Browse the corpus

e PCR reaction with cycle threshold values for the IS481 gene ≤35-<40 and negative for the ptxS1 gene. f One specimen also tested positive for influenza A/H1N1. g Two specimens also tested positive for influenza A/H3N2. One participant tested pertussis PCR-positive in two different specimens collected 22 days apart, only the first episode was included. Our results suggest that influenza vaccination had a protective impact in the rates of B. pertussis infection in adult women. This novel observation deserves further investigation on the possible mechanisms in the upper-respiratory tract that can lead to the synergy between the two pathogens. Also, the possible impact of vaccinating mothers against influenza in the transmission of B. pertussis to their young infants warrants further consideration, as most severe pertussis disease occurs prior to them completing their immunization against pertussis, and household contacts, particularly mothers have been identified as major sources of infection to the infants5. Combined influenza and pertussis vaccination during pregnancy might have a cumulative benefit against B. pertussis infection. Acknowledgements The authors would like to thank all the study participants, the staff of the Departments of Obstetrics, Neonatology, and Paediatrics at Chris Hani Baragwanath Academic hospital, Soweto, South Africa, for their dedication to their patients, including our trial participants; the study midwives, nurses, laboratory staff, counsellors and data capturers.

all the study participants, the staff of the Departments of Obstetrics, Neonatology, and Paediatrics at Chris Hani Baragwanath Academic hospital, Soweto, South Africa, for their dedication to their patients, including our trial participants; the study midwives, nurses, laboratory staff, counsellors and data capturers. This study was supported by the Bill & Melinda Gates Foundation (grant number OPP1002747). There was also partial support from the South African Research Chairs Initiative of the Department of Science and Technology and National Research Foundation in Vaccine Preventable Diseases; and the Medical Research Council: Respiratory and Meningeal Pathogens Research Unit. The contents of this report are solely the responsibility of the authors and do not necessarily represent the official views of their institutions or organizations or of the sponsors. The funders did not participate in any aspect of the study, including study-conduct, data collection, analyses of the data or the write-up of the manuscript. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMc1705208..

Introduction Mycobacterium tuberculosis (M.tb) causes more deaths worldwide than any other infectious agent and is increasingly characterized by antimicrobial resistance1. New preventative tools are essential to end the tuberculosis (TB) epidemic1. Vaccines that prevent pulmonary TB in adolescents and young adults would have major impact on control of drug-sensitive and multidrug-resistant TB by interrupting transmission2. Development of new TB vaccines is hampered by lack of validated preclinical models and human immune correlates of protection to provide evidence for advancing candidates into late-stage trials. M.tb exposure may result in early elimination of bacteria by innate or adaptive immunity; or establishment of infection, which may remain asymptomatic (latent) in most individuals or progress to active disease3. Vaccine-mediated prevention of M.tb infection (POI) could be an important signal of efficacy against TB disease. Further, size and duration of a POI trial are less than for a trial of disease prevention, since M.tb infection occurs more frequently4-6.

Background Despite evidence of success of rotavirus vaccines, over 90 million infants still lack access to a rotavirus vaccine (1, 2). Barriers to global implementation include cost, sub-optimal efficacy in low-income countries and lingering safety concerns (3, 4). An oral rotavirus vaccine administered at birth has potential to address these challenges. Rotavirus disease occurs early in life in infants in low-income countries (5). A birth dose rotavirus vaccine would provide early protection and maximize the opportunity to complete a full vaccine schedule (6). Birth presents a unique opportunity that may assist the uptake of an oral vaccine as gastric acid is limited and environmental enteropathy not yet established (7, 8). As intussusception is rare in newborns, a birth dose administration may offer a safety advantage (9). RV3-BB vaccine was developed from the human neonatal rotavirus strain, RV3 (G3P[6]), identified in the stool of asymptomatic infants (10). Wild-type infection with RV3 provided protection from severe gastroenteritis in the first 3 years of life, with strong heterotypic serological responses to community rotavirus strains (11, 12). RV3 appears to be naturally attenuated and adapted to the newborn gut, replicating well despite the presence of maternal antibodies and breastfeeding (13). RV3-BB vaccine aims to take advantage of the intrinsic characteristics of this novel strain to target a birth dose vaccination strategy. RV3-BB was well tolerated and immunogenic when delivered in a neonatal or infant schedule in a phase IIa trial in New Zealand (14).

ence of maternal antibodies and breastfeeding (13). RV3-BB vaccine aims to take advantage of the intrinsic characteristics of this novel strain to target a birth dose vaccination strategy. RV3-BB was well tolerated and immunogenic when delivered in a neonatal or infant schedule in a phase IIa trial in New Zealand (14). The primary objective of this study was to assess the efficacy of three doses of RV3-BB against severe rotavirus gastroenteritis to 18 months of age. Secondary objectives included assessment of efficacy, immunogenicity and safety of RV3-BB when delivered in a neonatal schedule (first dose 0-5 days of age), or an infant schedule (first dose 8-10 weeks of age), compared with placebo, efficacy to 12 months, against rotavirus gastroenteritis of any severity and all-cause severe gastroenteritis.

uded assessment of efficacy, immunogenicity and safety of RV3-BB when delivered in a neonatal schedule (first dose 0-5 days of age), or an infant schedule (first dose 8-10 weeks of age), compared with placebo, efficacy to 12 months, against rotavirus gastroenteritis of any severity and all-cause severe gastroenteritis. Methods Trial design and oversight This phase IIb, randomized, double-blind, placebo-controlled trial involving 1649 participants was conducted from January 2013 to July 2016 in primary health centers and hospitals in Central Java and Yogyakarta, Indonesia. Indonesia is a low-middle income country with an under-5 mortality rate in Yogyakarta and Central Java of 30-38 per 1000 live births and per capita gross regional product of USD $2,164-$2,326 (15, 16). Authors from Murdoch Childrens Research Institute (MCRI) and Universitas Gadjah Mada (UGM) designed the trial. The protocol was approved by the ethics committees of UGM, Royal Children's Hospital Melbourne and National Agency of Drug and Food Control, Republic of Indonesia. Use of a placebo was deemed acceptable as rotavirus vaccines are not implemented in the Indonesian National Immunization Program and cost limits private purchase (17). The study was conducted in accordance with International Council for Harmonisation Good Clinical Practice guidelines and monitored by an independent contract research organization (Quintiles Pty Ltd). The study sponsor was MCRI and Indonesian sponsor was PT Bio Farma. An independent Data Safety Monitoring Board regularly reviewed safety data. Data management was performed by Biophics, Thailand. Statistical analysis was conducted by INC Research, Australia and an independent Statistical Consultant (WR). The National Health and Medical Research Council, Bill and Melinda Gates Foundation and PT Bio Farma funded the trial but had no role in study design, data collection or interpretation, or the decision to submit for publication. The second and third authors led clinical data collection. The first author wrote the first draft of the manuscript. All authors provided review and vouch for the accuracy and completeness of the data and analysis, and for the fidelity of the trial to the protocol (available at NEJM.org).

ion to submit for publication. The second and third authors led clinical data collection. The first author wrote the first draft of the manuscript. All authors provided review and vouch for the accuracy and completeness of the data and analysis, and for the fidelity of the trial to the protocol (available at NEJM.org). Participants, Randomization and Blinding Preliminary written informed consent was obtained from pregnant women prior to cord blood collection. Final written informed consent was obtained following birth prior to confirming eligibility. Eligible infants (healthy, full term babies 0-5 days of age, birth weight of 2.5-4.0 kg) were randomized into one of three groups (neonatal vaccine group, infant vaccine group, or placebo group) in a 1:1:1 ratio according to a computer generated code (block size =6) stratified by province. Investigational product (IP) (RV3-BB or placebo) doses were drawn into syringes for dispensing by an unblinded pharmacist at the central Pharmacy in each province. Investigators, study staff, families, monitors, data managers and statisticians remained blinded throughout the study.

lock size =6) stratified by province. Investigational product (IP) (RV3-BB or placebo) doses were drawn into syringes for dispensing by an unblinded pharmacist at the central Pharmacy in each province. Investigators, study staff, families, monitors, data managers and statisticians remained blinded throughout the study. Participants received four 1ml oral doses of IP according to their treatment allocation, with doses administered at 0-5 days (IP dose 1), 8-10 weeks (IP dose 2), 14-16 weeks (IP dose 3) and 18-20 weeks of age (IP dose 4) (Figure 1a). IP doses 2, 3 and 4 were preceded by a 2ml dose of antacid solution (Mylanta® Original). Feeding was withheld for 30 minutes before and after each dose. IP was co-administered at the same time as vaccines in the Indonesian NIP. Participants were followed by weekly telephone contacts and monthly visits to 18 months. All participants received oral polio vaccine, except for a subset of 282 participants included in the immunogenicity analysis of RV3-BB co-administered with inactivated polio vaccine. Vaccine RV3-BB clinical trial lots were prepared at Meridian Life Sciences (Memphis, USA) to a titre of 8.3 - 8.7 x 106 FFU/mL in serum free media supplemented with 10% sucrose. Placebo contained the same media with 10% sucrose and was visually indistinguishable. Vials were stored at -70°C until thawed within 6 hours prior to administration.

trial lots were prepared at Meridian Life Sciences (Memphis, USA) to a titre of 8.3 - 8.7 x 106 FFU/mL in serum free media supplemented with 10% sucrose. Placebo contained the same media with 10% sucrose and was visually indistinguishable. Vials were stored at -70°C until thawed within 6 hours prior to administration. Efficacy Severe rotavirus gastroenteritis was defined as rotavirus gastroenteritis with a modified Vesikari score of ≥ 11. Rotavirus gastroenteritis was defined as gastroenteritis with rotavirus antigen detected in the stool by enzyme linked-absorbent assay (ProSpecT Rotavirus Microplate Assay; Oxoid Ltd, UK). A modified Vesikari score was applied where intravenous, naso-gastric rehydration or 6-hours of supervised oral rehydration was scored as hospitalization, whether administered within a primary health center or hospital. Gastroenteritis of any severity, defined as three or more stools looser than normal for that child within a 24-hour period.

re was applied where intravenous, naso-gastric rehydration or 6-hours of supervised oral rehydration was scored as hospitalization, whether administered within a primary health center or hospital. Gastroenteritis of any severity, defined as three or more stools looser than normal for that child within a 24-hour period. Vaccine Take and Immunogenicity Vaccine take was assessed in the first cohort recruited (n=282). Blood was collected from the cord (baseline for neonatal schedule comparison), immediately prior to IP dose 2 (baseline for infant schedule comparison), 28 days after IP dose 3 and 28 days after IP dose 4. Serum rotavirus immunoglobulin A (IgA) antibody titers and serum neutralizing antibody titers were measured using previously described methods (14, 18). RV3-BB shedding in stool was detected using a rotavirus VP6 specific reverse transcription-polymerase chain reaction assay and confirmed by sequence analysis (14). Positive vaccine take was defined as a serum immune response (≥3 fold increase in titer from baseline in anti-rotavirus IgA or serum neutralising antibodies) 28 days following IP administration, or RV3-BB shedding on days 3-7 following IP administration. Cumulative vaccine take was defined as a positive vaccine take following IP dose of 1, 2 or 3 for the neonatal vaccine group, and following IP doses 2, 3 or 4 for the infant vaccine group.

or serum neutralising antibodies) 28 days following IP administration, or RV3-BB shedding on days 3-7 following IP administration. Cumulative vaccine take was defined as a positive vaccine take following IP dose of 1, 2 or 3 for the neonatal vaccine group, and following IP doses 2, 3 or 4 for the infant vaccine group. Safety Vital signs were assessed prior to, and in the 30 minutes after IP administration. Parents reported temperature and solicited gastrointestinal and systemic symptoms on diary cards for 7 days following each IP dose. All unsolicited adverse events (AEs) occurring up to 28 days after administration of IP doses were recorded. Serious AEs (SAEs) were defined as an AE that resulted in death, new or prolonged hospitalization or considered to be medically significant or life threatening occurring up to 28 days following the last dose of IP. Causality and severity grading of AEs were determined by the local Indonesian investigators.

re recorded. Serious AEs (SAEs) were defined as an AE that resulted in death, new or prolonged hospitalization or considered to be medically significant or life threatening occurring up to 28 days following the last dose of IP. Causality and severity grading of AEs were determined by the local Indonesian investigators. Statistical Methods The primary efficacy analysis compared the proportion of participants with an episode of severe rotavirus gastroenteritis from two weeks after IP dose 4 to 18 months in the combined vaccine group (neonatal and infant vaccine schedules) with that observed in the placebo group in the per protocol (PP) population, using a Pearson’s Chi square test. The PP population included only participants who received all 4 doses of IP within visit windows. A secondary analysis was conducted in the intention-to-treat (ITT) population (all randomized participants), comparing events from randomisation to 18 months. Vaccine efficacy is presented as 1-risk ratio x 100 with its exact 95% confidence interval based on the Clopper-Pearson method (19).

of IP within visit windows. A secondary analysis was conducted in the intention-to-treat (ITT) population (all randomized participants), comparing events from randomisation to 18 months. Vaccine efficacy is presented as 1-risk ratio x 100 with its exact 95% confidence interval based on the Clopper-Pearson method (19). Efficacy and vaccine take was assessed for the neonatal vaccine group (from two weeks after IP dose 3 to 12 and 18 months) and the infant vaccine group (from two weeks after IP dose 4 to 12 and 18 months). This resulted in two different presentations of data in the placebo group (denoted neonatal placebo and infant placebo). For the vaccine take analysis a participant was defined as missing only if all components of the outcome were missing. A Kaplan–Meier curve was used to estimate the cumulative hazard of a first severe rotavirus gastroenteritis episode from randomization, with group comparisons via the logrank test. All statistical tests were two-sided.

ake analysis a participant was defined as missing only if all components of the outcome were missing. A Kaplan–Meier curve was used to estimate the cumulative hazard of a first severe rotavirus gastroenteritis episode from randomization, with group comparisons via the logrank test. All statistical tests were two-sided. Based on local data we assumed 3% of placebo participants would experience an episode of severe rotavirus gastroenteritis during the study (20, 21) and calculated a sample size of 549 participants in each group would provide 80% power to reject the null hypothesis of no difference between the combined vaccine and placebo groups if the true efficacy was 60% (one-sided test with alpha of 0.1), allowing for 10% non-adherence. We calculated 282 participants were required to reject the null hypothesis of no difference in the proportion with a positive vaccine take (two-sided test with alpha of 0.05) assuming 25% of placebo participants would be exposed to rotavirus (14) and 50% in each vaccine group would have a positive vaccine take, allowing for 10% non-adherence. Results Of the 1649 newborns randomized, 1640 received at least one dose of IP (safety population) and 1588 (96%) were followed to 18 months. The primary efficacy analysis was performed on 1513 (92%) in the PP population (Figure 1b). The demographic characteristics of the study population and age of receipt of first dose of IP were similar across all groups (Appendix: Table S1).

st one dose of IP (safety population) and 1588 (96%) were followed to 18 months. The primary efficacy analysis was performed on 1513 (92%) in the PP population (Figure 1b). The demographic characteristics of the study population and age of receipt of first dose of IP were similar across all groups (Appendix: Table S1). Vaccine Efficacy Severe rotavirus gastroenteritis occurred in 28/504 (5.6%) participants in the placebo group compared with 21/1009 (2.1%) participants in the combined vaccine group, resulting in a vaccine efficacy of 63% at 18 months in the primary (PP) analysis (95% CI, 34, 80; p<0.001), with similar results in the ITT analysis (60%; 95%CI 31, 76; <0.001) (Table 1). Table 1 Efficacy of RV3-BB vaccine against severe rotavirus gastroenteritis to 18 months Per Protocol analysis Intention-to-treat analysis N No. participants with an episode (%) Efficacy* 95% CI p value N No. participants with an episode (%) Efficacy* 95% CI p value Placebo 504 28 (5.6%) 550 31 (5.6%) Combined vaccine group 1009 21 (2.1%) 63% 34, 80 <0.001 1099 25 (2.3%) 60% 31, 76 <0.001 Neonatal vaccine group 498 7 (1.4%) 75% 44, 91 <0.001 549 10 (1.8%) 68% 35, 86 0.001 Infant vaccine group 511 14 (2.7%) 51% 7, 76 0.03 550 15 (2.7%) 52% 11, 76 0.02 * When compared to respective placebo participants

) 550 31 (5.6%) Combined vaccine group 1009 21 (2.1%) 63% 34, 80 <0.001 1099 25 (2.3%) 60% 31, 76 <0.001 Neonatal vaccine group 498 7 (1.4%) 75% 44, 91 <0.001 549 10 (1.8%) 68% 35, 86 0.001 Infant vaccine group 511 14 (2.7%) 51% 7, 76 0.03 550 15 (2.7%) 52% 11, 76 0.02 * When compared to respective placebo participants When administered in the neonatal schedule, three doses of RV3-BB was associated with an efficacy of 75% against severe rotavirus gastroenteritis to 18 months (95% CI 44, 91; p<0.001) (Table 1) and 94% to 12 months (95% CI 56, 99; p=0.006) (Appendix: Table S2). Efficacy against rotavirus gastroenteritis of any severity to 18 months in the neonatal vaccine group was 63% (95% CI 37, 81; p<0.001), (Appendix: Table S2). In the infant vaccine group, efficacy against severe rotavirus gastroenteritis to 18 months was 51% (95% CI 7, 76; p=0.03) (Table 1) and 77% to 12 months (95% CI 31, 92; p=0.008) (Appendix: Table S2). Efficacy against rotavirus gastroenteritis of any severity to 18 months when RV3-BB was administered in the infant schedule was 45% (95% CI 12, 69; p=0.01) (Appendix: Table S2). The time from randomization to first episode of severe rotavirus gastroenteritis differed in participants receiving RV3-BB compared to placebo (Figure 2). Forty-six of 49 participants with severe rotavirus gastroenteritis had G3P[8] rotavirus detected in the stool.

In the infant vaccine group, efficacy against severe rotavirus gastroenteritis to 18 months was 51% (95% CI 7, 76; p=0.03) (Table 1) and 77% to 12 months (95% CI 31, 92; p=0.008) (Appendix: Table S2). Efficacy against rotavirus gastroenteritis of any severity to 18 months when RV3-BB was administered in the infant schedule was 45% (95% CI 12, 69; p=0.01) (Appendix: Table S2). The time from randomization to first episode of severe rotavirus gastroenteritis differed in participants receiving RV3-BB compared to placebo (Figure 2). Forty-six of 49 participants with severe rotavirus gastroenteritis had G3P[8] rotavirus detected in the stool. Vaccine Take and Immunogenicity Cumulative vaccine take following three doses of RV3-BB was detected in 78/83 (94%) of neonatal vaccine group and 83/84 (99%) of infant vaccine group (difference in proportions: neonatal vaccine group compared with neonatal placebo 0.52 [95%CI 0.39, 0.64; p<0.001]; infant vaccine group compared to infant placebo 0.52 [95%CI 0.40, 0.63; p<0.001]) (Figure 3; Appendix: Table S3). Cumulative serum immune response was observed after three doses of RV3-BB in 76% of neonatal vaccine group and 87% infant vaccine group. A serum IgA response was identified in 66% of the neonatal vaccine group and 81% of the infant vaccine group. Following two doses, cumulative vaccine take was identified in 87% of infant vaccine group compared with 28% in the infant placebo group (difference in proportions 0.59; 95% CI 0.45, 0.71: p<0.001). This comparison could not be assessed in the neonatal vaccine group as no blood was drawn at that time-point. Vaccine virus shedding was detected in 69% of the neonatal vaccine group and 75% of the infant vaccine group.

d with 28% in the infant placebo group (difference in proportions 0.59; 95% CI 0.45, 0.71: p<0.001). This comparison could not be assessed in the neonatal vaccine group as no blood was drawn at that time-point. Vaccine virus shedding was detected in 69% of the neonatal vaccine group and 75% of the infant vaccine group. Safety RV3-BB was well tolerated with the incidence of SAEs (Table 2) and unsolicited and solicited AEs similar across groups (Appendix: Table S4). All 11 deaths (neonatal vaccine group n=5, placebo n=6) were assigned as unrelated to IP by the investigator and are listed in Appendix Table S5. One case of intussusception occurred 114 days after the third dose of vaccine (infant vaccine group) and was assessed as unrelated to vaccine.

ss groups (Appendix: Table S4). All 11 deaths (neonatal vaccine group n=5, placebo n=6) were assigned as unrelated to IP by the investigator and are listed in Appendix Table S5. One case of intussusception occurred 114 days after the third dose of vaccine (infant vaccine group) and was assessed as unrelated to vaccine. Table 2 Adverse Events Discussion The human neonatal vaccine RV3-BB provided protection against severe rotavirus gastroenteritis and was well tolerated. When administered in the neonatal schedule, RV3-BB had a vaccine efficacy of 94% at 12 months and 75% at 18 months, providing proof of principle for the use of RV3-BB in a birth dose vaccination schedule. These results compare very favourably with licensed vaccines studied in similar high disease burden, low- and low-middle income countries. In a two dose schedule, Rotarix (Glaxo-SmithKline) had a combined one and two year efficacy of 34% in Malawi (22). In a three dose schedule, the combined one and two year efficacy for Rotarix was 42.3% (Malawi)(22), RotaTeq (Merck) was 17.6 to 63.9% (Mali, Bangladesh, Vietnam, Ghana, and Kenya) (23, 24) and Rotavac (Bharat Biotech) was 55.1% (India) (25). Three doses of Rotasil (Serum Institute of India) had an efficacy of 66.7% at a mean follow up of 9.8 months of age in Niger (26). If the 75% protective efficacy for the neonatal schedule of RV3-BB translates into effectiveness throughout Indonesia, it has the potential to avert an estimated 5,450 deaths, 117,110 hospitalizations and >300,000 outpatient clinic visits each year due to rotavirus gastroenteritis in children under 5 years (27).

(26). If the 75% protective efficacy for the neonatal schedule of RV3-BB translates into effectiveness throughout Indonesia, it has the potential to avert an estimated 5,450 deaths, 117,110 hospitalizations and >300,000 outpatient clinic visits each year due to rotavirus gastroenteritis in children under 5 years (27). The concept of a birth dose strategy for vaccination is not new. Birth is an established immunization time-point in many countries. A neonatal dose was investigated in the early phase of rotavirus vaccine development but not pursued due to concerns regarding inadequate immune responses and safety (28-30). The VP4 protein of human neonatal P[6] strains have specific residues at the basal surface of VP8* that may allow them to adhere to cell surface receptors in the newborn gut (31). This may provide an advantage for a birth dose schedule. The P[6] VP4 of RV3-BB may also offer an advantage in Africa and Asia where the Lewis-negative phenotype is common (32). Lewis (FUT3) and secretor (FUT2) genes appear to mediate susceptibility to rotavirus infection (32). P[8] rotaviruses only infect individuals who are Lewis-positive and secretor-positive whereas P[6] rotaviruses infect individuals irrespective of their Lewis and secretor status (33). This may explain the high proportion of disease caused by P[6] rotaviruses in Africa and the lower efficacy of vaccines with a P[8] genotype in these region (34). RV3-BB is currently the only vaccine with a P[6] VP4.

itive whereas P[6] rotaviruses infect individuals irrespective of their Lewis and secretor status (33). This may explain the high proportion of disease caused by P[6] rotaviruses in Africa and the lower efficacy of vaccines with a P[8] genotype in these region (34). RV3-BB is currently the only vaccine with a P[6] VP4. Unlike IgG, IgA is not transferred via the placenta, and the newborn may not mount a significant serum IgA response following the birth dose of an oral vaccine, such as RV3-BB, despite evidence that the neonatal schedule is efficacious (35). Similar dissonance has been demonstrated with other vaccines administered in the newborn period (36). An equine-like G3P[8] strain was responsible for most episodes of severe gastroenteritis in this study and reflects the global emergence of this strain (37). Based on the strong heterotypic serological responses to community strains (G1,G2 dominant) provided by the parent strain RV3 (11), it is anticipated that RV3-BB will also offer protection against a range of circulating rotavirus strain but this could not be assessed in this study. Despite the success of rotavirus vaccines remaining challenges to global implementation need to be overcome if all infants are to be protected against rotavirus disease. RV3-BB was efficacious, immunogenic and well-tolerated when administered in a neonatal or infant schedule. In particular, the high protective efficacy in the neonatal schedule suggests that RV3-BB could make a significant contribution to the global prevention of rotavirus disease.

e protected against rotavirus disease. RV3-BB was efficacious, immunogenic and well-tolerated when administered in a neonatal or infant schedule. In particular, the high protective efficacy in the neonatal schedule suggests that RV3-BB could make a significant contribution to the global prevention of rotavirus disease. Supplementary Appendix Supplementary Material Acknowledgements We would sincerely like to thank the infants and their families for participating in this study and the members of the UGM Paediatric Research Office who assisted in this trial: Nia Milastuti Triatmojo, Rony Trilaksono, Pramitha Esha Nirmala Dewi and all research assistants. This work was assisted by the RV3 Trial Site Co-ordinators: Dr Fauziah, Dr Samad, Bu Inayati Hasanah Evita Dewi, Dr Cahyo Widodo, Dr Roni and Dr Agus. We are also very grateful for the support from the Director, Paediatricians, Head of Research and Training Unit and all staff at Soeradji Tirtonegoro Hospital Klaten and District Hospital at Sleman, and all affiliated private hospitals and clinics. We would like to thank the heads of Health District Office of Klaten and Sleman, and all study physicians and midwives in Primary Health Centers in the Klaten and Sleman regions for their contributions to this study. For their support of this study, we would like to thank the UGM Dean of Faculty of Medicine, the Head of Pediatric Department Faculty of Medicine, Head of the Microbiology Laboratory Dr Abu Tholib and trial advisors: Professor Iwan Dwiprahasto, Professor Mohammad Hakimi Dr. Mei Neni, Dr. Ekawati Luthfia Haksari and Dr. Ahmad Mahmudi. We are extremely grateful for the support from PT Bio Farma and President Director Dr Iskandar, Dr Sugeng Raharso, and Dr Adriansjah Azhari.

he Microbiology Laboratory Dr Abu Tholib and trial advisors: Professor Iwan Dwiprahasto, Professor Mohammad Hakimi Dr. Mei Neni, Dr. Ekawati Luthfia Haksari and Dr. Ahmad Mahmudi. We are extremely grateful for the support from PT Bio Farma and President Director Dr Iskandar, Dr Sugeng Raharso, and Dr Adriansjah Azhari. We are indebted to the members of the DSMB: Professors Peter Richmond (Chair), Beatrice de Vos, Kristine Macartney, Michael Law, Sri Rejeki Hadinegoro and the RV3 Rotavirus Vaccine Scientific Advisory Committee: chaired by Professor Sir Gustav Nossal A.O., including Karen Kotloff, Duncan Steele, Tilman Ruff, John Matthews, Kim Mulholland, Marie-Paule Kieny and Don Roberton. We would like to thank the RV3 Clinical Reference Committee including Margaret Danchin and Francesca Orsini. We are extremely grateful for the guidance provided by consultants to this study by Nicole Kruger (NMK Consulting), Wasima Rida (independent biostatistics consultant) and Mark Sullivan (Medicines Development for Global Health). We thank also our service providers Quintiles (study monitoring), Biophics Thailand (data management) and INC Research (Statistical analysis). This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1706804.

We are indebted to the members of the DSMB: Professors Peter Richmond (Chair), Beatrice de Vos, Kristine Macartney, Michael Law, Sri Rejeki Hadinegoro and the RV3 Rotavirus Vaccine Scientific Advisory Committee: chaired by Professor Sir Gustav Nossal A.O., including Karen Kotloff, Duncan Steele, Tilman Ruff, John Matthews, Kim Mulholland, Marie-Paule Kieny and Don Roberton. We would like to thank the RV3 Clinical Reference Committee including Margaret Danchin and Francesca Orsini. We are extremely grateful for the guidance provided by consultants to this study by Nicole Kruger (NMK Consulting), Wasima Rida (independent biostatistics consultant) and Mark Sullivan (Medicines Development for Global Health). We thank also our service providers Quintiles (study monitoring), Biophics Thailand (data management) and INC Research (Statistical analysis). This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1706804. Declaration Conflict of interests C.D.K, G.L.B. and R.F.B. and the MCRI hold a patent for the RV3-BB vaccine. C.D.K. (pre-2015) and J.E.B. (after 2015) have been/is the lead of the Australian Rotavirus Surveillance Program, which is supported by research grants from the vaccine companies Commonwealth Serum Laboratories and GlaxoSmithKline, as well as the Australian Commonwealth Department of Health and Aging. From 2015 C.D.K. has been an employee of the Bill and Melinda Gates Foundation who provided funds to support this trial. J.E.B. is chair of a clinical events committee for a trial in Mali conducted by the University of Maryland and supported by Merck, she receives no payment although MCRI is compensated for her time. J.P.B. chairs influenza vaccine data safety monitoring boards for Sequiris Pty Ltd, the Australian distributor of RV5, he receives no payment although Monash Health is compensated for his time. N.S.B. is an employee of Bio Farma PT who contributed funds to UGM to support the conduct of this trial and plan to manufacture the RV3-BB vaccine. All other authors have no interest to declare.

Evidence from epidemiological studies and challenge experiments in animal models have demonstrated that influenza virus infection can enhance the susceptibility to infection by bacteria, such as Streptococcus, Haemophilus influenzae and Staphylococcus aureus1. The relationship between influenza virus and atypical bacteria, namely Bordetella pertussis has, however not been well characterized2. In a randomized, double-blind, placebo-controlled trial conducted in South Africa in 2011 and 2012 we have shown that influenza vaccination of HIV-uninfected pregnant women was 50.4% efficacious in preventing polymerase-chain-reaction (PCR) confirmed influenza infection from the time of enrolment to 24-weeks post-partum3. Here we present the results from the retrospective testing by PCR for B. pertussis infection of the maternal pharyngeal specimens collected at the time of respiratory illnesses. The PCR protocol and results interpretation have been described4.

med influenza infection from the time of enrolment to 24-weeks post-partum3. Here we present the results from the retrospective testing by PCR for B. pertussis infection of the maternal pharyngeal specimens collected at the time of respiratory illnesses. The PCR protocol and results interpretation have been described4. Overall 2116 women were enrolled in the study including 1062 in the influenza vaccine- group and 1054 in the placebo-group. A total of 3583 respiratory specimens were collected from 1361 participants from enrolment to 24-weeks post-partum, and of these 3125 (87.2%) specimens were tested for B. pertussis. Eleven vaccine-recipients tested pertussis PCR- positive compared to 26 placebo-recipients (risk-ratio 0.4 [95%CI: 0.2, 0.8]). Further, there were 10 and 16 women in the vaccine and placebo groups, respectively, who were pertussis PCR-indeterminate on testing, Table. The overall risk-ratio, including the PCR-indeterminate episodes was 0.5 (95%CI: 0.2, 0.7). Thirty-three pertussis episodes among the women occurred post-delivery, with infants testing pertussis PCR-positive at the sometime as the mother on five occasions (three among the vaccine-group), within 22-days of maternal episode on two occasions (both in the placebo-group), and between 52 to 73 days after the maternal episode on three occasions (all in the placebo-group). No pertussis episodes were observed in the mothers a posteriori from the infants’.

other on five occasions (three among the vaccine-group), within 22-days of maternal episode on two occasions (both in the placebo-group), and between 52 to 73 days after the maternal episode on three occasions (all in the placebo-group). No pertussis episodes were observed in the mothers a posteriori from the infants’. Table Detection rate of Bordetella pertussis in women who participated in a randomized, double-blind, placebo-controlled trial of trivalent inactivated influenza vaccine 95%CI: 95% confidence interval; PCR: polymerase chain reaction. Archived DNA samples independently of clinical presentation were tested by real-time PCR for the presence of the multicopy pertussis insertion sequence (IS) IS4814; if IS481 cycle threshold values were <40, total nucleic acids were extracted from the corresponding achieved respiratory specimen using a NucliSENS easyMAG (bioMerieux) platform and re-tested for IS481 and in a duplex reaction for hIS1001 and pIS1001, and in a singleplex reaction for the pertussis toxin subunit S1 (ptxS1). Primers and probes used, and PCR results interpretation have been described4.

the corresponding achieved respiratory specimen using a NucliSENS easyMAG (bioMerieux) platform and re-tested for IS481 and in a duplex reaction for hIS1001 and pIS1001, and in a singleplex reaction for the pertussis toxin subunit S1 (ptxS1). Primers and probes used, and PCR results interpretation have been described4. Influenza vaccine N=1062a Placebo N=1054a Risk-Ratio (95%CI) P value Participants with at least one specimen collected, no (%)b 675 (63.6) 686 (65.1) 1.0 (0.9, 1.0) 0.46 Participants with at least one specimen tested by PCR for pertussis, no (%)b 635 (59.8) 652 (61.9) 1.0 (0.9, 1.0) 0.33 Respiratory specimens collected 1713 1870 - - Respiratory specimens tested by PCR for pertussis, no (%)c 1494 (87.2) 1631 (87.2) 1.0 (0.9, 1.0) 0.99 Pertussis PCR-positive cases, no (%)d 11 (1.0)f 26 (2.5)g 0.4 (0.2, 0.8) 0.012 Pertussis PCR-indeterminate cases, no (%)e 10 (0.9) 16 (1.5) 0.6 (0.3, 1.4) 0.23 Overall pertussis cases, no (%) 21 (2.0) 42 (4.0) 0.5 (0.2, 0.7) 0.007 a Participants were followed-up by weekly active surveillance for any respiratory illness from the time of enrolment through pregnancy to 24- weeks post-partum. Pharyngeal specimens were collected by study staff at the time of respiratory illness visits to the study clinic. b Percentage calculated from total participants enrolled. Respiratory specimens only collected in participants presenting with a respiratory illness. c Percentage calculated from total specimens collected. d PCR reaction with cycle threshold values for the IS481 gene <35 or ≤35-<40 plus positive for the ptxS1 gene.

Influenza vaccine N=1062a Placebo N=1054a Risk-Ratio (95%CI) P value Participants with at least one specimen collected, no (%)b 675 (63.6) 686 (65.1) 1.0 (0.9, 1.0) 0.46 Participants with at least one specimen tested by PCR for pertussis, no (%)b 635 (59.8) 652 (61.9) 1.0 (0.9, 1.0) 0.33 Respiratory specimens collected 1713 1870 - - Respiratory specimens tested by PCR for pertussis, no (%)c 1494 (87.2) 1631 (87.2) 1.0 (0.9, 1.0) 0.99 Pertussis PCR-positive cases, no (%)d 11 (1.0)f 26 (2.5)g 0.4 (0.2, 0.8) 0.012 Pertussis PCR-indeterminate cases, no (%)e 10 (0.9) 16 (1.5) 0.6 (0.3, 1.4) 0.23 Overall pertussis cases, no (%) 21 (2.0) 42 (4.0) 0.5 (0.2, 0.7) 0.007 a Participants were followed-up by weekly active surveillance for any respiratory illness from the time of enrolment through pregnancy to 24- weeks post-partum. Pharyngeal specimens were collected by study staff at the time of respiratory illness visits to the study clinic. b Percentage calculated from total participants enrolled. Respiratory specimens only collected in participants presenting with a respiratory illness. c Percentage calculated from total specimens collected. d PCR reaction with cycle threshold values for the IS481 gene <35 or ≤35-<40 plus positive for the ptxS1 gene. e PCR reaction with cycle threshold values for the IS481 gene ≤35-<40 and negative for the ptxS1 gene. f One specimen also tested positive for influenza A/H1N1. g Two specimens also tested positive for influenza A/H3N2. One participant tested pertussis PCR-positive in two different specimens collected 22 days apart, only the first episode was included.

tomatic (latent) in most individuals or progress to active disease3. Vaccine-mediated prevention of M.tb infection (POI) could be an important signal of efficacy against TB disease. Further, size and duration of a POI trial are less than for a trial of disease prevention, since M.tb infection occurs more frequently4-6. Acquisition, persistence and clearance of asymptomatic M.tb infection cannot be measured directly. Diagnosis of M.tb infection is based on immunological sensitization to M.tb antigens, assessed by tuberculin skin test (TST), which cross-reacts with other mycobacteria including Bacille Calmette-Guerin (BCG) vaccine. IFNγ release assays, including the QuantiFERON- TB Gold In-tube assay (QFT), are more specific for M.tb, but may yield false positive/negative results due to assay variability and uncertainty around the optimal assay cut-off7,8. Although neither TST nor QFT distinguishes between infection and disease, recent infection, diagnosed by TST or QFT conversion from negative to positive, is associated with increased disease risk, compared to non-conversion or remote conversion8-11. Human and animal studies suggest TST reversion is associated with early containment of M.tb infection and lower risk of TB disease4,12-14, perhaps indicating sterilization. QFT can also revert from positive to negative11. Although the clinical significance of QFT reversion remains to be established11, we propose that sustained QFT conversion is more likely associated with sustained M.tbb infection and progression to disease than transient QFT conversion. These observations suggest that vaccine-mediated prevention of initial or sustained M.tb infection could be critical steps towards TB control.

to be established11, we propose that sustained QFT conversion is more likely associated with sustained M.tbb infection and progression to disease than transient QFT conversion. These observations suggest that vaccine-mediated prevention of initial or sustained M.tb infection could be critical steps towards TB control. Observational studies indicate that primary BCG vaccination may offer partial protection against M.tb infection15-18. BCG revaccination may also protect against sustained M.tb infection, but this hypothesis has not previously been tested in a prospective, randomized, placebo-controlled trial19. Two large randomized trials showed no benefit of BCG revaccination for protection against TB disease20-22, but neither trial enrolled based on M.tb infection status or measured infection acquisition during follow-up. H4:IC31® is a candidate subunit vaccine, consisting of mycobacterial antigens Ag85B and TB10.4, which do not cross-react with QFT, together with the IC31® adjuvant (see Supplementary Appendix). H4:IC31® has shown protection in pre-clinical models23-25 and acceptable safety and immunogenicity in humans26,27.

follow-up. H4:IC31® is a candidate subunit vaccine, consisting of mycobacterial antigens Ag85B and TB10.4, which do not cross-react with QFT, together with the IC31® adjuvant (see Supplementary Appendix). H4:IC31® has shown protection in pre-clinical models23-25 and acceptable safety and immunogenicity in humans26,27. Demonstration of efficacy in a POI trial would provide strong impetus for larger trials to test H4:IC31® or BCG revaccination efficacy in preventing TB disease in M.tb-uninfected populations, allow identification of immune correlates of vaccine-mediated protection, and confirm the utility of the POI trial design to identify promising TB vaccine candidates. This clinical trial evaluated safety, immunogenicity, and prevention of initial and sustained QFT conversion by H4:IC31® or BCG revaccination in healthy South African adolescents in a high TB transmission setting11. Methods Trial design This phase II, randomized, three-arm, placebo-controlled, partially-blinded clinical trial aimed to enroll 990 healthy, HIV-uninfected, QFT-negative, 12- to 17-year-old adolescents, BCG- vaccinated in infancy, at two South African sites (Table 1). Adolescents with previously treated or current TB, a household TB contact, substance use, or pregnancy were excluded. Adolescents provided written informed assent and parents/legal guardians written informed consent. Regulatory approvals, consent procedures and inclusion/exclusion criteria are detailed in the Supplementary Appendix.

th previously treated or current TB, a household TB contact, substance use, or pregnancy were excluded. Adolescents provided written informed assent and parents/legal guardians written informed consent. Regulatory approvals, consent procedures and inclusion/exclusion criteria are detailed in the Supplementary Appendix. Table 1 Participant baseline characteristics (safety population) Variable Statistic Placebo (n=329) H4:IC31® (n=330) BCG (n=330) Total (n=989) Site SATVI n (%) 306 (93.0) 306 (92.7) 305 (92.4) 917 (92.7) Emavundleni n (%) 23 (7.0) 24 (7.3) 25 (7.6) 72 (7.3) Age (years)1 Median (min, max) 14 (12, 17) 14 (12, 17) 14 (12, 17) 14 (12, 17) Self-declared Race1 Asian n (%) 1 (0.3) 1 (0.3) 1 (0.3) 3 (0.3) Black African n (%) 120 (36.5) 120 (36.4) 126 (38.2) 366 (37.0) Caucasian n (%) 1 (0.3) 1 (0.3) 3 (0.9) 5 (0.5) Cape Mixed Ancestry n (%) 207 (62.9) 208 (63.0) 200 (60.6) 615 (62.2) Sex (females)1 n (%) 169 (51.4) 189 (57.3) 162 (49.1) 520 (52.6) Body mass index (kg/m2)1 Median (min, max) 19.9 (14.3, 36.8) 19.6 (13.8, 38.3) 19.4 (13.1, 36.9) 19.6 (13.1, 38.3) 1 Numbers are presented for both sites combined Eligible participants were enrolled into two sequential cohorts, each randomized 1:1:1 to receive intramuscular saline placebo or H4:IC31® (15μg H4 polyprotein, Sanofi Pasteur, and 500nmol IC31®, Statens Serum Institut) on Day 0 (D0) and D56, or intradermal BCG (2- 8x105 CFU, Statens Serum Institut) at D0.

