reproductive health, and life skills. • Prevention of violence against women and children: disclosure, reporting, community-based crisis intervention, and support. Activities

• Child health and nutrition: feeding of infants and young children, immunisation, treatment of severe acute malnutrition, and seeking care for illness. • Sexual and reproductive health: family planning, adolescent sexual and reproductive health, and life skills. • Prevention of violence against women and children: disclosure, reporting, community-based crisis intervention, and support. Activities • Home visits: each community organiser visited ten homes per day in which they identified and provided information for family health needs, referred people to appropriate institutions, provided help with access, and negotiated with family members to facilitate appropriate choices. Home visits were open-ended, but health needs, providing appropriate information, and accessing services for women of reproductive age and their children were prioritised. • Group meetings: community organisers facilitated daily group meetings with between five and 15 people, including pregnant women, mothers of malnourished children, men, and adolescents, in the SNEHA centre, an open space, or a member's home to address women's and children's health concerns through information exchange, peer learning, and discussion, based on established participatory group approaches.13

ve and 15 people, including pregnant women, mothers of malnourished children, men, and adolescents, in the SNEHA centre, an open space, or a member's home to address women's and children's health concerns through information exchange, peer learning, and discussion, based on established participatory group approaches.13 • Day care for malnourished children: a day-care centre annex to the SNEHA centre was established in each cluster, staffed by a teacher and an aide trained in early childhood development activities by Mumbai Mobile Creches. Children with acute malnutrition (anthropometric wasting), classified as weight for length or height more than 2 SD below the WHO median for age and sex for moderate or more than 3 SD for severe, were identified and enrolled by community organisers and were screened by a clinician. Children with severe acute malnutrition were prescribed local medical nutrition therapy for 56 days with ready-to-use therapeutic food. Those with moderate acute malnutrition were prescribed four meals per day, including milk, egg, fruit, lunch from home, and the take-home ration provided by the Integrated Child Development Services. Children attending day-care centres were seen weekly by a SNEHA clinician and were referred to a paediatrician if they did not improve or they developed illness. Other day-care activities included early child development stimulation, rest, attention to hygiene, and a monthly parents' meeting.

hild Development Services. Children attending day-care centres were seen weekly by a SNEHA clinician and were referred to a paediatrician if they did not improve or they developed illness. Other day-care activities included early child development stimulation, rest, attention to hygiene, and a monthly parents' meeting. • Community events: events aimed at increasing awareness and creating an environment conducive to women's and children's health. These included puppet shows, street plays, rallies, games, competitions, “flash mobs”, and cooking demonstrations. Traditional events, such as the Godbharai baby shower and Ushtavan initiation of complementary feeding, were complemented by events celebrating improvements in child health and nutrition that provided opportunities to discuss health behaviours. International events, such as Breastfeeding and Nutrition Week and 16 Days Activism against Violence, were highlighted.

by shower and Ushtavan initiation of complementary feeding, were complemented by events celebrating improvements in child health and nutrition that provided opportunities to discuss health behaviours. International events, such as Breastfeeding and Nutrition Week and 16 Days Activism against Violence, were highlighted. • Service provision: counsellors were available to support survivors of physical, emotional, sexual, or economic violence by intimate or non-intimate partners. Women reporting violence were offered participation in an extensive support and response programme that included crisis intervention with counselling, psychotherapy, and family intervention, and support with police complaints and legal redress. Visiting clinicians who rotated weekly across SNEHA centres could provide some over-the-counter allopathic medications (anthelmintics, oral rehydration salts, antimalarials, eye and ear drops, oral antimicrobials, oral analgesics, and micronutrient supplements) and make referrals to paediatricians. Community organisers provided family planning directly (condoms and oral contraceptive pills) or by referral to the Family Planning Association of India under a formal memorandum.