Table 1 Participant baseline characteristics (safety population) Variable Statistic Placebo (n=329) H4:IC31® (n=330) BCG (n=330) Total (n=989) Site SATVI n (%) 306 (93.0) 306 (92.7) 305 (92.4) 917 (92.7) Emavundleni n (%) 23 (7.0) 24 (7.3) 25 (7.6) 72 (7.3) Age (years)1 Median (min, max) 14 (12, 17) 14 (12, 17) 14 (12, 17) 14 (12, 17) Self-declared Race1 Asian n (%) 1 (0.3) 1 (0.3) 1 (0.3) 3 (0.3) Black African n (%) 120 (36.5) 120 (36.4) 126 (38.2) 366 (37.0) Caucasian n (%) 1 (0.3) 1 (0.3) 3 (0.9) 5 (0.5) Cape Mixed Ancestry n (%) 207 (62.9) 208 (63.0) 200 (60.6) 615 (62.2) Sex (females)1 n (%) 169 (51.4) 189 (57.3) 162 (49.1) 520 (52.6) Body mass index (kg/m2)1 Median (min, max) 19.9 (14.3, 36.8) 19.6 (13.8, 38.3) 19.4 (13.1, 36.9) 19.6 (13.1, 38.3) 1 Numbers are presented for both sites combined Eligible participants were enrolled into two sequential cohorts, each randomized 1:1:1 to receive intramuscular saline placebo or H4:IC31® (15μg H4 polyprotein, Sanofi Pasteur, and 500nmol IC31®, Statens Serum Institut) on Day 0 (D0) and D56, or intradermal BCG (2- 8x105 CFU, Statens Serum Institut) at D0. In the first cohort of 90 participants, approximately 30 per arm, additional safety tests and immunogenicity assays were performed (Supplementary Appendix). The follow-up schedule of individual participants was contingent on QFT results at D84 and Months 6, 12, 18 and 24 (Figure 1A). An 84-day ‘wash-out’ period was stipulated to exclude participants who may have been M.tb-infected at baseline, but not yet QFT-positive. Participants who tested QFT- positive at D84 were followed for 6 months after last vaccination for safety but excluded from efficacy evaluations (Figure 1B). An Independent Data Monitoring Committee (IDMC) reviewed D7 and D84 safety data from the first cohort and safety and efficacy data for all participants throughout follow-up (Supplementary Appendix).

e at D84 were followed for 6 months after last vaccination for safety but excluded from efficacy evaluations (Figure 1B). An Independent Data Monitoring Committee (IDMC) reviewed D7 and D84 safety data from the first cohort and safety and efficacy data for all participants throughout follow-up (Supplementary Appendix). Figure 1: Study design and CONSORT diagram (A) Study design. Each participant followed a schedule of evaluations according to study arm and QFT test results. An 84-day wash-out period was implemented to account for participants who may already have been M.tb-infected at enrollment but had not yet QFT converted. After the primary analysis, the IDMC recommended that participants who converted at Month 6 or 12 should return for an additional end-of-study visit to evaluate sustained QFT conversion. Safety outcomes were assessed at each study visit, including evaluation of symptoms of TB disease. QFT, QuantiFERON-TB Gold In-tube; EoS, end of study. CONSORT diagram. Amongst screened individuals (n=2976), 1986 were excluded for one or more reasons. The most common reason for ineligibility was a positive QFT test (n=1405, 71%); other common reasons for exclusion were: abnormal blood results (n=244, 12%), body mass index out of range (n=122, 6%), previous TB or household TB contact (n=55, 3%). ITT, intent-to-treat; mITT, modified ITT; PP, per protocol. South African guidelines do not recommend preventive antimicrobials for HIV-negative, M.tb-infected persons >5 years old; therefore therapy was not provided to converters28.

CONSORT diagram. Amongst screened individuals (n=2976), 1986 were excluded for one or more reasons. The most common reason for ineligibility was a positive QFT test (n=1405, 71%); other common reasons for exclusion were: abnormal blood results (n=244, 12%), body mass index out of range (n=122, 6%), previous TB or household TB contact (n=55, 3%). ITT, intent-to-treat; mITT, modified ITT; PP, per protocol. South African guidelines do not recommend preventive antimicrobials for HIV-negative, M.tb-infected persons >5 years old; therefore therapy was not provided to converters28. Safety Outcomes Solicited adverse events (AEs) were recorded for 7 days, unsolicited AEs for 28 days and injection site AEs for 28 days after placebo and H4:IC31®, or 84 days after BCG. Serious adverse events (SAEs) and adverse events of special interest (AESIs) were recorded for the entire study period (Supplementary Appendix). Immunogenicity Outcomes Immunogenicity was evaluated by intracellular cytokine staining (ICS)29 and flow cytometry (Supplementary Appendix; Table S1). Efficacy Outcomes We prioritized efficacy assessment using the modified intent-to-treat (mITT) population, defined as those who received at least one injection and had not converted to QFT-positive at D84. The primary efficacy endpoint was initial QFT conversion using the threshold of IFNγ ≥0.35 International Units (IU)/mL at any time after D84 and was compared for the H4:IC31® or BCG revaccination arms versus placebo.

Efficacy Outcomes We prioritized efficacy assessment using the modified intent-to-treat (mITT) population, defined as those who received at least one injection and had not converted to QFT-positive at D84. The primary efficacy endpoint was initial QFT conversion using the threshold of IFNγ ≥0.35 International Units (IU)/mL at any time after D84 and was compared for the H4:IC31® or BCG revaccination arms versus placebo. The QFT assay was conducted according to the manufacturer’s instructions, with additional, more stringent parameters to reduce variability and improve reproducibility8 (Supplementary Appendix). The secondary efficacy endpoint was sustained QFT conversion without reversion through 6 months after initial QFT conversion, i.e., three consecutive positive QFT results after D84 (Figure 1A). In this study, QFT conversion and sustained QFT conversion were considered surrogate endpoints for M.tb infection and sustained M.tb infection, respectively. Exploratory efficacy endpoints included evaluation of sustained conversion through end of study (EoS) and alternative QFT threshold values for initial and sustained conversion, including IFNγ <0.2IU/mL to >0.7IU/mL8, and IFNγ <0.35IU/mL to >4IU/mL30, as detailed in Randomization and blinding Group allocation was concealed by an interactive web response system. Assignment was based on block randomization to placebo, H4:IC31® or BCG (1:1:1), stratified by school (Worcester site) or residential area (Emavundleni site).

Exploratory efficacy endpoints included evaluation of sustained conversion through end of study (EoS) and alternative QFT threshold values for initial and sustained conversion, including IFNγ <0.2IU/mL to >0.7IU/mL8, and IFNγ <0.35IU/mL to >4IU/mL30, as detailed in Randomization and blinding Group allocation was concealed by an interactive web response system. Assignment was based on block randomization to placebo, H4:IC31® or BCG (1:1:1), stratified by school (Worcester site) or residential area (Emavundleni site). Blinding was partial because BCG causes a recognizable injection site reaction and is administered once. However, randomization to H4:IC31® and placebo was double-blinded: syringe contents were masked, injection volumes were identical, and injections were administered by a research nurse who did not perform post-enrollment study procedures or data collection. Laboratory personnel were blinded to all three treatment groups. Primary and secondary efficacy endpoints were analyzed using two log-rank statistics (H4:IC31® or BCG versus placebo, α=0.1, one-sided) without adjustment for multiplicity. Efficacy estimates were based on hazard ratios from a Cox regression model. Analyses and endpoints are detailed in the Supplementary Appendix. All other analyses were two-sided (α=0.05).

s were analyzed using two log-rank statistics (H4:IC31® or BCG versus placebo, α=0.1, one-sided) without adjustment for multiplicity. Efficacy estimates were based on hazard ratios from a Cox regression model. Analyses and endpoints are detailed in the Supplementary Appendix. All other analyses were two-sided (α=0.05). Results Baseline characteristics Between 1 April 2014 and 25 May 2015, 2,976 participants were screened and 990 were enrolled. Most (1405/1986, 71%) exclusions were due to positive QFT (Figure 1). Baseline characteristics did not differ among arms (Table 1). The final visit occurred on 28 August 2017). Loss to follow-up was 4% (41/990) through EoS (Figure 1). Safety Safety was assessed in all participants who received at least one injection. Each vaccine had an acceptable safety profile (Table S2); 550 participants experienced at least one AE. H4:IC31® and placebo had similar AE profiles. AEs were more frequent in the BCG arm; 98.8% experienced at least one event. These were predominantly local injection site AEs of mild-to-moderate severity, consistent with BCG’s known reactogenicity profile31. Upper respiratory tract infections occurred less frequently in the BCG arm compared to placebo and H4:IC31 arms (2.1%, 7.9%, and 9.4%, respectively; p<0.001). In total, 4 severe AEs, 19 SAEs, and no AESIs or severe related AEs were observed. There was no clinically significant difference in the rate of severe AEs or SAEs between study arms. One participant in the placebo arm died from suicide.

ompared to placebo and H4:IC31 arms (2.1%, 7.9%, and 9.4%, respectively; p<0.001). In total, 4 severe AEs, 19 SAEs, and no AESIs or severe related AEs were observed. There was no clinically significant difference in the rate of severe AEs or SAEs between study arms. One participant in the placebo arm died from suicide. Immunogenicity Frequencies of cytokine-expressing antigen-specific T cells were assessed at baseline and D70 by ICS (Figure 2). Ag85B- and TB10.4-specific CD4 T cell responses were low before vaccination and H4:IC31® induced significant increases in these responses. By contrast, high levels of pre-vaccination BCG-specific CD4 T cell responses were observed in all arms. BCG revaccination boosted the BCG-specific CD4 T cell responses significantly (Figure 2 and Figure S1). Figure 2: Immunogenicity Vaccine immunogenicity measured by PBMC intracellular cytokine staining (ICS) and flow cytometry following stimulation with Ag85B or TB10.4 peptide pools (summed response is shown) or BCG. Paired responses of CD4 T cells expressing IFNγ and/or IL2 for each individual (between 23 and 28 participants were included at each time point) at D0 (circles) and D70 (diamonds) randomized to placebo (blue), H4:IC31® (red) or BCG (green). Changes in response between D0 and D70 were compared by Wilcoxon Signed-Rank Test.

Paired responses of CD4 T cells expressing IFNγ and/or IL2 for each individual (between 23 and 28 participants were included at each time point) at D0 (circles) and D70 (diamonds) randomized to placebo (blue), H4:IC31® (red) or BCG (green). Changes in response between D0 and D70 were compared by Wilcoxon Signed-Rank Test. Efficacy Sixty (6.1%) participants were excluded from the mITT population (Figure 1). There were 134 initial QFT conversions (14.4% or 9.9/100 person-years; Figure S2A) in the mITT population, with a high QFT reversion rate (50 of 133 with repeated QFT, 37.6%). There were 82 sustained converters (8.8% of all participants; 62.6% of initial converters with non- missing QFT results; Figure 3A). Median time to initial QFT conversion among converters was 15.0 months. No TB disease cases were identified.

pulation, with a high QFT reversion rate (50 of 133 with repeated QFT, 37.6%). There were 82 sustained converters (8.8% of all participants; 62.6% of initial converters with non- missing QFT results; Figure 3A). Median time to initial QFT conversion among converters was 15.0 months. No TB disease cases were identified. Figure 3: Vaccine efficacy (A) Longitudinal quantitative IFNγ values measured by QFT by study arm, aligned to initial QFT conversion time point (month 0). Each line represents one individual; those who never converted and those with missing QFT results after initial conversion are not shown. Solid lines denote participants who met the secondary efficacy endpoint (sustained QFT conversion, top row) and dashed lines denote participants with initial QFT conversion who then reverted (bottom row). The solid horizontal line denotes the manufacturer’s recommended threshold (0.35IU/mL); the shaded horizontal area denotes the uncertainty zone (0.2-0.7IU/mL); the horizontal line at 4.0IU/mL denotes an alternative QFT threshold applied in exploratory analyses. Values <0.01IU/mL were set to 0.01 to enable plotting on the log scale. (B) Kaplan-Meier curves representing time to initial QFT conversion (primary efficacy endpoint) after first vaccination by study arm in the mITT population. Statistics are reported in Table 2.

Figure 3: Vaccine efficacy (A) Longitudinal quantitative IFNγ values measured by QFT by study arm, aligned to initial QFT conversion time point (month 0). Each line represents one individual; those who never converted and those with missing QFT results after initial conversion are not shown. Solid lines denote participants who met the secondary efficacy endpoint (sustained QFT conversion, top row) and dashed lines denote participants with initial QFT conversion who then reverted (bottom row). The solid horizontal line denotes the manufacturer’s recommended threshold (0.35IU/mL); the shaded horizontal area denotes the uncertainty zone (0.2-0.7IU/mL); the horizontal line at 4.0IU/mL denotes an alternative QFT threshold applied in exploratory analyses. Values <0.01IU/mL were set to 0.01 to enable plotting on the log scale. (B) Kaplan-Meier curves representing time to initial QFT conversion (primary efficacy endpoint) after first vaccination by study arm in the mITT population. Statistics are reported in Table 2. (C) Kaplan-Meier curves representing time after first vaccination to initial QFT conversion in participants with sustained conversion (secondary efficacy endpoint) by study arm in the mITT population. Inset depicts time to QFT reversion within 6 months of initial conversion in participants with QFT values at three and six months post-conversion. Statistics for conversion endpoints are reported in Table 2.

ion in participants with sustained conversion (secondary efficacy endpoint) by study arm in the mITT population. Inset depicts time to QFT reversion within 6 months of initial conversion in participants with QFT values at three and six months post-conversion. Statistics for conversion endpoints are reported in Table 2. Neither H4:IC31® vaccination nor BCG revaccination met the primary efficacy criterion, based on initial QFT conversion rates (Table 2; Figure 3B). However, H4:IC31® efficacy point estimate for prevention of sustained QFT conversion, the secondary endpoint, was 30.5% (one-sided p=0.08; 95% CI: -15.8, 58.3%; Table 2; Figure 3C), with 17/43 (39.5%) reversions among converters with non-missing QFT data. H4:IC31® efficacy for prevention of sustained QFT conversion at EoS was 34.2% (one-sided p=0.05; 95% CI: -10.4, 60.7%; Table 2).

ained QFT conversion, the secondary endpoint, was 30.5% (one-sided p=0.08; 95% CI: -15.8, 58.3%; Table 2; Figure 3C), with 17/43 (39.5%) reversions among converters with non-missing QFT data. H4:IC31® efficacy for prevention of sustained QFT conversion at EoS was 34.2% (one-sided p=0.05; 95% CI: -10.4, 60.7%; Table 2). Table 2 Vaccine efficacy Arms Placebo H4:IC31® BCG Endpoint QFT conversion threshold n/N (%) n/N (%) Vaccine efficacy n/N (%) Vaccine efficacy Point Est (%) 80% CI 95% CI p-val Point Est (%) 80% CI 95% CI p-val Primary endpoint QFT conversion1 ≥ 0.35IU/mL 49/310 (15.8) 44/308 (14.3) 9.42 -18.3, 30.6 -36.2, 39.7 0.323 41/312 (13.1) 20.12 -4.8, 39.1 -21.0, 47.2 0.143 Secondary endpoint Sustained QFT conversion4 ≥ 0.35IU/mL 36/310 (11.6) 25/308 (8.1) 30.52 3.0, 50.2 -15.8, 58.3 0.083 21/312 (6.7) 45.42 22.3, 61.6 6.4, 68.1 0.013 Exploratory endpoint Sustained QFT conversion5 <0.2 to >0.7IU/mL 31/310 (10.0) 24/308 (7.8) 23.22 -8.8, 45.8 -30.9, 54.9 0.163 19/312 (6.1) 41.62 15.2, 59.8 -3.3, 67.0 0.033 End-of Study sustained QFT conversion6 ≥ 0.35IU/mL 36/310 (11.6) 24/308 (7.8) 34.22 7.7, 53.0 -10.4, 60.7 0.053 20/312 (6.4) 48.22 25.9, 63.8 10.5, 70.0 0.0083 QFT conversion7 > 4IU/mL 33/310 (10.6) 22/308 (7.1) 34.58 6.8, 54.2 -12.1, 62.3 0.139 19/312 (6.1) 45.18 20.5, 62.2 3.8, 69.3 0.049 1 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive ( 2 ≥ 0.35 IU/mL) at any time point through end of study. 2 Vaccine efficacy point estimates and 80% CI and 95% CI are based on the hazard ratio estimated from the Cox regression model (two-sided).

Table 2 Vaccine efficacy Arms Placebo H4:IC31® BCG Endpoint QFT conversion threshold n/N (%) n/N (%) Vaccine efficacy n/N (%) Vaccine efficacy Point Est (%) 80% CI 95% CI p-val Point Est (%) 80% CI 95% CI p-val Primary endpoint QFT conversion1 ≥ 0.35IU/mL 49/310 (15.8) 44/308 (14.3) 9.42 -18.3, 30.6 -36.2, 39.7 0.323 41/312 (13.1) 20.12 -4.8, 39.1 -21.0, 47.2 0.143 Secondary endpoint Sustained QFT conversion4 ≥ 0.35IU/mL 36/310 (11.6) 25/308 (8.1) 30.52 3.0, 50.2 -15.8, 58.3 0.083 21/312 (6.7) 45.42 22.3, 61.6 6.4, 68.1 0.013 Exploratory endpoint Sustained QFT conversion5 <0.2 to >0.7IU/mL 31/310 (10.0) 24/308 (7.8) 23.22 -8.8, 45.8 -30.9, 54.9 0.163 19/312 (6.1) 41.62 15.2, 59.8 -3.3, 67.0 0.033 End-of Study sustained QFT conversion6 ≥ 0.35IU/mL 36/310 (11.6) 24/308 (7.8) 34.22 7.7, 53.0 -10.4, 60.7 0.053 20/312 (6.4) 48.22 25.9, 63.8 10.5, 70.0 0.0083 QFT conversion7 > 4IU/mL 33/310 (10.6) 22/308 (7.1) 34.58 6.8, 54.2 -12.1, 62.3 0.139 19/312 (6.1) 45.18 20.5, 62.2 3.8, 69.3 0.049 1 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive ( 2 ≥ 0.35 IU/mL) at any time point through end of study. 2 Vaccine efficacy point estimates and 80% CI and 95% CI are based on the hazard ratio estimated from the Cox regression model (two-sided). 3 P-values are based on a one-sided log-rank test compared to placebo. No multiplicity adjustment done for p-values.

Table 2 Vaccine efficacy Arms Placebo H4:IC31® BCG Endpoint QFT conversion threshold n/N (%) n/N (%) Vaccine efficacy n/N (%) Vaccine efficacy Point Est (%) 80% CI 95% CI p-val Point Est (%) 80% CI 95% CI p-val Primary endpoint QFT conversion1 ≥ 0.35IU/mL 49/310 (15.8) 44/308 (14.3) 9.42 -18.3, 30.6 -36.2, 39.7 0.323 41/312 (13.1) 20.12 -4.8, 39.1 -21.0, 47.2 0.143 Secondary endpoint Sustained QFT conversion4 ≥ 0.35IU/mL 36/310 (11.6) 25/308 (8.1) 30.52 3.0, 50.2 -15.8, 58.3 0.083 21/312 (6.7) 45.42 22.3, 61.6 6.4, 68.1 0.013 Exploratory endpoint Sustained QFT conversion5 <0.2 to >0.7IU/mL 31/310 (10.0) 24/308 (7.8) 23.22 -8.8, 45.8 -30.9, 54.9 0.163 19/312 (6.1) 41.62 15.2, 59.8 -3.3, 67.0 0.033 End-of Study sustained QFT conversion6 ≥ 0.35IU/mL 36/310 (11.6) 24/308 (7.8) 34.22 7.7, 53.0 -10.4, 60.7 0.053 20/312 (6.4) 48.22 25.9, 63.8 10.5, 70.0 0.0083 QFT conversion7 > 4IU/mL 33/310 (10.6) 22/308 (7.1) 34.58 6.8, 54.2 -12.1, 62.3 0.139 19/312 (6.1) 45.18 20.5, 62.2 3.8, 69.3 0.049 1 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive ( 2 ≥ 0.35 IU/mL) at any time point through end of study. 2 Vaccine efficacy point estimates and 80% CI and 95% CI are based on the hazard ratio estimated from the Cox regression model (two-sided). 3 P-values are based on a one-sided log-rank test compared to placebo. No multiplicity adjustment done for p-values. 4 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive (≥ 0.35 IU/mL) at any time point through end of study, and without a change in QFT from positive to negative through 6 months after QFT conversion (excluding end-of-study call-back visit for participants who converted at Month 6 or 12).

3 P-values are based on a one-sided log-rank test compared to placebo. No multiplicity adjustment done for p-values. 4 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive (≥ 0.35 IU/mL) at any time point through end of study, and without a change in QFT from positive to negative through 6 months after QFT conversion (excluding end-of-study call-back visit for participants who converted at Month 6 or 12). 5 QFT conversion from negative at Day 84 to positive at any time point through end of study, using an alternative threshold of < 0.2 IU/mL at any time point prior to conversion and > 0.7 IU/mL at conversion and maintained through 6 months after initial conversion (excluding end-of-study call-back visitfor participants who converted at Month 6 or 12). 6 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive (≥ 0.35 IU/mL) at any time point through end of study, and without a change inQFT from positive to negative through 6 months after QFT conversion as well as the end-of-study call-back visit for participants who converted at Month 6 or 12. 7 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive (> 4.0 IU/mL) at any time point through end of study. 8 Vaccine efficacy point estimates and 95% CI are calculated based on the conditional binomial procedure (Clopper-Pearson method with mid-p correction). 9 Two-sided p-values are based on Pearson Chi-square test. Abbreviations: Point est = point estimate; CI = confidence interval; p-val = p-value

7 QFT conversion from negative (< 0.35 IU/mL) at Day 84 to positive (> 4.0 IU/mL) at any time point through end of study. 8 Vaccine efficacy point estimates and 95% CI are calculated based on the conditional binomial procedure (Clopper-Pearson method with mid-p correction). 9 Two-sided p-values are based on Pearson Chi-square test. Abbreviations: Point est = point estimate; CI = confidence interval; p-val = p-value BCG revaccination efficacy for prevention of sustained QFT conversion was 45.4% (one- sided p=0.01; 95% CI: 6.4, 68.1%; Table 2; Figure 3C); 48.2% efficacy was observed at EoS (one-sided p=0.008; 95% CI: 10.5, 70.0%; Table 2). This BCG-induced effect was explained by a near-two-fold higher 6-month QFT reversion rate after conversion, compared to placebo recipients (19/41, 46.3% vs 12/49, 24.5%). 88% of all reversions occurred by 3 months post- conversion (Figure 3C inset). In exploratory analyses, efficacy of BCG revaccination for sustained QFT conversion was 41.6% using a stringent QFT conversion threshold (<0.2IU/mL to >0.7IU/mL) (one-sided p=0.03; 95% CI: -3.3, 67.0%); no significant effect of H4:IC31 was noted. BCG efficacy for initial conversion using the most stringent QFT threshold, >4.0IU/mL, was 45.1% (two-sided p=0.04; 95% CI: 3.8, 69.3%; Table 2; Figure S2B); no significant effect of H4:IC31 was noted at the 95% confidence level, although it did show significance at the less stringent confidence level (80% CI: 6.8, 54.2%). Additional exploratory analyses are reported in Table S3.

FT threshold, >4.0IU/mL, was 45.1% (two-sided p=0.04; 95% CI: 3.8, 69.3%; Table 2; Figure S2B); no significant effect of H4:IC31 was noted at the 95% confidence level, although it did show significance at the less stringent confidence level (80% CI: 6.8, 54.2%). Additional exploratory analyses are reported in Table S3. Estimates of efficacy based on primary and secondary endpoints were not affected by sex, race or study site in post-hoc analysis (data not shown). Discussion We performed the first randomized controlled prevention of M.tb infection trial and showed that vaccination can reduce the rate of sustained M.tb infection in a high-transmission setting. Neither H4:IC31® nor BCG revaccination prevented initial QFT conversion. H4:IC31® showed 30.5% efficacy against sustained QFT conversion, which met the pre-defined significance threshold (one-sided p<0.1) as the first proof-of-concept efficacy signal ever observed for a subunit TB vaccine candidate. Although this modest effect did not meet stringent statistical criteria for demonstration of efficacy typically used in a licensure trial (95% CI), it provides an indication that subunit vaccines comprising few M.tb antigens may have biological effect and supports clinical evaluation of next-generation subunit vaccine candidates.

is modest effect did not meet stringent statistical criteria for demonstration of efficacy typically used in a licensure trial (95% CI), it provides an indication that subunit vaccines comprising few M.tb antigens may have biological effect and supports clinical evaluation of next-generation subunit vaccine candidates. BCG revaccination demonstrated 45.4% efficacy against sustained QFT conversion and met the more stringent statistical criterion. The durability of this important finding and potential public health significance for protection against TB disease warrants modeling and further clinical evaluation. We showed that vaccine-mediated protection against sustained QFT conversion may inform clinical development of vaccine candidates before entry into large- scale prevention of disease efficacy trials. Our findings, and availability of stored biospecimens, also provide an opportunity to discover immune responses that correlate with protection against infection, which would enable new TB vaccine design and evaluation. The efficacy signal for BCG revaccination against sustained QFT conversion was also observed using more stringent QFT thresholds8. Importantly, BCG showed protection against initial conversion at IFNγ >4.0IU/mL, which was associated with increased risk of TB disease in infants30, consistent with predictions from animal models32.

cy signal for BCG revaccination against sustained QFT conversion was also observed using more stringent QFT thresholds8. Importantly, BCG showed protection against initial conversion at IFNγ >4.0IU/mL, which was associated with increased risk of TB disease in infants30, consistent with predictions from animal models32. A meta-analysis of observational studies of primary BCG vaccination reported a pooled estimate of 27% efficacy against initial M.tb infection and 71% efficacy against TB disease18. Primary BCG vaccine efficacy against disease is highly variable in different populations, is greatest in mycobacteria-naïve individuals33and may last for 10 years33,34. Our findings suggest BCG revaccination of QFT-negative adolescents may provide additional benefit19. Two large cluster-randomized trials evaluated prevention of disease by BCG revaccination and did not demonstrate efficacy21,22. Neither trial enrolled based on M.tb or HIV infection status, or tested for prior mycobacterial sensitization or acquisition of M.tb infection. In Brazilian children aged 7-14 years, efficacy of BCG revaccination against TB disease was 9% after 5 years21 and 12% after 9 years, both estimates not statistically significant20. The trial was cluster-randomized, open-label, with no placebo, and the TB disease endpoint was determined from health service records21. However, a modest statistically significant efficacy signal (33%) was observed in children revaccinated at <11 years of age at one of two sites20. The second trial, a double-blind, randomized placebo-controlled trial of BCG revaccination among more than 46,000 people aged 3 months – 70 years showed no significant efficacy against confirmed TB disease (incidence rate ratio 1.43)22, in a Malawian community in which a trial of primary BCG vaccination had also shown no efficacy35. Based on our results and given the substantial differences in trial methodology, TB epidemiology and study populations, a trial of BCG revaccination for prevention of disease in M.tb-uninfected adolescents is justified in high TB incidence settings. Such a trial would also validate the POI strategy to de-risk TB vaccine development and allow identification of immune correlates of protection against disease. From a public health perspective, the potential risk of BCG disease in adolescents at high risk for HIV infection should be balanced against the potential benefits.

a trial would also validate the POI strategy to de-risk TB vaccine development and allow identification of immune correlates of protection against disease. From a public health perspective, the potential risk of BCG disease in adolescents at high risk for HIV infection should be balanced against the potential benefits. A successful TB vaccine might function by several mechanisms, including prevention of initial M.tb infection, sustained infection, or progression to disease. Our results indicate that vaccination did not avert initial colonization, likely mediated by innate immunity, but allowed antigen trafficking to lymphoid tissues to trigger adaptive immunity (measured by initial QFT conversion). Rather, we hypothesize that vaccine-mediated QFT reversion is associated with enhanced bacterial control or clearance, likely mediated by collaborative adaptive and innate immune responses, which have been associated with sterilization of individual granulomas in non-human primates36,37. Although antigen-specific memory T cells measured by QFT can persist after bacterial clearance22, there is a positive correlation between M.tb replication in animal models and the magnitude of IFNγ responses to M.tb-specific antigens25. Indeed, transient TST conversion has been shown in humans and guinea pigs to be associated with reduced risk of TB disease compared with sustained conversion12-14. Further studies are required to understand clinical significance of QFT reversion and underlying immunological determinants. Comprehensive analyses are required to elucidate immune responses and mechanisms that correlate with protection, to guide new TB vaccine evaluation and design.

red with sustained conversion12-14. Further studies are required to understand clinical significance of QFT reversion and underlying immunological determinants. Comprehensive analyses are required to elucidate immune responses and mechanisms that correlate with protection, to guide new TB vaccine evaluation and design. Definitive interpretation of our findings is limited because there is no gold standard test for acquisition, persistence or clearance of M.tb infection. QFT has technical limitations, which we addressed by implementing optimized assay procedures8, utilizing alternative threshold definitions and serial testing. Testing only for initial QFT conversion in this trial would not have demonstrated efficacy; thus future POI trials should evaluate prevention of sustained conversion to avoid rejection of a potentially efficacious vaccine candidate. The POI trial design has potential to miss an impactful vaccine that prevents TB disease but not M.tb infection18. Conversely, a vaccine that prevented sustained infection mainly in the ~90% of M.tb-infected individuals who naturally never develop disease would have little impact on TB prevention4,5. These findings confirm model predictions that vaccine efficacy against M.tb infection can be observed in a very high transmission setting4. It is unclear if our observations are generalizable to settings with lower M.tb transmission rates21,22.

Definitive interpretation of our findings is limited because there is no gold standard test for acquisition, persistence or clearance of M.tb infection. QFT has technical limitations, which we addressed by implementing optimized assay procedures8, utilizing alternative threshold definitions and serial testing. Testing only for initial QFT conversion in this trial would not have demonstrated efficacy; thus future POI trials should evaluate prevention of sustained conversion to avoid rejection of a potentially efficacious vaccine candidate. The POI trial design has potential to miss an impactful vaccine that prevents TB disease but not M.tb infection18. Conversely, a vaccine that prevented sustained infection mainly in the ~90% of M.tb-infected individuals who naturally never develop disease would have little impact on TB prevention4,5. These findings confirm model predictions that vaccine efficacy against M.tb infection can be observed in a very high transmission setting4. It is unclear if our observations are generalizable to settings with lower M.tb transmission rates21,22. Our results raise important questions around the significance of prevention of M.tb infection for control of TB disease and provide a promising signal for BCG, other live mycobacterial and adjuvanted subunit vaccines. These encouraging findings provide impetus to re- evaluate BCG revaccination of M.tb-uninfected populations for prevention of disease19 and accelerate new TB vaccine development and illustrate the value of conducting human trials of TB vaccine candidates.

r BCG, other live mycobacterial and adjuvanted subunit vaccines. These encouraging findings provide impetus to re- evaluate BCG revaccination of M.tb-uninfected populations for prevention of disease19 and accelerate new TB vaccine development and illustrate the value of conducting human trials of TB vaccine candidates. Supplementary Material Supplementary Appendix Acknowledgements We thank the study participants and their families for taking part in this trial and the Worcester and Emavundleni Research Site staff for conduct of the clinical activities. We also thank Jacqueline Shea, Danilo Casimiro, Chris Karp, Chris Wilson and Jim Tartaglia for valuable discussions. Hassan Mahomed, Peter Donald, Wasima Rida, Gil Price, Matthew Downs, James Balsley, Bernard Fourie were members of Independent Data Monitoring Committee (IDMC); Anthony Hawkridge and Zainab Waggie were the Local Medical Monitors. We also thank the Department of Education, Western Cape Government, South Africa.

Supplementary Material Supplementary Appendix Acknowledgements We thank the study participants and their families for taking part in this trial and the Worcester and Emavundleni Research Site staff for conduct of the clinical activities. We also thank Jacqueline Shea, Danilo Casimiro, Chris Karp, Chris Wilson and Jim Tartaglia for valuable discussions. Hassan Mahomed, Peter Donald, Wasima Rida, Gil Price, Matthew Downs, James Balsley, Bernard Fourie were members of Independent Data Monitoring Committee (IDMC); Anthony Hawkridge and Zainab Waggie were the Local Medical Monitors. We also thank the Department of Education, Western Cape Government, South Africa. Other information WAH, SGS, RR, SG, CADG, PA, IK, TE, RDE, BL, AMG, TJS and MH designed the study. Data were gathered by EN, HG, VR, FR, NB, SM, LM, ME, AT, HM, LGB, DAH, TJS, MH and the C-040-404 Study Team. Data management and statistical analyses were performed by a contract research organization (IQVIA) and KTR. All authors vouch for the accuracy and completeness of the data, and for the fidelity of the study to the protocol, which is available with the full text of this article at NEJM.org. The first draft of the manuscript was written by EN, VR, KTR, RH, AMG, TJS and MH. All authors participated in the writing of subsequent drafts. The decision to publish the paper was made jointly by the sponsor and investigators. All authors signed a confidentiality agreement with the sponsor. Registration: ClinicalTrials.gov Identifier: NCT02075203

Other information WAH, SGS, RR, SG, CADG, PA, IK, TE, RDE, BL, AMG, TJS and MH designed the study. Data were gathered by EN, HG, VR, FR, NB, SM, LM, ME, AT, HM, LGB, DAH, TJS, MH and the C-040-404 Study Team. Data management and statistical analyses were performed by a contract research organization (IQVIA) and KTR. All authors vouch for the accuracy and completeness of the data, and for the fidelity of the study to the protocol, which is available with the full text of this article at NEJM.org. The first draft of the manuscript was written by EN, VR, KTR, RH, AMG, TJS and MH. All authors participated in the writing of subsequent drafts. The decision to publish the paper was made jointly by the sponsor and investigators. All authors signed a confidentiality agreement with the sponsor. Registration: ClinicalTrials.gov Identifier: NCT02075203 Protocol: The full protocol and Statistical Analysis Plan are included as Supplementary Appendix and can also be accessed at ClinicalTrials.gov Funding: The study was co-funded by Aeras and Sanofi Pasteur. Aeras was the trial sponsor and contributed to the study design and data analysis. H4:IC31® was supplied by Sanofi- Pasteur (Toronto, Canada; H4 antigen) and Statens Serum Institut (Copenhagen, Denmark; IC31® adjuvant). BCG Vaccine SSI was sourced by the clinical trial sites. EN is an ISAC Marylou Ingram Scholar, VR was supported by the Swiss National Foundation. Conflict of interest TJS and MH report grants from Aeras. RR was a Sanofi Pasteur employee. SG and CADG are employees and share-holders at Sanofi Pasteur.

Funding: The study was co-funded by Aeras and Sanofi Pasteur. Aeras was the trial sponsor and contributed to the study design and data analysis. H4:IC31® was supplied by Sanofi- Pasteur (Toronto, Canada; H4 antigen) and Statens Serum Institut (Copenhagen, Denmark; IC31® adjuvant). BCG Vaccine SSI was sourced by the clinical trial sites. EN is an ISAC Marylou Ingram Scholar, VR was supported by the Swiss National Foundation. Conflict of interest TJS and MH report grants from Aeras. RR was a Sanofi Pasteur employee. SG and CADG are employees and share-holders at Sanofi Pasteur. PA and IK report annual fees and milestone payments from Sanofi Pasteur and collaboration with Aeras and SATVI in other TB vaccine trials. PA has a patent WO2010/006607 "Vaccines comprising TB10.4" with royalties paid from Sanofi Pasteur to SSI. TE, BL, DAH were employed by Aeras during the trial. KTR, RH and AMG report grants from Bill and Melinda Gates Foundation, grants from UK DFID, trial co-funding and in-kind support from Sanofi Pasteur, grants from DGIS (Dutch Government), during the conduct of the study; trial co-funding and in-kind support from GSK, outside the submitted work. EN, HG, VR, FR, NB, SM, LM, ME, AT, HM, LGB, WAH, SGS, RDE have nothing to disclose. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoaXXXX..

INTRODUCTION Lymphatic filariasis (LF), an infectious disease caused by mosquito-borne nematode parasites, is characterized by lymphedema of the extremities (“elephantiasis”), hydroceles and chronic disability. The life cycle of the parasite requires uptake of microfilariae (Mf) by mosquitoes with their blood meal and development of Mf in the mosquito to infective larvae that are the transmission stage for new infections in humans 1. The filarial species Wuchereria bancrofti and to a lesser extent, Brugia spp., infect more than 100 million people in 73 countries with another one billion at risk 2. The World Health Organization (WHO) has targeted LF for global elimination by 2020 by means of mass drug administration (MDA)3 that uses one of three anti-filarial drug regimens: i) DEC/ALB in LF endemic areas outside Africa and in countries within Africa that do not have onchocerciasis or loiasis, ii) IVM/ALB in African countries that have both LF and onchocerciasis iii) ALB alone in countries that have both LF and loiasis. MDA is intended to reduce the Mf reservoir below a level that is required to sustain transmission of the infection by mosquitoes. Because a single dose of these treatments fails to sterilize or kill all adult filarial worms and reduce the community Mf reservoir to sufficiently low levels 4-7, many rounds of MDA are required to interrupt transmission.8 Although this approach has successfully eliminated LF in some countries, a treatment that is more effective for killing or sterilizing adult worms could greatly accelerate efforts to eliminate LF by reducing the number of doses and annual cycles of MDA required to interrupt transmission.

red to interrupt transmission.8 Although this approach has successfully eliminated LF in some countries, a treatment that is more effective for killing or sterilizing adult worms could greatly accelerate efforts to eliminate LF by reducing the number of doses and annual cycles of MDA required to interrupt transmission. We recently reported results of a small pilot study that compared the pharmacokinetics and efficacy of a single dose of co-administered IVM/DEC/ALB versus DEC/ALB for bancroftian filariasis in Papua New Guinea 9. Triple drug therapy achieved 100% clearance of Mf at 12 and 24 months after treatment compared to 8% clearance after DEC/ALB, suggesting that IVM/DEC/ALB may have killed or permanently sterilized adult filarial worms. There were no severe or serious adverse events (AEs). The current randomized clinical trial aimed to evaluate the triple drug treatment compared to the standard DEC/ALB in a larger number of infected adult residents of an area of Papua New Guinea where LF is highly endemic and associated with high Mf burdens 4.

worms. There were no severe or serious adverse events (AEs). The current randomized clinical trial aimed to evaluate the triple drug treatment compared to the standard DEC/ALB in a larger number of infected adult residents of an area of Papua New Guinea where LF is highly endemic and associated with high Mf burdens 4. METHODS STUDY DESIGN AND PARTICIPANTS A randomized, controlled, study was performed with participants recruited from 12 villages in Dreikikir district, East Sepik Province, Papua New Guinea. None of the participants had received previous treatment for LF. Institutional Review Boards at University Hospitals Cleveland Medical Center, Cleveland, OH (#04-12-33) and the Papua New Guinea Institute of Medical Research (#1220) and Medical Research Advisory Committee (#12.35) of Papua New Guinea approved the study protocols and documents. All participants provided written informed consent. 319 circulating filarial antigen (CFA) test positive individuals were screened for blood Mf levels. 182 participants met the inclusion criteria of >50 Mf/mL, age 18-65 years, no recent illness by history, non-pregnant, no prior treatment with DEC or ALB, no significant biochemical or hematologic abnormalities and no significant proteinuria, hematuria or glucosuria (Figure 1).

iduals were screened for blood Mf levels. 182 participants met the inclusion criteria of >50 Mf/mL, age 18-65 years, no recent illness by history, non-pregnant, no prior treatment with DEC or ALB, no significant biochemical or hematologic abnormalities and no significant proteinuria, hematuria or glucosuria (Figure 1). Figure 1. Enrollment and follow-up of participants in the treatment trial. RANDOMIZATION AND BLINDING Eligible and consenting participants were randomized 1:1:1 to one of three treatment arms using a computer-generated randomization table: i) DEC 6mg/kg (Sanofi S.A., Gentilly, France) + ALB 400mg (GlaxoSmithKline, Uxbridge, United Kingdom) administered once at study initiation; ii) DEC 6mg/kg + ALB 400mg at study initiation and at 12 months, and iii) IVM 200 µg/kg (Merck & Co., Inc., Kenilworth, NJ, USA) + DEC 6mg/kg + ALB 400mg once at study initiation. A designated individual administered medications under direct observation to assure all pills were swallowed. Participants were unaware of the treatment arm. Blood samples were labelled only with ID numbers. Individuals counting Mf and evaluating adverse events (AEs) were blinded to treatment assignment.

dy initiation. A designated individual administered medications under direct observation to assure all pills were swallowed. Participants were unaware of the treatment arm. Blood samples were labelled only with ID numbers. Individuals counting Mf and evaluating adverse events (AEs) were blinded to treatment assignment. PROCEDURES Screening and initial treatment were performed at the Dreikikir health center under direct observation for ten hours and monitored for AEs over the next two days. A symptom questionnaire was administered and vital signs obtained and repeated after treatment. A symptom-directed physical examination was performed if moderate or severe subjective AEs were reported. New or worsening symptoms, changes in vital signs and new abnormal physical examination findings were considered to be drug-related AEs and scored using a modified version of the National Cancer Institute Common Terminology Criteria for Adverse Events, v4.0. Microfilaremia was assessed by passing two mL of heparinized blood (collected by venipuncture between 9 p.m. and 1 a.m.) through two 5 μm polycarbonate filters (one mL per filter, EMD Millipore Corp). Filters were washed, placed on glass slides, dried, stained with Giemsa and read by microscopy for the presence of Mf as previously described 10.

d by passing two mL of heparinized blood (collected by venipuncture between 9 p.m. and 1 a.m.) through two 5 μm polycarbonate filters (one mL per filter, EMD Millipore Corp). Filters were washed, placed on glass slides, dried, stained with Giemsa and read by microscopy for the presence of Mf as previously described 10. Circulating filarial antigen levels were measured by ELISA at baseline, 12 and 24 months, as previously described 11,12. Analysis was limited to participants with CFA levels above 15ng/mL at baseline and for whom samples were available for all time points (N=48,45,42 for the DEC/ALBx1, DEC/ALBx2, IVM/DEC/ALBx1 arms respectively). Individuals with lower baseline CFA levels were excluded from the percent reduction analysis since measurement of CFA levels is not accurate near the lower limit of detection (6.8 ng/mL). OUTCOMES The primary outcome was percent of individuals with total Mf clearance at 24 months post- treatment. Secondary outcomes were percent Mf clearance at 12 months, reduction in Mf counts, percent of individuals who cleared CFA and percent reduction in CFA relative to baseline.

Circulating filarial antigen levels were measured by ELISA at baseline, 12 and 24 months, as previously described 11,12. Analysis was limited to participants with CFA levels above 15ng/mL at baseline and for whom samples were available for all time points (N=48,45,42 for the DEC/ALBx1, DEC/ALBx2, IVM/DEC/ALBx1 arms respectively). Individuals with lower baseline CFA levels were excluded from the percent reduction analysis since measurement of CFA levels is not accurate near the lower limit of detection (6.8 ng/mL). OUTCOMES The primary outcome was percent of individuals with total Mf clearance at 24 months post- treatment. Secondary outcomes were percent Mf clearance at 12 months, reduction in Mf counts, percent of individuals who cleared CFA and percent reduction in CFA relative to baseline. STATISTICAL ANALYSIS The primary hypothesis was that IVM/DEC/ALB given once would achieve 75% Mf clearance at 36 months compared to 50% Mf clearance with a single dose DEC/ALB. A second hypothesis is that single dose of IVM/DEC/ALB would be non-inferior to DEC/ALB given annually with a confidence margin of 15%. For alpha=0.05 and power of 0.8 we estimated 46 and 54 individuals would be required for each arm to test the first and second hypotheses respectively. Additional participants were recruited to account for potential dropout. We conducted an unplanned 24- month interim analysis based on the unexpectedly high efficacy of IVM/DEC/ALB at 24 months observed in a separate pilot study 9. Because of the greater efficacy of IVM/DEC/ALB and the potential importance of these results for the Global Programme to Eliminate Lymphatic Filariasis, we decided to report the results of the 24-month interim analysis based on recommendations of our technical advisory committee. We were also concerned about risk of re-infection in study participants that resided in communities where MDA for LF had been delayed. We performed an intent-to-treat (ITT) analysis for all individuals for whom a sample was collected at 24 months. Mf counts were expressed as Mf/mL+1 and log transformed; geometric mean values (GM) were used as measures of central tendency. Baseline characteristics and Mf clearance rates by treatment group as well as differences in Mf counts and circulating antigen levels at 12 and 24 months after treatment relative to baseline were compared using the chi-squared test and the Kruskal-Wallis H test. A generalized estimating equation (SAS v 9.2) compared the Mf clearance relative to baseline among treatment arms and evaluated the independent effects of age, sex, baseline Mf counts, and village location on Mf clearance rate.

relative to baseline were compared using the chi-squared test and the Kruskal-Wallis H test. A generalized estimating equation (SAS v 9.2) compared the Mf clearance relative to baseline among treatment arms and evaluated the independent effects of age, sex, baseline Mf counts, and village location on Mf clearance rate. ROLE OF THE FUNDING SOURCE The study sponsor had no role in study design, data collection, analysis, interpretation, or writing of the report. RESULTS ENROLLMENT AND FOLLOW-UP Participants were enrolled between June 11 and December 13, 2014. Baseline demographics, Mf counts, and CFA levels were similar among the three treatment groups (Table 1). Pretreatment CFA levels correlated positively with pre-treatment Mf counts (Spearman’s rho: 0.42, p=0.02). 95% and 91% of subjects were available for follow-up at 12 and 24 months post- treatment. Reasons for loss-to-follow-up are shown in Figure 1. Three participants died from causes unrelated to the study. These were probable liver cancer (DEC/ALB x 1), snakebite (IVM/DEC/ALB), and probable suicide (DEC/ALB x 2). Table 1 Characteristics of Study Participants at Baseline Treatment Groups

RESULTS ENROLLMENT AND FOLLOW-UP Participants were enrolled between June 11 and December 13, 2014. Baseline demographics, Mf counts, and CFA levels were similar among the three treatment groups (Table 1). Pretreatment CFA levels correlated positively with pre-treatment Mf counts (Spearman’s rho: 0.42, p=0.02). 95% and 91% of subjects were available for follow-up at 12 and 24 months post- treatment. Reasons for loss-to-follow-up are shown in Figure 1. Three participants died from causes unrelated to the study. These were probable liver cancer (DEC/ALB x 1), snakebite (IVM/DEC/ALB), and probable suicide (DEC/ALB x 2). Table 1 Characteristics of Study Participants at Baseline Treatment Groups DEC/ALB x 1 DEC/ALB x 2 IVM/DEC/ALB x 1 N 61 61 60 Age Median (range) 34 (18-62) 37 (18-61) 40 (19-60) Sex (M/F) 34/27 30/31 28/32 Hemoglobin gm/dL (mean ± SD) 11.2 (1.8) 11.2 (1.7) 11.4 (1.8) Weight kg (mean ± SD) 51 (5) 52 (7) 50 (6) Microfilaria/mL geomean (Range) 744 (52-8,290) 596 (61-9,656) 699 (55-15,621) Circulating Filarial Antigen ng/mL geomean (Range) 79 (18-340) 81 (15-325) 72 (17-348) DEC – diethylcarbamazine, ALB – albendazole, IVM – ivermectin

gm/dL (mean ± SD) 11.2 (1.8) 11.2 (1.7) 11.4 (1.8) Weight kg (mean ± SD) 51 (5) 52 (7) 50 (6) Microfilaria/mL geomean (Range) 744 (52-8,290) 596 (61-9,656) 699 (55-15,621) Circulating Filarial Antigen ng/mL geomean (Range) 79 (18-340) 81 (15-325) 72 (17-348) DEC – diethylcarbamazine, ALB – albendazole, IVM – ivermectin EFFECTS OF TREATMENT ON MICROFILAREMIA With respect to the primary outcome at 24 months (Figure 2), a single dose of IVM/DEC/ALB at baseline completely cleared Mf in 52 of 54 participants (96%, [95% CI, 92%, 100%]) compared to 31 of 55 participants (56% [46%, 66%]) treated with a single dose of DEC/ALB once at baseline (relative risk was 0.08 (95% CI, 0.02-0.34, p<0.001). DEC/ALB administered twice (at baseline and 12 months post-treatment) cleared blood Mf in 42 of 56 participants (75%, 95% CI 67%, 83%). This clearance rate was significantly lower than that after a single dose of IVM/DEC/ALB (relative risk was 0.15 (0.03-0.62, 95% CI, p=0.009). At 12 months IVM/DEC/ALB completely cleared Mf in 57 of 59 participants (96%, [95% CI, 92%, 100%]) compared to 18 of 56 participants (32% [22%, 41%]) and 20 of 59 (34% [25%, 43%]) following a single dose of DEC/ALB at baseline in the two other treatment arms. There were significant differences in Mf clearance by treatment group at 24 months using a generalized estimating equation adjusted for location, age, sex and pretreatment Mf levels (odds ratios of 46 and 30 relative to DEC/ALBx1 and DEC/ALBx2 respectively, p<0.0001, Table S1). In this model, higher pre-treatment Mf counts were associated with 3% reduced likelihood of completely clearing Mf at 24 months (p=0.004). Mf clearance was not significantly associated with age or village of residence, however women were 52% more likely to be Mf negative compared to men (p=0.014).

p<0.0001, Table S1). In this model, higher pre-treatment Mf counts were associated with 3% reduced likelihood of completely clearing Mf at 24 months (p=0.004). Mf clearance was not significantly associated with age or village of residence, however women were 52% more likely to be Mf negative compared to men (p=0.014). Figure 2. Percent of participants with complete Mf clearance at 12 and 24 months post- treatment with DEC/ALB x 1 (hatched bars), DEC/ALB x 2 (light bars), and IVM/DEC/ALB x 1 (dark solid bars). Mf clearance rates were significantly higher for the IVM/DEC/ALB x 1 group compared to the other two groups at both 12 and 24 months (***p<0.001, chi-square). Complete Mf clearance at 24 months was more common in the DEC/ALB x 2 group compared to DEC/ALB x 1, p=0.004. With respect to missing data, if we assumed all missing participants in the IVM/DEC/ALB arm were Mf positive (complete clearance in 54 of 60 participants, 90%) and that all individual missed in the two DEC/ALB arms were Mf negative (complete clearance in 36 of 61 participants [59%] after DEC/ALBx1 and in 47 of 61 [77%] after DEC/ALBx2), IVM/DEC/ALB would still have had significantly higher Mf clearance rates at 24 months than either DEC/ALB treatment arms (p<0.001. p=0.04, respectively, chi-square).

missed in the two DEC/ALB arms were Mf negative (complete clearance in 36 of 61 participants [59%] after DEC/ALBx1 and in 47 of 61 [77%] after DEC/ALBx2), IVM/DEC/ALB would still have had significantly higher Mf clearance rates at 24 months than either DEC/ALB treatment arms (p<0.001. p=0.04, respectively, chi-square). The geometric mean (GM) Mf count in individuals with persistent Mf at 24 months following a single dose of DEC/ALB was 12 Mf/mL (range 1-671); four participants of the total number of persons tested at 24 months (7%) had >50 Mf/mL (Figure 3). The GM Mf count in individuals with persistent Mf at 24 months following two annual doses of DEC/ALB was 5 Mf/mL (range 1-39). At 24 months post IVM/DEC/ALB, the two Mf+ participants, one with 2 Mf/mL and the other with 44 Mf/mL (Figure 3). The same two individuals were Mf positive at 12 months with 1 Mf/ml each. Figure 3. Reductions in Mf counts at 12 and 24 months post-treatment. Note the log scale + 1 for Mf counts. A single dose of IVM/DEC/ALB was significantly more effective for reducing Mf counts than either of the two DEC/ALB treatments at 12 and 24 months, p<0.001, Mann- Whitney U test. The DEC/ALB x 2 group had greater reductions in Mf counts than the DEC/ALB x 1 group at 24 months, p=0.004.

the log scale + 1 for Mf counts. A single dose of IVM/DEC/ALB was significantly more effective for reducing Mf counts than either of the two DEC/ALB treatments at 12 and 24 months, p<0.001, Mann- Whitney U test. The DEC/ALB x 2 group had greater reductions in Mf counts than the DEC/ALB x 1 group at 24 months, p=0.004. EFFECTS OF TREATMENT ON CIRCULATING FILARIAL ANTIGENEMIA CFA levels decreased significantly by 12 months after treatment relative to baseline in all treatment groups with further reductions between 12 and 24 months. The DEC/ALB once and DEC/ALB twice treatment groups had similar reductions in CFA levels of 58%-59% and 70%- 71% at 12 and 24 months, respectively. These values were less than the 67% and 75% reductions observed at 12 and 24 months after IVM/DEC/ALB treatment, but the differences were not statistically significant. More people in the IVM/DEC/ALB treatment arm had CFA levels reduced to below the assay’s limit of detection at 24 months (14 of 42 or 32%) than those in the other treatment groups (10 of 48 or 21% in the DEC/ALB x 1 group and 11 of 45 or 24% in the DEC/ALB x 2 treatment group), but these differences were not significant. Relative CFA levels after treatment were not significantly lower in individuals who completely cleared Mf after treatment compared to individuals with persistent Mf at 12 and 24 months (p = 0.07 and 0.30).

ALB x 1 group and 11 of 45 or 24% in the DEC/ALB x 2 treatment group), but these differences were not significant. Relative CFA levels after treatment were not significantly lower in individuals who completely cleared Mf after treatment compared to individuals with persistent Mf at 12 and 24 months (p = 0.07 and 0.30). SAFETY All individuals were actively observed for up to 10 hours post-treatment for AE, and 73% of participants were assessed for AEs between 24 to 36 hours after returning to their villages (Table 2). Five participants experienced AEs during the initial 10 hours observation period; three AEs were mild, one was moderate, and one was severe. The individual with a severe AE was a 42-year-old woman with a pre-treatment Mf count of 792/mL who experienced headache, nausea and chills starting six hours after taking IVM/DEC/ALB. Physical examination revealed a temperature of 41.1°C, orthostatic hypotension, and tachycardia. She was successfully treated with oral fluids and acetaminophen and returned to her pre-treatment state of health the following day. Objective findings of fever (temperature >37.8°C) and hemodynamic changes following initial treatment tended to be higher in participants receiving IVM/DEC/ALB; however, these differences were not significant (Table 2). The frequency of subjective AEs was greater in participants who received triple drug treatment versus DEC/ALB. The difference was most pronounced in individuals who had AEs with severity >1. The most common AEs reported were headache, fatigue and nausea both for persons with grade 1 AEs (data not shown) and for persons who experienced AEs with severity >1 (Table 2). Higher pre-treatment Mf counts were associated with a greater frequency and severity of AEs. Using a logistic regression model, the odds of a grade 2 AE increased by 19% for each 200 increase in Mf/mL count (OR 1.19 [95% CI 1.09,1.36], p=0.01). This association was greatest among individuals with >500 Mf/mL. Of note, no participant experienced an AE with severity greater than grade 1 and none had fever or changes in blood pressure after their second dose of DEC/ALB. Consequently, only AEs associated with the initial treatment are included in Table 2, and AEs for the two DEC/ALB treatment arms at baseline are combined.

/mL. Of note, no participant experienced an AE with severity greater than grade 1 and none had fever or changes in blood pressure after their second dose of DEC/ALB. Consequently, only AEs associated with the initial treatment are included in Table 2, and AEs for the two DEC/ALB treatment arms at baseline are combined. Table 2 Adverse Events (AEs) Following Treatment for Lymphatic Filariasis DEC/ALB (N=91) IVM/DEC/ALB (N=41) Number of participants with AEs (percent) At least one AE 37 (41) 24 (59) Individuals with two or more AEs 24 (26) 19 (46)* Fever§ 19 (21) 14 (34) Hemodynamic changes¶ 4 (4) 5 (12) Overall Grade 1 AEs (subjective)† 36 (40) 22 (54) Overall Grade 2 and 3 AEs with severity grade >1† 5 (5) 11 (27)*** Frequency of AEs with severity grade >1 Fatigue 5 (5) 8 (20) Headache 3 (3) 7 (17) Nausea/vomiting 2 4 (10) Itch/rash 0 2 (5) Muscle ache 3 (3) 5 (12) Eye swelling 0 1 (2) Scrotal pain/swelling 2 (2) 4 (10) Dyspnea 0 2 (5) * p<0.05 § Auricular temperature ≥37.5°C. The highest temperature recorded post-treatment was 41.1°C. ¶ Defined as a change in blood pressure of 30 mm Hg systolic or 20 mm Hg diastolic compared to the pre-treatment recording. Three of five individuals in the IVM+DEC+ALB group had reduced blood pressure. All four individuals in the DEC+ALB group had reduced blood pressure. † Only grade 2 or 3 AEs are listed by symptoms. All but one participant with grade 2 symptoms had more than one AE. *** p<0.001 by chi-square

¶ Defined as a change in blood pressure of 30 mm Hg systolic or 20 mm Hg diastolic compared to the pre-treatment recording. Three of five individuals in the IVM+DEC+ALB group had reduced blood pressure. All four individuals in the DEC+ALB group had reduced blood pressure. † Only grade 2 or 3 AEs are listed by symptoms. All but one participant with grade 2 symptoms had more than one AE. *** p<0.001 by chi-square DISCUSSION These results show that a single dose of the new triple drug regimen consisting of IVM/DEC/ALB was much more effective for clearing W. bancrofti blood Mf than treatment with standard MDA of DEC/ALB in LF-endemic areas outside of sub-Saharan Africa. Participants in this clinical trial had not been previously treated for LF, and they had moderate to very high blood levels of W. bancrofti Mf and filarial antigenemia. A single dose of triple drug therapy cleared Mf from almost all participants, and the effects persisted for at least 24 months. This regimen was superior for clearing Mf compared to a single dose or two annual doses of DEC/ALB. Results observed after DEC/ALB treatment were consistent with those reported from previous trials 9,13. Although a single dose of triple therapy did not completely clear Mf in every subject, residual Mf counts were reduced to levels unlikely to support mosquito-borne transmission 14,15. Both IVM/DEC/ALB and the two drug regimen of DEC/ALB had potent macrofilaricidial effects based on >70% reductions in CFA levels 24 months after treatment commenced. These data are consistent with prior studies that documented partial macrofilaricidal effects of DEC/ALB 16,17,18, whereas IVM has little or no ability to kill adult worms 19. The addition of IVM to DEC/ALB had only a marginal impact on reducing CFA levels, but the triple drug combination appears to be very effective for sterilizing adult worms, an effect that may be permanent based on observations to date. Limitations are that an open-labeled study could bias assessment of adverse events and blood Mf detected at the time of follow-up could be due to reinfection in some cases. However, we believe this is unlikely because of high rates of bed net use in study communities.

hat may be permanent based on observations to date. Limitations are that an open-labeled study could bias assessment of adverse events and blood Mf detected at the time of follow-up could be due to reinfection in some cases. However, we believe this is unlikely because of high rates of bed net use in study communities. Adverse events were more frequent after treatment with IVM/DEC/ALB. This result is consistent with observations in a pilot study performed by our group 9. Since AEs are triggered by Mf death, it is not surprising that AEs were more common in people treated with two potent microfilaricidal drugs (DEC/IVM). The single severe AE that occurred was self-limited, similar to AEs reported in earlier studies after individuals were treated with DEC/IVM or DEC alone 20. AE frequency and severity after triple drug treatment are likely to be much lower in community MDA settings where infection rates and blood Mf levels are lower than those in this clinical trial. Indeed, a recently completed, multicenter community safety study performed in LF-endemic areas found the same rates and severity of AEs after MDA with either IVM/DEC/ALB or DEC/ALB (authors’ unpublished data). Simulation-modeling studies suggest that MDA with IVM/DEC/ALB should significantly reduce the number of rounds of MDA required to reach elimination targets 21. Thus, triple drug MDA with IVER/DEC/ALB provides a potential road to success for countries that are not currently on track to eliminate LF by the current target year of 2020 2,22.

studies suggest that MDA with IVM/DEC/ALB should significantly reduce the number of rounds of MDA required to reach elimination targets 21. Thus, triple drug MDA with IVER/DEC/ALB provides a potential road to success for countries that are not currently on track to eliminate LF by the current target year of 2020 2,22. Supplementary Materials Supplementary Materials ACKNOWLEDGEMENTS We gratefully acknowledge cooperation of the study participants and study investigators who provided essential technical, community engagement and clinical support. We sincerely acknowledge the support of the Papua New Guinea National Department of Health Lymphatic Filariasis Control Program, in particular Dr Sibauk Bieb, Dr Lucy John, Mr Leo Makita and Ms Mary Yogahu and WHO Papua New Guinea Neglected Tropical Disease representative Dr James Wangi. Mr. Kurt Curtis at Washington University in St. Louis performed filarial antigen testing. Dr. Daniel Tisch helped with generalized estimating equation analysis. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1706854.. Funding by the Bill and Melinda Gates Foundation grant GH5342.

Introduction MORDOR I (Macrolides Oraux pour Réduire les Décès avec un Oeil sur la Résistance) found that biannual azithromycin distributions reduced childhood mortality by 14% in communities in Niger, Malawi, and Tanzania. The greatest observed benefit was seen in Niger, with 18% fewer deaths in communities randomized to azithromycin compared to those randomized to placebo. This observed effect could decrease or increase over time for a number of reasons. For example, a beneficial effect of azithromycin might wane with selection of antibiotic-resistant bacteria. This is certainly possible, since mass azithromycin distributions in trachoma programs have selected for macrolide-resistant strains of Streptococcus pneumoniae and Escherichia coli, and since azithromycin clearly selected for resistance in Niger during MORDOR I.1-9 Or azithromycin might delay the death of a frail child, but not ultimately prevent it. Such an effect could occur if antibiotic distributions diminished the development of protective immunity in a population by reducing its exposure to pathogens. On the other hand, the observed efficacy actually increased with each distribution during the 2 years of MORDOR I, suggesting the possibility of an enhanced effect with additional treatment.10 Such an effect could be explained by cumulative reduction in pathogens with each distribution, or if resistant bacteria were less fit. Moreover, efficacy could improve over time due to better implementation with experience.

years of MORDOR I, suggesting the possibility of an enhanced effect with additional treatment.10 Such an effect could be explained by cumulative reduction in pathogens with each distribution, or if resistant bacteria were less fit. Moreover, efficacy could improve over time due to better implementation with experience. Here in MORDOR II, we provided biannual azithromycin to both the original placebo-treated and azithromycin-treated arms in the Nigerien communities of MORDOR I for an additional year. This resulted in a randomized comparison of the first year to the third year of mass azithromycin treatment. As azithromycin affects transmissible diseases, treating an individual may influence others in the same community. Thus randomization and intervention were at the community level, and inference of efficacy was made at the community level. MORDOR II continued the large simple trial paradigm of MORDOR I, with a straightforward intervention and primary outcome.11

ases, treating an individual may influence others in the same community. Thus randomization and intervention were at the community level, and inference of efficacy was made at the community level. MORDOR II continued the large simple trial paradigm of MORDOR I, with a straightforward intervention and primary outcome.11 Methods Eligibility This continuation study was planned only in the Niger districts of MORDOR I, not in the lower mortality sites of Malawi or Tanzania. The Niger component of MORDOR I was conducted in the departments of Boboye and Loga. The randomization unit was the grappe, and those with a population between 200 and 2,000 inhabitants on the most recent pre-MORDOR I census were eligible for enrollment. Communities remained in the continuation study MORDOR II even if the population had drifted out of this range. All children aged 1-59 months (truncated to month) and weighing at least 3,800 grams were eligible for treatment. Radomizataion and masking The original MORDOR I randomization and interventions were performed at the community level. The randomization list was generated by TCP using R (R Foundation for Statistical Computing, Vienna, Austria) and implemented by TCP, KJR, and necessary members of the Pfizer team. While all communities received azithromycin in MORDOR II, participants and observers remained masked to the original treatment arm from MORDOR I.

ndomization list was generated by TCP using R (R Foundation for Statistical Computing, Vienna, Austria) and implemented by TCP, KJR, and necessary members of the Pfizer team. While all communities received azithromycin in MORDOR II, participants and observers remained masked to the original treatment arm from MORDOR I. Census A house-to-house census was performed during the 2 additional 6-month periods in the same manner as in MORDOR I.10 All households in the community were entered into a custom- built mobile application (Conexus Inc., Los Gatos, CA), with the head of household and the GPS coordinates facilitating identification of the household at the subsequent census. All children in the household aged 1-59 months were enumerated. The vital status (alive, dead, or unknown) and residence (moved within community, moved outside community, or unknown) were recorded for each child. The vital status of children enrolled in the preceding census who had aged past 59 months was also assessed, although these children were not included in the next study period. Pregnant women and children under the age of 1 month were recorded in the application in anticipation of enrollment in the subsequent census. Communities were censused in the same general order each period. Data were uploaded to the Salesforce Cloud Database Service (Salesforce.com, San Francisco, CA), and data cleaning was performed in Salesforce.com, Stata (Statacorp, College Station, TX), and R.

tion in anticipation of enrollment in the subsequent census. Communities were censused in the same general order each period. Data were uploaded to the Salesforce Cloud Database Service (Salesforce.com, San Francisco, CA), and data cleaning was performed in Salesforce.com, Stata (Statacorp, College Station, TX), and R. Intervention Each child aged 1-59 months at the census was offered a single, directly observed dose of oral azithromycin (Pfizer, Inc., New York, NY). A volume of suspension corresponding to at least 20 mg/kg was given by height-stick approximation according to Niger’s trachoma program guidelines, or by weight for those children unable to stand (typically those under 1 year of age). No azithromycin tablets were used, only suspension, and children known to be allergic to macrolides were not treated. Treatment was administered at the census and during additional visits in an attempt to achieve at least 80% coverage. Administration of study medication was documented for each child in the mobile application, with community coverage calculated relative to the census data. Serious adverse events other than death within 2 weeks of the outcome were reported to study personnel.

ional visits in an attempt to achieve at least 80% coverage. Administration of study medication was documented for each child in the mobile application, with community coverage calculated relative to the census data. Serious adverse events other than death within 2 weeks of the outcome were reported to study personnel. Primary outcome The pre-specified primary outcome was the community-level, all-cause mortality rate determined by biannual census. Each inter-census period was treated separately, with a mortality event counting only when a child was recorded as being alive and living in the household at one census, and recorded as having died while residing in the community at the subsequent census. By design, no attempt was made to track down the status of a child after they had moved outside the community. Person-time at risk was calculated as days between consecutive censuses, with children who moved, died, or had an unknown follow-up status contributing half the inter-census period. All children documented as alive and present in the household at the initial census of each inter-census period were included in the analysis. No changes to trial methods or outcomes were made after the continuation trial had commenced.

died, or had an unknown follow-up status contributing half the inter-census period. All children documented as alive and present in the household at the initial census of each inter-census period were included in the analysis. No changes to trial methods or outcomes were made after the continuation trial had commenced. Sample size and statistical analysis plan This continuation study was pre-specified before the initiation of the original MORDOR I trial, contingent on finding a significant beneficial effect. The original sample size for MORDOR-I Niger estimated that 624 clusters would provide 80% power to detect a 15% reduction in mortality, assuming a community size of 668, 17% of the population in the target range of 1-59 months, a death rate of 2% per year, a coefficient of variation (CV) of 0.51, and loss to follow-up of 10% per year (Statistical Analysis Plan). Updating calculations based on results from MORDOR I-Niger, using the observed CV of 0.34 and a death rate of 2.5% per year, resulted in 80% power to detect a 15.5% effect size for MORDOR II. Because all communities were treated biannually, no interim efficacy or futility stopping rule was included, although a Data and Safety Monitoring Committee of 3 individuals reviewed the data on the completion of the year (Supplemental Material). For the randomized comparison of communities receiving their first year of azithromycin distributions versus those receiving their third year of antibiotics, the pre-specified primary analysis was negative binomial regression of the number of deaths per community, with treatment arm as a predictor and person- time at risk as an offset. Hypothesis testing was 2-sided, with an alpha of 0.05. A P-value was determined by Monte Carlo permutation testing (10,000 replications). Intra-cluster correlations were accounted for by using community-level data and community-level heterogeneity taken into account by the dispersion parameter in the negative binomial regression.

testing was 2-sided, with an alpha of 0.05. A P-value was determined by Monte Carlo permutation testing (10,000 replications). Intra-cluster correlations were accounted for by using community-level data and community-level heterogeneity taken into account by the dispersion parameter in the negative binomial regression. Secondary outcomes Mortality was compared longitudinally within each arm using similar methods, contrasting the first 2 years of treatment (placebo or azithromycin) with the third year of treatment (azithromycin in both arms), clustering on community. All statistical analyses were conducted in R. Ethics Approval for the study was obtained from the ethical committees of the Niger Ministry of Health, the UCSF Institutional Review Board, and Emory University. Informed consent was obtained from the local Ministry of Health, village leaders, and guardians of children. No incentives were offered for participation. The study was undertaken in accordance with the Declaration of Helsinki. The study was designed by authors JDK, AMA, PME, TCP, and TML, data gathered by JDK, AMA, RM, NB, SEA, MMA, CC, EL, KSO, TD, CEO, EKC, and TML, and data analyzed by JDK, YL, KJR, TCP, and TML. The initial draft was written by TML, with all coauthors participating in editing and agreeing to publication.

ion of Helsinki. The study was designed by authors JDK, AMA, PME, TCP, and TML, data gathered by JDK, AMA, RM, NB, SEA, MMA, CC, EL, KSO, TD, CEO, EKC, and TML, and data analyzed by JDK, YL, KJR, TCP, and TML. The initial draft was written by TML, with all coauthors participating in editing and agreeing to publication. Results Participant flow As displayed in Figure 1, all 594 communities from MORDOR I were followed in MORDOR II. No communities were lost to follow-up. Census periods were from February 2017 to August 2017, September 2017 to January 2018, and Feburary 2018 to August 2018. Demographic characteristics of communities in both arms at 24 months are displayed in Table 1. Table 1. Demographic characteristics of communities and participants at the start of MORDOR II Characteristic Treatment Arm Placebo Azithromycin Communities (number) 291 303 Children 1-59 months (number) 33,294 37,497 Children per community (mean ± sd) 114±80 124±86 Male sex 51.3% 51.2% Age group 1-5 mo 5.9% 6.3% 6-11 mo 9.6% 9.5% 12-23 mo 23.1% 23.0% 24-59 mo 61.4% 61.2% Figure 1 In MORDOR I, communities were enrolled and randomly assigned to 4 biannual distributions of azithromycin or placebo. In MORDOR II, these same communities were followed, with both arms offered 2 biannual distributions of azithromycin. Distribution by randomization unit is expressed as the estimated mean (±SD) for the population. No communities were lost to follow-up during MORDOR I or MORDOR II.

stributions of azithromycin or placebo. In MORDOR II, these same communities were followed, with both arms offered 2 biannual distributions of azithromycin. Distribution by randomization unit is expressed as the estimated mean (±SD) for the population. No communities were lost to follow-up during MORDOR I or MORDOR II. In MORDOR II, azithromycin coverage averaged 92.0% (standard deviation ±6.6%) in the original azithromycin-treated communities and 91.3% (±7.2%) in the original placebo-treated communities (Supplementary Table 1). The census status was recorded as moved or unknown in 4079 of 64,225 cases (6.4%) in those communities receiving their first year of azithromycin, and as moved or unknown in 4685 of 72,108 cases (6.5%) in those communities receiving their third year of azithromycin, with no significant difference between arms (p=0.48, Supplementary Table 2). Table 2. Childhood mortality rate over time Distribution year Mortality Rate in children 1-59 months Deaths per 1000 person-years (95% CI) Difference between arms (randomized comparison) Placebo Azithromycin MORDOR I 1 26.3 (24.2—28.8) 21.9 (20.2—24.2) 16.0% (5.7%—25.1%) P=0.003 2 28.0 (25.8—30.5) 22.4 (20.6—24.3) 20.3% (10.6%—28.9%) P=0.0001 Azithromycin Azithromycin MORDOR II 3 24.0 (22.1—26.3) 23.3 (21.4—25.5) 3.5% (-8.3%—14%) P=0.55 Difference between MORDOR I and II (longitudinal comparison) 13.3% (5.8%—20.2%) P=0.007 -3.6% (-12.3%—4.5%) P=0.50

Distribution year Mortality Rate in children 1-59 months Deaths per 1000 person-years (95% CI) Difference between arms (randomized comparison) Placebo Azithromycin MORDOR I 1 26.3 (24.2—28.8) 21.9 (20.2—24.2) 16.0% (5.7%—25.1%) P=0.003 2 28.0 (25.8—30.5) 22.4 (20.6—24.3) 20.3% (10.6%—28.9%) P=0.0001 Azithromycin Azithromycin MORDOR II 3 24.0 (22.1—26.3) 23.3 (21.4—25.5) 3.5% (-8.3%—14%) P=0.55 Difference between MORDOR I and II (longitudinal comparison) 13.3% (5.8%—20.2%) P=0.007 -3.6% (-12.3%—4.5%) P=0.50 Primary outcome Mortality rates by treatment arm are displayed by inter-census period (Figure 2) and by year (Table 2). Communities randomized to the first year of azithromycin distribution experienced 24.0 deaths per 1,000 person-years (95% CI, 22.1 to 26.3), and those randomized to the third year of azithromycin distribution 23.3 deaths (95% CI, 21.4 to 25.5) per 1,000 person- years. We found no evidence that the first year of treatment had a greater effect than the third year of treatment, with a relative 3.5% (95% CI -8.3% to 14%) more deaths in those communities randomized to the first year of treatment (p=0.55, Table 2).

n distribution 23.3 deaths (95% CI, 21.4 to 25.5) per 1,000 person- years. We found no evidence that the first year of treatment had a greater effect than the third year of treatment, with a relative 3.5% (95% CI -8.3% to 14%) more deaths in those communities randomized to the first year of treatment (p=0.55, Table 2). Figure 2 All-cause mortality rate in 1-59 month old children over time in communities randomized to 2 years of treatment with biannual placebo and the third year with biannual azithromycin (blue, with 95% CI in lighter blue), and in communities randomized to 3 years of biannual azithromycin (red, with 95% CI in lighter red). In MORDOR II, we were unable to show a statistically signficant difference between the 2 arms in year 3 (p=0.55). Mortality did decrease significantly in the originally placebo-treated communities (-13.0%, 95% CI, -21.5% to -3.7%, p=0.008). In the communities originally receiving azithromycin, mortality was not significantly different in a third year of azithromycin (2.1%, 95% CI, -7.6 to 12.6%, p=0.69). Note that the annual mortality rates used in this study are expected to be several fold lower than the Under 5 Mortality Rate (U5MR), which is the number of live births that do not survive till their fifth birthday.

y was not significantly different in a third year of azithromycin (2.1%, 95% CI, -7.6 to 12.6%, p=0.69). Note that the annual mortality rates used in this study are expected to be several fold lower than the Under 5 Mortality Rate (U5MR), which is the number of live births that do not survive till their fifth birthday. Secondary outcomes In the communities originally receiving 2 years of biannual placebo distribution, mortality decreased (-13.3%, 95% CI, 5.8% to -20.2%, p=0.007) over the next year when treated with biannual azithromycin. In the communities originally receiving azithromycin, the mortality reduction was not significantly different in the third year compared to the first 2 years (-3.6%, 95% CI, -12.3% to 4.5%, p=0.50). Serious adverse events Mortality is as reported, and medical review was unable to declare that any additional serious adverse events were possibly caused by azithromycin.