antimalarials, eye and ear drops, oral antimicrobials, oral analgesics, and micronutrient supplements) and make referrals to paediatricians. Community organisers provided family planning directly (condoms and oral contraceptive pills) or by referral to the Family Planning Association of India under a formal memorandum. • Liaison: we worked with the Municipal Corporation of Greater Mumbai and the Indian Government Child Development Services to improve communication with communities and strengthen outreach and uptake. Regular meetings were held with front-line and supervisory groups to share data, plan, and deliver activities, such as outreach immunisation camps, growth monitoring, follow-up on referrals, and supplementary nutrition provided in Indian Government Child Development Services Anganwadi centres. Periodic meetings with senior officials focused on sharing concerns and developing strategies to improve access to and quality of public services. SNEHA=Society for Nutrition, Education and Health Action. Panel 2 Outcome definitions Primary outcomes • Met need for family planning among women aged 15–49 years, derived from the unmet need ascribed to married, widowed, separated, or divorced women who were pregnant but said that they did not want to be, who said they did not want any more children but were not using contraception, or who were using family planning but not with modern methods • Proportion of children aged 12–23 months fully immunised (BCG, diphtheria, pertussis, and tetanus [three doses], polio, hepatitis B virus [three doses], and measles29)

• Met need for family planning among women aged 15–49 years, derived from the unmet need ascribed to married, widowed, separated, or divorced women who were pregnant but said that they did not want to be, who said they did not want any more children but were not using contraception, or who were using family planning but not with modern methods • Proportion of children aged 12–23 months fully immunised (BCG, diphtheria, pertussis, and tetanus [three doses], polio, hepatitis B virus [three doses], and measles29) • Proportion of children younger than 5 years with anthropometric wasting, defined as weight for length or height more than 2 SD below the median WHO value for age and sex Secondary outcomes • Number of consultations for violence against women or children, including physical (slapping, arm twisting, pushing, punching, kicking, choking, or use of implements or weapons), emotional (jealousy, accusations of infidelity, prevention of association with friends or family, aggressive monitoring of whereabouts, lack of trust with money, humiliation in front of others, threats of violence, or insults), or sexual (forced sex, coercion into undesired sexual acts) • Proportion of home births in the preceding year • Proportion of pregnancies in the preceding 2 years in women younger than 20 years • Proportion of children younger than 5 years with anthropometric stunting • Proportion of children younger than 5 years with anthropometric underweight

• Proportion of children younger than 5 years with anthropometric stunting • Proportion of children younger than 5 years with anthropometric underweight • Proportion of children born in the preceding 2 years who received services from Government of India Integrated Child Development Services (food supplements, health checks, early childhood development intervention, or measurement of weight) • Proportion of children meeting Infant and Young Child Feeding core indicators: early initiation of breastfeeding, defined as breastfeeding within 1 h of birth; exclusive breastfeeding of children younger than 6 months, defined as receiving only breastmilk during the previous day; continued breastfeeding at 1 year, defined as children aged 12–15 months receiving breastmilk in the previous day; introduction of solid, semisolid, or soft foods in children, defined as children aged 6–8 months receiving such foods in the previous day; minimum dietary diversity, defined as children aged 6–23 months receiving foods from four or more food groups; minimum meal frequency, defined as breastfed and non-breastfed children aged 6–23 months receiving solid, semisolid, or soft foods (including milk feeds for non-breastfed children); minimum acceptable diet, defined as children aged 6–23 months receiving at least minimum dietary diversity and minimum meal frequency during the previous day in breastfed children and at least minimum dietary diversity (excluding milk feeds) and minimum meal frequency plus at least two milk feeds during the previous day in non-breastfed children; and consumption of iron-rich or iron-fortified foods in children aged 6–23 months30

Introduction The combination of fetal growth restriction, underweight, stunting, and wasting in later childhood, suboptimal breastfeeding, and micronutrient deficiencies have been estimated to cause more than 3 million child deaths annually, equivalent to 45% of the global total.1 Among these factors, the association between undernutrition and mortality is confounded by the effects of deprivation but is probably at least partly causal as evidenced by the greatly elevated hospital case fatality rates of undernourished children compared with better nourished children.2 The Millennium Development Goals (MDGs) adopted underweight as a key indicator for MDG1, but stunting has since been adopted as the preferred indicator because it offers a more stable index of long-term malnutrition. Latest estimates suggest that rates of stunting have been declining in most regions, but there remain 159 million children with stunting worldwide.3 The prevalence of stunting has declined most slowly in sub-Saharan Africa, and as a consequence of population growth the absolute number of children with stunting has increased.3