Secondary outcomes In the communities originally receiving 2 years of biannual placebo distribution, mortality decreased (-13.3%, 95% CI, 5.8% to -20.2%, p=0.007) over the next year when treated with biannual azithromycin. In the communities originally receiving azithromycin, the mortality reduction was not significantly different in the third year compared to the first 2 years (-3.6%, 95% CI, -12.3% to 4.5%, p=0.50). Serious adverse events Mortality is as reported, and medical review was unable to declare that any additional serious adverse events were possibly caused by azithromycin. Discussion In MORDOR I, 2 years of biannual oral azithromycin distribution to post-neonatal preschool children significantly reduced all-cause mortality in Niger by 18%.10 Here in MORDOR II, both arms received an additional year of biannual azithromycin, resulting in a randomized comparison of a third year of treatment to the first year of treatment. We found no evidence that the benefit of azithromycin waned in the third year. Some had hypothesized a decrease in efficacy with more distributions due to the selection of antibiotic resistant bacteria.12-14 Repeated mass azithromycin distributions for trachoma have indeed selected for macrolide resistance in nasopharyngeal S. pneumoniae and rectal E. coli.1-3,6,15 Resistance was clearly selected for in the nasopharynx and stool in Niger in MORDOR I.9 Resistance emerging during mass azithromycin distributions could theoretically have curbed or even reversed any potential survival benefit. We also found no evidence that the effect of azithromycin was enhanced with additional distributions. Enhancement was possible since the overall benefit in the 3 sites of MORDOR I increased with each of the first 4 biannual distributions from 7% to 22%, although that apparent increase was not statistically significant.10 Here, the randomized comparison between the first and third year of treatments did not support either an increasing or decreasing effect on mortality with additional rounds distributions of azithromycin. Even longer follow-up will be necessary to determine whether the mortality effect is sustained past the third year of distributions.

ized comparison between the first and third year of treatments did not support either an increasing or decreasing effect on mortality with additional rounds distributions of azithromycin. Even longer follow-up will be necessary to determine whether the mortality effect is sustained past the third year of distributions. The communities receiving their first year of treatment had 13% lower mortality than they had in the previous two years of receiving placebo. While this longitudinal analysis was not a randomized comparison and is therefore subject to confounding, the result does support the original MORDOR I finding of a 14% reduction in the 3-country analysis. The mortality rate fell with the first of the two additional distributions, suggesting that cumulative treatments are not necessary to achieve efficacy. This is consistent with a secondary analysis of MORDOR I in which deaths were relatively lower in the first 3 months after a biannual distribution.16 The convergence of mortality rates in the two arms in MORDOR II—when both arms received the same treatment—adds some support that the difference in MORDOR I was indeed due to intervention and not from imbalanced randomization.

in which deaths were relatively lower in the first 3 months after a biannual distribution.16 The convergence of mortality rates in the two arms in MORDOR II—when both arms received the same treatment—adds some support that the difference in MORDOR I was indeed due to intervention and not from imbalanced randomization. The study has several limitations. As a large simple trial, little information was collected on each child and community.11 Deaths were determined by consecutive censuses. Children who were born and died between censuses contributed neither to the death count nor person-time at risk for the primary outcome. Death rates may have differed in children who moved or had an unknown census status. While the randomized comparison assessed whether a community’s prior treatment history affected the results, it was not designed to analyze an individual’s prior treatment history. Cluster-randomized trials run the risk of contamination between arms, which could dampen the observed effect. While the intervention itself was not subject to contamination since all communities were given the same treatment, infections could spread between nearby communities and cause contamination. Although this could theoretically explain the MORDOR II findings, contamination did not prevent a highly significant result in MORDOR I, so invoking this explanation would require contamination in the third year only. No child in MORDOR had ever received azithromycin as part of a trachoma program, but macrolide use outside of the study was not recorded. As distributions were offered only biannually, a child’s first treatment might not be until 7 months of age. Supplementary treatments given during a scheduled vaccination visit to a health clinic might prove to be a more reliable way of reaching younger infants. The longitudinal comparisons of years 1 and 2 versus year 3 were not randomized. Conditions may have changed between these time periods. This study did not investigate whether morbidity increased or decreased with azithromycin.

visit to a health clinic might prove to be a more reliable way of reaching younger infants. The longitudinal comparisons of years 1 and 2 versus year 3 were not randomized. Conditions may have changed between these time periods. This study did not investigate whether morbidity increased or decreased with azithromycin. The study also did not evaluate the mechanism by which azithromycin reduced mortality, although its antimicrobial effect presumably plays a role since a majority of child deaths in this area are attributed to infectious disease.17 Smaller parallel trials with detailed microbiological and anthropometric assessments were conducted, and may provide insight into mechanism of action.18, 19 Azithromycin has been linked to cardiac death in adults, although epidemiological results are mixed and may not be relevant to children in this setting.20-23 Later development of atopic disease has been associated with infant antibiotic use in general, and macrolides in particular.24 Rare side effects, or those only apparent later in life would be difficult to assess with this study design.

ological results are mixed and may not be relevant to children in this setting.20-23 Later development of atopic disease has been associated with infant antibiotic use in general, and macrolides in particular.24 Rare side effects, or those only apparent later in life would be difficult to assess with this study design. The International Trachoma Initiative has now distributed over 800 million doses of oral azithromycin in the trachoma control programs sponsored by the World Health Organization.25 Azithromycin has proven quite effective in reducing the prevalence of, and in some cases completely elimininating the strains of ocular chlamydia that cause the disease.2,4,6,26-28 The number of annual trachoma distributions is now declining as countries continue to meet control criteria.25 Many regions with high childhood mortality are either no longer endemic for trachoma, or never were. Thus the majority of children now being born into areas with the highest under-5 mortality will not receive azithromycin as part of trachoma programs.29 The treatment regimens were different for trachoma and MORDOR: annual mass azithromcyin treatment of ages 6 months and older for trachoma, and biannual distributions targeted to ages 1- 59 months in MORDOR I and II. MORDOR distributed approximately one third as many doses of azithromycin per community per year as would a trachoma program. If azithromycin for childhood mortality were targeted to areas with very high mortality such as Niger, only a fraction of the total antibiotic used in trachoma programs would be required.

and II. MORDOR distributed approximately one third as many doses of azithromycin per community per year as would a trachoma program. If azithromycin for childhood mortality were targeted to areas with very high mortality such as Niger, only a fraction of the total antibiotic used in trachoma programs would be required. In summary, a randomized comparison found no evidence that the beneficial effect of mass biannual azithromycin distribution on childhood mortality wanes in the third year of distribution compared to the first. Biannual oral azithromycin distribution did significantly reduce mortality compared to the 2 previous years of biannual placebo distributions. This longitudinal observation supports the original MORDOR I community-randomized trial results. Selection of antibiotic resistant strains of pathogenic bacteria may eventually reduce efficacy and needs to continue to be monitored with longer follow-up. Supplementary appendix Supplementary appendix Acknowledgments We thank That Man May See and Research to Prevent Blindness. We also thank the Biblioteca Angelica and their staff (MIBAC). Funding The Bill and Melinda Gates Foundation provided the funding for the trial (OPP1032340). Pfizer Inc. (New York City) provided both the azithromycin and the placebo oral suspensions. The Salesforce Foundation provided user licenses to Salesforce.com and cloud storage. Registration Clinicaltrials.gov NCT02047981

Supplementary appendix Supplementary appendix Acknowledgments We thank That Man May See and Research to Prevent Blindness. We also thank the Biblioteca Angelica and their staff (MIBAC). Funding The Bill and Melinda Gates Foundation provided the funding for the trial (OPP1032340). Pfizer Inc. (New York City) provided both the azithromycin and the placebo oral suspensions. The Salesforce Foundation provided user licenses to Salesforce.com and cloud storage. Registration Clinicaltrials.gov NCT02047981 This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1817213.

Mass drug administration with azithromycin reduced childhood mortality in MORDOR I and II in Niger, although antibiotic resistance remains a major concern.1-3 MORDOR I was a trial that did not itself include morbidity assessments.1 However, 30 communities in Niger were randomly selected from the same pool as the mortality trial, and randomly assigned to either azithromycin or placebo distributed biannually to pre-school children as in MORDOR I (see protocol and SAP at nejm.org for full study details). The mean (±SD) placebo and azithromycin coverage over the four twice-yearly distributions was 82 ± 6% and 79 ± 8%, respectively. Ethical approval was obtained from the University of California San Francisco Committee for Human Research and the Ethical Committee of the Niger Ministry of Health. We obtained oral consents from guardians prior to treatment and swab collection. No incentives were offered. All analyses were at the community level.

ly. Ethical approval was obtained from the University of California San Francisco Committee for Human Research and the Ethical Committee of the Niger Ministry of Health. We obtained oral consents from guardians prior to treatment and swab collection. No incentives were offered. All analyses were at the community level. Here, we compare the proportion of macrolide-resistant pneumococcus in pre-school children between azithromycin and placebo-treated communities, using broth dilution assays on pneumococcus isolated from nasopharyngeal swabs collected at 24 months (approximately 6 months after the fourth biannual treatment). Pneumococcus isolation and resistance testing were performed according to standard protocols at ARUP, a CLIA-certified reference laboratory, using breakpoints as documented in the Clinical and Laboratory Standard Institute guidelines (see Supplementary Appendix). For this study, intermediate and resistant cases were considered as resistant. Because the gut is a reservoir for antibiotic resistance genes, we also evaluated the resistome from rectal samples at 24 months using metagenomic DNA sequencing and compared the non-host sequences against a curated antibiotic resistance database.4,5

study, intermediate and resistant cases were considered as resistant. Because the gut is a reservoir for antibiotic resistance genes, we also evaluated the resistome from rectal samples at 24 months using metagenomic DNA sequencing and compared the non-host sequences against a curated antibiotic resistance database.4,5 At 24 months, the proportion of macrolide resistance in nasopharyngeal S. pneumoniae at the community level was higher in the azithromycin-treated communities (mean 12.3%, 95% confidence interval 5.7—20.0%) than in the placebo-treated communities (2.9%, 0—6.1%, p=0.02, see Table 1 and Table S1 in the Supplementary Appendix). Similarly, macrolide resistance determinants in the gut were more prevalent in the azithromycin-treated communities (68.1%, 60.4—74.8% vs 46.3%, 35.9—52.9%; p <0.001). We next evaluated whether mass oral azithromycin administration was associated with an increase in resistance to other antibiotics. The proportion of isolated pneumococcus resistant to penicillin was similar between treatment arms: 18.7%, (8.2%—30.6%) in the azithromycin group versus 22.3% (10.2%—37.8%) in the placebo group (p=0.72). As with the nasopharyngeal samples, we found no evidence of a difference between arms for the rectal samples in the non-macrolide classes (Table 1 and Table S2 in the Supplementary Appendix). Table 1 Nasopharyngeal pneumococcal and gut antibiotic resistance of pre-school children at 24 months.

At 24 months, the proportion of macrolide resistance in nasopharyngeal S. pneumoniae at the community level was higher in the azithromycin-treated communities (mean 12.3%, 95% confidence interval 5.7—20.0%) than in the placebo-treated communities (2.9%, 0—6.1%, p=0.02, see Table 1 and Table S1 in the Supplementary Appendix). Similarly, macrolide resistance determinants in the gut were more prevalent in the azithromycin-treated communities (68.1%, 60.4—74.8% vs 46.3%, 35.9—52.9%; p <0.001). We next evaluated whether mass oral azithromycin administration was associated with an increase in resistance to other antibiotics. The proportion of isolated pneumococcus resistant to penicillin was similar between treatment arms: 18.7%, (8.2%—30.6%) in the azithromycin group versus 22.3% (10.2%—37.8%) in the placebo group (p=0.72). As with the nasopharyngeal samples, we found no evidence of a difference between arms for the rectal samples in the non-macrolide classes (Table 1 and Table S2 in the Supplementary Appendix). Table 1 Nasopharyngeal pneumococcal and gut antibiotic resistance of pre-school children at 24 months. Antibiotic Mean Proportion, % (95% Confidence Interval) Streptococcus pneumoniae resistance (phenotypic) Placebo Azithromycin Erythromycin 2.9 (0 to 6.1) 12.3 (5.7 to 20.0) Clindamycin 1.7 (0 to 4.3) 9.0 (4.3 to 14.1) Penicillin 22.3 (10.2 to 37.8) 18.7 (8.2 to 30.6) TMP-SMX 77.1 (65.4 to 88.1) 84.7 (76.4 to 92.4) Doxycycline 50.1 (33.7 to 66.0) 60.1 (50.8 to 70.5) Linezolid 0 (0 to 2.3) 0 (0 to 2.3) Ceftriaxone 0 (0 to 2.3) 0 (0 to 2.3) Vancomycin 0 (0 to 2.3) 0 (0 to 2.3) Levofloxacin 0 (0 to 2.3) 0 (0 to 2.3) Meropenem 0 (0 to 2.3) 0 (0 to 2.3) Antibiotic Mean Prevalence, % (95% Confidence Interval) Genetic resistance determinants in stool Placebo Azithromycin Macrolides 46.3 (35.9 to 52.9) 68.1 (60.4 to 74.8) Aminocoumarins 5.5 (2.9 to 8.6) 9.2 (5.2 to 13.1) Aminoglycosides 30.9 (25.2 to 37.8) 37.6 (30.4 to 45.3) Bacitracin 17.5 (10.8 to 26.1) 17.7 (12.8 to 25.9) β-lactam 64.0 (57.5 to 72.9) 67.6 (59.7 to 74.5) Cationic 33.2 (25.5 to 41.8) 35.2 (29.3 to 42.0) Elfamycins 47.0 (39.4 to 54.0) 48.0 (36.0 to 54.7) Fluoroquinolones 28.3 (20.9 to 36.7) 27.4 (20.2 to 35.2) Fosfomycin 0 (0 to 2.3) 0.6 (0 to 1.8) Glycopeptides 1.2 (0 to 3.0) 1.3 (0 to 3.2) Metronidazole 21.9 (15.5 to 28.5) 31.8 (21.7 to 41.3) Multi-drug-resistance 44.9 (36.6 to 54.4) 43.6 (37.9 to 50.3) Phenicol 4.4 (1.9 to 8.0) 5.6 (1.9 to 13.9) Rifampin 13.8 (9.2 to 19.5) 16.8 (10.8 to 24.7) Sulfonamide 23.2 (17.0 to 30.4) 16.7 (9.9 to 26.8) Tetracycline 74.0 (68.5 to 79.6) 75.4 (68.7 to 81.0) Trimethoprim 49.5 (40.2 to 58.2) 50.8 (43.1 to 59.1) In conclusion, targeted biannual mass oral azithromycin to pre-school children in Niger increases resistance to macrolides (and related clindamycin resistance), but we found no evidence of increased resistance to other classes of antibiotics at 2 years. This is consistent with results from trachoma programs, which typically distribute annually in ages 6 months through adults. The longer-term effects of prolonged mass azithromycin distributions to pre-school children remain to be determined.3 Any policy for implementation of mass antibiotic administration should be coupled with careful monitoring for antibiotic resistance.

which typically distribute annually in ages 6 months through adults. The longer-term effects of prolonged mass azithromycin distributions to pre-school children remain to be determined.3 Any policy for implementation of mass antibiotic administration should be coupled with careful monitoring for antibiotic resistance. Trial Registration. Clinicaltrials.gov NCT02047981 Supplementary Material Click here for additional data file. Funding This work was funded by the Bill and Melinda Gates Foundation, the Peierls Foundation, Research to Prevent Blindness Career Development Award, and an unrestricted grant from Research to Prevent Blindness (RPB, New York, NY).

Bedaquiline improves survival among individuals with multidrug-resistant tuberculosis (MDR-TB).1 We report a 65-year old HIV-negative South African male diagnosed in 2013 with MDR-TB (resistant to rifampicin and isoniazid; phenotypically susceptible to a fluoroquinolone and amikacin). Baseline X-ray showed bilateral TB disease with left apex cavitation. He initiated standardised treatment including moxifloxacin, pyrazinamide, kanamycin, ethionamide, isoniazid and terizidone. After initial sputum culture conversion (month 3) and clinical improvement, the patient reconverted to culture positive and developed bilateral cavitation. Following detection of phenotypic ofloxacin resistance (month 6), treatment was revised (month 8) to include high-dose isoniazid, ethambutol, pyrazinamide, terizidone, linezolid, para-aminosalicylic acid and kanamycin (Figure 1). Bedaquiline was added 22 days later and administered for 6 months.2 The patient remained culture positive (treatment failure) and treatment was stopped 15 months after revision of the regimen. The patient died 7 months later. Figure 1 Chronology of the diagnosis and treatment of the case study

Bedaquiline improves survival among individuals with multidrug-resistant tuberculosis (MDR-TB).1 We report a 65-year old HIV-negative South African male diagnosed in 2013 with MDR-TB (resistant to rifampicin and isoniazid; phenotypically susceptible to a fluoroquinolone and amikacin). Baseline X-ray showed bilateral TB disease with left apex cavitation. He initiated standardised treatment including moxifloxacin, pyrazinamide, kanamycin, ethionamide, isoniazid and terizidone. After initial sputum culture conversion (month 3) and clinical improvement, the patient reconverted to culture positive and developed bilateral cavitation. Following detection of phenotypic ofloxacin resistance (month 6), treatment was revised (month 8) to include high-dose isoniazid, ethambutol, pyrazinamide, terizidone, linezolid, para-aminosalicylic acid and kanamycin (Figure 1). Bedaquiline was added 22 days later and administered for 6 months.2 The patient remained culture positive (treatment failure) and treatment was stopped 15 months after revision of the regimen. The patient died 7 months later. Figure 1 Chronology of the diagnosis and treatment of the case study Summary of treatment provision, genotypic drug resistance (based on whole genome sequencing, WGS), phenotypic bedaquiline drug susceptibility testing (DST, MGIT), targeted deep sequencing and treatment monitoring during standardised treatment and a subsequent individualised bedaquilinecontaining regimen. Overall, eight isolates (A-H) collected 4.7 months after initiation of standard treatment regimen until 6 months after all TB treatment was stopped underwent WGS, targeted deep sequencing of Rv0678 and phenotypic bedaquiline DST. The patient was initially diagnosed with MDRTB with low-level isoniazid resistance using Genotype MTBDRplus, and treated with a standardised MDR-TB treatment regimen but remained culture positive. As per guidelines, subsequent isolates were phenotypically characterized for ofloxacin and amikacin susceptibility. Ofloxacin resistance was first noted 6 months after treatment initiation. All isolates remained susceptible to second-line injectables. At 8.1 months a revised regimen was initiated with the subsequent addition of bedaquiline (22 days after initiation of revised regimen) and withdrawal of pyrazinamide and ethambutol (2 months after initiation of revised regimen). Bedaquiline was administered for 6 months. The patient refused kanamycin at month 6 of the revised regimen for a duration of 2.4 months. The individualized regimen was continued until the outcome of treatment failure at 15 months. Phenotypic DST showed that all isolates with a variant frequency of >1% in Rv0678 were resistant to bedaquiline at 1µg/ml in MGIT.

. The patient refused kanamycin at month 6 of the revised regimen for a duration of 2.4 months. The individualized regimen was continued until the outcome of treatment failure at 15 months. Phenotypic DST showed that all isolates with a variant frequency of >1% in Rv0678 were resistant to bedaquiline at 1µg/ml in MGIT. Abbreviations: MDR-TB=multi-drug resistant tuberculosis; INH=isoniazid; Z=pyrazinamide; KAN=kanamycin; MXF=moxifloxacin; ETH=ethionamide; TZD=terizidone; hdIND=high dose isoniazid; KAN=kanamycin; LZD=linezolid; E=ethambutol; PAS=para-aminosalicyclic acid; BDQ=bedaquiline; WGS=whole genome sequencing; DST=drug susceptibility testing; ins=insertion; R=resistant; S=susceptible

INH=isoniazid; Z=pyrazinamide; KAN=kanamycin; MXF=moxifloxacin; ETH=ethionamide; TZD=terizidone; hdIND=high dose isoniazid; KAN=kanamycin; LZD=linezolid; E=ethambutol; PAS=para-aminosalicyclic acid; BDQ=bedaquiline; WGS=whole genome sequencing; DST=drug susceptibility testing; ins=insertion; R=resistant; S=susceptible Overall, eight M. tuberculosis isolates (A-H) underwent whole genome sequencing (WGS), targeted deep sequencing3 of Rv0678 and phenotypic bedaquiline resistance testing. WGS of isolate A collected 4.7 months after standard MDR-TB treatment initiation revealed a Beijing strain with mutations conferring resistance to rifampicin, isoniazid, ethambutol, ethionamide, fluoroquinolones, pyrazinamide and streptomycin (Figure 1). WGS of isolate C, collected 2 months after treatment revision, suggested that bedaquiline (to which the isolate was phenotypically susceptible) was added to a regimen with 5 potentially effective drugs. Targeted deep sequencing of isolate C showed a base pair insertion in Rv06784 at a variant frequency of 0.05% (position 192), which was not present in isolate B taken before bedaquiline treatment. Isolate D, collected after bedaquiline cessation, showed the presence of this insertion in >90% of the bacterial population. The frequency of the Rv0678 192 insertion decreased in subsequent isolates, but two different insertions in Rv0678 emerged (GA and G at position 138, isolates F and G, respectively). The G insertion at position 138 became fixed after all treatment was stopped (isolates G and H). Isolates D, E, F, G, and H were phenotypically resistant to bedaquiline.

insertion decreased in subsequent isolates, but two different insertions in Rv0678 emerged (GA and G at position 138, isolates F and G, respectively). The G insertion at position 138 became fixed after all treatment was stopped (isolates G and H). Isolates D, E, F, G, and H were phenotypically resistant to bedaquiline. This case demonstrates the emergence of bedaquiline resistance despite the presence of five potentially effective drugs and good adherence (based on clinical notes). The emergence of Rv0678 variants, after completion of 6 months of bedaquiline treatment, demonstrates the risk of resistance amplification after cessation of a drug with a long half-life (5.5 months for bedaquiline).5 Supplementary Material Click here for additional data file.

This case demonstrates the emergence of bedaquiline resistance despite the presence of five potentially effective drugs and good adherence (based on clinical notes). The emergence of Rv0678 variants, after completion of 6 months of bedaquiline treatment, demonstrates the risk of resistance amplification after cessation of a drug with a long half-life (5.5 months for bedaquiline).5 Supplementary Material Click here for additional data file. Acknowledgements This work is supported by the National Research Foundation, the South African Medical Research Council (SA MRC) and the Stellenbosch University Faculty of Medicine Health Sciences. The content is solely the responsibility of the authors and does not necessarily represent the official views of the SA MRC. The work is supported by National Institutes of Health (NIH) grant R01AI131939. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. HC acknowledges support from a Wellcome Trust fellowship (ref 099818/Z/12/Z). MdV, AvR and SL are supported by TORCH funding through the Flemish Fund for Scientific Research (FWO G0F8316N). SL acknowledges support by the Swiss National Science Foundation (P2BSP3_165379). GT acknowledges support from the EDCTP2 programme supported by the European Union (grant number SF1401 – OPTIMAL DIAGNOSIS). The sequencing work was performed at the Translational Genomics Research Institute and supported by the Bill & Melinda Gates Foundation Grant OPP1115887 (M Schito) for the ReSeqTB sequencing platform.

knowledges support from the EDCTP2 programme supported by the European Union (grant number SF1401 – OPTIMAL DIAGNOSIS). The sequencing work was performed at the Translational Genomics Research Institute and supported by the Bill & Melinda Gates Foundation Grant OPP1115887 (M Schito) for the ReSeqTB sequencing platform. Disclaimer Use of trade names is for identification only and does not constitute endorsement by the U.S. Department of Health and Human Services, the U.S. Public Health Service, or the Centers for Disease Control and Prevention. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention. Conflict of interest The authors declare that there is no conflict of interest regarding the publication of the manuscript.

Introduction In 2017, ~37 million people were living with HIV worldwide, with 1.8 million new infections.1 HIV incidence is declining worldwide, but is unlikely to reach the UNAIDS target of <500,000 new infections by 2020.2 Steep reductions in incidence are needed to curb the HIV/AIDS epidemic. Universal testing and treatment (UTT) has been proposed as an important component of HIV combination prevention programs.3,4 The HPTN 052 trial showed that early antiretroviral therapy (ART) initiation dramatically reduced HIV transmission among couples5,6, and the PARTNER study showed that viral suppression (<200 copies/ml) prevented HIV sexual transmission.7,8 Mathematical modeling predicted that HIV incidence would fall steeply if HIV testing were delivered throughout a population and ART initiated immediately after diagnosis.9-11 Early ART also confers individual health benefits.12,13 In 2015, the World Health Organization updated its guidelines recommending immediate ART for all HIV-positive individuals14, and UNAIDS proposed 90-90-90 HIV testing and treatment targets (by 2020: 90% of HIV-positive individuals should know their status; 90% of those individuals should be on ART; and 90% of those on ART should be virally suppressed).15

ion updated its guidelines recommending immediate ART for all HIV-positive individuals14, and UNAIDS proposed 90-90-90 HIV testing and treatment targets (by 2020: 90% of HIV-positive individuals should know their status; 90% of those individuals should be on ART; and 90% of those on ART should be virally suppressed).15 While there is compelling evidence supporting UTT for HIV prevention, it was not clear whether UTT could be implemented effectively at population level and impact HIV incidence. Four community-randomized trials (CRTs) in sub-Saharan Africa addressed these questions; two (TasP and SEARCH) reported no impact of UTT on HIV incidence; a third (Ya Tsie) reported a 30% reduction in incidence, of borderline statistical significance.16-18 The fourth study, HPTN 071 (PopART), is the largest HIV prevention trial ever conducted. Here, we present the primary results of HPTN 071 (PopART); we also describe the uptake of the PopART intervention and its impact on viral suppression.

orted a 30% reduction in incidence, of borderline statistical significance.16-18 The fourth study, HPTN 071 (PopART), is the largest HIV prevention trial ever conducted. Here, we present the primary results of HPTN 071 (PopART); we also describe the uptake of the PopART intervention and its impact on viral suppression. Methods The study was designed by members of the Study Team with input from the sponsor, funders and government and non-governmental partners in Zambia and South Africa, listed in the Acknowledgements. The data were collected by staff of Zambart and the Desmond Tutu TB Centre in collaboration with LSHTM and the HPTN Statistical and Data Management Center. The data were analyzed by the analytic authors identified at the beginning of the manuscript who vouch for the integrity of the analysis. All authors vouch for the integrity of the data, contributed to the preparation and review of the manuscript and agreed to its publication. The initial draft was written by the first author. The sponsor required no agreements restricting access to the data or freedom to publish the study findings. The study design has been described previously19 and is summarized below.

Methods The study was designed by members of the Study Team with input from the sponsor, funders and government and non-governmental partners in Zambia and South Africa, listed in the Acknowledgements. The data were collected by staff of Zambart and the Desmond Tutu TB Centre in collaboration with LSHTM and the HPTN Statistical and Data Management Center. The data were analyzed by the analytic authors identified at the beginning of the manuscript who vouch for the integrity of the analysis. All authors vouch for the integrity of the data, contributed to the preparation and review of the manuscript and agreed to its publication. The initial draft was written by the first author. The sponsor required no agreements restricting access to the data or freedom to publish the study findings. The study design has been described previously19 and is summarized below. Study population HPTN 071 (PopART) was conducted between 2013-2018, in 21 urban/peri-urban communities in Zambia and Western Cape Province, South Africa (total population ~1 million; average ~50,000/community). Each community was the catchment population of a government clinic. Communities were arranged in seven triplets matched on geographical location and estimated HIV prevalence. Communities in each triplet were randomly allocated to three study arms in simultaneous public ceremonies. Restricted randomization was used to ensure balance across study arms on population size, baseline ART coverage and HIV prevalence.19

n seven triplets matched on geographical location and estimated HIV prevalence. Communities in each triplet were randomly allocated to three study arms in simultaneous public ceremonies. Restricted randomization was used to ensure balance across study arms on population size, baseline ART coverage and HIV prevalence.19 The three study arms are shown in Figure S1. Arm A communities received the PopART intervention (see below) with universal ART. Arm B communities received the PopART intervention with ART provided according to local guidelines. Arm C communities did not receive the PopART intervention, but received standard-of-care at government clinics, including HIV testing and ART offered according to local guidelines. Intervention The PopART intervention, delivered to Arm A and B communities only, was a combination prevention package (Figure S2). Specially trained community health workers (Community HIV-care Providers, CHiPs) delivered services at annual household visits (see supplementary text). CHiPs worked in pairs, each pair responsible for ~500 households. Data collected by CHiPs were used primarily to support service delivery but also to evaluate intervention coverage.

ned community health workers (Community HIV-care Providers, CHiPs) delivered services at annual household visits (see supplementary text). CHiPs worked in pairs, each pair responsible for ~500 households. Data collected by CHiPs were used primarily to support service delivery but also to evaluate intervention coverage. At each visit, CHiPs offered HIV counseling and rapid testing, and provided support for linkage to care and ART adherence for HIV-positive clients. They referred uncircumcised HIV-negative men for voluntary medical male circumcision and HIV-positive pregnant women for antenatal care including prevention of mother-to-child HIV transmission. CHiPs also screened clients for symptoms of tuberculosis and sexually transmitted infections, with referral for diagnosis and treatment, and promoted and provided condoms. In all 21 communities, HIV care and ART were provided at local government clinics. In Arm A, these clinics offered ART irrespective of CD4 count throughout the trial, with written consent for those initiating ART outside of local guidelines until universal ART became standard. In Arms B and C, the clinics provided ART initially at a CD4 threshold of 350 cells/ml, which increased to 500 cells/ml in 2014. Universal ART was offered from April 2016 (Zambia) and October 2016 (S Africa) (Figure S3).

n consent for those initiating ART outside of local guidelines until universal ART became standard. In Arms B and C, the clinics provided ART initially at a CD4 threshold of 350 cells/ml, which increased to 500 cells/ml in 2014. Universal ART was offered from April 2016 (Zambia) and October 2016 (S Africa) (Figure S3). Outcome evaluation The effect of the intervention on population-level HIV incidence was measured in a Population Cohort (PC) (enrolled December 2013-March 2015) that included one randomly-selected adult aged 18-44 years from a random sample of households in each community (Figure S4). PC participants were surveyed at baseline (PC0) and after 12, 24 and 36 months (PC12/PC24/PC36). Because the original enrollment target (2,500 adults/community) was not reached in PC0, additional participants were enrolled at 12 months (PC12N) and in arms A and C only at 24 months (PC24N), excluding households sampled previously. At each visit, PC participants were interviewed by a field research assistant (separate from the CHiPs) using a structured questionnaire that included collection of demographic, socio-economic and behavioral data, as well as data related to HIV prevention, diagnosis and treatment; data were collected electronically. Following the interview, blood was collected by a research nurse, who also offered HIV rapid testing to all PC participants.

questionnaire that included collection of demographic, socio-economic and behavioral data, as well as data related to HIV prevention, diagnosis and treatment; data were collected electronically. Following the interview, blood was collected by a research nurse, who also offered HIV rapid testing to all PC participants. The pre-defined primary study outcome was HIV incidence between PC12 and PC36, comparing Arm A and Arm B to Arm C. This approach provided one year to fully establish the study intervention before measuring study outcomes. Other outcomes reported here include viral suppression (VS, < 400 copies HIV RNA/ml) and the estimated coverage of HIV testing and ART based on CHiPs data from Arms A and B. Laboratory methods Laboratory-based HIV testing was performed for all PC participants at all visits. Central laboratories in South Africa and Zambia performed a single 4th generation HIV test. The HPTN Laboratory Center (LC, Baltimore, MD USA) performed additional testing to determine HIV status (see supplementary text). If seroconversion was confirmed, testing was performed to determine if the participant had acute infection at the prior visit. HIV viral load testing was performed at the HPTN LC for selected samples: all HIV-positive participants at PC24, and a random subset of ~75 HIV-positive participants per community at PC0, PC12 and PC36. HIV Viral load testing was performed using the Abbott Realtime HIV-1 assay (Abbott Molecular Inc, Des Plaines, Il) utilizing a <400 HIV RNA/ml threshold.

at the HPTN LC for selected samples: all HIV-positive participants at PC24, and a random subset of ~75 HIV-positive participants per community at PC0, PC12 and PC36. HIV Viral load testing was performed using the Abbott Realtime HIV-1 assay (Abbott Molecular Inc, Des Plaines, Il) utilizing a <400 HIV RNA/ml threshold. Statistical considerations Sample size calculations were informed by initial projections of intervention effect based on mathematical modeling19,20 which suggested that HIV incidence might be reduced by up to 60% in Arm A and 25% in Arm B, compared with Arm C. Assuming HIV incidence in Arm C of 1.0 to 1.5 per 100 person-years (py), a between-community coefficient of variation (k) within matched triplets of 0.15-0.20, 2,500 PC participants/community with 85% HIV-negative at baseline, and 25% loss to follow-up over three years, study power would exceed 75% or 85% for effects of 35% or 40%, respectively.

rm C of 1.0 to 1.5 per 100 person-years (py), a between-community coefficient of variation (k) within matched triplets of 0.15-0.20, 2,500 PC participants/community with 85% HIV-negative at baseline, and 25% loss to follow-up over three years, study power would exceed 75% or 85% for effects of 35% or 40%, respectively. Analysis methods are described in detail in the Statistical Analysis Plan, completed before data unblinding.21 Briefly, HIV incidence was measured in PC participants who were HIV-negative at enrollment; HIV infection was assumed to occur at the mid-point between the last HIV-negative and the first HIV-positive sample, or at a visit where acute infection was identified. When timing of infection was unclear because of missed visits, the time of infection was estimated by imputation (see supplementary text and Table S1). For the primary outcome, statistical inference used a two-stage approach recommended for CRTs with <15 clusters/arm.22,23 At the first stage, Poisson regression on data from all three study arms was used to compute E, the expected number of events (incident HIV infections) in each community, after adjusting for age, sex and baseline HIV prevalence, assuming a null intervention effect. At the second stage, two-way analysis of variance was carried out on log(O/E) (log ratio-residuals), where O was the observed number of events in each community, with matched triplet and study arm as factors. The test statistic is the estimated difference in means of log(O/E) between study arms, with two-sided p-values and 95% confidence intervals (CI) computed using the t-distribution. The corresponding rate ratios and 95%CI for the comparison of Arms A and C, and Arms B and C, were calculated with exponentiation. Similar methods were used for the analysis of viral suppression, except that logistic regression was used at the first stage without adjustment for HIV prevalence. The robustness of the above analyses was assessed using a permutation test based on the restricted randomization scheme.

ere calculated with exponentiation. Similar methods were used for the analysis of viral suppression, except that logistic regression was used at the first stage without adjustment for HIV prevalence. The robustness of the above analyses was assessed using a permutation test based on the restricted randomization scheme. Because the analysis plan did not include a method for controlling type I error when conducting treatment comparisons for subgroup and post-hoc analyses, treatment effects are reported with point estimates and 95% confidence intervals (which have not been adjusted for multiplicity and should not be used to infer treatment effects). In Arms A and B, CHiPs data were used to estimate the proportion of HIV-positive community members who knew their HIV status and were on ART, using methods and assumptions described in supplementary text. Ethical considerations PC participants provided written informed consent before enrollment. Community members visited by CHiPs provided verbal consent for participation in the intervention and data collection. In Arm A, clinic patients provided written informed consent when ART was initiated outside of prevailing local guidelines (2013-2016). Ethical approval for the study was granted by ethics committees at the London School of Hygiene and Tropical Medicine, University of Zambia, and Stellenbosch University.