estimates suggest that rates of stunting have been declining in most regions, but there remain 159 million children with stunting worldwide.3 The prevalence of stunting has declined most slowly in sub-Saharan Africa, and as a consequence of population growth the absolute number of children with stunting has increased.3 Stunting rates fall rapidly as countries pass through the economic transition, but the key elements of progress that alleviate growth faltering are poorly understood, thus limiting the design of interventions and the targeting of health and development inputs in populations that remain impoverished. In this study, we analyse a longitudinal dataset spanning almost four decades of growth monitoring in three rural African villages that have received an unprecedented level of health-orientated interventions. A meta-analysis4 of previous interventions for water, sanitation, and hygiene (WASH) has not yielded strong grounds for optimism regarding the likely efficacy of such investments at the levels currently offered. The analysis of randomised trials included more than 4600 children studied over 9–12 months of intervention, and its findings showed no evidence of any beneficial effect on weight-for-age or weight-for-height and only a marginally significant effect on height-for-age of less than a tenth of a standard deviation (0·08 Z score, 95% CI 0·00–0·16). Additionally, the Lancet Series on Maternal and Child Nutrition5 reinforced the conclusion that nutrition interventions alone will have little effect on childhood undernutrition and estimated that, even if scaled up to 90% coverage, the implementation of all of the currently identified evidence-based interventions relating to nutrition would eliminate only about 20% of stunting globally. The results of ongoing trials to test the effect on growth of WASH interventions are keenly awaited.6, 7

on and estimated that, even if scaled up to 90% coverage, the implementation of all of the currently identified evidence-based interventions relating to nutrition would eliminate only about 20% of stunting globally. The results of ongoing trials to test the effect on growth of WASH interventions are keenly awaited.6, 7 Research in context Evidence before this study

on and estimated that, even if scaled up to 90% coverage, the implementation of all of the currently identified evidence-based interventions relating to nutrition would eliminate only about 20% of stunting globally. The results of ongoing trials to test the effect on growth of WASH interventions are keenly awaited.6, 7 Research in context Evidence before this study We searched PubMed and subsequent reference lists of relevant articles, with combinations of the terms “secular trends in growth”, “growth faltering”, “undernutrition”, “wasting”, “stunting”, “underweight”, “African children”, “rural”, “underfive”, and “infants” between June 1, 2012, and Feb 28, 2015. All studies published between Jan 1, 1980, and Feb 28, 2015, that had the relevant search terms (irrespective of language) were included. The quality of the evidence was inadequate for the research questions that we posed, including what the secular trends were in growth in rural African children younger than 2 years during the past four decades, and how the effect of seasonality on the growth of these children has changed during the past four decades. A small number of longitudinal studies from east and central sub-Saharan Africa described the growth patterns in cohorts of young African children over a period of less than a decade, assessing the effect of seasonality, immunisation uptake, and maternal health factors on the patterns of growth faltering. These findings showed that weight declined after the first 3 months in infants and that improved growth in infancy was associated with immunisation status and indices of adequate maternal nutritional status, whereas the rainy season was associated with reduced growth velocity. However, none of these studies described secular trends in these growth patterns. Additionally, the multicountry analyses used cross-sectional data, making interpretation of trends in growth faltering over time within individual populations difficult. Several studies from southern Africa assessed the secular trends in growth in children of school age and older children (older than 8 years), whereas other researchers combined different cohorts in their analyses, making it difficult to contextualise the associated trends in the social environmental and health interventions within the respective populations.

essed the secular trends in growth in children of school age and older children (older than 8 years), whereas other researchers combined different cohorts in their analyses, making it difficult to contextualise the associated trends in the social environmental and health interventions within the respective populations. Added value of this study To our knowledge, this study is the first to describe in fine detail the secular trends in longitudinal and seasonal growth patterns of children in a rural sub-Saharan African community with a constant sampling frame. We have documented the introduction of a series of nutrition-specific and nutrition-sensitive interventions resulting in an unprecedented level of health care in these villages. Simultaneous socioeconomic transitions have occurred with increased access to formal education, employment, and income through remittances from family members overseas. Families have become much less reliant on subsistence farming for their income and nutritional needs. These changes have resulted in reduction of mortality to a tenth of its former level in children younger than 5 years, and major reductions in diarrhoeal and other morbidity. Growth has improved but, despite these profound health and socioeconomic changes, the patterns of childhood growth faltering persist with stunting prevalence remaining at 30%. Our findings indicate that communities must exceed a very high threshold for health and environmental change before growth faltering will be eliminated. Implications of all the available evidence

To our knowledge, this study is the first to describe in fine detail the secular trends in longitudinal and seasonal growth patterns of children in a rural sub-Saharan African community with a constant sampling frame. We have documented the introduction of a series of nutrition-specific and nutrition-sensitive interventions resulting in an unprecedented level of health care in these villages. Simultaneous socioeconomic transitions have occurred with increased access to formal education, employment, and income through remittances from family members overseas. Families have become much less reliant on subsistence farming for their income and nutritional needs. These changes have resulted in reduction of mortality to a tenth of its former level in children younger than 5 years, and major reductions in diarrhoeal and other morbidity. Growth has improved but, despite these profound health and socioeconomic changes, the patterns of childhood growth faltering persist with stunting prevalence remaining at 30%. Our findings indicate that communities must exceed a very high threshold for health and environmental change before growth faltering will be eliminated. Implications of all the available evidence Children in resource limited settings, particularly in sub-Saharan Africa, continue to have suboptimal growth patterns despite access to public health interventions such as immunisation, clean water, and sanitation. Our analysis suggests that mitigation of growth faltering will need these public health interventions to be combined with many other improvements in children's environments, perhaps including improved housing with the provision of piped water directly into the home. Evidence from countries that have passed through the economic transition suggests that poverty reduction promotes such improvements and is accompanied by rapid declines in stunting.

ther improvements in children's environments, perhaps including improved housing with the provision of piped water directly into the home. Evidence from countries that have passed through the economic transition suggests that poverty reduction promotes such improvements and is accompanied by rapid declines in stunting. The implication therefore is that there is a very high threshold for improvements in living conditions, disease elimination, dietary sufficiency, and access to health care that must be exceeded to eliminate malnutrition. On this basis, we predict that current WASH interventions might not be sufficiently intensive to yield a substantial improvement in child growth, and that greater efforts will be required to meet the new UN Sustainable Development Goals (SDGs).8 In this study we assessed the aggregate improvements in child growth associated with progressive improvements in a wide range of nutrition-specific and nutrition-sensitive interventions in three rural Gambian villages that have been under continuous growth monitoring for almost 4 decades. Methods Study design and participants We did a retrospective cohort study using routine growth monitoring data for all children whose date of birth had been recorded to assess trends in growth faltering in children younger than 2 years in the West Kiang region of The Gambia during the past four decades.

The implication therefore is that there is a very high threshold for improvements in living conditions, disease elimination, dietary sufficiency, and access to health care that must be exceeded to eliminate malnutrition. On this basis, we predict that current WASH interventions might not be sufficiently intensive to yield a substantial improvement in child growth, and that greater efforts will be required to meet the new UN Sustainable Development Goals (SDGs).8 In this study we assessed the aggregate improvements in child growth associated with progressive improvements in a wide range of nutrition-specific and nutrition-sensitive interventions in three rural Gambian villages that have been under continuous growth monitoring for almost 4 decades. Methods Study design and participants We did a retrospective cohort study using routine growth monitoring data for all children whose date of birth had been recorded to assess trends in growth faltering in children younger than 2 years in the West Kiang region of The Gambia during the past four decades. Three rural villages in this region (Keneba, Manduar, and Kantong Kunda) have benefited from free health care provided by the UK Medical Research Council for the past 40 years. Since the 1970s there have been increasing levels of support and interventions (panel) such that these villages have benefited from unprecedented levels of nutrition-specific and nutrition-sensitive interventions compared with other such communities in rural low-income settings. Growth monitoring was done on a monthly basis in the 1970s but from 1983 onward, measurements were made at birth, 6 weeks, 3 months and then every 3 months thereafter. Diseases were recorded both at regular child ‘well baby’ clinics and when mothers presented with a sick child, and here we focus on clinical diagnoses for pneumonia, chest infections, diarrhoea, and malaria. Malaria diagnoses were based on positive blood films and, since 2007, on rapid diagnostic tests.9

hs thereafter. Diseases were recorded both at regular child ‘well baby’ clinics and when mothers presented with a sick child, and here we focus on clinical diagnoses for pneumonia, chest infections, diarrhoea, and malaria. Malaria diagnoses were based on positive blood films and, since 2007, on rapid diagnostic tests.9 As described elsewhere,9 the climate in the intervention area has a long, dry harvest season (November to May) and a wet so-called ‘hungry’ season (late June to mid-October) when agricultural work, depletion of food supply, and infectious diseases are at their peak. Ethics approval for the demographic surveillance of the three villages was granted by the Joint Gambian Government/Medical Research Council Unit The Gambia Ethics Committee. Procedures Standard anthropometric measurements were done in the clinic by trained clinic staff and Z scores were calculated against the WHO 2006 growth standards.10 We defined stunting, wasting, and underweight as height-for-age, weight-for-length, and weight-for-age of less than 2 SDs (Z scores) below the WHO reference median. Further details are provided in the appendix.