Ethical considerations PC participants provided written informed consent before enrollment. Community members visited by CHiPs provided verbal consent for participation in the intervention and data collection. In Arm A, clinic patients provided written informed consent when ART was initiated outside of prevailing local guidelines (2013-2016). Ethical approval for the study was granted by ethics committees at the London School of Hygiene and Tropical Medicine, University of Zambia, and Stellenbosch University. Results Enrollment and follow-up The CONSORT diagram (Figure 1) shows the enrollment and follow-up of PC participants; 38,474 adults were enrolled at baseline (PC0), with 5,014 and 4,813 additional enrollments at PC12N and PC24N, respectively (total enrolled: 48,301). At PC12, and again at PC24, 13% of PC participants were terminated from the study, most because of confirmed permanent relocation out of the study community (Table S2), and were censored from further observation; ~75% of remaining participants completed each visit. The final survey (PC36) reached 72% of eligible participants. Retention was similar across study arms at PC36 (73%, 73% and 71% in Arms A, B and C, respectively). Figure 1 CONSORT diagram showing enrollment and follow-up of the Population Cohort.

At PC12, and again at PC24, 13% of PC participants were terminated from the study, most because of confirmed permanent relocation out of the study community (Table S2), and were censored from further observation; ~75% of remaining participants completed each visit. The final survey (PC36) reached 72% of eligible participants. Retention was similar across study arms at PC36 (73%, 73% and 71% in Arms A, B and C, respectively). Figure 1 CONSORT diagram showing enrollment and follow-up of the Population Cohort. HPTN 071 (PopART) included 21 communities that were matched in seven sets of three communities each; the three communities in each triplet were randomized to Study Arms A, B, and C. The purpose of the Population Cohort (PC) was to enrol and follow a representative sample of residents to assess the impact of the PopART intervention on HIV incidence and viral suppression. Participants in the PC were enrolled from randomly selected households in the community; with one member aged 18-44 selected at random for eligibility assessment. The diagram shows the number of participants enrolled at the start of the study (PC0). Additional participants were enrolled in PC12N in communities with fewer than 2000 PC0 participants; additional participants were enrolled in Arms A and C in PC24N to preserve power for this comparison. The status of participants at each survey year (PC12, PC24, PC36) is reported. Individuals who missed yearly follow-up visits were eligible for subsequent annual surveys, individuals who were terminated were not. The percentage retained is the proportion of participants who completed a visit amongst those eligible for the visit.

of participants at each survey year (PC12, PC24, PC36) is reported. Individuals who missed yearly follow-up visits were eligible for subsequent annual surveys, individuals who were terminated were not. The percentage retained is the proportion of participants who completed a visit amongst those eligible for the visit. Baseline comparisons More women (71%) than men (29%) were enrolled in PC0, with 40% of participants aged <24 years (Table 1). Socio-demographic and behavioral characteristics were similar across study arms. Approximately 17% of men reported having undergone medical circumcision. Baseline HIV prevalence was 22% (women: 26%, men: 12%) and baseline Herpes simplex virus type-2 (HSV-2) prevalence was 46% (women: 54%, men: 24%). The prevalence of both infections was similar across study arms (HIV: 21% Arm A, 21% Arm B, 22% Arm C; HSV-2: 46% Arm A; 46% Arm B, 45% Arm C). Reported ART coverage was slightly higher in Arm B (33% Arm A, 41% Arm B, 35% Arm C), but the proportion of HIV-positive participants with VS at PC0 was similar across study arms (56% Arm A, 57% Arm B, 54% Arm C). Table 1 Characteristics of population cohort at baseline (PC0) The table shows baseline characteristics of the PC in the three study arms. The table is restricted to PC participants enrolled at PC0. Data are pooled across all seven communities in each study arm.

Baseline HIV prevalence was 22% (women: 26%, men: 12%) and baseline Herpes simplex virus type-2 (HSV-2) prevalence was 46% (women: 54%, men: 24%). The prevalence of both infections was similar across study arms (HIV: 21% Arm A, 21% Arm B, 22% Arm C; HSV-2: 46% Arm A; 46% Arm B, 45% Arm C). Reported ART coverage was slightly higher in Arm B (33% Arm A, 41% Arm B, 35% Arm C), but the proportion of HIV-positive participants with VS at PC0 was similar across study arms (56% Arm A, 57% Arm B, 54% Arm C). Table 1 Characteristics of population cohort at baseline (PC0) The table shows baseline characteristics of the PC in the three study arms. The table is restricted to PC participants enrolled at PC0. Data are pooled across all seven communities in each study arm. Baseline Variable Arm A Arm B Arm C Total enrolled (PC0) 12671 13404 12399 Sex Male 3595 (28%) 3906 (29%) 3701 (30%) Female 9042 (72%) 9458 (71%) 8639 (70%) Missing 34 40 59 Age (years) 18-24 5065 (40%) 5179 (39%) 4981 (40%) 25-34 4928 (39%) 5170 (39%) 4688 (38%) 35-44 2643 (21%) 3015 (23%) 2667 (22%) Missing 35 40 63 Marital status Married/living as married 5363 (43%) 5210 (39%) 4693 (38%) Never married 6292 (50%) 6923 (52%) 6644 (54%) Divorced/separated 708 (6%) 892 (7%) 656 (5%) Widowed 197 (2%) 208 (2%) 206 (2%) Missing 111 171 200 Nights spent away from community (past 3m) None 11623 (94%) 10650 (87%) 10864 (89%) 1-7 556 (4%) 890 (7%) 886 (7%) 8-14 97 (1%) 228 (2%) 178 (1%) 15+ 149 (1%) 418 (3%) 245 (2%) Missing 246 1218 226 Number of sexual partners (past 12m) 0 3160 (27%) 4266 (33%) 3188 (27%) 1 8032 (68%) 7663 (60%) 7913 (66%) 2-4 496 (4%) 753 (6%) 722 (6%) 5+ 70 (1%) 121 (1%) 81 (1%) Missing 913 601 495 Male circumcision (self-report)1 Not circumcised 1725 (51%) 1974 (53%) 1904 (55%) Medical 567 (17%) 613 (16%) 646 (19%) Traditional 1113 (33%) 1171 (31%) 895 (26%) Missing 190 148 256 ART coverage2 Yes 788 (33%) 1048 (41%) 878 (35%) No 1587 (67%) 1534 (59%) 1648 (65%) Missing 208 152 161 HIV prevalence Negative 9594 (79%) 10235 (79%) 9301 (78%) Positive 2583 (21%) 2734 (21%) 2687 (22%) Not determined3 494 435 411 HSV-2 prevalence Negative 6506 (53%) 7005 (54%) 6585 (55%) Positive 5667 (46%) 5959 (46%) 5357 (45%) Indeterminate 64 (1%) 55 (<1%) 74 (1%) Not determined4 434 385 383 HIV viral suppression5 Yes 295 (56%) 300 (57%) 267 (54%) No 228 (44%) 225 (43%) 227 (46%) 1 For male circumcision the denominator is the number of men.

HSV-2 prevalence Negative 6506 (53%) 7005 (54%) 6585 (55%) Positive 5667 (46%) 5959 (46%) 5357 (45%) Indeterminate 64 (1%) 55 (<1%) 74 (1%) Not determined4 434 385 383 HIV viral suppression5 Yes 295 (56%) 300 (57%) 267 (54%) No 228 (44%) 225 (43%) 227 (46%) 1 For male circumcision the denominator is the number of men. 2 ART coverage is the proportion of HIV-positive participants self-reporting current ART use. The denominator is the number of HIV-positive participants. 3 HIV status not determined occurred when a participant did not consent to specimen collection, no sample was available or when lab testing did not result in a determination of infection status. 4 HSV-2 status not determined occurred when a participant did not consent to specimen collection, or no sample was available. 5 Viral suppression was assessed in a random sample of ~75 HIV-positive participants per community in PC0. Missing data are excluded from % calculations which are based on data pooled across communities. Baseline comparisons between arms include only PC0 participants as this best represents the balance between arms in the communities prior to the delivery of the intervention. Impact of the intervention on HIV incidence Estimated effects of the intervention on HIV incidence are shown in Table 2 and Figure 2. Table 2 Effect of PopART intervention on HIV incidence and HIV viral suppression

Baseline comparisons between arms include only PC0 participants as this best represents the balance between arms in the communities prior to the delivery of the intervention. Impact of the intervention on HIV incidence Estimated effects of the intervention on HIV incidence are shown in Table 2 and Figure 2. Table 2 Effect of PopART intervention on HIV incidence and HIV viral suppression The table shows the HIV incidence rate between PC12 and PC36 and proportion of HIV-positive participants with viral suppression at PC24 for each triplet and overall, and for men and women, in each study arm. The table also shows the unadjusted and adjusted rate ratios for incidence and viral suppression overall, and for men and women. Viral suppression was defined as HIV viral load <400 copies/mL.

IV-positive participants with viral suppression at PC24 for each triplet and overall, and for men and women, in each study arm. The table also shows the unadjusted and adjusted rate ratios for incidence and viral suppression overall, and for men and women. Viral suppression was defined as HIV viral load <400 copies/mL. Outcome Arm A Arm B Arm C HIV incidence rate (PC12-PC36) No. of events/total person-years (rate per 100 person-years)1 Triplet 1 28/1687 (1.64) 19/1979 (0.94) 24/2054 (1.17) Triplet 2 33/2086 (1.57) 29/2408 (1.20) 33/2262 (1.48) Triplet 3 23/1695 (1.36) 22/1687 (1.30) 29/1811 (1.63) Triplet 4 41/2013 (2.04) 19/1698 (1.13) 37/1561 (2.39) Triplet 5 36/1507 (2.35) 33/1811 (1.80) 28/1304 (2.15) Triplet 6 26/1808 (1.43) 26/2078 (1.24) 32/1375 (2.31) Triplet 7 13/2195 (0.57) 10/2488 (0.40) 14/2195 (0.59) Overall IR2 198/12990 (1.45) 157/14149 (1.06) 198/12563 (1.55) Arm A vs Arm C Arm B vs Arm C Unadjusted rate ratio (95% CI) 0.94 (0.77, 1.15) 0.68 (0.56, 0.84) 1 P value3 0.505 0.002 Adjusted rate ratio4 (95% CI) 0.93 (0.74, 1.18) 0.70 (0.55, 0.88) 1 P value5 0.509 0.006 Men Overall IR2 36/3766 (0.77) 23/4301 (0.45) 39/4115 (0.92) Adjusted rate ratio4 (95% CI) 0.88 (0.41, 1.88) 0.52 (0.24, 1.12) 1 Women Overall IR2 162/9225 (1.71) 134/9848 (1.26) 159/8448 (1.79) Adjusted rate ratio4 (95% CI) 0.96 (0.72, 1.28) 0.73 (0.55, 0.97) 1 P value for interaction by sex 0.794 0.401 Arm A Arm B Arm C Viral suppression (PC24) No. VS/total no.

39/4115 (0.92) Adjusted rate ratio4 (95% CI) 0.88 (0.41, 1.88) 0.52 (0.24, 1.12) 1 Women Overall IR2 162/9225 (1.71) 134/9848 (1.26) 159/8448 (1.79) Adjusted rate ratio4 (95% CI) 0.96 (0.72, 1.28) 0.73 (0.55, 0.97) 1 P value for interaction by sex 0.794 0.401 Arm A Arm B Arm C Viral suppression (PC24) No. VS/total no. HIV-positive (%) Triplet 1 140/175 (80.0%) 183/244 (75.0%) 212/290 (73.1%) Triplet 2 204/311 (65.6%) 276/371 (74.4%) 179/271 (66.1 %) Triplet 3 225/295 (76.3%) 177/255 (69.4%) 174/284 (61.3%) Triplet 4 356/518 (68.7%) 219/324 (67.6%) 354/476 (74.4%) Triplet 5 270/389 (69.4%) 275/381 (72.2%) 211/315 (67.0%) Triplet 6 250/355 (70.4%) 126/202 (62.4%) 338/506 (66.8%) Triplet 7 86/116 (74.1%) 62/114 (54.4%) 12/41 (29.3%) Overall prevalence6 1531/2159 (71.9%) 1318/1891 (67.5%) 1480/2183 (60.2%) Arm A vs Arm C Arm B vs Arm C Unadjusted VS prevalence ratio (95% CI) 1.19 (0.97, 1.47) 1.12 (0.91, 1.38) 1 P value3 0.090 0.258 Adjusted VS prevalence ratio7 (95% CI) 1.16 (0.99, 1.36) 1.08 (0.92, 1.27) 1 P value3 0.071 0.297 Men Overall prevalence6 183/294 (63.0%) 153/244 (60.8%) 179/330 (40.0%) Adjusted VS prevalence ratio7 (95% CI) 1.46 (0.86, 2.48) 1.41 (0.83, 2.41) 1 Women Overall prevalence6 1348/1865 (73.3%) 1165/1647 (68.4%) 1301/1853 (65.8%) Adjusted VS prevalence ratio7 (95% CI) 1.10 (1.00, 1.22) 1.03 (0.93, 1.13) 1 P value for interaction by sex 0.220 0.164 Abbreviations: IR = incidence rate; VS = viral suppression (<400 copies/mL).

% CI) 1.46 (0.86, 2.48) 1.41 (0.83, 2.41) 1 Women Overall prevalence6 1348/1865 (73.3%) 1165/1647 (68.4%) 1301/1853 (65.8%) Adjusted VS prevalence ratio7 (95% CI) 1.10 (1.00, 1.22) 1.03 (0.93, 1.13) 1 P value for interaction by sex 0.220 0.164 Abbreviations: IR = incidence rate; VS = viral suppression (<400 copies/mL). 1 Imputation was used to estimate missing timing of HIV infection in seroconverting participants who missed PC12 or PC24 (See supplementary material) 2 Overall IR is geometric mean of individual community IR 3 P-value compared to t-distribution with 12 degrees of freedom. 4 Adjusted for age, sex, baseline HIV prevalence 5 P-value compared to t-distribution with 11 degrees of freedom. 6 Overall prevalence is geometric mean of individual community proportions with viral suppression 7 Adjusted for age, sex Figure 2 Estimates of HIV incidence and log ratio-residuals for the seven study triplets. The plots show estimates of HIV incidence (plotted per 100 person-years upper panels) and log ratio-residuals (observed/expected HIV infections adjusted for age, sex and baseline HIV prevalence, lower panels) for Arm A vs. Arm C and Arm B vs. Arm C. Data are shown for the study period included in the primary endpoint analysis (PC12 to PC36). Colored lines represent each of the seven triplets (numbered 1 to 7). For HIV incidence, the size of the colored dot at the end of each line represents the number of events contributing to the incidence estimate for each community. Abbreviations: Z: Zambia; SA: South Africa.

The plots show estimates of HIV incidence (plotted per 100 person-years upper panels) and log ratio-residuals (observed/expected HIV infections adjusted for age, sex and baseline HIV prevalence, lower panels) for Arm A vs. Arm C and Arm B vs. Arm C. Data are shown for the study period included in the primary endpoint analysis (PC12 to PC36). Colored lines represent each of the seven triplets (numbered 1 to 7). For HIV incidence, the size of the colored dot at the end of each line represents the number of events contributing to the incidence estimate for each community. Abbreviations: Z: Zambia; SA: South Africa. Between PC12 and PC36 (primary outcome), 553 incident HIV infections were observed during 39,702py follow-up (1.4/100py; women: 1.7/100py; men: 0.8/100py). Incidence in Arm C (geometric mean across communities) was 1.6/100py overall (Table 2). Incidence in Arm A was 1.5/100py; the adjusted rate ratio (AdjRR) compared with Arm C was 0.93 (95%CI: 0.74-1.18, p=0.51). Incidence in Arm B was 1.1/100py; the AdjRR compared with Arm C was 0.70 (95%CI: 0.55-0.88, p=0.006). HIV incidence was lower in Arm B vs. Arm C in all seven matched triplets, while incidence was lower in Arm A vs. Arm C in only four triplets (Figure 2). A permutation test based on the restricted randomization scheme showed even stronger evidence of an effect in Arm B vs. Arm C (p=0.001), but not in Arm A vs. Arm C (p=0.48). The findings were essentially similar when the analysis was restricted to PC participants enrolled at PC0 (Table S6).

ly four triplets (Figure 2). A permutation test based on the restricted randomization scheme showed even stronger evidence of an effect in Arm B vs. Arm C (p=0.001), but not in Arm A vs. Arm C (p=0.48). The findings were essentially similar when the analysis was restricted to PC participants enrolled at PC0 (Table S6). In Arm B vs. Arm C, subgroup analyses by sex (Table 2) and age and sex (Table S3) showed a greater effect on HIV incidence in men (AdjRR: 0.52, 95%CI: 0.24-1.12) than women (AdjRR: 0.73, 95%CI: 0.55-0.97), although this difference in effect could have occurred by chance (p for interaction = 0.40); there was also evidence of a greater effect in older participants (aged 25+; AdjRR: 0.58, 95%CI: 0.43-0.76) than younger participants (18-24y; AdjRR: 0.92, 95%CI: 0.70-1.20; p for interaction = 0.044). HIV incidence and estimated effects for individual years of follow-up, and for the entire study period (PC0-PC36) are shown in Tables S4 and S5. HIV incidence decreased in Arm C by 12% (95%CI:0%-23%) per year (Figure S5).

.76) than younger participants (18-24y; AdjRR: 0.92, 95%CI: 0.70-1.20; p for interaction = 0.044). HIV incidence and estimated effects for individual years of follow-up, and for the entire study period (PC0-PC36) are shown in Tables S4 and S5. HIV incidence decreased in Arm C by 12% (95%CI:0%-23%) per year (Figure S5). Impact of the intervention on viral suppression Proportions of HIV-positive PC24 participants with VS were 71.9% in Arm A, 67.5% in Arm B and 60.2% in Arm C (Table 2). The adjusted VS prevalence ratios were 1.16 (95%CI: 0.99-1.36, p=0.07) for Arm A vs. Arm C and 1.08 (95%CI: 0.92-1.27, p=0.30) for Arm B vs. Arm C. In Arms A and B, VS at PC24 was higher in women than in men, and considerably higher in those aged ≥25 years than those aged 18-24 years (Table 2 and Table S7). VS in Arms A and B increased steeply from ~55% at PC0 to ~75% at PC36 (Table S8). VS in participants who self-reported ART use was consistently high in Arms A and B (86-91%, Table S9). Coverage of the intervention Based on CHiPs data, the estimated proportions of all HIV-positive adults who were on ART at the end of the study were 81% in Arm A and 80% in Arm B (Table S10). Figure 3 shows estimated ART coverage by age and sex, indicating similar coverage in Arms A and B, lower coverage in men than women, and lower coverage among younger compared with older individuals. ART coverage was also similar in Arms A and B in most triplets (Figure S6).

re 81% in Arm A and 80% in Arm B (Table S10). Figure 3 shows estimated ART coverage by age and sex, indicating similar coverage in Arms A and B, lower coverage in men than women, and lower coverage among younger compared with older individuals. ART coverage was also similar in Arms A and B in most triplets (Figure S6). Figure 3 Estimated ART coverage at the end of the study, by age and sex and study arm; estimated from the CHiPs data and extrapolated to total population aged ≥15 years The plot shows estimated ART coverage among the total population aged ≥15 years in Arm A and B communities at the end of the study, by sex, age-group and study arm. Coverage estimates are shown in black solid lines for Arm A and in blue dashed lines for Arm B. Lines for men are shown with a square symbol, and for women with a circle symbol. The UNAIDS 90-90-90 target for ART coverage (81%) is shown in red. The estimated number of HIV-positive men who were resident in the community at the time that CHIPs first visited their household during the third (and last) annual round of intervention, and remained resident in the study community at the end of the study, was 8,388 in Arm A and 8,948 in Arm B, and the estimated number of HIV-positive women was 15,936 in Arm A and 17,586 in Arm B. The estimated number of HIV-positive men on ART was 6,286 in Arm A and 6,378 in Arm B, and the estimated number of HIV-positive women on ART was 13,600 in Arm A and 14,481 in Arm B.

the study, was 8,388 in Arm A and 8,948 in Arm B, and the estimated number of HIV-positive women was 15,936 in Arm A and 17,586 in Arm B. The estimated number of HIV-positive men on ART was 6,286 in Arm A and 6,378 in Arm B, and the estimated number of HIV-positive women on ART was 13,600 in Arm A and 14,481 in Arm B. Discussion This study provides evidence that UTT can reduce HIV incidence at population level. In Arm B, HIV incidence was reduced by 30% compared to the standard-of-care control arm; surprisingly, there was no evidence of such an effect in Arm A.

the study, was 8,388 in Arm A and 8,948 in Arm B, and the estimated number of HIV-positive women was 15,936 in Arm A and 17,586 in Arm B. The estimated number of HIV-positive men on ART was 6,286 in Arm A and 6,378 in Arm B, and the estimated number of HIV-positive women on ART was 13,600 in Arm A and 14,481 in Arm B. Discussion This study provides evidence that UTT can reduce HIV incidence at population level. In Arm B, HIV incidence was reduced by 30% compared to the standard-of-care control arm; surprisingly, there was no evidence of such an effect in Arm A. The Arm B effect was consistent with pre-study model projections and was observed in both countries.20 Reduction in incidence was seen in all seven matched triplets in Arm B; this effect was very unlikely to have occurred by chance. UTT is hypothesized to reduce HIV transmission by increasing the proportion of HIV-positive community members who know their HIV status, the proportion of those individuals who are on ART, and the proportion of those on ART who are virally suppressed. Data from this study indicate that the UNAIDS 90-90-90 targets were achieved by the end of the 3-year intervention in both Arm A and B communities. High levels of VS were observed among HIV-positive PC participants after 24m (~72%, increased from the baseline level of ~55%). This corresponds to a ~35% drop in the proportion of HIV-positive participants not virally suppressed, from ~45% to ~28%, consistent with the observed 30% reduction in HIV incidence in Arm B. The greater reduction in HIV incidence among men likely reflects greater uptake of the intervention and higher VS in women (thus protecting their male partners); a similar explanation applies for higher effectiveness in those aged over 25 years.

28%, consistent with the observed 30% reduction in HIV incidence in Arm B. The greater reduction in HIV incidence among men likely reflects greater uptake of the intervention and higher VS in women (thus protecting their male partners); a similar explanation applies for higher effectiveness in those aged over 25 years. There are several possible explanations for the lack of an effect on HIV incidence when the PopART intervention was combined with universal ART (Arm A vs. Arm C). First, written informed consent was required for initiation of ART outside local guidelines from the start of the trial until 2016 (see supplementary text). This requirement for “research consent” may inadvertently have discouraged ART initiation, although this is not supported by data that show similar ART coverage and VS in Arms A and B. Second, wide-scale ART delivery in Arm A may have led to sexual disinhibition or de-emphasis of primary prevention messaging by CHiPs, offsetting the observed increase in VS. Data on self-reported risk behaviors do not support this hypothesis; further analyses are planned once data on HSV-2 seroconversion (a proxy for sexual risk behavior) become available. Third, while the three study arms appeared well matched with respect to baseline data, there may have been unrecognized differences across triplets in socio-demographic or other factors, such as mobility and migration resulting in exposure to HIV-positive partners from other communities.

isk behavior) become available. Third, while the three study arms appeared well matched with respect to baseline data, there may have been unrecognized differences across triplets in socio-demographic or other factors, such as mobility and migration resulting in exposure to HIV-positive partners from other communities. While these urban communities had high mobility, analysis of available data do not suggest any appreciable differences in migration across study arms. Further analyses of qualitative and quantitative data from the study communities, and data from an ongoing phylogenetic study, may shed light on the unexpected Arm A result. Strengths of the study included the large sample size, enrollment of a randomly-sampled cohort to measure HIV incidence and VS at community level, delivery of ART through routine services at government clinics, the availability of extensive process data used to improve and refine delivery of the intervention, and strong community engagement. While our study communities were not chosen to be representative of Zambia or South Africa as a whole, conduct of the study in large urban communities with high rates of mobility should increase the generalizability of the findings to other urban areas of Southern Africa with generalized HIV epidemics.

nity engagement. While our study communities were not chosen to be representative of Zambia or South Africa as a whole, conduct of the study in large urban communities with high rates of mobility should increase the generalizability of the findings to other urban areas of Southern Africa with generalized HIV epidemics. A limitation of the study was the relatively small number of randomized communities (seven/arm). The difference in observed effects in Arm A vs C and B vs C may thus be a chance finding, given the similar levels of ART coverage and VS in Arms A and B, and the similar nature of the Arm A and B interventions during most of the primary analysis period. To evaluate the overall effect of the PopART intervention vs standard of care, we therefore conducted a post-hoc analysis combining Arms A and B and found an estimated rate ratio of 0.81 (95%CI:0.66-0.99) compared with Arm C, consistent with a 20% reduction in incidence. Another limitation is that data on uptake of interventions among HIV-positive participants in the PC may be subject to a Hawthorne effect, because participants had regular contact with research staff offering HIV testing and providing referral to care. We would expect the Hawthorne effect to be greatest in Arm C, where PC participants did not have access to CHiP services for testing and referral. Thus, for uptake estimates we rely mainly on intervention data, which were only available from Arm A and B communities. Lastly, men were under-represented in the PC, and a substantial number of PC participants moved out of the community during follow-up and were censored from further observation. Thus, we cannot rule out selection bias although there was no evidence that these factors differed between study arms. Since men were under-represented in the PC, and a greater effect of the intervention was observed in men, the population-level effect may have been underestimated.

re censored from further observation. Thus, we cannot rule out selection bias although there was no evidence that these factors differed between study arms. Since men were under-represented in the PC, and a greater effect of the intervention was observed in men, the population-level effect may have been underestimated. The results of HPTN 071 (PopART) are consistent with programmatic and survey data24-27, and should be considered alongside those of the other three trials that measured the effect of UTT on HIV incidence in Africa, all of which were smaller and undertaken in largely rural communities. The TasP trial16 found no effect on HIV incidence, which may have reflected the similar HIV testing services provided in the intervention and control arms, with low levels of ART coverage in both arms. The SEARCH trial17 also found no effect on HIV incidence, which may have reflected intensive baseline community-based HIV testing in both intervention and control arms. The Ya Tsie trial18 observed a 30% reduction in incidence, which was of borderline statistical significance given the relatively small numbers of HIV seroconversion events.

so found no effect on HIV incidence, which may have reflected intensive baseline community-based HIV testing in both intervention and control arms. The Ya Tsie trial18 observed a 30% reduction in incidence, which was of borderline statistical significance given the relatively small numbers of HIV seroconversion events. Our finding of a 20-30% reduction in HIV incidence at population level was measured against a background of decreasing incidence in Arm C, possibly attributable to gradually increasing coverage of ART in the general population. This indicates that combination prevention including UTT can make a substantial contribution to HIV epidemic control. Importantly, the effects seen in our study, Ya Tsie study and others28 were achieved by delivering intensive household-based HIV-testing services; this may have played a more important role than changes in ART guidelines. The universal “test” component of a “test-and-treat” strategy is vital, as is continued attention to primary HIV prevention interventions.29,30 Results from planned cost-effectiveness and modeling studies will provide information on the value-for-money and long-term impact of such strategies which will help to inform policy and practice. ART coverage data from HPTN 071 (PopART), like data from other studies, draws special attention to the challenges in achieving ART coverage targets in young people, men, and communities with high mobility.31-33 If HIV transmission is concentrated in these subgroups, impact of UTT on HIV transmission may be compromised. Special efforts will be needed to address these coverage gaps to realize the full impact of UTT on HIV epidemic control.

eving ART coverage targets in young people, men, and communities with high mobility.31-33 If HIV transmission is concentrated in these subgroups, impact of UTT on HIV transmission may be compromised. Special efforts will be needed to address these coverage gaps to realize the full impact of UTT on HIV epidemic control. Supplementary Material Click here for additional data file. Acknowledgements HPTN 071 (PopART) is sponsored by the National Institute of Allergy and Infectious Diseases (NIAID) under Cooperative Agreements UM1-AI068619, UM1-AI068617, and UM1-AI068613, with funding from the U.S. President's Emergency Plan for AIDS Relief (PEPFAR). Additional funding is provided by the International Initiative for Impact Evaluation (3ie) with support from the Bill & Melinda Gates Foundation, as well as by NIAID, the National Institute on Drug Abuse (NIDA) and the National Institute of Mental Health (NIMH), all part of the U.S. National Institutes of Health (NIH). RH, SFl, KSa and DM are jointly funded by the UK Medical Research Council (MRC) and the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement, which is also part of the EDCTP2 programme supported by the European Union. Grant Ref: MR/R010161/1 KH acknowledges Centre funding from the UK Medical Research Council and Department for International Development, MRC Centre for Global Infectious Disease Analysis, reference R/R015600/1, and from the National Institute for Health Research Health Protection Research Unit (NIHR HPRU) with Public Health England, Grant/Award Number: HPRU‐2012‐10080.

ng from the UK Medical Research Council and Department for International Development, MRC Centre for Global Infectious Disease Analysis, reference R/R015600/1, and from the National Institute for Health Research Health Protection Research Unit (NIHR HPRU) with Public Health England, Grant/Award Number: HPRU‐2012‐10080. SF acknowledges funding from the Imperial College National Institute for Health Research Biomedical Research Centre We wish to acknowledge partners in South Africa including PEPFAR partners (Kheth’Impilo, ANOVA Healthcare and the SACTWU Worker Health Program) and City of Cape Town and Western Cape Government department of health colleagues who have worked to implement the HPTN 071 (PopART) trial activities, as well as partners in Zambia including the Zambian Ministry of Health, CIDRZ, ZPCT II and JSI. The team further acknowledges the work of the administrative and support teams at the institutions involved in this trial and the hundreds of field staff who delivered the intervention and collected the research data. The team extends its sincere appreciation to the study’s Community Advisory Boards (CABs), in-country Trial Steering/Management Committees, International Advisory Group and Data and Safety Monitoring Board for their oversight and consultation during study conduct. The team thanks all the communities and participants which took part in the study, and without whom the work would not have been possible.

Ministry of Health, CIDRZ, ZPCT II and JSI. The team further acknowledges the work of the administrative and support teams at the institutions involved in this trial and the hundreds of field staff who delivered the intervention and collected the research data. The team extends its sincere appreciation to the study’s Community Advisory Boards (CABs), in-country Trial Steering/Management Committees, International Advisory Group and Data and Safety Monitoring Board for their oversight and consultation during study conduct. The team thanks all the communities and participants which took part in the study, and without whom the work would not have been possible. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIAID, NIMH, NIDA, PEPFAR, 3ie, or the Bill & Melinda Gates Foundation. First draft of paper written by: Dr. Hayes Analysis of the data was performed by: D. Donnell, R. Hayes, S. Floyd, T. Skalland, A. Schaap, D. Macleod and E. Wilson

The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIAID, NIMH, NIDA, PEPFAR, 3ie, or the Bill & Melinda Gates Foundation. First draft of paper written by: Dr. Hayes Analysis of the data was performed by: D. Donnell, R. Hayes, S. Floyd, T. Skalland, A. Schaap, D. Macleod and E. Wilson Study Team: Richard J. Hayes London School of Hygiene and Tropical Medicine Sarah Fidler Imperial College London and Imperial College National Institute for Health Research Biomedical Research Centre Nulda Beyers Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Helen Ayles London School of Hygiene and Tropical Medicine & Zambart Peter Bock Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Wafaa El-Sadr ICAP at Columbia University Myron Cohen University of North Carolina School of Medicine at Chapel Hill Susan H. Eshleman Johns Hopkins University School of Medicine Yaw Agyei Johns Hopkins University School of Medicine Estelle Piwowar-Manning Johns Hopkins University School of Medicine Virginia Bond London School of Hygiene and Tropical Medicine and Zambart Graeme Hoddinott Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Deborah Donnell Fred Hutchinson Cancer Research Center Sian Floyd London School of Hygiene and Tropical Medicine Ethan Wilson Fred Hutchinson Cancer Research Center Lynda Emel Fred Hutchinson Cancer Research Center Heather Noble Fred Hutchinson Cancer Research Center Dave Macleod London School of Hygiene and Tropical Medicine David N. Burns Division of AIDS, National Institute of Allergy and Infectious Diseases Christophe Fraser University of Oxford Anne Cori Imperial College London Nirupama Deshmane Sista FHI 360 Sam Griffith FHI 360 Ayana Moore FHI 360 Tanette Headen FHI 360 Rhonda White FHI 360 Eric Miller FHI 360 James R.

pical Medicine David N. Burns Division of AIDS, National Institute of Allergy and Infectious Diseases Christophe Fraser University of Oxford Anne Cori Imperial College London Nirupama Deshmane Sista FHI 360 Sam Griffith FHI 360 Ayana Moore FHI 360 Tanette Headen FHI 360 Rhonda White FHI 360 Eric Miller FHI 360 James R. Hargreaves London School of Hygiene and Tropical Medicine Katharina Hauck Imperial College London Ranjeeta Thomas Imperial College London Mohammed Limbada ZAMBART Justin Bwalya ZAMBART Michael Pickles University of Oxford Kalpana Sabapathy London School of Hygiene and Tropical Medicine Ab Schaap London School of Hygiene and Tropical Medicine & Zambart Rory Dunbar Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Kwame Shanaube ZAMBART Blia Yang Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Musonda Simwinga ZAMBART Peter C. Smith Imperial College London Business School Sten H. Vermund Yale School of Public Health Nomtha Mandla Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Nozizwe Makola Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Anneen van Deventer Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Anelet James Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University Karen Jennings City Health Department, City of Cape Town James Kruger Department of Health, Western Cape Mwelwa Phiri ZAMBART Barry Kosloff London School of Hygiene and Tropical Medicine & Zambart Lawrence Mwenge ZAMBART Sarah Kanema ZAMBART Rafael Sauter University of Oxford Will Probert University of Oxford Ramya Kumar ZAMBART Ephraim Sakala ZAMBART Andrew Silumesi Ministry of Health, Zambia Timothy Skalland Fred Hutchinson Cancer Research Center Krista Yuhas Fred Hutchinson Cancer Research Center

ne & Zambart Lawrence Mwenge ZAMBART Sarah Kanema ZAMBART Rafael Sauter University of Oxford Will Probert University of Oxford Ramya Kumar ZAMBART Ephraim Sakala ZAMBART Andrew Silumesi Ministry of Health, Zambia Timothy Skalland Fred Hutchinson Cancer Research Center Krista Yuhas Fred Hutchinson Cancer Research Center This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1814556.

INTRODUCTION Typhoid fever is a systemic illness caused by the Salmonella enteric serovar Typhi. An estimated 11 to 21 million cases of febrile illness, and 117,000 to 161,000 deaths are attributed to the disease each year1–5. Typhoid fever is a major public health problem in Kathmandu, Nepal6,7 where S. Typhi accounts for up to 45% of all positive blood cultures and is the leading cause of blood-stream infections among pediatric patients 8–10. Typhoid is seasonal in Kathmandu, with a high season in July/August and lower incidence in winter. Annual population incidence of typhoid and paratyphoid combined has been recently estimated as 449 (95% CI, 383, 521) per 100,000 2. Antibiotic-resistant S. Typhi is increasingly common in South Asia. Extensively drug-resistant (XDR) variants of S. Typhi have recently emerged in other nearby South Asian countries such as India and Bangladesh, and a large outbreak is ongoing in Pakistan, leading to a situation in which the disease in South Asian populations is becoming increasingly difficult to treat11,12. The WHO recommended the use of typhoid vaccines in 200813 but, vaccine-based control programs have not been widely implemented. Oral live attenuated Ty21a vaccine and Vi-polysaccharide vaccine (Vi-PS) were available but are either not tolerated (Ty21a) or poorly immunogenic in the youngest children and therefore deemed unsuitable for widespread use. A prototype TCV, Vi-rEPA (Vi conjugated to recombinant Pseudomonas aeruginosa exotoxin A) had over 90% efficacy in children aged 2-5 years in clinical trials in 2001 but is not available.

are either not tolerated (Ty21a) or poorly immunogenic in the youngest children and therefore deemed unsuitable for widespread use. A prototype TCV, Vi-rEPA (Vi conjugated to recombinant Pseudomonas aeruginosa exotoxin A) had over 90% efficacy in children aged 2-5 years in clinical trials in 2001 but is not available. More recently, new generation typhoid conjugate vaccines (TCV), containing Vi polysaccharide conjugated to a tetanus-toxoid protein carrier, have become available. In a phase III safety and immunogenicity study, TCV was found to be highly immunogenic and safe in young children14. Furthermore, in a stringent typhoid controlled infection challenge model among adults in a non-endemic setting, TCV had a protective efficacy of 54.6% (95% CI, 26.8%, 71.8%)15. In October 2017, based on these immunogenicity and human challenge study results, the WHO SAGE recommended the use of TCV over the other available typhoid vaccines in view of its improved immunological properties, suitability for use in infants and young children, and expected longer duration of protection13. Gavi, the Vaccine Alliance, also approved a funding window for 2019-2020 to support the introduction of TCVs in developing countries. To aid Gavi-eligible countries to accelerate the introduction of TCVs, the Typhoid Vaccine Acceleration Consortium (TyVAC) was formed16. We conducted the first individually randomized phase III trial of the efficacy of TCV in an endemic population, to inform vaccine implementation strategies. Herein, we report the interim results of this trial after one-year of follow-up.

duction of TCVs, the Typhoid Vaccine Acceleration Consortium (TyVAC) was formed16. We conducted the first individually randomized phase III trial of the efficacy of TCV in an endemic population, to inform vaccine implementation strategies. Herein, we report the interim results of this trial after one-year of follow-up. METHODS Study Design and Participants A phase III, participant- and observer-blind randomized controlled trial was conducted in Lalitpur Metropolitan City of Kathmandu Valley, Nepal. Full methodology has been previously described 17,18. Briefly, children aged 9 months to <16 years living in the study catchment area, who were in good health at the time of enrolment, and whose parents/ legal guardian were willing and competent to provide informed consent were eligible to participate in the study. The lower age limit of 9 months was chosen to align with the potential future programmatic use of TCV given with measles vaccine at 9 months of age. The study (ISRCTN43385161, https://doi.org/10.1186/ISRCTN43385161) was approved by the Oxford Tropical Research Ethics Committee (OxTREC 15–17) and the Nepal Health Research Council (Ref. no. 170/2017). Vaccines Vi polysaccharide-tetanus toxoid conjugate vaccine (TCV, Typbar-TCV Bharat-Biotech, Hyderabad, India) containing 25 µg of Vi-polysaccharide per 0·5 mL dose was used as the trial vaccine for all age groups. Meningococcal capsular Group A conjugate vaccine (MenA; MenAfriVac, Serum Institute of India PVT Ltd) was the control vaccine (see supplementary file).

ugate vaccine (TCV, Typbar-TCV Bharat-Biotech, Hyderabad, India) containing 25 µg of Vi-polysaccharide per 0·5 mL dose was used as the trial vaccine for all age groups. Meningococcal capsular Group A conjugate vaccine (MenA; MenAfriVac, Serum Institute of India PVT Ltd) was the control vaccine (see supplementary file). Randomization and Blinding Participants received either TCV or the control vaccine using 1:1 stratified block randomization with block sizes randomly varying from 2-6. Stratification was done by age (9 months to ≥5 years old or >5 years old to <16 years). Participants were randomized after consent and general examination using a bespoke randomization application loaded on an electronic tablet device. A sub-set of children were further randomized on a 2:1 basis (1000 TCV: 500 control) to have blood drawn for immunogenicity. Parents, guardians, participants, clinicians, and trial staff were blinded to vaccine allocation. Only the unblinded vaccinating staff were aware of the vaccine given and were not subsequently involved in participant follow-up.