e in the clinic by trained clinic staff and Z scores were calculated against the WHO 2006 growth standards.10 We defined stunting, wasting, and underweight as height-for-age, weight-for-length, and weight-for-age of less than 2 SDs (Z scores) below the WHO reference median. Further details are provided in the appendix. Statistical analysis We fitted the effects of age and season on repeated growth parameters using random effects models. Models for boys and girls and each decade were fitted separately. To describe secular changes in rates of stunting, wasting, and underweight, we fitted random effects logistic regression of the binary variable on the first four orthogonal polynomials in age and the first pair of Fourier terms for season (appendix). To describe the effect of season on growth, we obtained seasonal patterns of body size by Fourier regression, as described in the appendix.11 To describe the changes in body size with age, we produced plots of mean Z score versus age by fitting age with ten-knot cubic regression splines and controlling for season by including the first pair of Fourier terms. We quantified growth faltering as the drop in Z score during the 18 month interval between 3 months and 21 months of age. These estimates are all simple linear combinations of the regression coefficients and their standard errors calculated using the variance–covariance matrix for the regression coefficients (ie, the Fisher information matrix), using Stata's post-estimation command lincom. We did not do any formal statistical hypothesis tests. With such large volumes of observational data almost any difference examined would be significant, so statistical significances poorly discriminate between important and trivial patterns in the data. Instead, we focused on estimation of effect sizes and their confidence intervals. All analyses were done with Stata 12.

With such large volumes of observational data almost any difference examined would be significant, so statistical significances poorly discriminate between important and trivial patterns in the data. Instead, we focused on estimation of effect sizes and their confidence intervals. All analyses were done with Stata 12. Role of the funding source The UK Medical Research Council has provided sustained support for our unique cohort over many years and approved our general research plans (including longitudinal data collection) every 5 years. MRC played no other role in interpretation of the data or preparation of the manuscript. The corresponding author had access to all of the data and had final responsibility for the decision to submit for publication. Results From May 1, 1976, to Feb 29, 2012, 4474 children younger than 2 years from these villages were seen at the child clinics in Keneba. Children were included in this analysis if their date of birth was known accurately and they visited the clinic on six or more occasions, giving a total of 3659 children eligible for the study. Those ineligible included 24 with unknown date of birth and 791 visitors who attended the clinic on five or fewer occasions. The median number of visits per child was 16 (IQR 13–26), resulting in a total of 59 371 visits at which anthropometric measurements were made. Most deliveries occurred at home in the presence of a traditional birth attendant but a trained midwife completed a baby check including anthropometric measurements within 5 days of delivery (mostly within 72 h).

d was 16 (IQR 13–26), resulting in a total of 59 371 visits at which anthropometric measurements were made. Most deliveries occurred at home in the presence of a traditional birth attendant but a trained midwife completed a baby check including anthropometric measurements within 5 days of delivery (mostly within 72 h). We analysed secular trends in birth size, because it is an important determinant of postnatal growth and attained size. Data about birth size were available for 2728 (75%) babies (figure 1, table 1). We excluded length data because there were more missing data than for weight and head circumference and because birth lengths measured with a length mat in the babies' homes are inherently less reliable than the other measurements. During the four decades of the study period, birthweight Z score increased by 0·26 (95% CI 0·18 to 0·34) from a starting point of −0·85 (figure 1, table 2). Head circumference at birth Z score increased by 0·58 (95% CI 0·33 to 0·83) from −0·36, thus ending up slightly above the WHO standards at 0·22 (95% CI 0·11 to 0·33). A small part of this increase might be attributable to a steady increase in maternal height totalling 28 mm (95% CI 18 to 38; table 1).