A sub-set of children were further randomized on a 2:1 basis (1000 TCV: 500 control) to have blood drawn for immunogenicity. Parents, guardians, participants, clinicians, and trial staff were blinded to vaccine allocation. Only the unblinded vaccinating staff were aware of the vaccine given and were not subsequently involved in participant follow-up. Outcomes: Assessment of Vaccine Efficacy Blood cultures were taken from any study participant with ≥2 days of self-reported fever AND/OR a temperature of ≥ 38°C presenting to Patan hospital or 18 community-based study fever clinics. Trained physicians attended to patients, and consent was obtained for blood culture. Three-monthly follow-up phone calls were used to capture additional possible typhoid fever cases in participants who attended non-study facilities. Where available, medical records were reviewed to capture blood culture-confirmed typhoid diagnoses made at non-study hospitals and clinics. Self-treated typhoid cases, cases treated but without a blood culture taken, and cases not reported to the study team will not be captured in these study data. The primary outcome was blood culture-confirmed typhoid fever.

Outcomes: Assessment of Vaccine Efficacy Blood cultures were taken from any study participant with ≥2 days of self-reported fever AND/OR a temperature of ≥ 38°C presenting to Patan hospital or 18 community-based study fever clinics. Trained physicians attended to patients, and consent was obtained for blood culture. Three-monthly follow-up phone calls were used to capture additional possible typhoid fever cases in participants who attended non-study facilities. Where available, medical records were reviewed to capture blood culture-confirmed typhoid diagnoses made at non-study hospitals and clinics. Self-treated typhoid cases, cases treated but without a blood culture taken, and cases not reported to the study team will not be captured in these study data. The primary outcome was blood culture-confirmed typhoid fever. Assessment of Safety Participants were observed at the vaccination site for at least 20 minutes after the vaccine was administered. All participants were given a diary to capture local and systemic adverse events. Participants' parents/guardians were then contacted by telephone at Day 7, to record any vaccine-related adverse events and all serious adverse events (SAEs). SAEs continue to be captured through ongoing three-monthly follow-up calls and visits.

ticipants were given a diary to capture local and systemic adverse events. Participants' parents/guardians were then contacted by telephone at Day 7, to record any vaccine-related adverse events and all serious adverse events (SAEs). SAEs continue to be captured through ongoing three-monthly follow-up calls and visits. Immunogenicity Anti-Vi IgG titres were measured from plasma samples collected at Day 0 and Day 28, at the Oxford Vaccine Group Laboratory, University of Oxford, using a commercial ELISA kit (VaccZyme, The Binding Site, Birmingham, UK) according to the manufacturer's instructions. Further blood samples will be collected at 18 months and 2 years of follow-up. Interim Analysis The target sample size for the study was 20,000 children (see supplementary files for further details).

Immunogenicity Anti-Vi IgG titres were measured from plasma samples collected at Day 0 and Day 28, at the Oxford Vaccine Group Laboratory, University of Oxford, using a commercial ELISA kit (VaccZyme, The Binding Site, Birmingham, UK) according to the manufacturer's instructions. Further blood samples will be collected at 18 months and 2 years of follow-up. Interim Analysis The target sample size for the study was 20,000 children (see supplementary files for further details). Over the two-year trial follow-up period 45 cases of typhoid fever were expected if the assumptions underlying the sample size held true (see supplementary files for further details). While this was originally designed as a two-year study, given the public health significance of the results, an interim analysis was planned after at least one year of follow-up had been completed, if 45 cases were observed by this time. The interim analysis therefore has full statistical power for the primary outcome. The protocol was amended to include the interim analysis when it became clear that the 45 cases may be reached before 2 years of follow-up. The interim analysis was agreed by the international data safety and monitoring board on August 1st, 2018, approximately 9 months into the study and received ethical approval. Study participants and staff were not unblinded as part of the interim analysis and follow-up continues.

be reached before 2 years of follow-up. The interim analysis was agreed by the international data safety and monitoring board on August 1st, 2018, approximately 9 months into the study and received ethical approval. Study participants and staff were not unblinded as part of the interim analysis and follow-up continues. Statistical Analysis The primary analysis of blood culture-confirmed typhoid fever included only those cases that occurred at least 14 days after vaccination. Additionally, secondary outcomes reported in this interim analysis include adverse events within the first 7 days after vaccination, SAEs within 6 months of vaccination and immunogenicity in the first 28 days. Full analysis of all study outcomes will be reported at the end of the study. For the interim analysis of the primary outcome, the incidence of typhoid fever was estimated as the number of cases divided by the total number of person-years of follow-up. Vaccine efficacy (VE) was calculated as (1 – IRR) x 100%, where IRR is the incidence rate ratio (the ratio of the incidence in the TCV arm compared to the control arm). All p-values were 2-sided; a p value < 0.05 was considered significant in efficacy assessment. Serious adverse events, local and systemic vaccine reactions, and baseline characteristics were not compared statistically.

For the interim analysis of the primary outcome, the incidence of typhoid fever was estimated as the number of cases divided by the total number of person-years of follow-up. Vaccine efficacy (VE) was calculated as (1 – IRR) x 100%, where IRR is the incidence rate ratio (the ratio of the incidence in the TCV arm compared to the control arm). All p-values were 2-sided; a p value < 0.05 was considered significant in efficacy assessment. Serious adverse events, local and systemic vaccine reactions, and baseline characteristics were not compared statistically. The cumulative incidence of typhoid fever is presented using the Kaplan-Meier method. A detailed statistical analysis plan covering all analyses was agreed and signed by investigators prior to unblinding of study data for analysis and further details are included in the supplementary files. Author Contributions Study design and co-ordination: AJP, KMN, MV, KTN, SS, BuBa, MS, DP, RCJ, NS, SB; Data collection and management: AA, BiBa, MG, SK, OM, YF, ST, and the TyVAC Nepal Study Team (see supplementary file); Laboratory analysis: JC, JH, SD, AK; Data analysis: MV, XL; MV vouches for the data and analysis; Final decision to submit for publication: AJP; MS prepared the first draft manuscript, which was reviewed and approved by all authors. RESULTS Study Participants From November 20, 2017 to April 9, 2018, 20119 children were screened, and 20,019 participants were randomized to receive the TCV or control vaccine (Figure S1). The baseline characteristics were similar in the TCV recipients and the control vaccine recipients (Table 1).

Author Contributions Study design and co-ordination: AJP, KMN, MV, KTN, SS, BuBa, MS, DP, RCJ, NS, SB; Data collection and management: AA, BiBa, MG, SK, OM, YF, ST, and the TyVAC Nepal Study Team (see supplementary file); Laboratory analysis: JC, JH, SD, AK; Data analysis: MV, XL; MV vouches for the data and analysis; Final decision to submit for publication: AJP; MS prepared the first draft manuscript, which was reviewed and approved by all authors. RESULTS Study Participants From November 20, 2017 to April 9, 2018, 20119 children were screened, and 20,019 participants were randomized to receive the TCV or control vaccine (Figure S1). The baseline characteristics were similar in the TCV recipients and the control vaccine recipients (Table 1). Table 1 Baseline characteristics of randomised participants Characteristics TCV (N=10,005) Men A (N=10 014) Total (N= 20019) Gender Male N (%) 5106 (51.0%) 5158 (51.5%) 10264 (51.3%) Age at enrolment (years) Mean (SD) 7.9 (4.1) 7.8 (4.0) 7.9 (4.1) Median [Range] 7.7 [0.8 – 16.1] 7.7 [0.7 – 16.1] 7.7 [0.7 – 16.1] * < 5 years N (%) 2907 (29.1%) 2905 (29.0%) ≥ 5 years N (%) 7098 (70.9%) 7109 (71.0%) Self-reported medical history of Typhoid fever** N (%) 345 (3.5%) 395 (4.0%) 740 (3.7%) * 6 participants are outside the age range for eligibility (9 months to 15 years + 364 days) ** Self-reported history of typhoid infection prior to the beginning of the study, reported at baseline intake TCV = Typhoid Conjugate Vaccine. Men A = Group A meningococcal vaccine (control)

Characteristics TCV (N=10,005) Men A (N=10 014) Total (N= 20019) Gender Male N (%) 5106 (51.0%) 5158 (51.5%) 10264 (51.3%) Age at enrolment (years) Mean (SD) 7.9 (4.1) 7.8 (4.0) 7.9 (4.1) Median [Range] 7.7 [0.8 – 16.1] 7.7 [0.7 – 16.1] 7.7 [0.7 – 16.1] * < 5 years N (%) 2907 (29.1%) 2905 (29.0%) ≥ 5 years N (%) 7098 (70.9%) 7109 (71.0%) Self-reported medical history of Typhoid fever** N (%) 345 (3.5%) 395 (4.0%) 740 (3.7%) * 6 participants are outside the age range for eligibility (9 months to 15 years + 364 days) ** Self-reported history of typhoid infection prior to the beginning of the study, reported at baseline intake TCV = Typhoid Conjugate Vaccine. Men A = Group A meningococcal vaccine (control) Vaccine Efficacy Between Dec 6, 2017 and March 9, 2019, 46 cases of blood culture-confirmed typhoid fever were recorded. One case occurred within 2 weeks of vaccination and was excluded from analyses. All cases recovered; 5 were admitted to hospital (TCV: 2, Control: 3). Blood culture-confirmed typhoid fever was diagnosed in 0.07% (n=7 of 10,005) of participants in the TCV group and 0.38% (n=38 of 10,013) in the control group. The protective efficacy of TCV was 81.6% (95% CI, 58.8%, 91.8%, P<0.001) (Figure 1, Table 2). Table 2 Occurrence of blood culture-confirmed typhoid fever and protective efficacy of typhoid conjugate vaccine Outcome TCV (N=10005) Incidence per 100,000 person-years (95% CI) Men A (N=10014) Incidence per 100,000 person-years (95% CI) Vaccine Efficacy (95% CI) p value (Log-rank) Person-years of follow-up a 8903 8885

Table 2 Occurrence of blood culture-confirmed typhoid fever and protective efficacy of typhoid conjugate vaccine Outcome TCV (N=10005) Incidence per 100,000 person-years (95% CI) Men A (N=10014) Incidence per 100,000 person-years (95% CI) Vaccine Efficacy (95% CI) p value (Log-rank) Person-years of follow-up a 8903 8885 Blood culture-confirmed typhoid fever in first 14 days after vaccination 1 Blood culture-confirmed typhoid fever after 14 days b 7 79 (37, 165) 38 428 (311, 588) 81.6% (58.8%, 91.8%) <0.001 Detected through fever clinics 5 27 Detected through active follow-up and medical record review 2 11 Blood culture-confirmed typhoid fever in those with at least 3 days of fever prior to blood culture (fever clinics) c 3 34 (11, 105) 20 226 (146, 350) 85.1% (49.7%, 95.6%) <0.001 a Participants with no follow-up contact contribute half a day follow-up in calculations. Participants who move away from Lalitpur no longer contribute to person-years of follow-up time b For all reported culture positive cases reported from medical records review, isolates were checked, when available, to reconfirm diagnostic results. c From fever clinic cases only. Data not available from cases detected through medical records review. TCV = Typhoid conjugate vaccine. Men A = Group A meningococcal vaccine Figure 1 Kaplan-Meier cumulative incidence of blood culture-positive typhoid fever by randomised vaccine group

b For all reported culture positive cases reported from medical records review, isolates were checked, when available, to reconfirm diagnostic results. c From fever clinic cases only. Data not available from cases detected through medical records review. TCV = Typhoid conjugate vaccine. Men A = Group A meningococcal vaccine Figure 1 Kaplan-Meier cumulative incidence of blood culture-positive typhoid fever by randomised vaccine group There were 23 cases of blood culture-confirmed typhoid fever in those presenting to fever clinics with at least 3 days of fever prior to their blood draw for culture; the WHO recommended threshold for blood cultures in typhoid surveillance programs19. Vaccine efficacy in those with at least 3 days of fever was similar to the overall estimate: 85.1% (95%CI, 49.7%, 95.6%). Immunogenicity 1343 participants provided at least one sample for immunogenicity analysis. At baseline, 268 (31.6%) participants in the TCV group and 122 (26.5%) participants in the Men A group had detectable Vi-IgG antibody levels. The geometric mean titre of anti-Vi antibody at day 28 was 2038 EU/mL (95% CI, 1905, 2180) for the TCV group and 7.0 EU/mL (95% CI, 6.2, 7.9) for the Men A group (p<0.001). Seroconversion (≥ four-fold rise in antibody titre 28 days after vaccination) was 99% in the TCV group and 2% in the control group (Table 3). Table 3 Vi-IgG levels at baseline and 28 days after randomization in immunogenicity cohort Visit TCV Men A p value* N N (%) above LLD GMC (95%CI) or Median [IQR] or % N N (%) above LLD GMC (95%CI) or Median [IQR] or %

Immunogenicity 1343 participants provided at least one sample for immunogenicity analysis. At baseline, 268 (31.6%) participants in the TCV group and 122 (26.5%) participants in the Men A group had detectable Vi-IgG antibody levels. The geometric mean titre of anti-Vi antibody at day 28 was 2038 EU/mL (95% CI, 1905, 2180) for the TCV group and 7.0 EU/mL (95% CI, 6.2, 7.9) for the Men A group (p<0.001). Seroconversion (≥ four-fold rise in antibody titre 28 days after vaccination) was 99% in the TCV group and 2% in the control group (Table 3). Table 3 Vi-IgG levels at baseline and 28 days after randomization in immunogenicity cohort Visit TCV Men A p value* N N (%) above LLD GMC (95%CI) or Median [IQR] or % N N (%) above LLD GMC (95%CI) or Median [IQR] or % Day 0 849 268 (31.6%) 7.2 (6.7, 7.8) a 460 122 (26.5%) 6.5 (5.9, 7.1) a 3.7 [3.7, 13.4] b 3.7 [3.7, 8.9] b 0.07 Day 28 709 708 (99.9%) 2038 (1905, 2180) a 388 112 (28.9%) 7.0 (6.2, 7.9) a 2221 [1297, 3726] b [3.7, 10.5] b <0.001 Both Day 0 & Day 28 683 380 ≥ 4-fold rise from Day 0 677 99.1% 8 2.1% LLD: The lower limit of quantification of the assay (7.4 EU/mL). Values below this limit were substituted with 3.7 EU/mL for analysis. GMC = geometric mean concentration * p value from non-parametric two-sided Wilcoxon Rank Sum Test. TCV = Typhoid conjugate vaccine. Men A = Group A meningococcal vaccine (control). a geometric mean concentration (95% confidence interval); b median [interquartile range]

Day 0 849 268 (31.6%) 7.2 (6.7, 7.8) a 460 122 (26.5%) 6.5 (5.9, 7.1) a 3.7 [3.7, 13.4] b 3.7 [3.7, 8.9] b 0.07 Day 28 709 708 (99.9%) 2038 (1905, 2180) a 388 112 (28.9%) 7.0 (6.2, 7.9) a 2221 [1297, 3726] b [3.7, 10.5] b <0.001 Both Day 0 & Day 28 683 380 ≥ 4-fold rise from Day 0 677 99.1% 8 2.1% LLD: The lower limit of quantification of the assay (7.4 EU/mL). Values below this limit were substituted with 3.7 EU/mL for analysis. GMC = geometric mean concentration * p value from non-parametric two-sided Wilcoxon Rank Sum Test. TCV = Typhoid conjugate vaccine. Men A = Group A meningococcal vaccine (control). a geometric mean concentration (95% confidence interval); b median [interquartile range] Reactogenicity Adverse vaccine reactions in the first 7 days after vaccination were assessed in 18,743 (93.6%) children. 5.9% of children experienced pain at the vaccination site (TCV: 5.1%, Men A: 6.7%) which was mostly mild (92.6%). 6.7% of children reported being generally unwell (TCV: 6.4%, Men A: 7.1%). 5.2% of children had a fever (by parental self-report) in the first 7 days (TCV: 5.0%, Men A: 5.4%). Vomiting and diarrhea occurred in 1.4% and 1.8% of children respectively (Vomiting: TCV: 1.2%, Men A: 1.6%; Diarrhea: TCV: 1.7%, Men A 1.8%), and of those reporting these symptoms, 20.5% and 25.9% of instances were moderate or severe. 1.9% of children were eating less than usual (TCV: 1.8%, Men A: 1.9%). All other reactions were rare, occurring in less than 1% of children (Table S1).

tively (Vomiting: TCV: 1.2%, Men A: 1.6%; Diarrhea: TCV: 1.7%, Men A 1.8%), and of those reporting these symptoms, 20.5% and 25.9% of instances were moderate or severe. 1.9% of children were eating less than usual (TCV: 1.8%, Men A: 1.9%). All other reactions were rare, occurring in less than 1% of children (Table S1). Serious Adverse Events In the first 28 days after vaccination, 18 SAEs were reported in 17 participants; 7 participants in the TCV group and 10 in the Men A group (Tables S2 and S3). One SAE was identified as vaccine-related; a high-grade fever within 24 hours of vaccination. The participant was admitted to the local hospital and given antipyretics. The fever subsided after 12 hours, investigations were within normal limits and the participant was discharged without an alternative diagnosis. The participant remains blinded (Tables S2 and S3). SAEs occurring in the 6 months after vaccination were reported by 121 participants who experienced 132 events. (Table S4). SAEs occurring more than once per group are summarised by MedDRA codes in Table S5. The most common SAEs were pneumonia/lower respiratory tract infection and pyrexia. There was one death due to staphylococcal sepsis, occurring 7 months after vaccination, deemed unrelated to vaccination (see Supplementary file).

SAEs occurring in the 6 months after vaccination were reported by 121 participants who experienced 132 events. (Table S4). SAEs occurring more than once per group are summarised by MedDRA codes in Table S5. The most common SAEs were pneumonia/lower respiratory tract infection and pyrexia. There was one death due to staphylococcal sepsis, occurring 7 months after vaccination, deemed unrelated to vaccination (see Supplementary file). DISCUSSION This is the first large-scale field trial to assess the efficacy of a WHO prequalified typhoid conjugate vaccine in children in an endemic setting and shows that a single dose of TCV is safe, immunogenic, efficacious, and has the potential to save thousands of lives. Incidence was 428 per 100,000 in our control group, confirming the high burden of disease in children in this setting. Large-scale vaccination strategies using TCV can potentially reduce the burden of enteric fever, an important goal given the global increase in antimicrobial resistance. The rise in extensively drug resistant (XDR) typhoid severely limits treatment options. Over 5000 cases of XDR typhoid have been reported in Pakistan since the outbreak began in 2016, with cases also being reported in travelers returning from Pakistan. Deployment of the vaccine in Pakistan, as is being done, and beyond, is of paramount importance to curb the spread of the drug resistant strain regionally and transcontinentally.

typhoid have been reported in Pakistan since the outbreak began in 2016, with cases also being reported in travelers returning from Pakistan. Deployment of the vaccine in Pakistan, as is being done, and beyond, is of paramount importance to curb the spread of the drug resistant strain regionally and transcontinentally. A single dose of TCV resulted in a reduction in typhoid fever by 81.6% in children in our study. This protective efficacy is higher than that of Vi-PS which was estimated to have 35% to 65% efficacy in trials in Pakistan and India respectively 20,21, and higher than live attenuated oral typhoid vaccines22. The results are similar to the 91.1% (95% CI, 77.1%, 96.6%) efficacy seen with two doses of Vi-rEPA in Vietnam in 199723. The results are also consistent with the seroefficacy estimates (85%, 95% CI 80%, 88%) of TCV extrapolated from serological responses in the phase 3 trial in India24.The vaccine efficacy was 54.6% (95%CI, 26.8%, 71.8%) in a human challenge study conducted in Oxford15. However, the challenge model used a composite definition of typhoid fever that included self-resolving asymptomatic bacteraemia not detected in the field, adults rather than children, and a probable high challenge dose (following neutralization of gastric acid), which could provide some explanation as to why vaccine efficacy was lower compared with our results.

te definition of typhoid fever that included self-resolving asymptomatic bacteraemia not detected in the field, adults rather than children, and a probable high challenge dose (following neutralization of gastric acid), which could provide some explanation as to why vaccine efficacy was lower compared with our results. TCV is highly immunogenic, eliciting a strong antibody response one month after vaccination. This is consistent with previous findings in immunogenicity trials 14,15,23. Immunogenicity trials in children and adults in India reported seroconversion rates of over 90% across different age strata at day 42 post-vaccination compared to baseline titres and a four-fold rise in anti Vi- antibody titre occurred 2 to 5 times more often in the TCV group in comparison with the Vi-PS group14. The Vi-rEPA study reported that Vi-IgG increased by a factor of more than 575 (P<0.001) four weeks after administration of the conjugate Vi-rEPA vaccine, although using a different assay23. Conjugate vaccines are T-cell dependent and are expected to provide long-term protection as demonstrated with the Vi-rEPA vaccine unlike the protection provided by polysaccharide vaccines which generally last for only 2-3 years25.

fter administration of the conjugate Vi-rEPA vaccine, although using a different assay23. Conjugate vaccines are T-cell dependent and are expected to provide long-term protection as demonstrated with the Vi-rEPA vaccine unlike the protection provided by polysaccharide vaccines which generally last for only 2-3 years25. TCV was safe and clinically acceptable in this study. Our data on reactogenicity to the vaccine are consistent with those from the phase III trial in India 14, and the human challenge model study15. In our study, one SAE was deemed to be a vaccine-related fever without any alternative diagnosis, but remains blinded to group allocation. Reported adverse events were similar for both TCV and the control vaccine, indicating an acceptable safety profile in comparison with another widely used conjugate vaccine. These data were part of a package reviewed by the WHO Global Advisory Committee on Vaccine Safety in December 2018, leading to an endorsement of the safety of this vaccine26.

e similar for both TCV and the control vaccine, indicating an acceptable safety profile in comparison with another widely used conjugate vaccine. These data were part of a package reviewed by the WHO Global Advisory Committee on Vaccine Safety in December 2018, leading to an endorsement of the safety of this vaccine26. These results provide strong evidence that TCV can play an important role in the control of typhoid fever in endemic settings. TCV is highly cost-effective in high transmission settings and should be taken into account in country decision-making 27. However, further data are still required to demonstrate vaccine efficacy in the medium- and long-term, the indirect effect and herd immunity achieved from large-scale vaccination, and the effectiveness in different age groups and populations. The full analysis of data from this trial, as well as data from on-going trials in Malawi and Bangladesh, will be available within the next two years to address these outstanding questions28,29. Our findings uphold the WHO's recent recommendations to use TCV to control typhoid in high burden settings through immunization of children from 9 months to 15 years of age5. Inclusion of the conjugate vaccine in routine immunization schedules in high burden countries could prevent a large burden of a disease that has been disproportionately affecting children. Supplementary Material Click here for additional data file.

Our findings uphold the WHO's recent recommendations to use TCV to control typhoid in high burden settings through immunization of children from 9 months to 15 years of age5. Inclusion of the conjugate vaccine in routine immunization schedules in high burden countries could prevent a large burden of a disease that has been disproportionately affecting children. Supplementary Material Click here for additional data file. Acknowledgements The authors acknowledge The Bill & Melinda Gates Foundation (OPP1151153) for funding the Typhoid Vaccine Acceleration Consortium, including this trial, the Wellcome Trust and the Bill & Melinda Gates Foundation for funding of The Strategic Typhoid Alliance across Africa and Asia, which supported the surveillance that underpins this trial. The Typhoid Vaccine Acceleration Consortium (TyVAC), a partnership between the Center for Vaccine Development and Global Health at the University of Maryland School of Medicine, the Oxford Vaccine Group at the University of Oxford, and PATH, an international nonprofit, aims to accelerate the introduction of new typhoid conjugate vaccines (TCVs) as part of an integrated approach to reducing the burden of morbidity and mortality from typhoid in countries eligible for support from Gavi, the Vaccine Alliance. The authors acknowledge the support of the Wellcome Trust in development of the typhoid human challenge model which supported the rationale for this study, and the support of the NIHR Oxford Biomedical Research Centre.

Acknowledgements The authors acknowledge The Bill & Melinda Gates Foundation (OPP1151153) for funding the Typhoid Vaccine Acceleration Consortium, including this trial, the Wellcome Trust and the Bill & Melinda Gates Foundation for funding of The Strategic Typhoid Alliance across Africa and Asia, which supported the surveillance that underpins this trial. The Typhoid Vaccine Acceleration Consortium (TyVAC), a partnership between the Center for Vaccine Development and Global Health at the University of Maryland School of Medicine, the Oxford Vaccine Group at the University of Oxford, and PATH, an international nonprofit, aims to accelerate the introduction of new typhoid conjugate vaccines (TCVs) as part of an integrated approach to reducing the burden of morbidity and mortality from typhoid in countries eligible for support from Gavi, the Vaccine Alliance. The authors acknowledge the support of the Wellcome Trust in development of the typhoid human challenge model which supported the rationale for this study, and the support of the NIHR Oxford Biomedical Research Centre. The authors would like to thank the volunteers who participated in the study and their families; Patan Hospital and Patan Academy of Health Sciences; Child Health Division – Nepal Committee on Immunization Practices, for their support of the study; Lalitpur Metropolitan City representatives including Mayor, Ward chairpersons, Ward representatives and Tole Health Promoters, Ward Health implementation committee; and the team at Nepal Family Development Foundation for field support in conducting the study. We are grateful to Roma Chilengi (Chair) and the members of the International DSMB who are overseeing the ongoing trial. We would also like to acknowledge Bharat Biotech International Limited for supplying the investigational vaccine. We thank our clinical team including Dr. Rashmi Shrestha, Dr. Prabina Aryal, Dr. Pankaj Giri, Dr. Sumnima Shrestha, Dr. Anita Banjade, Dr. Arjun Gautam, Dr. Ayush Jung Pandey, Dr. Anuradha Twayana, Dr. Aryan Shah, Dr. Sailesh Pathak, Dr. Ashmita Ghimire, Dr. Raveena Yadav and all the Community Medical Assistants (CMAs) for their work.

nical team including Dr. Rashmi Shrestha, Dr. Prabina Aryal, Dr. Pankaj Giri, Dr. Sumnima Shrestha, Dr. Anita Banjade, Dr. Arjun Gautam, Dr. Ayush Jung Pandey, Dr. Anuradha Twayana, Dr. Aryan Shah, Dr. Sailesh Pathak, Dr. Ashmita Ghimire, Dr. Raveena Yadav and all the Community Medical Assistants (CMAs) for their work. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1905047. Funding This publication is based on research funded by a grant from the Bill & Melinda Gates Foundation (OPP1151153). Conflict of Interest AJP is Chair of UK Dept. Health and Social Care's (DHSC) Joint Committee on Vaccination & Immunisation (JCVI) & the European Medicine Agency (EMA) scientific advisory group on vaccines, and is a member of the WHO's Strategic Advisory Group of Experts. KMN is a member of the WHO's Strategic Advisory Group of Experts. All other authors report no conflicts of interest. The views expressed in this article do not necessarily represent the views of DHSC, JCVI, EMA, or WHO.

Introduction Schistosomiasis is a parasitic neglected tropical disease (NTD), estimated to currently infect over 140 million people.1,2 The disease burden is greatest (at least 90%) in sub-Saharan Africa (SSA), where the main species causing human schistosomiasis are Schistosoma mansoni (intestinal schistosomiasis) and S. haematobium (urogenital schistosomiasis), transmitted through faeces and urine, respectively.3,4 Symptoms of schistosomiasis morbidity include anaemia, stunting, fever, genital lesions, and irreversible organ damage.5–7 Preventive chemotherapy (PC) with Praziquantel is the World Health Organization (WHO)-recommended strategy for the control of schistosomiasis and is primarily distributed to school-aged children (SAC) aged 5–15 years, who carry the highest infection burden and who can be reached efficiently through schools.8 The PC strategy is indicated by prevalence (estimated by initial parasitological assessment) at implementation unit level, usually district. Prevalence of infection less than 10% requires triennial PC, 10% to 49% biennial treatment, and 50% or greater annual treatment.9

who can be reached efficiently through schools.8 The PC strategy is indicated by prevalence (estimated by initial parasitological assessment) at implementation unit level, usually district. Prevalence of infection less than 10% requires triennial PC, 10% to 49% biennial treatment, and 50% or greater annual treatment.9 The success of morbidity control in some countries 10 has led to a more ambitious vision of "a world free of schistosomiasis".11 The WHO has set goals for controlling schistosomiasis morbidity (defined as prevalence of heavy-intensity infection <5% aggregated across sentinel sites) by 2020 and achieving elimination as a public health problem (EPHP, defined as prevalence of heavy-intensity infection <1% in all sentinel sites) in all endemic countries by 2025. Complete interruption of transmission is a target in selected regions by 2025 (Figure 1).11–13 The WHO strategic plan provides guidance on how programmes can progress from control of schistosomiasis to EPHP and interruption of transmission.11 r

sity infection <1% in all sentinel sites) in all endemic countries by 2025. Complete interruption of transmission is a target in selected regions by 2025 (Figure 1).11–13 The WHO strategic plan provides guidance on how programmes can progress from control of schistosomiasis to EPHP and interruption of transmission.11 r In practice, it is unlikely that the time lines for transitioning between goals will be uniform for all countries due to their epidemiological heterogeneity (Figure 1). Hence, there exists a need to analyse quantitative data, captured through programme monitoring, to validate and update these guidelines. Recent theoretical mathematical modelling work projects that the 2020 goal of morbidity control is likely obtainable for low and moderate prevalence settings, but will be missed in high-intensity settings with current treatment guidelines.14 We empirically addressed whether countries have already reached the 2020 and 2025 goals and if so, how many treatment rounds were required. Nationally representative cross-sectional epidemiological data for both S. mansoni and S. haematobium from nine countries were used. These data were made available by the national Ministries of Health of endemic countries. This study represents the first multi-country and multi-year empirical study to assess whether a one-size-fits-all approach is appropriate for guiding schistosomiasis treatment strategies to reach the WHO defined threshold criteria on morbidity control and EPHP.

y the national Ministries of Health of endemic countries. This study represents the first multi-country and multi-year empirical study to assess whether a one-size-fits-all approach is appropriate for guiding schistosomiasis treatment strategies to reach the WHO defined threshold criteria on morbidity control and EPHP. Materials and methods Data collation Data were collated from the Schistosomiasis Control Initiative (SCI)-supported multi-year, cross-sectional treatment impact surveys in nine countries, which took place approximately six weeks prior to the following treatment round (i.e. just less than one year after the last treatment round for annual PC programmes and just less than two years after the last treatment round for biennial PC). . The inclusion criteria for were: i) countries where Ministries of Health were supported by the SCI; ii) having more than 2 years of impact survey data post baseline; and iii) cross-sectional data comprising SAC aged 5–15 years (Figure 2 and Table S1). Only epidemiological data available at SCI were analysed, so any further data points on the country programmes available from other sources were not included. Further details on the original surveys for this study can be found in the Supplementary Appendix.

l data comprising SAC aged 5–15 years (Figure 2 and Table S1). Only epidemiological data available at SCI were analysed, so any further data points on the country programmes available from other sources were not included. Further details on the original surveys for this study can be found in the Supplementary Appendix. Data analysis The methods used to calculate sample sizes in each country programme were as currently employed at the SCI (Table S1). This provided the number of sentinel sites (schools) and children to be sampled within each site, powered to detect a pre-set difference in prevalence at a given administrative level for the country, accounting for clustering (a design effect) at the sentinel site level. Survey methods were standardised across countries. Standard Kato-Katz and urine filtration methods were used to detect S. mansoni and S. haematobium infection, respectively. The infection intensity category, i.e. the proportion of individuals with a given number of schistosome eggs per gram of faeces (epg) for S. mansoni (light intensity: 1-99 epg, moderate intensity: 100-399 epg and heavy intensity: ≥400 epg) or per 10 ml of urine for S. haematobium (light intensity: 1-50 eggs/10ml and heavy intensity: >50 eggs/10ml), and 95% confidence intervals (95% CIs), were calculated by treatment round, schistosome species, and country programme.9 Mean prevalence and 95% CIs were calculated to account for the clustering of the data at sentinel site level, using the R survey package.15 The point (mean) prevalence estimates were used for the comparison against WHO guidelines, since the guidelines do not suggest calculations of 95% CIs. We assessed whether the mean prevalence of heavy-intensity infection across sentinel sites fell to <5%, indicative of morbidity control, and/or <1% in all sentinel sites, indicative of EPHP.11 The overall prevalence and prevalence of moderate- plus heavy-intensity infection (S. mansoni only) was also estimated and compared with trends of heavy-intensity infection prevalence.

y infection across sentinel sites fell to <5%, indicative of morbidity control, and/or <1% in all sentinel sites, indicative of EPHP.11 The overall prevalence and prevalence of moderate- plus heavy-intensity infection (S. mansoni only) was also estimated and compared with trends of heavy-intensity infection prevalence. While the WHO guidelines only use prevalence of heavy-intensity infection as an indirect measure of morbidity (assuming morbidity is proportional to infection intensity), we included the combined measure of prevalence of moderate- plus heavy-intensity infection due to uncertainty in the appropriateness of egg count thresholds for intensity and because some degree of morbidity is likely to be caused by lighter infections.16 Results Baseline endemicity varied by species and country. S. haematobium ranged from 9.8% [95% CI: 6.0-15.5] prevalence in Malawi to 82.1% [95% CI: 70.1-90.0] in Mali-Segou.9 Prevalence for S. mansoni varied from 1.9% [95% CI: 0.5-6.9] in Malawi to 45.4% [95% CI: 35.6-55.7] in Uganda. Despite this heterogeneity, infection intensity in all countries fell following the first round of treatment to below, or within 0.8% of the 5% prevalence of heavy intensity threshold for control for S. mansoni infection and within 3.3% for S. haematobium (Figures 3 and 4 and Supplementary Appendix Figures S1 and S2).

Despite this heterogeneity, infection intensity in all countries fell following the first round of treatment to below, or within 0.8% of the 5% prevalence of heavy intensity threshold for control for S. mansoni infection and within 3.3% for S. haematobium (Figures 3 and 4 and Supplementary Appendix Figures S1 and S2). Treatment reduced the prevalence of heavy-intensity infection for both species to below 5% in all countries except Niger (5.4% [95% CI: 2.0-13.8]), which only marginally missed the metric for S. haematobium in the first treatment round (Figures 3 and 4 and Table 1). The more ambitious target of EPHP was only achieved for S. mansoni infection, and only in half of the country programmes. Moreover, Malawi had already reached EPHP for S. mansoni at baseline. Table 1 Rounds of treatment required to reduce Schistosoma mansoni and Schistosoma haematobium infection to reach the World Health Organization's (WHO's) goal of morbidity control (<5% prevalence of heavy-intensity infection, aggregated across all sentinel sites) and elimination as a public health problem (EPHP, <1% prevalence of heavy-intensity infection in all sentinel sites). Baseline endemicity levels refer to the WHO prevalence category at country-level and the 95% CIs were calculated accounting for clustering of the data at the level of the sentinel sites.

sentinel sites) and elimination as a public health problem (EPHP, <1% prevalence of heavy-intensity infection in all sentinel sites). Baseline endemicity levels refer to the WHO prevalence category at country-level and the 95% CIs were calculated accounting for clustering of the data at the level of the sentinel sites. Species Baseline endemicity levels Mean baseline prevalence % (95% CI) Baseline prevalence of heavy-intensity infection % (95% CI) Country Frequency of treatment Goal/s reached§ No. of treatment rounds (post-baseline) No. of treatment rounds for moderate- plus heavy-intensity prevalence Schistosoma mansoni Low 6.5 (1.8-2.7) 0.7 (0.2-3.0) Burkina Faso Biennial Control 0 0 EPHP 2 3 Low 6.0 (2.4-14.2) 0.5 (0.2-1.3) Burundi National Annual Control 0 0 EPHP 2 Not yet reached Low 12.7 (4.6-30.9) 1.5 (0.4-4.9) Burundi Pilot Annual Control 0 0 EPHP 3 Not yet reached Low 1.9 (0.5-6.9) 0.1 (0.0-0.9) Malawi Annual Control 0 0 EPHP 0 1 Low 12.9 (4.6-31.2) 1.1 (0.2-5.7) Rwanda Annual Control 0 0 EPHP 1 1 Low 9.2 (6.4-13.0) 0.6 (0.3-1.2) Yemen Biennial Control 0 0 EPHP Not yet reached Not yet reached Moderate 28.8 (8.9-62.6) 9.6 (2.5-27.5) Mali-Segou Annual Control 1 1 EPHP Not yet reached Not yet reached Moderate 38.8 (17.2-66.0) 10.6 (3.6-27.5) Mali-Bamako/Koulikoro Annual/Biennial Control 1 Not yet reached EPHP Not yet reached Not yet reached Moderate 26.6 (7.4-62.1) 7.7 (2.1-24.7) Tanzania Annual Control 1 Not yet reached EPHP Not yet reached Not yet reached Moderate 45.4 (35.6-55.7) 17.7 (11.7-25.8) Uganda Annual Control 2 Not yet reached EPHP Not yet reached Not yet reached Schistosoma haematobium Low 9.8 (6.0-15.5) 2.2 (1.0-4.5) Malawi Annual Control 0 NA EPHP Not yet reached NA Moderate 24.1 (14.1-38.0) 6.9 (3.2-14.4) Tanzania Annual Control 1 NA EPHP Not yet reached NA Moderate 10.6 (6.5-16.8) 3.6 (2.1-6.3) Yemen Biennial Control 0 NA EPHP Not yet reached NA High 56.2 (32.4-77.4) 25.2 (14.3-40.3) Burkina Faso Biennial Control 1 NA EPHP Not yet reached NA High 70.0 (54.2-82.2) 20.8 (12.1-33.5) Niger Annual Control Not yet reached NA EPHP Not yet reached NA High 82.1 (70.1-90.0) 44.0 (27.3-62.3) Mali-Segou Annual Control 2 NA EPHP Not yet reached NA High 47.6 (33.5-62.1) 11.5 (6.8-18.6) Mali-Bamako/Koulikoro Annual/Biennial Control 1 NA EPHP Not yet reached NA Schistosoma mansoni All ten country programmes reached the control of morbidity threshold after two rounds of treatment or fewer (Figure 3 and Table 1, Supplementary Appendix Figure S1).