gure 1, table 2). Head circumference at birth Z score increased by 0·58 (95% CI 0·33 to 0·83) from −0·36, thus ending up slightly above the WHO standards at 0·22 (95% CI 0·11 to 0·33). A small part of this increase might be attributable to a steady increase in maternal height totalling 28 mm (95% CI 18 to 38; table 1). Figure 2 captures the characteristic growth patterns of these rural infants. They are born small and continue to fall away from the WHO standard length centiles throughout the first 2 years of life. Their weight shows early catchup while the infants are still fully breastfed and largely protected from infections; this trend is magnified in their weight-for-length due to the simultaneous decline in length. Mid-upper-arm and head circumferences show a similar resilience in very early infancy. The figure also illustrates the secular trends in growth during the four decades. Length shows a consistent, but limited, improvement. At 2 years, length-for-age Z score had improved by 0·74 (95% CI 0·59 to 0·89) from a starting point of −2·10 (table 2). Weight and head circumference showed an initial improvement by the second decade but little further gain. Weight-for-length showed absolutely no change in the second year of life. Mid-upper-arm circumference increased by a quarter of a Z score (table 2). The prevalence of stunting at 2 years almost halved from 57% to 30% and the prevalence of underweight decreased from 39% to 22% (table 2, figure 3). There was no change in the prevalence of wasting.

The figure also illustrates the secular trends in growth during the four decades. Length shows a consistent, but limited, improvement. At 2 years, length-for-age Z score had improved by 0·74 (95% CI 0·59 to 0·89) from a starting point of −2·10 (table 2). Weight and head circumference showed an initial improvement by the second decade but little further gain. Weight-for-length showed absolutely no change in the second year of life. Mid-upper-arm circumference increased by a quarter of a Z score (table 2). The prevalence of stunting at 2 years almost halved from 57% to 30% and the prevalence of underweight decreased from 39% to 22% (table 2, figure 3). There was no change in the prevalence of wasting. Growth failure is markedly seasonal in this environment (figure 2), with greater deficits occurring in the rainy season (July to November) when infections are more common and maternal care declines due to the pressures of farming. Figure 4 shows that there has been a substantial attenuation of the seasonality of growth during the four decades studied. When assessed as the amplitude of Z score fluctuation, this measure was significant for all indices (table 2) in the order of a tenth of a Z score.

al care declines due to the pressures of farming. Figure 4 shows that there has been a substantial attenuation of the seasonality of growth during the four decades studied. When assessed as the amplitude of Z score fluctuation, this measure was significant for all indices (table 2) in the order of a tenth of a Z score. We defined growth faltering on the basis of the differences in Z score between 3 months and 21 months post partum. In the 1970s, Z scores for length-for-age, weight-for-age, weight-for-length, and head circumference all fell by between 0·79 and 0·95 (table 2). Over time, this fall was slightly attenuated for length-for-age (Z score 0·17, 95% CI 0·11–0·23) and weight-for-length (0·09, 0·01–0·17), and more markedly attenuated for head circumference (0·28, 0·18–0·38; figure 5). The decline in Z score for weight-for-age and mid-upper-arm circumference did not change during the period studied. The incidence of diarrhoea, malaria, and bronchiolitis in the children younger than 12 months fell by 80% during the four decades studied. Conversely, the incidence of pneumonia seemed to increase during the four decades (figure 6).

We defined growth faltering on the basis of the differences in Z score between 3 months and 21 months post partum. In the 1970s, Z scores for length-for-age, weight-for-age, weight-for-length, and head circumference all fell by between 0·79 and 0·95 (table 2). Over time, this fall was slightly attenuated for length-for-age (Z score 0·17, 95% CI 0·11–0·23) and weight-for-length (0·09, 0·01–0·17), and more markedly attenuated for head circumference (0·28, 0·18–0·38; figure 5). The decline in Z score for weight-for-age and mid-upper-arm circumference did not change during the period studied. The incidence of diarrhoea, malaria, and bronchiolitis in the children younger than 12 months fell by 80% during the four decades studied. Conversely, the incidence of pneumonia seemed to increase during the four decades (figure 6). Discussion Goal 2 of the SDGs, “to end hunger, achieve food security and improved nutrition, and promote sustainable agriculture”, is accompanied by the target to achieve the internationally agreed goals for stunting and wasting in children younger than 5 years by 2025. For stunting, this goal would require a 40% reduction from the current estimate of 159 million stunted children to reach the target of less than 100 million. In Africa there has been a disappointing decline in the prevalence of stunting from 42% in 1990 to 32% in 201512 and, because of population growth, the absolute numbers of children with stunting actually increased from 47 million to 58 million during this period. The prevalence of stunting is now predicted to stabilise at that level because continued population growth offsets a slower-than-required decline in prevalence. By comparison, during the same period the prevalence of stunting in Asia decreased from 48% to 25% and the total number of children with stunting declined from 189 million to 84 million.12