P Not yet reached NA High 47.6 (33.5-62.1) 11.5 (6.8-18.6) Mali-Bamako/Koulikoro Annual/Biennial Control 1 NA EPHP Not yet reached NA Schistosoma mansoni All ten country programmes reached the control of morbidity threshold after two rounds of treatment or fewer (Figure 3 and Table 1, Supplementary Appendix Figure S1). This included Uganda which had a relatively high baseline prevalence. However, a subsequent gradual increase in the prevalence of heavy-intensity infection to just over the 5% threshold was observed in Uganda after the third and fourth treatment rounds. Burkina Faso, Burundi (pilot and national programme) and Rwanda reached the EPHP threshold after three rounds or fewer (but note that these sites had a baseline mean prevalence of heavy-infection intensity already below the 5% morbidity control threshold). When using the more conservative criterion of <1% and <5% prevalence of moderate- plus heavy-intensity infection to represent morbidity control (S. mansoni only), six country programmes were already below the <5% prevalence threshold at baseline and one further country programme (Mali-Segou) met this target after one round of treatment (Figure 3 and Table 1). Three country programmes achieved EPHP and only three out of ten country programmes failed to reach any target (control or EPHP) in the relatively short treatment period of the data currently available.

e and one further country programme (Mali-Segou) met this target after one round of treatment (Figure 3 and Table 1). Three country programmes achieved EPHP and only three out of ten country programmes failed to reach any target (control or EPHP) in the relatively short treatment period of the data currently available. Schistosoma haematobium All countries had a baseline S. haematobium prevalence of heavy-intensity infection above 5%, except for Malawi and Yemen and, by the second treatment round, all except for Niger were below this threshold, meeting the control of morbidity criteria (Figure 4 and Table 1, Supplementary Appendix Figure S2). The prevalence of heavy-intensity infection in Niger fell following a single treatment round, from 20.8% [95% CI: 12.1-33.5] to 5.4% [95% CI: 2.0-13.8], only just missing the control of morbidity target. Although three country programmes reached <1% heavy-intensity infection prevalence aggregated across sentinel sites (Figure 4), no countries reached this threshold in every sentinel site for S. haematobium, and thus did not meet the EPHP requirement.

Schistosoma haematobium All countries had a baseline S. haematobium prevalence of heavy-intensity infection above 5%, except for Malawi and Yemen and, by the second treatment round, all except for Niger were below this threshold, meeting the control of morbidity criteria (Figure 4 and Table 1, Supplementary Appendix Figure S2). The prevalence of heavy-intensity infection in Niger fell following a single treatment round, from 20.8% [95% CI: 12.1-33.5] to 5.4% [95% CI: 2.0-13.8], only just missing the control of morbidity target. Although three country programmes reached <1% heavy-intensity infection prevalence aggregated across sentinel sites (Figure 4), no countries reached this threshold in every sentinel site for S. haematobium, and thus did not meet the EPHP requirement. Discussion The WHO provides guidance on the expected number of years of treatment to reach morbidity control and EPHP (5-10 years plus an additional 3-6 years, respectively). We demonstrate that these thresholds are often reached much sooner, whether with annual or biennial treatment. With the exception of S. haematobium in Niger, all programmes reached the morbidity control thresholds in two or fewer treatments rounds (between 1 and 2 years, depending on the frequency of PC). Notably, six country programmes started with a prevalence of heavy infection below 5% for S. mansoni, indicating that they were already at 'control' at baseline. The goal of EPHP for S. mansoni was reached by five programmes and required three or fewer treatment rounds (1 to 3 years). This prompts the question of what strategy to adopt in cases where the baseline prevalence of heavy-intensity infection already meets the control target. The S. haematobium areas had higher overall baseline infection levels and none reached the EPHP goals within the time horizon of this study. Endemically low prevalence countries, which also have lower baseline prevalence of heavy-intensity infection (≤1.5% for S. mansoni), achieved EPHP sooner than proposed by the guidelines. Through the analysis of extensive and nationally representative datasets from multiple countries, these results provide a crucial complement to recent theoretical modelling work which projects that achievement of control is possible in low and moderate prevalence settings and EPHP is possible in low prevalence areas.14 We recommend that this combination of programmatic data and mathematical models should be a blueprint for future activities as it utilizes the power of theoretical modelling to inform concretely programmatic goals and targets.

sible in low and moderate prevalence settings and EPHP is possible in low prevalence areas.14 We recommend that this combination of programmatic data and mathematical models should be a blueprint for future activities as it utilizes the power of theoretical modelling to inform concretely programmatic goals and targets. The case of Uganda illustrates that goals may be reached but are reversible (precise reasons for the Ugandan rebound are beyond the scope of the current study, but may reflect factors such as those relating to the influx of refugees, reduced compliance and/or changes in drug efficacy17). This is particularly relevant where programme stability is impeded, due to, for example, civil unrest in Burundi, war in Yemen. Figures 3 and 4 also highlight the variability between sentinel sites in each country, which need to be taken into consideration when looking at the country-level control of morbidity target proposed by the WHO. Additionally, the effectiveness of programme implementation may vary through time. It is, thus, important to define time periods over which control and elimination targets should be sustained to declare success and to be particularly vigilant to recrudescence of disease if elimination of transmission has not yet been achieved.

ally, the effectiveness of programme implementation may vary through time. It is, thus, important to define time periods over which control and elimination targets should be sustained to declare success and to be particularly vigilant to recrudescence of disease if elimination of transmission has not yet been achieved. As expected, there was a strong positive association between overall infection prevalence and either the prevalence of heavy-intensity infection, or the prevalence of moderate- plus heavy-intensity infection (Supplementary Appendix Figure S3). There was substantial variation of data points between countries and treatment rounds which is likely caused by the heterogeneity of underlying adult parasite loads (perhaps due to variation in exposure among human hosts) such that the disease/morbidity prevalence varies substantially in settings with similar prevalence of infection. The magnitude of the change in infection following treatment varied substantially between country programmes. This emphasises the need for research on appropriate morbidity indicators. Once identified, they should be applied consistently across programmes and in guidelines. It is not just heavy infections which lead to morbidity,16 therefore moderate plus heavy intensity infections were combined to form a more conservative metric of morbidity. When considering the aims of morbidity control and EPHP, since control thresholds may be reached relatively quickly, it would be worth considering this metric in the meantime, to include a larger population group potentially suffering from morbidity.

ctions were combined to form a more conservative metric of morbidity. When considering the aims of morbidity control and EPHP, since control thresholds may be reached relatively quickly, it would be worth considering this metric in the meantime, to include a larger population group potentially suffering from morbidity. An important limitation of this study is the absence of information on specific treatment coverage which can, in general, vary substantially among national scale PC programmes (see Supplementary Appendix). Other information such as migration patterns, school-enrolment and attendance rates may also influence the effectiveness of a PC programme and explain variation among study areas. Detailed information on these factors has not been routinely collected by the SCI. This is common to NTD programmes and we recommend that, in future, the scope of data collection should be enhanced and incorporated into routine data collection protocols. This additional information would support the interpretation of epidemiological data when evaluating the impact and effectiveness of PC programmes.

This is common to NTD programmes and we recommend that, in future, the scope of data collection should be enhanced and incorporated into routine data collection protocols. This additional information would support the interpretation of epidemiological data when evaluating the impact and effectiveness of PC programmes. More than half of the programmes had sentinel sites of mixed S. mansoni and S. haematobium infections. In this study, the species were analysed independently; however, some areas may have had higher infection prevalence when both are combined, or underlying interactions occurring.18 These issues require further clarification in the guidelines. China (endemic species S. japonicum) and Brazil (S. mansoni) have shown great progress towards achieving interruption of transmission, particularly considering the added challenge of multiple animal reservoirs of S. japonicum. This highlights that further integration of other practices such as clean water and sanitation, treatment of the animal population and snail control are required, all of which are still lagging in the much of the SSA settings.

, particularly considering the added challenge of multiple animal reservoirs of S. japonicum. This highlights that further integration of other practices such as clean water and sanitation, treatment of the animal population and snail control are required, all of which are still lagging in the much of the SSA settings. It is necessary to mention briefly some of the key factors which highlight the need for the guidelines to be updated. The metrics and definitions for control and elimination of schistosomiasis-related morbidity use the egg-count intensity cut-offs. However, these metrics, definitions and egg-count cut-offs are an imperfect measure of infection and the relationship between morbidity and egg counts needs to be carefully and urgently addressed. The adult and pre-SAC population are generally not actively monitored and data on SAC are currently used as a proxy for the situation in the wider community. However, it is unrealistic to declare EPHP (and in some cases even control) with this unmonitored reservoir in the population. Suitability of the currently recommended diagnostic tools, upon which the guidelines are based, also need to be promptly assessed. Studies are actively looking at the feasibility and cost-effectiveness of alternative and more accurate diagnostics for large-scale use, particularly as diagnostic sensitivity decreases with reduced infection intensity.19–24 Another critical area is the hotspot phenomenon – a blanket catchall term used to describe areas of persistent infection despite multiple rounds of treatment. Guidance on defining and managing hotspot areas, especially in countries which are otherwise on target for the 2020/2025 goals, are presently lacking, but studies are aiming to address this.25–27 Additional analysis of available data, the ability to collect such information routinely as part of large-scale control programmes, and further research are required to establish a robust evidence base for these (or updated) targets, which will be critical especially as countries move towards interruption of transmission. What has not yet been addressed is that there will still be true morbidity in the community even after the targets have been reached, since schistosomiasis morbidity can continue many years after infection has ceased (e.g. genital schistosomiasis, hepatosplenomegaly, etc.).

es move towards interruption of transmission. What has not yet been addressed is that there will still be true morbidity in the community even after the targets have been reached, since schistosomiasis morbidity can continue many years after infection has ceased (e.g. genital schistosomiasis, hepatosplenomegaly, etc.). Achieving true morbidity control and elimination would thus require the redefining of morbidity control (not to be completely dependent on egg-output) and incorporation of new strategies that address long-term morbidity, such as the SAFE (Surgery, Antibiotics, Facial Cleanliness and Environmental improvement) adopted as the recommended strategy for trachoma elimination.28

thus require the redefining of morbidity control (not to be completely dependent on egg-output) and incorporation of new strategies that address long-term morbidity, such as the SAFE (Surgery, Antibiotics, Facial Cleanliness and Environmental improvement) adopted as the recommended strategy for trachoma elimination.28 Our study analyses the most extensive datasets available to assess the timeframes in the WHO's guidelines for control and elimination of schistosomiasis. In conclusion, if the indicative timelines to transition to the next control or elimination goal are accurate, these will take many countries beyond the 2020 and 2025 targets. Where countries have <5% heavy-intensity prevalence at baseline, it is unclear whether they should aim immediately for EPHP or continue to treat as per guidelines for 5-10 years. The study's key messages are that countries often achieve morbidity control following very few treatment rounds, and that the universal timeline currently recommended is not appropriate for all programmes, and will be affected by baseline endemicity, schistosome species, and context-specific relationship between infection and morbidity. Outputs from analysis of empirical and modelling data can be used to update these timelines. This will allow more useful programmatic decision-making tools and more accurate projections of progress against schistosomiasis at the national, regional, and global level as we move towards the 2020 and 2025 goals. Supplementary Material Click here for additional data file.

Our study analyses the most extensive datasets available to assess the timeframes in the WHO's guidelines for control and elimination of schistosomiasis. In conclusion, if the indicative timelines to transition to the next control or elimination goal are accurate, these will take many countries beyond the 2020 and 2025 targets. Where countries have <5% heavy-intensity prevalence at baseline, it is unclear whether they should aim immediately for EPHP or continue to treat as per guidelines for 5-10 years. The study's key messages are that countries often achieve morbidity control following very few treatment rounds, and that the universal timeline currently recommended is not appropriate for all programmes, and will be affected by baseline endemicity, schistosome species, and context-specific relationship between infection and morbidity. Outputs from analysis of empirical and modelling data can be used to update these timelines. This will allow more useful programmatic decision-making tools and more accurate projections of progress against schistosomiasis at the national, regional, and global level as we move towards the 2020 and 2025 goals. Supplementary Material Click here for additional data file. Acknowledgements First and foremost, we would like to thank all the children and teachers who participated in and supported the impact surveys of the national control programmes. These surveys would not have taken place without the significant drive of national Ministry of Health staff who have and continue to manage and implement these successful programmes. A debt of gratitude also goes to the Ministry of Health's for granting the authors permission to access and re-analyse these data for the study. We would like to thank past and present Directors, Programme Managers, Biostatisticians, Field Operations staff from the Schistosomiasis Control Initiative (SCI) team for their technical support to the countries for the delivery of the programmes and the impact surveys. Finally we would also like to thank the Children's Investment Fund Foundation (grant #239) who supported this study and The Bill and Melinda Gates Foundation (grants #13122; 36202), the Department for International Development: Integrated Control of Schistosomiasis and Intestinal Helminths in Sub- Saharan Africa Phase 1 & 2 [GB-1-200706] and Geneva Global through the END Fund for financially supporting the primary data collection as part of programmatic monitoring and evaluation. Without all of these partners, this study would not have been possible.

ed Control of Schistosomiasis and Intestinal Helminths in Sub- Saharan Africa Phase 1 & 2 [GB-1-200706] and Geneva Global through the END Fund for financially supporting the primary data collection as part of programmatic monitoring and evaluation. Without all of these partners, this study would not have been possible. Contributors AKD conducted the literature search, conceived the first draft of the manuscript. AKD, MDF, FF and JPW conceptualised the study and AKD, FF, MDF, MW, MGB and JPW developed the manuscript with essential revisions. AKD, BCU, MDF, MW and MGB contributed to the data analysis strategy. AKD led and conducted the data analysis, BCU formatted the datasets for analysis and provided the first phase analysis and MW provided essential input for the analysis. AKD, BCU, MDF, MW, MGB, FF, JWP, VB, IG, ET, SJ, UJM, AA, ST, MT and ER all contributed to the final draft of the manuscript. Our partners at the ministries of health and the Schistosomiasis Control Initiative conducted the original data collection. All co-authors approved the final draft. Declaration of interests We declare no competing interests.

Contributors AKD conducted the literature search, conceived the first draft of the manuscript. AKD, MDF, FF and JPW conceptualised the study and AKD, FF, MDF, MW, MGB and JPW developed the manuscript with essential revisions. AKD, BCU, MDF, MW and MGB contributed to the data analysis strategy. AKD led and conducted the data analysis, BCU formatted the datasets for analysis and provided the first phase analysis and MW provided essential input for the analysis. AKD, BCU, MDF, MW, MGB, FF, JWP, VB, IG, ET, SJ, UJM, AA, ST, MT and ER all contributed to the final draft of the manuscript. Our partners at the ministries of health and the Schistosomiasis Control Initiative conducted the original data collection. All co-authors approved the final draft. Declaration of interests We declare no competing interests. Ethics approval and consent to participate The data used in this study were collected as part of the M&E activities of the schistosomiasis control programmes taking place in these endemic countries. Ethical approval was granted by the St Mary's Hospital Local Ethics Research Committee, R&D office (part of the Imperial College Research Ethics Committee (ICREC 8.2.2, EC No. 03.36, R&D No. 03/SB/003E)), as a constituent part of the ongoing Schistosomiasis Control Initiative (SCI) activities, and by the Ministries of Health ethical review boards in these countries.

INTRODUCTION Combination HIV prevention (CHP) is the concurrent implementation of multiple interventions to reduce HIV incidence.1 Most CHP packages include antiretroviral therapy (ART) and medical male circumcision (MC), along with provision of HIV testing and counseling, condom promotion, and other behavioral interventions.2 CHP scale-up has been an intense focus of global health over the past decade.3 Modeling studies indicate that high coverage of ART and MC could substantially reduce HIV incidence to low-endemic levels,4 5 and potentially even lead to its elimination.6 7 However, the effectiveness of CHP remains uncertain due to challenges in increasing CHP coverage and in accurately measuring changes in population-level HIV incidence.8 9 Demonstrating the population-level effectiveness of CHP is critical to understanding whether the current evidence98 based interventions are sufficient for HIV mitigation and to guide resource allocation.

e to challenges in increasing CHP coverage and in accurately measuring changes in population-level HIV incidence.8 9 Demonstrating the population-level effectiveness of CHP is critical to understanding whether the current evidence98 based interventions are sufficient for HIV mitigation and to guide resource allocation. While prior research from South Africa has shown that increasing community ART coverage reduces individual-level HIV risk, population-level HIV incidence declines were not demonstrated.10 11 Other research from North America suggests that ART scale-up has reduced HIV incidence, but these studies relied on modeled incidence and sentinel surveillance data.9 12-14 The “gold standard” for assessing HIV incidence is the longitudinal measurement of HIV seroconversions in a population-based cohort.8 9 However, these studies are rare despite the urgency to demonstrate relationships between changes in CHP coverage and HIV incidence over time.4 5 15 To assess the impact of CHP on HIV incidence, we analyzed long-term trends in HIV incidence based on observed seroconversions and their associations with ART and MC scale-up, population-level viral load suppression, and sexual behaviors in Rakai, Uganda.

etween changes in CHP coverage and HIV incidence over time.4 5 15 To assess the impact of CHP on HIV incidence, we analyzed long-term trends in HIV incidence based on observed seroconversions and their associations with ART and MC scale-up, population-level viral load suppression, and sexual behaviors in Rakai, Uganda. METHODS Cohort Description The Rakai Community Cohort Study (RCCS), conducted by the Rakai Health Sciences Program (RHSP), is an open, population-based, multi-community cohort of individuals aged 15-49 years.16 The RCCS is situated in Rakai District (population ~518,000) which is mostly rural with scattered trading centers.17 This study uses data from thirty RCCS communities which were continuously surveyed from April 6, 1999 to September 2, 2016 over a total of twelve surveys (Supplemental Fig.1). To identify eligible participants, a household census enumerates all persons by gender, age, and duration of residence, regardless of whether they are present or currently absent. After the census, the RCCS surveys all present, age-eligible residents providing written informed consent. Participants are interviewed to assess demographics, sexual behaviors, ART use, and MC status. Venous blood is obtained for HIV testing at each survey (Supplemental RCCS laboratory methods). Funded by the U.S. President’s Emergency Plan for AIDS Relief (PEPFAR),18 CHP scale-up began in earnest in the mid-2000’s (Supplemental CHP scale-up).

rviewed to assess demographics, sexual behaviors, ART use, and MC status. Venous blood is obtained for HIV testing at each survey (Supplemental RCCS laboratory methods). Funded by the U.S. President’s Emergency Plan for AIDS Relief (PEPFAR),18 CHP scale-up began in earnest in the mid-2000’s (Supplemental CHP scale-up). Statistical Analysis CHP coverage was assessed using person-visit data at each survey with descriptive statistics and logistic regression. Specifically, ART coverage was defined as the proportion of all HIV positive participants who self-reported ART use, regardless of ART eligibility criteria, and was assessed overall and separately by gender. Self-reported ART use in the cohort has been validated previously by plasma detection of antiretroviral drugs showing a specificity and sensitivity of 99% (95%CI: 97-100%) and 77% (95%CI: 70-83%), respectively, with no differences by gender.19 MC coverage at a given visit was defined as the proportion of men who self-reported being circumcised. Self-reported circumcision status has been previously validated from clinical records with 100% specificity.20 Viral suppression was defined using a cutoff of 1000 copies/ml as per WHO recommendations.21

by gender.19 MC coverage at a given visit was defined as the proportion of men who self-reported being circumcised. Self-reported circumcision status has been previously validated from clinical records with 100% specificity.20 Viral suppression was defined using a cutoff of 1000 copies/ml as per WHO recommendations.21 The unit of exposure for HIV incidence were person-intervals of follow-up between surveys in initially HIV-negative individuals who participated in at least two surveys. HIV incident cases were persons who tested HIV-seropositive for the first time with an HIV seronegative test result at the prior RCCS visit, allowing for up to one missed visit. Incident infections were assumed to occur at the mid-point of the interval and changes in HIV incidence per 100 person years (py) were estimated using Poisson multivariate regression with generalized estimating equations and an exchangeable correlation structure and were reported as incidence rate ratios (IRR) with 95% confidence intervals (CI).

sumed to occur at the mid-point of the interval and changes in HIV incidence per 100 person years (py) were estimated using Poisson multivariate regression with generalized estimating equations and an exchangeable correlation structure and were reported as incidence rate ratios (IRR) with 95% confidence intervals (CI). To assess the impact of CHP, mean incidence at each visit interval after 2004 (6th survey) was compared to mean HIV incidence over the entire period prior to ART and MC availability. The final multivariate model included individual-level information on demographics (gender, age, marital status, education) and sexual behaviors (sexual partners in the last year, sex with partners outside the community of residence, sex with non-marital partners, condom use and self-reported genital ulceration). A categorical term for community-level HIV prevalence was included to adjust for variation in exposure. Secondary analyses were stratified by gender and conducted separately for circumcised and uncircumcised men. HIV incidence and individual risk was also assessed in relation to community-level measures of ART and MC coverage and prevalence of HIV viremia (Supplemental statistical methods). Sensitivity of results to both selective participation and loss to follow-up were evaluated using inverse probability weights (Supplemental statistical methods). To assess the potential impact of birth cohort effects on HIV incidence trends, a term for each five-year birth cohort was included in the multivariate model. HIV incidence was also assessed by gender for each five year age group.

w-up were evaluated using inverse probability weights (Supplemental statistical methods). To assess the potential impact of birth cohort effects on HIV incidence trends, a term for each five-year birth cohort was included in the multivariate model. HIV incidence was also assessed by gender for each five year age group. RESULTS Survey participation Table 1 shows eligibility and participation summary statistics for the twelve surveys. Overall, 33,937 individual participants contributed 103,011 person-visits, including an incidence cohort of 17,870 initially HIV-negative persons followed for 94,427 person-years. The mean participation rate among all eligible persons censused was 64% and did not vary substantially between surveys (range: 59%-67%); however, reasons for non-participation and study drop-out (e.g. refusal, travel) changed over time (Supplemental Tables.1a-c, 2a-c). The proportion of individuals who refused participation steadily declined from 21% to 0.5% over the analysis period, whereas the proportions absent due to work or school increased from 18% to 31%. The most common reasons for loss to follow-up were out-migration from study communities (ranging from 42-63% of losses) and travel for work or school (ranging from 25-33% of losses).

adily declined from 21% to 0.5% over the analysis period, whereas the proportions absent due to work or school increased from 18% to 31%. The most common reasons for loss to follow-up were out-migration from study communities (ranging from 42-63% of losses) and travel for work or school (ranging from 25-33% of losses). Table 1. Summary of eligibility, participation and follow-up in the RCCS by survey round, 1999-2016 Survey Interview Date Census Eligibleα Eligible and present for surveyβ Percent eligible who participated in surveyγ Percent eligible and present who participated in survey HIV-negative participants eligible for incidence cohort* Percent of eligible HIV-negative participants who outmigrated prior to the subsequent survey Incidence cohort Percent of age-eligible HIV-negative participants followed Percent of age and resident eligible HIV-negative participants followed** Years since prior survey visit

ve participants eligible for incidence cohort* Percent of eligible HIV-negative participants who outmigrated prior to the subsequent survey Incidence cohort Percent of age-eligible HIV-negative participants followed Percent of age and resident eligible HIV-negative participants followed** Years since prior survey visit Median (range) no. no. Percent (no.) Percent no. Percent (no.) no. Percent Percent median (IQR) 1 Oct.1999 (Apr.1999-Feb.2000) 9869 8125 61% (5992) 74% - - - - - - 2 Oct.2000 (Feb. 2000-Feb.2001) 10448 8567 64% (6732) 79% 5183 11% (546) 3760 73% 93% 1.0 (1.0,1.0) 3 Jan.2002 (Apr.2001-May.2002) 11316 9176 65% (7340) 80% 7277 23% (1677) 4540 62% 82% 1.3 (1.1,1.3) 4 Apr.2003 (Jul.2002-Aug.2003) 11436 8603 60% (6856) 80% 7905 27% (2167) 4555 58% 80% 1.2 (1.2,1.3) 5 Jul.2004 (Sep.2003-Nov.2004) 11860 8436 59% (7038) 83% 8014 28% (2206) 4693 59% 81% 1.3 (1.2,1.3) 6 Jan.2006 (Feb.2005-Jun.2006) 12528 9137 65% (8097) 89% 7768 28% (2159) 4867 63% 87% 1.5 (1.4,1.6) 7 Oct.2007 (Aug.2006-Jun.2008) 13636 9130 63% (8645) 95% 8624 30% (2585) 5001 58% 83% 1.7 (1.6,1.8) 8 Jul.2009 (Jun.2008-Dec.2009) 13293 9009 65% (8691) 96% 9679 30% (2952) 5611 58% 84% 1.7 (1.6,1.8) 9 Jan.2011 (Jan.2010-Jun.2011) 14629 9949 66% (9643) 97% 9686 30% (2894) 5742 59% 85% 1.6 (1.6,1.6) 10 Jun.2012 (Aug.2011-May.2013) 16007 10846 66% (10588) 98% 10300 29% (3032) 6176 60% 85% 1.6 (1.5,1.7) 11 Jul.2014 (Jul.2013-Jan.2015) 17477 11566 65% (11379) 98% 11419 34% (3875) 6277 55% 83% 2.0 (1.9,2.1) 12 Jan.2016 (Jan.2015-Sep.2016) 18065 12308 66% (12010) 98% 12908 31% (4017) 7122 55% 80% 1.6 (1.4,2.0) α Residents aged 15-19 in the census.

3) 16007 10846 66% (10588) 98% 10300 29% (3032) 6176 60% 85% 1.6 (1.5,1.7) 11 Jul.2014 (Jul.2013-Jan.2015) 17477 11566 65% (11379) 98% 11419 34% (3875) 6277 55% 83% 2.0 (1.9,2.1) 12 Jan.2016 (Jan.2015-Sep.2016) 18065 12308 66% (12010) 98% 12908 31% (4017) 7122 55% 80% 1.6 (1.4,2.0) α Residents aged 15-19 in the census. β Eligible census population present at time of survey, γ Eligible census population present and participated in survey. * Includes all age-eligible HIV-negative participants from prior survey and any HIV-negative participants from two surveys prior if participant was absent at the most recent survey. ** Calculation excludes HIV-negative persons who out-migrated prior to survey. Table 2 HIV incidence and unadjusted and adjusted incidence rate ratios comparing HIV incidence in each visit interval during combination HIV prevention (CHP) scale-up to mean HIV incidence in the entire period prior to scale-up. IRR=Incidence Rate Ratio; adjIRR=Adjusted incidence rate ratio; Final adjusted model included age, gender (full cohort only), marital status, level of education, number of sexual partners in past year, sex with partners outside community, self-reported genital ulcer disease, condom use with casual partners, community residence type (trading, agrarian), and community HIV prevalence.

Final adjusted model included age, gender (full cohort only), marital status, level of education, number of sexual partners in past year, sex with partners outside community, self-reported genital ulcer disease, condom use with casual partners, community residence type (trading, agrarian), and community HIV prevalence. HIV incidence Cohort (N=17,780) Survey(s) Incident HIV cases person-years HIV incidence per 100 py (95%CI) IRR (95%CI) p-value adjIRR (95% CI) p-value Pre-CHP (2-5) 254 21765 1.17 (1.03,1.32) Ref. - Ref. - Jan.2006 (6) 86 7773 1.11 (0.89,1.36) 0.95 (0.74,1.21) 0.66 0.94 (0.73,1.2) 0.61 Oct.2007 (7) 105 8769 1.2 (0.98,1.44) 1.02 (0.82,1.29) 0.84 1.00 (0.79,1.26) 0.99 Jul.2009 (8) 125 10201 1.23 (1.02,1.45) 1.05 (0.85,1.3) 0.67 0.95 (0.76,1.18) 0.62 Jan.2011 (9) 105 9815 1.07 (0.88,1.29) 0.91 (0.73,1.15) 0.44 0.94 (0.74,1.19) 0.60 Jun.2012 (10) 86 10352 0.83 (0.67,1.02) 0.71 (0.55,0.91) 0.006 0.72 (0.56,0.93) 0.012 Jul.2014 (11) 87 13159 0.66 (0.53,0.81) 0.56 (0.44,0.72) <0.001 0.60 (0.47,0.78) <0.001 Jan.2016 (12) 83 12593 0.66 (0.53,0.81) 0.56 (0.44,0.72) <0.001 0.58 (0.45,0.76) <0.001 Women (N=9,709) Survey(s) Incident HIV cases person-years HIV incidence per 100 py (95%CI) IRR (95%CI) p-value adjIRR (95% CI) p-value Pre-CHP (2-5) 145 12409 1.17 (0.99,1.37) Ref. - Ref. - Jan.2006 (6) 50 4425 1.13 (0.84,1.47) 0.97 (0.7,1.33) 0.84 0.98 (0.71,1.35) 0.90 Oct.2007 (7) 65 4978 1.31 (1.01,1.65) 1.12 (0.83,1.5) 0.46 1.10 (0.82,1.48) 0.52 Jul.2009 (8) 67 5610 1.19 (0.93,1.5) 1.02 (0.76,1.36) 0.89 0.91 (0.68,1.23) 0.55 Jan.2011 (9) 61 5319 1.15 (0.88,1.46) 0.98 (0.73,1.32) 0.90 0.98 (0.72,1.33) 0.89 Jun.2012 (10) 50 5587 0.89 (0.67,1.17) 0.77 (0.56,1.05) 0.10 0.75 (0.54,1.05) 0.096 Jul.2014 (11) 55 7090 0.78 (0.59,1) 0.66 (0.49,0.90) 0.010 0.65 (0.47,0.91) 0.012 Jan.2016 (12) 56 6689 0.84 (0.64,1.08) 0.72 (0.53,0.97) 0.034 0.68 (0.50,0.94) 0.021 Men (N=8,161) Survey(s) Incident HIV cases person-years HIV incidence per 100 py (95%CI) IRR (95%CI) p-value adjIRR (95% CI) p-value Pre-CHP (2-5) 109 9356 1.17 (0.96,1.4) Ref. - Ref.

59,1) 0.66 (0.49,0.90) 0.010 0.65 (0.47,0.91) 0.012 Jan.2016 (12) 56 6689 0.84 (0.64,1.08) 0.72 (0.53,0.97) 0.034 0.68 (0.50,0.94) 0.021 Men (N=8,161) Survey(s) Incident HIV cases person-years HIV incidence per 100 py (95%CI) IRR (95%CI) p-value adjIRR (95% CI) p-value Pre-CHP (2-5) 109 9356 1.17 (0.96,1.4) Ref. - Ref. - 6 36 3348 1.08 (0.76,1.47) 0.92 (0.63,1.34) 0.661 0.90 (0.61,1.33) 0.60 7 40 3791 1.06 (0.76,1.42) 0.90 (0.63,1.29) 0.572 0.89 (0.62,1.3) 0.56 8 58 4591 1.26 (0.97,1.62) 1.08 (0.78,1.48) 0.641 1.02 (0.73,1.42) 0.93 9 44 4497 0.98 (0.72,1.3) 0.83 (0.59,1.18) 0.309 0.92 (0.63,1.33) 0.64 10 36 4765 0.76 (0.53,1.03) 0.64 (0.44,0.94) 0.022 0.70 (0.47,1.05) 0.083 11 32 6069 0.53 (0.37,0.73) 0.45 (0.3,0.67) <0.001 0.54 (0.36,0.83) 0.005 12 27 5904 0.46 (0.31,0.65) 0.39 (0.25,0.59) <0.001 0.46 (0.29,0.73) 0.001 Uncircumcised men only (N=5,660) Survey(s) Incident HIV cases person-years HIV incidence per 100 py (95%CI) IRR (95%CI) p-value adjIRR (95% CI) p-value Pre-CHP (2-5) 94 7773 1.21 (0.98,1.47) Ref. - Ref. - Jan.2006 (6) 30 2456 1.22 (0.83,1.71) 1.01 (0.67,1.52) 0.98 0.96 (0.63,1.46) 0.85 Oct.2007 (7) 31 2590 1.2 (0.82,1.67) 0.98 (0.66,1.48) 0.94 0.92 (0.61,1.4) 0.70 Jul.2009 (8) 40 2927 1.37 (0.99,1.83) 1.12 (0.78,1.63) 0.54 1.00 (0.68,1.47) 0.99 Jan.2011 (9) 32 2571 1.24 (0.86,1.73) 1.02 (0.68,1.53) 0.92 1.00 (0.67,1.51) 0.98 Jun.2012 (10) 24 2493 0.96 (0.63,1.4) 0.79 (0.50,1.24) 0.30 0.77 (0.49,1.22) 0.27 Jul.2014 (11) 17 2779 0.61 (0.36,0.95) 0.50 (0.30,0.84) 0.009 0.46 (0.26,0.81) 0.007 Jan.2016 (12) 15 2303 0.65 (0.37,1.04) 0.53 (0.31,0.92) 0.024 0.51 (0.29,0.88) 0.016 Participation and follow-up rates were significantly lower among younger individuals, men, and persons living in trading centers, but these associations were stable over time. Individuals with high-risk sexual behaviors were somewhat more likely to be lost to follow-up but this was also constant over time. (Supplemental Figs.2-4). The population growth rate, calculated from the censused resident population irrespective of age, was 3.4% per year.

but these associations were stable over time. Individuals with high-risk sexual behaviors were somewhat more likely to be lost to follow-up but this was also constant over time. (Supplemental Figs.2-4). The population growth rate, calculated from the censused resident population irrespective of age, was 3.4% per year. Temporal trends in sexual behaviors Figure 1 shows age-specific sexual behaviors by survey for HIV-negative men and women. The most substantive changes in sexual behaviors were in adolescents aged 15-19, among whom the proportion self-reporting no initiation of sex increased from 30% in 1999 to 55% in 2016 (p<0.0001) overall, and from 35% (n=194/553) to 56% (n=679/1207) in men, and 28% (n=209/757) to 55% (n=646/1165) over the same time period (p<0.001 for both). Adolescent men who initiated sex were also significantly less likely to report multiple sexual partners in the last survey (40% in 1999 versus 19% in 2016, p<0.001). There were no substantial changes in female multiple partnerships. Overall ages, levels of self-reported condom use with casual partners remained largely unchanged (Figures 1E and 1F).

ex were also significantly less likely to report multiple sexual partners in the last survey (40% in 1999 versus 19% in 2016, p<0.001). There were no substantial changes in female multiple partnerships. Overall ages, levels of self-reported condom use with casual partners remained largely unchanged (Figures 1E and 1F). Figure 1. Sexual Behaviors in the Rakai Community Cohort Study, 1999-2016. Figure shows proportion of HIV-negative men and women by age-group and overall ages reporting the following sexual behaviors A-B) never initiating sex (i.e. delayed sexual debut), C-D) multiple sexual partnerships among sexually active persons, and E-F) consistent condom use among those reporting casual (i.e. non-marital) sexual partnerships. The most substantial changes in sexual behaviors occurred among adolescent men and women aged 15-19 years reporting never initiating sex and adolescent men reporting multiple partnerships. Scale up of biomedical HIV interventions and changes in population HIV viral load The scale-up of biomedical HIV prevention interventions is shown in Figure 2. Self-reported ART use among all HIV-positive persons increased from 12% in 2006 to 69% in 2016 (p<0.001). ART coverage was consistently higher among women (p<0.001); however, the proportional increase in coverage was similar in both genders. By 2016, 61% of HIV-positive men (n=285/465) and 72% (n=766/1060) of women self-reported ART use (Supplemental Table A). ART coverage was highest among older age groups in all surveys (Supplemental Fig.5).

istently higher among women (p<0.001); however, the proportional increase in coverage was similar in both genders. By 2016, 61% of HIV-positive men (n=285/465) and 72% (n=766/1060) of women self-reported ART use (Supplemental Table A). ART coverage was highest among older age groups in all surveys (Supplemental Fig.5). Figure 2. Scale-up of antiretroviral therapy, viral suppression in HIV-positive participants and male circumcision, 1999-2016. 2A shows scale-up of ART coverage measured by selfreport in men, women and all HIV-positive RCCS participants beginning in 2006. Figure 2B show the proportion of all HIV-positive persons by gender and overall virologically suppressed (<1000 HIV copies/ml) in 2009 and 2016. 2C shows scale-up of MC coverage in men irrespective of religion by HIV status and overall beginning in 2004. 2D shows community-level MC coverage vs. community-level ART coverage for all 30 communities at each survey during CHP scale-up. A smoothing-spline was fit to the smooth curve to assess trend. Scale-up of interventions occurred simultaneously and increased significantly in all communities.

nd overall beginning in 2004. 2D shows community-level MC coverage vs. community-level ART coverage for all 30 communities at each survey during CHP scale-up. A smoothing-spline was fit to the smooth curve to assess trend. Scale-up of interventions occurred simultaneously and increased significantly in all communities. HIV viral load assays were obtained for 96% (1115/1160) of HIV-positive participants in 2009 and for 99.9% (1525/1526) of HIV-positive participants in 2016. Viral load suppression (<1000 cps mL) among those self-reporting ART use was 94% (n=1228/1312) and did not differ by gender (p=0.382) or survey visit (p=0.525). HIV viral load suppression in all HIV-positive participants increased concomitant with increasing ART coverage. By 2016, 75% (n=1151/1526) of all HIV-positive persons, regardless of whether or not they reported ART use, were virally suppressed compared with 42% (n=464/1115) in 2009 (p<0.001) (Figure 2B). Population coverage of MC also significantly increased from 15% (n=374/2518) in 1999 to 59% (n=3177/5361) in 2016 among all men (p<0.001) (Figure 2C), and from 3.5% (n=77/2217) to 53% (n=2492/4666, p<0.001) among non-Muslim men who are not traditionally circumcised at birth. MC coverage increased among both HIV-positive and HIV-negative men with highest coverage in younger men (Supplemental Table 3B, Supplemental Fig.5).