now predicted to stabilise at that level because continued population growth offsets a slower-than-required decline in prevalence. By comparison, during the same period the prevalence of stunting in Asia decreased from 48% to 25% and the total number of children with stunting declined from 189 million to 84 million.12 Elimination of stunting creates a complex and paradoxical challenge, which suggests that one or more key causative factors remain unknown. On the one hand, nutrition-specific interventions have repeatedly shown very limited efficacy even when implemented under the optimal conditions of randomised trials,2, 13, 14, 15, 16, 17 whereas on the other hand, stunting resolves rapidly as wealth and living conditions improve in countries passing through the economic transition.11, 18

rition-specific interventions have repeatedly shown very limited efficacy even when implemented under the optimal conditions of randomised trials,2, 13, 14, 15, 16, 17 whereas on the other hand, stunting resolves rapidly as wealth and living conditions improve in countries passing through the economic transition.11, 18 The longitudinal data presented in this study add to this challenge. During almost four decades the Medical Research Council has made sustained investments in health care and nutrition-related infrastructure within our core study villages; these inputs are unparalleled across rural Africa and would be prohibitively expensive for governments of low-income countries to roll out nationwide. These villages have access to antenatal and postnatal care, and round-the-clock access to clinicians and nurses in a well equipped and efficient primary health-care clinic. All health services are free of charge. All children are fully vaccinated, receive vitamin A, mebendazole, and other health interventions as per WHO protocols. Breastfeeding rates are among the very best worldwide and are further supported by Baby Friendly Community Initiatives accompanied by regular messaging in support of exclusive breastfeeding for 6 months. Open defecation and water obtained from contaminated open wells have been universally replaced by latrines in all compounds and tube well water supplied through clean pipes to standpipes around the villages. These interventions have had a profound effect on mortality in children younger than 5 years9 and the incidence of most diseases, especially diarrhoea (which has been previously implicated as a major cause of growth failure; figure 6).19 Further, children attend regular well-baby checks with growth monitoring and we provide a dedicated treatment centre for severely malnourished children to treat those who do become malnourished. The remittance economy from village members who have migrated overseas, together with incomes from employment at the Medical Research Council, have greatly improved food security and attenuated the stress of the so-called hungry season as reflected in the reduction in the amplitude of seasonal growth faltering in figure 4. This increased wealth has also improved housing conditions and dispersed families over a wider area, reducing overcrowding. Child mortality has fallen, birth spacing has increased, and family size has decreased.

alled hungry season as reflected in the reduction in the amplitude of seasonal growth faltering in figure 4. This increased wealth has also improved housing conditions and dispersed families over a wider area, reducing overcrowding. Child mortality has fallen, birth spacing has increased, and family size has decreased. There is now free universal primary education with enrolment of about 97%, although this figure drops for secondary education particularly for girls to 30%.22 Furthermore we have, over the years, conducted and published a series of randomised trials of nutritional interventions targeted at pregnant and lactating mothers, infants, and children, with the main aim to improve growth (appendix). Our findings have shown at most modest improvements in infant growth, consistent with results from systematic reviews and meta-analyses.2, 13, 23 The modest increase of 2·8 cm in the mean maternal height is indicative of a small degree of improvement in maternal nutrition during the four decades. Meta-analysis of the relationship between maternal height and birthweight24 yielded an expected effect of 8 g more birthweight per cm of maternal height. The COHORTS group reported a similar value of 0·024 Z scores per cm.25 Therefore the increase in maternal height probably contributed only about 20–30 g of the observed 120 g increase in birthweight.

between maternal height and birthweight24 yielded an expected effect of 8 g more birthweight per cm of maternal height. The COHORTS group reported a similar value of 0·024 Z scores per cm.25 Therefore the increase in maternal height probably contributed only about 20–30 g of the observed 120 g increase in birthweight. A limitation for our data was the difficulty in deriving a consistent sampling frame for the population under study in an area undergoing rapid change, particularly in the later decades. Changes in the population structure during the past 40 years might have influenced the trends we have reported. We attempted to control for this factor by excluding the children who attended our clinic fewer than six times as likely visitors. However, exclusion of these children might have created a sampling bias, because infrequent attenders might represent resident children who engaged poorly with health care. This potential bias would only affect the trends displayed if the population prevalence of poor attenders changed during the period studied. Another limitation was missing data from the 1970s, particularly birth data, limiting our ability to evaluate the trends in these parameters. Additionally, we omitted birth length data because of poor reliability in the measurements and the small number of measurements that were available in all the four decades. Comparison of the trends noted in our core study villages with those in neighbouring villages receiving less intensive intervention would have been desirable, but such data were not available.

cause of poor reliability in the measurements and the small number of measurements that were available in all the four decades. Comparison of the trends noted in our core study villages with those in neighbouring villages receiving less intensive intervention would have been desirable, but such data were not available. Growth has improved during these four decades but, despite the unprecedented levels of investment, the prevalence of low birthweight (12%), childhood stunting (30%), and underweight (22%) remains high. The prevalence of wasting has not changed, and growth faltering between 3 months and 21 months has been only marginally attenuated. These data suggest that the refractory stunting must be caused by factors (beyond the improvements and interventions provided in the study villages) that are corrected as nations pass through the economic transition and advance from low-income and lower-middle-income status. Environmental enteropathy affecting almost all children in low-income settings has been proposed as the mechanism linking growth failure with WASH deficits.26 Our results, together with a previous analysis27 of associations between poor child growth and a range of indicators of socioeconomic status and living conditions in this same community, suggest that there is a very high threshold for WASH improvements that must be achieved before growth faltering can be eliminated. Improved housing conditions, possibly including the provision of piped water directly into the home, might be a necessary step in the global challenge to eliminate childhood malnutrition.

, suggest that there is a very high threshold for WASH improvements that must be achieved before growth faltering can be eliminated. Improved housing conditions, possibly including the provision of piped water directly into the home, might be a necessary step in the global challenge to eliminate childhood malnutrition. Our study villages of Keneba, Kantong Kunda, and Manduar are highly unusual (and possibly unique) in having the combination of intensive interventions over a protracted period accompanied by systematic growth monitoring; our results might therefore not be generalisable. However, before Medical Research Council inputs and in all other respects such as environment and farming practices they share many characteristics with countless other rural villages in sub-Saharan African in areas of low malaria endemicity. Therefore, we believe that our findings and suggestions for future interventions are likely to be applicable to other similar settings in rural Africa. Supplementary Material Supplementary appendix Acknowledgments We thank the staff at MRC Keneba that have served the community in the Keneba, Manduar and Kantong Kunda villages during the four decades, the children and their parents/carers. The MRC International Nutrition Group is supported by the UK Medical Research Council (grant no MC-A760-5QX00) and the UK Department for International Development under the MRC–DFID Concordat agreement.

ity in the Keneba, Manduar and Kantong Kunda villages during the four decades, the children and their parents/carers. The MRC International Nutrition Group is supported by the UK Medical Research Council (grant no MC-A760-5QX00) and the UK Department for International Development under the MRC–DFID Concordat agreement. Contributors AMP, AJF, SEM, and HMN designed the study; AJF and HMN analysed the data. HMN, AJF, SEM, and AMP prepared the manuscript and are responsible for the final content. All authors read and approved the final manuscript. Declaration of interests We declare no competing interests. Figure 1 Secular changes in weight and head circumference at birth Figures shows mean (SE) Z scores for both sexes combined, calculated using the WHO 2006 growth reference standards. Figure 2 Secular and seasonal trends in child growth Figure shows mean age and Z scores for sexes combined, calculated by comparison with the WHO 2006 growth reference standards. Length refers to length-for-age; weight refers to weight-for-age. Figure 3 Secular trends in stunting, underweight, and wasting at 2 years of age Stunting, underweight, and wasting are defined as proportion below −2 Z scores against WHO 2006 growth reference standards. Figure 4 Amplitude of the seasonality by decade Figure shows seasonal Z score amplitude for sexes combined, calculated by comparison with the WHO 2006 growth standards. Figure 5 Fall in Z scores between 3 and 21 months of age for each decade Figure shows fall in Z scores for sexes combined, calculated by comparison with the WHO 2006 growth standards. Figure 6 Disease episodes for each decade