) in 2016 among all men (p<0.001) (Figure 2C), and from 3.5% (n=77/2217) to 53% (n=2492/4666, p<0.001) among non-Muslim men who are not traditionally circumcised at birth. MC coverage increased among both HIV-positive and HIV-negative men with highest coverage in younger men (Supplemental Table 3B, Supplemental Fig.5). Scale-up of ART and MC occurred concurrently in all communities (Figure 2D) and by 2016 were high in all 30 RCCS communities: median community-level ART coverage was 70% (IQR: 61-75) and median community-level MC coverage was 61% (IQR:55-65%). Changes in HIV incidence over time Figure 3 shows HIV incidence in the whole population, women, men, and circumcised and uncircumcised men. HIV incidence remained stable prior to CHP scale-up and began to significantly decline in 2012 (Fig. 3, Supplementary Table.4a-e). In 2016, mean HIV incidence declined by 42% from 1.17 per 100 py prior to CHP to 0.66 per 100 py (IRR=0.56, 95%CI: 0.44- 0.72; adjIRR=0.58; 0.45-0.76). The same incidence trends were observed when restricting analyses to sexually active adults and individuals over the age of 20 years (Supplementary Tables.4,5).

.4a-e). In 2016, mean HIV incidence declined by 42% from 1.17 per 100 py prior to CHP to 0.66 per 100 py (IRR=0.56, 95%CI: 0.44- 0.72; adjIRR=0.58; 0.45-0.76). The same incidence trends were observed when restricting analyses to sexually active adults and individuals over the age of 20 years (Supplementary Tables.4,5). Figure 3. HIV incidence and prevalence trends in the Rakai Community Cohort Study, 1996-2016. Trends in HIV incidence and prevalence over the analysis period among all initially HIV-negative men and women in the incidence cohort (3A), women only (3B), men only (3C), and in men by circumcision status (3D). HIV incidence is only shown for circumcised men ginning in 2007 after the WHO recommendation for MC for HIV-negative men for HIV prevention. HIV prevalence is shown in red and HIV incidence and 95% CI for each visit interval are shown in blue (green for circumcised men). The ART and MC coverage plots are also included to show the temporal association between scale-up of CHP and declines in HIV incidence. Declines in incidence were greater in men (adjIRR=0.46; 95%CI: 0.29-0.73) than in women (adjIRR=0.68, 95%CI: 0.50-0.94). HIV incidence was lower in circumcised compared to uncircumcised men (adjIRR=0.61: 95%CI: 0.48-0.79), but incidence declined significantly in both circumcised men (adjIRR=0.43; 95%CI: 0.19-0.99) and uncircumcised men by 2016 (adjIRR: 0.51, 95%CI: 0.29-0.88) (Fig.3, Supplementary Tables.4b-e).

R=0.68, 95%CI: 0.50-0.94). HIV incidence was lower in circumcised compared to uncircumcised men (adjIRR=0.61: 95%CI: 0.48-0.79), but incidence declined significantly in both circumcised men (adjIRR=0.43; 95%CI: 0.19-0.99) and uncircumcised men by 2016 (adjIRR: 0.51, 95%CI: 0.29-0.88) (Fig.3, Supplementary Tables.4b-e). There were HIV incidence declines among the majority of male and female age groups, and among both genders residing in trading and agrarian communities (Supplementary Fig.6-7). In sensitivity analyses, inclusion of birth cohort or inverse probability weights for selective participation and follow-up did not change inferences (Supplementary Tables 7-8). Though CHP coverage concurrently increased across RCCS communities (Fig. 2D., Supplemental Fig.7-8), we also assessed HIV incidence and individual-level HIV risk as functions of ART coverage, population prevalence of viremia, and MC coverage at the community-level. These analyses showed declining incidence and lower individual-level risk with increasing community ART and MC coverage and declining population viremia (Supplemental Fig.9-11).

ncidence and individual-level HIV risk as functions of ART coverage, population prevalence of viremia, and MC coverage at the community-level. These analyses showed declining incidence and lower individual-level risk with increasing community ART and MC coverage and declining population viremia (Supplemental Fig.9-11). DISCUSSION In this study, HIV incidence significantly declined with CHP scale-up, providing some of the first empiric evidence that CHP can have substantial population-level impact. The declines in HIV incidence are likely due to ART and MC scale-up; reduced sexual activity in late adolescence may also have contributed. HIV incidence declined less in women compared to men, suggesting that the combined direct effects of MC and indirect effects of female ART use differentially benefited men. Additional efforts are needed to avert new infections in women such as further scale-up of ART in men and potentially introducing new primary prevention interventions (e.g. Pre-exposure Prophylaxis or PrEP). We previously found that community-levels of MC and female ART at modest coverage levels were associated with lower community HIV incidence in males.22 A study in rural South Africa reported lower risk of individual-level HIV acquisition associated with higher rates of ART coverage, but that study did not assess temporal declines in incidence or MC coverage.10 Our finding of a 42% reduction in HIV incidence to 0.66/100py is substantial, but still well above the 0.1/100py incidence rate estimated as the threshold for HIV elimination.6 23

HIV acquisition associated with higher rates of ART coverage, but that study did not assess temporal declines in incidence or MC coverage.10 Our finding of a 42% reduction in HIV incidence to 0.66/100py is substantial, but still well above the 0.1/100py incidence rate estimated as the threshold for HIV elimination.6 23 From 2009 to 2016, the proportion of HIV-positive persons with viral suppression increased by 46%, suggesting that HIV viral suppression via ART likely reduced HIV exposure to uninfected opposite sex partners, consistent with other studies.12 13 24-26 By 2016, the rate of virologic suppression among HIV-positive persons was 75%, meeting the 2020 goal of the UNAIDS 90-90-90 initiative which modeling suggests could end the HIV epidemic by 2030.27 Our results demonstrate that ambitious ART scale-up goals can be achieved. Similar viral suppression results have been reported in Botswana (71%), although beneficial effects on HIV incidence rates in Botswana have not yet been reported.28 29

0-90 initiative which modeling suggests could end the HIV epidemic by 2030.27 Our results demonstrate that ambitious ART scale-up goals can be achieved. Similar viral suppression results have been reported in Botswana (71%), although beneficial effects on HIV incidence rates in Botswana have not yet been reported.28 29 MC coverage steadily increased to 59% by 2016, but remained below UNAIDS targets of 80% coverage.30 Scale up of ART and MC were highly correlated (Fig. 2D) so it is difficult to disaggregate their effects. Nevertheless, we attempted to address this issue empirically by assessing HIV incidence trends separately in men and women and in uncircumcised and circumcised men. Prior mathematical modeling studies suggest that there are substantial, long term indirect effects of MC on both female partner HIV incidence and in uncircumcised men; however, these benefits are unlikely to be realized until at least a decade after HIV prevalence declines resulting from direct effects of MC.31 Therefore, the significant reductions of HIV incidence in women and uncircumcised men observed in this study most likely result from the population-level impact of increasing ART coverage on HIV incidence. Notably, circumcised men had the sharpest declines in HIV incidence, nearly twice as great as uncircumcised men, likely because they benefit from the direct protective effect of MC and from the indirect effect of female partners on ART. In comparison, women and uncircumcised men had more moderate declines in incidence, likely because they largely benefit from indirect reduced exposure afforded by their partner’s ART use. Rates of ART use were lower in HIV-positive men, which would further attenuate benefits for women.31

fect of female partners on ART. In comparison, women and uncircumcised men had more moderate declines in incidence, likely because they largely benefit from indirect reduced exposure afforded by their partner’s ART use. Rates of ART use were lower in HIV-positive men, which would further attenuate benefits for women.31 Statistically significant HIV incidence declines were first observed in 2012 when ART and MC coverage levels reached 36% and 43%, respectively. It would be tempting to conclude that these coverage levels represent threshold effects, but because interventions were scaled concurrently and the impact of interventions may be delayed, we cannot reliably make such inferences from these empiric data alone. Defining intervention thresholds would also depend on the proportion of infections introduced from outside the population of interest, a quantity which likely varies across settings. We found reductions in sexual activity in both males and females aged 15-19. Prior RHSP studies showed a decline in HIV incidence among 15-19 year-old girls associated with factors such as delayed sexual debut, coincident with increased school enrollment.32 However, this age group represents a small fraction of all incident HIV infections in the RCCS with limited behavioral changes in older age groups suggesting its impact on population HIV incidence are likely modest. Of note, there were no significant changes in condom use in any age group which is sobering given many years of condom promotion and provision.

all fraction of all incident HIV infections in the RCCS with limited behavioral changes in older age groups suggesting its impact on population HIV incidence are likely modest. Of note, there were no significant changes in condom use in any age group which is sobering given many years of condom promotion and provision. This observational study meets almost all of Hill’s criteria for causality including a strong temporal association between CHP scale-up and HIV incidence declines, a dose-response relationship (i.e., greater declines in HIV incidence with increasing CHP coverage), consistency with prior studies of ART and MC, and biological plausibility.33 However, the study has a number of limitations. ART and MC coverage, and sexual behaviors were self-reported and may be subject to social desirability and other biases. However, there are no clear indications that any biases changed over time, and self-reported ART has been validated with high specificity in this population.19 Viral load testing was conducted on stored sera which may be subject to RNA degradation over time, potentially resulting in overestimation of viral suppression in the earlier survey and an underestimation of the magnitude of viral suppression over time.34 While RCCS has relatively high participation rates compared to other African population-based cohorts, there was substantial mobility which reduced participation and follow-up.35 36 However, participation rates among those present in the community increased over time, and sensitivity analyses to assess potential selection bias did not change our inferences.

ation rates compared to other African population-based cohorts, there was substantial mobility which reduced participation and follow-up.35 36 However, participation rates among those present in the community increased over time, and sensitivity analyses to assess potential selection bias did not change our inferences. An important consideration is whether these CHP coverage and HIV incidence results can be generalized. RCCS demographic and behavioral data are largely consistent with Demographic and Health Surveys in the region.37 RCCS is an open population-based cohort with extensive in and out-migration which likely minimized, though did not eliminate, potential Hawthorne effects of repeat observations. RHSP has conducted CHP intervention and prevention studies which may have increased ART and MC coverage.38-40 All RCCS participants are offered HIV testing services resulting in high coverage (98% in 2015). Although conditions in Rakai may have been favorable for rapidly scaling CHP services, the impact of these interventions on population-level HIV incidence provides proof of concept and should be generalizable. Indeed, data from the Uganda Ministry of Health’s National AIDS Control Program data indicates dramatic scale-up of CHP was also occurring nationally: ART and MC coverage were 68% and 54%, respectively, in 2016 (Steven Wiersma, personal communication, 2017).

idence provides proof of concept and should be generalizable. Indeed, data from the Uganda Ministry of Health’s National AIDS Control Program data indicates dramatic scale-up of CHP was also occurring nationally: ART and MC coverage were 68% and 54%, respectively, in 2016 (Steven Wiersma, personal communication, 2017). In summary, data from this longitudinal cohort in Rakai, Uganda show a 42% decline in HIV incidence associated with CHP, providing evidence that HIV control efforts can have a substantial population-level impact. Differential declines in HIV incidence by gender indicate a need for strengthening CHP efforts to benefit women, including improving ART coverage in men and consideration of newer, primary prevention interventions such as PrEP. Intensification of CHP efforts for both women and men including key underserved populations such as migrants, as well as long-term surveillance, are needed to determine whether HIV incidence can be further reduced to the levels necessary for elimination. Supplementary Material Supplementary Appendix Acknowledgements Presented in part at the Conference on Retroviruses and Opportunistic Infections (CROI), Seattle, February 13-17, 2017.

In summary, data from this longitudinal cohort in Rakai, Uganda show a 42% decline in HIV incidence associated with CHP, providing evidence that HIV control efforts can have a substantial population-level impact. Differential declines in HIV incidence by gender indicate a need for strengthening CHP efforts to benefit women, including improving ART coverage in men and consideration of newer, primary prevention interventions such as PrEP. Intensification of CHP efforts for both women and men including key underserved populations such as migrants, as well as long-term surveillance, are needed to determine whether HIV incidence can be further reduced to the levels necessary for elimination. Supplementary Material Supplementary Appendix Acknowledgements Presented in part at the Conference on Retroviruses and Opportunistic Infections (CROI), Seattle, February 13-17, 2017. Supported by the National Institute of Mental Health (R01MH107275), the National Institute of Allergy and Infectious Diseases (R01AI110324, U01AI100031, U01AI075115, R01AI110324, R01AI102939, K01AI125086-01), the National Institute of Child Health and Development (RO1HD070769, R01HD050180), and Division of Intramural Research of the National Institute for Allergy and Infectious Diseases, the World Bank, the Doris Duke Charitable Foundation, the Bill & Melinda Gates Foundation (#08113, 22006.02), the Johns Hopkins University Center for AIDS Research (P30AI094189), and the President’s Emergency Plan for AIDS Relief through the Centers for Disease Control and Prevention (NU2GGH000817). The findings and conclusions in this report are those of the authors and do not represent the official position of the funding agencies.

ns Hopkins University Center for AIDS Research (P30AI094189), and the President’s Emergency Plan for AIDS Relief through the Centers for Disease Control and Prevention (NU2GGH000817). The findings and conclusions in this report are those of the authors and do not represent the official position of the funding agencies. We thank the cohort participants and the many staff and investigators who have made this study possible over these many years. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1702150..

Introduction Mass drug administration (MDA) with ivermectin-containing regimens is the main strategy for elimination of lymphatic filariasis and onchocerciasis. Although generally safe, ivermectin distribution has led to severe adverse events (SAEs) in central African countries. More than 500 cases of characteristic post-ivermectin encephalopathy1 including ~60 fatal case, occurred during MDA and have therefore been reported to the Mectizan Donation Program since 1990. Of note, these neurologic SAEs have occurred exclusively in individuals with peripheral blood Loa loa microfilarial densities >30,000 microfilariae (mf) per milliliter,1 2 and are presumed to be related to eosinophil-mediated inflammation around dying mf and/or micro-embolization with subsequent loss of central nervous system vascular integrity. Current WHO guidelines allow ivermectin-based MDA to be implemented in areas where onchocerciasis is meso- or hyperendemic because the potential benefits of MDA are felt to outweigh the risk of ivermectin-associated SAEs, although enhanced surveillance for adverse events (AEs) is required. However, areas endemic for loiasis and hypo-endemic for onchocerciasis are spread throughout Central Africa,3 and remain a serious problem. For these areas, a “Test and (not) Treat” (TaNT) strategy has been proposed, wherein individuals with high L. loa mf loads (at risk for SAEs) are excluded from ivermectin treatment and the remaining population (typically >95%) can be treated safely.

s are spread throughout Central Africa,3 and remain a serious problem. For these areas, a “Test and (not) Treat” (TaNT) strategy has been proposed, wherein individuals with high L. loa mf loads (at risk for SAEs) are excluded from ivermectin treatment and the remaining population (typically >95%) can be treated safely. Successful implementation of the TaNT strategy requires a rapid, point-of-contact, fieldfriendly and highly accurate method to quantify L. loa mf. To this end, a mobile phone-based videomicroscope – the LoaScope (previously CellScope Loa) – was developed.4 The LoaScope automatically counts L. loa mf in peripheral blood collected in disposable rectangular capillaries without the need for sample processing using a smartphone coupled to a simple optical device (Figure S1 and Movie S1).4 To advance O. volvulus elimination in L. Loa co-endemic countries in Central Africa, we tested the feasibility of this TaNT strategy in the Okola health district in Cameroon, where ivermectin distribution was halted in 1999 after the occurrence of Loa-related encephalopathy. As such, the TaNT strategy allowed the safe reintroduction of ivermectin in all of the communities in this health district without provoking a single SAE.

of this TaNT strategy in the Okola health district in Cameroon, where ivermectin distribution was halted in 1999 after the occurrence of Loa-related encephalopathy. As such, the TaNT strategy allowed the safe reintroduction of ivermectin in all of the communities in this health district without provoking a single SAE. METHODS Study site The Okola health district (Figure S2) includes 11 health areas where, in 1999 MDA was halted by the Ministry of Public Health following 23 cases of encephalopathy that occured during the first treatment campaign. MDA only resumed in 5/11 areas deemed hyper- or mesoendemic (>20% onchocercal nodule prevalence in adult males). In 2013, nodule surveys in the 6 excluded health areas had prevalences of 6% to 40% consistent with hypo- or meso-endemic onchocerciasis.5 The entire Okola health district is known to be highly endemic for L. loa.6 7 Study design The TaNT strategy was implemented in the 92 villages of the 6 health areas untreated since 1999 (Figure S2). The timeline of the TaNT project is depicted in Figure S3 and the process is described in the Supplementary Methods. All individuals aged >=5 years were invited to participate.

METHODS Study site The Okola health district (Figure S2) includes 11 health areas where, in 1999 MDA was halted by the Ministry of Public Health following 23 cases of encephalopathy that occured during the first treatment campaign. MDA only resumed in 5/11 areas deemed hyper- or mesoendemic (>20% onchocercal nodule prevalence in adult males). In 2013, nodule surveys in the 6 excluded health areas had prevalences of 6% to 40% consistent with hypo- or meso-endemic onchocerciasis.5 The entire Okola health district is known to be highly endemic for L. loa.6 7 Study design The TaNT strategy was implemented in the 92 villages of the 6 health areas untreated since 1999 (Figure S2). The timeline of the TaNT project is depicted in Figure S3 and the process is described in the Supplementary Methods. All individuals aged >=5 years were invited to participate. The TaNT process consisted of registration of consenting (or assenting) individuals >=5 years of age, LoaScope quantification of L. loa microfilarial density, treatment of eligible individuals with ivermectin (150μg/kg) and surveillance for AEs. Non-pregnant subjects excluded from ivermectin distribution because of high L. loa mf counts were given albendazole (400 mg) for intestinal deworming. Self-declared pregnant women were not treated with ivermectin or albendazole, but were offered iron and folic acid tablets. Each participant was given a card (Figure S4) with their L. loa mf count, the treatment received and a contact phone number for questions and/or reporting of AEs.

400 mg) for intestinal deworming. Self-declared pregnant women were not treated with ivermectin or albendazole, but were offered iron and folic acid tablets. Each participant was given a card (Figure S4) with their L. loa mf count, the treatment received and a contact phone number for questions and/or reporting of AEs. Quantification of L. loa microfilaremia The use of the LoaScope and its performance have been described previously.4 A threshold of 26,000 mf per milliliter was initially selected for ivermectin treatment, based on the lower 95% confidence interval around the 30,000 mf per milliliter threshold below which no neurologic SAEs were observed in prior studies8-10 and the calculated false negativity rate of 1 in 10 million (<0.00001%).4 Two weeks after the study start, a case of conjunctival hemorrhage, similar to those described previously,11 occurred in a subject with a LoaScope L. loa mf count of 24,599 mf per milliliter. For potential safety reasons, the exclusion threshold of the LoaScope was decreased to 20,000 mf per milliliter for the remainder of the trial. Calibrated (50 μL) thick smears were performed as a backup for samples unable to be analyzed with the LoaScope, to identify and quantify Mansonella perstans mf, and to corroborate the accuracy of the LoaScope. Smears were read by 2 different microscopists blinded to the LoaScope results. Dried blood spots collected on filter paper were archived and stored at -80°C.

as a backup for samples unable to be analyzed with the LoaScope, to identify and quantify Mansonella perstans mf, and to corroborate the accuracy of the LoaScope. Smears were read by 2 different microscopists blinded to the LoaScope results. Dried blood spots collected on filter paper were archived and stored at -80°C. Assessment of exposure to onchocerciasis Ov16 IgG4 antibodies positivity from eluted single blood spots (10 μl equivalent) was determined the SD Bioline Onchocerciasis IgG4 Rapid Test.13,14 Results were read and recorded at 24 hours. Monitoring of post-treatment adverse reactions Monitoring for AEs was performed by two surveillance teams, each composed of a physician and a driver, with the assistance of selected community members (pre-Community Drug Distributors (pre-CDDs)) and local nurses. The surveillance teams visited each village on days 1, 2, 3 and 6 post-treatment, examined all individuals complaining of AEs and provided symptomatic treatment if indicated. In addition, the team toured the entire community by car to identify additional individuals with AEs. All AEs were recorded using a standardized form (Figure S5). A Karnofsky performance score index was assigned to each patient examined. Clinical management was based on reference guidelines.15

c treatment if indicated. In addition, the team toured the entire community by car to identify additional individuals with AEs. All AEs were recorded using a standardized form (Figure S5). A Karnofsky performance score index was assigned to each patient examined. Clinical management was based on reference guidelines.15 Statistical analysis Medians and interquartile ranges were used as measurements of central tendency. Associations between individual factors (gender, age, L. Loa mf density (assessed by LoaScope), presence of Ov16 IgG4, presence of M. perstans mf (assessed by calibrated blood smear microscopy) and the occurrence of AEs were assessed using multivariable logistic regression. The logistic regression coefficients were used to calculate population attributable fractions.16 The calibrated thick smear was used as the reference test for assessment of the specificity and negative predictive value of the LoaScope. Ethical agreement This study was authorized by the National Ethics Committee of Cameroon (ethical clearance n° 2013/11/370/L/CNERSH/SP) and approved by the Division of Operational Research at the Ministry of Health (Administrative authorization n° D30- 571/L/MINSANTE/SG/DROS/CRSPE/BBM). All volunteers provided written signed consent (or parental consent in the case of minors) before undergoing blood sampling and again before receiving treatment.

SH/SP) and approved by the Division of Operational Research at the Ministry of Health (Administrative authorization n° D30- 571/L/MINSANTE/SG/DROS/CRSPE/BBM). All volunteers provided written signed consent (or parental consent in the case of minors) before undergoing blood sampling and again before receiving treatment. Authors’ contributions to the study Authors’ contributions to this study are as follows: JK, SDP, CDM, ADK, TBN and MB designed the study; JK, SDP, CBC, MHB, MVDA CDM, HCND, RGK, GRN, PN, JBTM, SW, DAF, ADK, TBN and MB gathered the data; SDP and CBC analyzed the data; TNB and MB vouch for the data and the analysis; MHB, MVDA, DAF provided diagnostic technology development and support; JK, SDP, CDM, HCND, WAS, DAF, ADK, TBN and MB wrote the paper; and JK, SDP, CDM, DAF, ADK, TBN and MB decided to publish the paper. SDP wrote the first draft of the manuscript. RESULTS Population Characteristics A total of 16,259 individuals were examined during the TaNT process (Figure 1). The median age of the examined populations ranged from 17 to 26 years in the different health areas and the sex distribution was relatively equal (48% male). The prevalence of Ov16 IgG4 antibody in the six health areas varied from 15.3% to 29.9%. The prevalence of L. loa microfilaremia varied from 15.3% to 22.8%, and the proportion of individuals with more than 20,000 Loa mf per milliliter as determined by LoaScope ranged from 1.3% in the Ngoya health area to 2.4% in the Nlong and Ekekam III health areas (Table 1, Table S2).

th areas varied from 15.3% to 29.9%. The prevalence of L. loa microfilaremia varied from 15.3% to 22.8%, and the proportion of individuals with more than 20,000 Loa mf per milliliter as determined by LoaScope ranged from 1.3% in the Ngoya health area to 2.4% in the Nlong and Ekekam III health areas (Table 1, Table S2). Figure 1. Flowchart of the population examined for L. loa microfilaremia and treated with ivermectin in the six health areas of the Okola district. Loa+++ indicates the number of individuals identified as at risk of post-ivermectin severe adverse events and excluded from ivermectin treatment.

th areas varied from 15.3% to 29.9%. The prevalence of L. loa microfilaremia varied from 15.3% to 22.8%, and the proportion of individuals with more than 20,000 Loa mf per milliliter as determined by LoaScope ranged from 1.3% in the Ngoya health area to 2.4% in the Nlong and Ekekam III health areas (Table 1, Table S2). Figure 1. Flowchart of the population examined for L. loa microfilaremia and treated with ivermectin in the six health areas of the Okola district. Loa+++ indicates the number of individuals identified as at risk of post-ivermectin severe adverse events and excluded from ivermectin treatment. Table 1. Demographics, onchocerciasis prevalence and L. loa microfilaremia levels in the population of the six health areas of the Okola district (Cameroon) . Census Median age (IQR) M/F O. volvulus antibody positivity (%) L. loa microfilaremia (%) Prevalence < 8000 mf per milliliter 8-20,000 mf per milliliter > 20,000 mf per milliliter Ekekam III 2753 26 (13-49) 0.51 18.8 22.8 91.7 5.9 2.4 Lobo 3041 23 (12-45) 0.50 29.9 21.3 93.1 5.3 1.6 Mvoua 4637 17 (10-43) 0.49 20.2 18.9 94.4 3.7 1.9 Ngoya 6612 17 (10-38) 0.47 15.3 15.3 95.4 3.3 1.3 Nlong 2007 20 (12-50) 0.51 23.8 20.3 94.5 3.1 2.4 Okola 7380 18 (11-38) 0.58 27 16.1 95.1 3.5 1.4 Total 26430 18 (11-42) 0.48 22.4 17.8 94.5 3.9 1.6 Table 2. Adverse events recorded during the post-treatment surveillance process. Adverse Events No of adverse events No. with Loa mf (%) No. with no Loa mf (%) P value Pruritus 564 188 (33.3) 376 (66.7) <0.001 Asthenia 389 171 (44) 218 (56) 0.002 Headache 326 149 (45.7) 177 (54.3) 0.14 Rash 274 52 (19) 222 (81) <0.001 Back Pain 257 128 (49.8) 129 (50.2) 0.97 Arthralgias 235 124 (52.8) 111 (47.2) 0.39 Edema 125 21 (16.8) 104 (83.2) <0.001 Myalgia 115 51 (44.4) 64 (55.6) 0.20 Vertigo 106 48 (45.3) 58 (54.7) 0.35 Anorexia 89 42 (47.2) 47 (52.8) 0.57 Abdominal pain 67 19 (28.4) 48 (71.6) <0.001 Blurred vision 66 27 (40.9) 39 (59) 0.15 Difficulty ambulating 58 27 (46.6) 31 (53.4) 0.65 Diarrhea 46 18 (39.1) 28 (60.9) 0.14 Difficulty in getting upright 37 20 (54) 17 (46) 0.63 Lymphadenopathy 23 8 (34.8) 15 (65.2) 0.17 Conjunctival hemorrhage 20 14 (68.4) 6 (31.6) 0.14 Conjunctival itching 13 7 (53.9) 6 (46.2) 0.77 Tinnitus 6 4 (66.7) 2 (33.3) Not tested Temporary hearing loss 2 0 (0) 2 (100) Not tested Total 2818 1118 (39.7) 1702 (60.4) <0.001 Test and Not Treat Between 50 and 100 participants were typically examined per village per day. The mean time from finger prick to LoaScope result was 2-3 minutes. The LoaScope results were immediately available for 16,099/16,259 individuals (99%) and were delayed for 160 individuals (1%) because of technical problems requiring determination of the mf count by calibrated thick smear.

examined per village per day. The mean time from finger prick to LoaScope result was 2-3 minutes. The LoaScope results were immediately available for 16,099/16,259 individuals (99%) and were delayed for 160 individuals (1%) because of technical problems requiring determination of the mf count by calibrated thick smear. Ivermectin was administered to 15,522 individuals (95.5%) with mf levels below the established threshold. Seven hundred and thirty seven (4.5%) subjects were excluded from ivermectin therapy. Of these, 340 (2.1%) were excluded because of a L. loa density above the risk-threshold, 228 (1.4%) because of poor health (signs or symptoms consistent with a serious acute or chronic concomitant illness) or inebriation, and 169 (1%) because of pregnancy or breastfeeding. The proportion of excluded individuals per village varied from 0% to 15.1% (Figure S6). All excluded individuals (except pregnant women), were treated with albendazole (400 mg). The median treatment coverage in the district was 55% of the total population (interquartile range between villages: 42.9–64.1%), and 64% of the targeted population. The prevalence of O. volvulus-specific antibody (Ov16 IgG4) was 22.0% in individuals who received ivermectin, 25.4% in those excluded for pregnancy or illness, and 33.5% in those with a L. loa density above the risk-threshold. Thus, individuals who were not treated because of Loa microfilaremia and who were potentially infected with O. volvulus represented only 0.7% of the examined population.

ividuals who received ivermectin, 25.4% in those excluded for pregnancy or illness, and 33.5% in those with a L. loa density above the risk-threshold. Thus, individuals who were not treated because of Loa microfilaremia and who were potentially infected with O. volvulus represented only 0.7% of the examined population. Frequency and types of AEs Among the 15,522 individuals treated with ivermectin, 934 (6%) had documented AEs. The incidence of AEs decreased slightly from 6.6% (464/7,065) to 5.6% (470/8,457) (p<0.0001) after reducing the exclusion threshold from 26,000 to 20,000 mf per milliliter. Dermatologic manifestations were most common, followed by systemic and rheumatologic manifestations (Table 2). Eight hundred and sixty-nine people (93%) had a Karnofsky score of 90, and 65 (7%) had a score of 80. All AEs resolved within one week without treatment or with basic supportive therapy (anti-histamines, non-steroidal anti-inflammatory drugs, or acetaminophen).

ed by systemic and rheumatologic manifestations (Table 2). Eight hundred and sixty-nine people (93%) had a Karnofsky score of 90, and 65 (7%) had a score of 80. All AEs resolved within one week without treatment or with basic supportive therapy (anti-histamines, non-steroidal anti-inflammatory drugs, or acetaminophen). Both L. loa microfilaremia and the presence of Ov16-specific IgG4 were assessed in 888 of the 934 individuals who developed an AE. Among these, 43.2% had neither L. loa mf nor Ov16 IgG4, 22.3% had only L. loa mf, 23.9% had only Ov16 IgG4, and 10.6% had both L. loa mf and Ov16 IgG4. Multivariable regression indicates that AEs were significantly more frequent in older individuals, females, and individuals with either L. loa mf or Ov16 IgG4 (Figure 2). The risk of AEs associated with presence of Ov16 IgG4 was similar to that associated with harboring 1-8000 Loa mf per milliliter (Odds ratio (OR)=1.61 and 1.71, respectively) and was about half the risk associated with harboring 8000-20,000 Loa mf per milliliter (OR=3.00). The risk of AEs associated with both L. loa microfilaremia and O. volvulus IgG4 was similarly increased in persons harboring 1-8000 L. loa mf per milliliter (OR=2.47) and in those harboring 8000-20,000 L. loa mf per milliliter. Population attributable fractions of AEs for L. loa mf density of 1-8000 mf per milliliter, 8000-20,000 mf per milliliter and Ov16 IgG4 were 8.0%, 8.3% and 12.2%, respectively.

similarly increased in persons harboring 1-8000 L. loa mf per milliliter (OR=2.47) and in those harboring 8000-20,000 L. loa mf per milliliter. Population attributable fractions of AEs for L. loa mf density of 1-8000 mf per milliliter, 8000-20,000 mf per milliliter and Ov16 IgG4 were 8.0%, 8.3% and 12.2%, respectively. Figure 2. Results of multivariate logistic regression of occurrence of post-ivermectin adverse events in relation to individual factors Dots with error bars represent odds-ratios (OR) and 95% confidence intervals. The dotted line (OR=1) 1 represents an absence of association. Agreement between LoaScope and calibrated blood smear microscopy Figure 3 shows that the distributions of L. loa mf density in the population using the LoaScope and thick smear microscopy were similar. The specificity and negative predictive values of the LoaScope to identify individuals with mf counts below 20,000 mf per milliliter (as assessed by microscopy) were 99.7% (95% confidence interval: 99.6 – 99.8) and 99.7% (99.6 – 99.7), respectively. Figure 3. Cumulative frequency distribution of L. loa microfilarial density in the population with tails of distribution censored for density above 50,000 mf per milliliter (mL). Dotted vertical lines (1), (2) and (3) correspond to the 8,000 (1), 20, 000 (2) and 26,000 )3) Loa mf/ml cutoffs used to determine treatment exclusion thresholds (2 and 3) and information relevant to the increased likelihood of adverse events (1) provided to each participant.

ensity above 50,000 mf per milliliter (mL). Dotted vertical lines (1), (2) and (3) correspond to the 8,000 (1), 20, 000 (2) and 26,000 )3) Loa mf/ml cutoffs used to determine treatment exclusion thresholds (2 and 3) and information relevant to the increased likelihood of adverse events (1) provided to each participant. Discussion Extension of ivermectin-based MDA to areas hypoendemic for onchocerciasis and coendemic for loiasis remains a significant obstacle to the success of onchocerciasis elimination programs in Africa. In the current study, a LoaScope-based TaNT strategy was used to safely treat more than 15,000 individuals with ivermectin in such an area. Although there was initial reticence to participate in some villages because of the memory of the SAEs (including deaths) that occurred in 1999, 16,259 of the 22,842 individuals aged >= 5 years old recorded during the initial census (71.1%) participated in the TaNT campaign. This suggests that TaNT is an acceptable strategy even in populations with a history of previous ivermectin-related SAEs. Though not formally assessed, it is likely that fear of SAEs was the main reason for non-participation.

d >= 5 years old recorded during the initial census (71.1%) participated in the TaNT campaign. This suggests that TaNT is an acceptable strategy even in populations with a history of previous ivermectin-related SAEs. Though not formally assessed, it is likely that fear of SAEs was the main reason for non-participation. During the first MDA campaign conducted in 1999 in Okola, 23 cases of neurological SAEs, including three fatalities, were recorded among the 6,000 individuals who received ivermectin before MDA was stopped.17 The incidences of post-ivermectin neurological SAEs and deaths were therefore 38/10,000 and 5/10,000, respectively. Extrapolating these data to the population enrolled in the present study, a minimum of 62 cases of neurological SAEs and 8 deaths were theoretically prevented by TaNT. Although some individuals (6%) complained of ivermectin-associated AEs during the TaNT campaign, the proportion was lower than that typically observed after ivermectin MDA for onchocerciasis in areas not endemic for loiasis: 13.1% in south-east Nigeria,18 12% and 20% in northern Cameroon,19 and 21.4% in eastern Sudan.20 It was also much lower than the 26.3% recorded in a neighboring Loa-endemic area of central Cameroon.2 The most likely explanation for the lower frequency of AEs recorded in the Okola district is that onchocerciasis is hypo- and mesoendemic in this area.

12% and 20% in northern Cameroon,19 and 21.4% in eastern Sudan.20 It was also much lower than the 26.3% recorded in a neighboring Loa-endemic area of central Cameroon.2 The most likely explanation for the lower frequency of AEs recorded in the Okola district is that onchocerciasis is hypo- and mesoendemic in this area. The LoaScope operators underwent a 1-hour training session 2 weeks before the field operations. This training was sufficient for the entire study, and the teams noted the ease of use and reliability of the device despite daily use and demanding field conditions. Because L. loa mf are diurnally periodic,21 LoaScope examinations (and treatment) started at 10 am and ended at 4 pm. During this TaNT campaign, up to 162 individuals were examined per village per day.

y, and the teams noted the ease of use and reliability of the device despite daily use and demanding field conditions. Because L. loa mf are diurnally periodic,21 LoaScope examinations (and treatment) started at 10 am and ended at 4 pm. During this TaNT campaign, up to 162 individuals were examined per village per day. Whereas the present study clearly shows that the TaNT procedure is safe and feasible at a district level, moving TaNT from the operational research arena to Central African-wide implementation will depend on a number of factors, including greater reliance on local personnel for the census and post-treatment surveillance and smaller teams (a “tester” and a “treater”) for the TaNT process itself. If only a single TaNT round were needed, this would have a major impact on the applicability of this approach at a larger scale. Since ivermectin has marked microfilaricidal and probable embryostatic activity against L. loa, marked and sustained reduction in L. loa mf density is expected for one year after the TaNT campaign. In fact, in a neighboring district, the average reduction in L. loa microfilarial load was >74% one year after a single dose of ivermectin, and no individual with a pre-treatment density <30,000 mf per milliliter had a count above this level after 12 months.22 Thus, it seems likely that a single community-wide round of TaNT will be necessary, with individual testing in subsequent years restricted to previously ivermectin-untreated individuals. This hypothesis, as well as operationality, performance and cost of a TaNT conducted by 3-member teams will be assessed in late 2017.

Thus, it seems likely that a single community-wide round of TaNT will be necessary, with individual testing in subsequent years restricted to previously ivermectin-untreated individuals. This hypothesis, as well as operationality, performance and cost of a TaNT conducted by 3-member teams will be assessed in late 2017. Given the low percentage (2.4%) of the total population excluded from ivermectin treatment and the proposed implementation of TaNT in areas hypo- and meso-endemic for onchocerciasis, it is unlikely that excluded individuals will be a significant reservoir of O. volvulus microfilariae at the community level. Nevertheless, some excluded individuals are likely to be infected with O. volvulus and, for ethical reasons, should be treated with effective and safe drug regimens, particularly in the setting of clinical manifestations of onchocerciasis. Although a 4-6-week course of doxycycline, a regimen known to be macrofilaricidal for O. volvulus23 but not L. loa24, is impractical at the community level, it could be used safely in this context.

e treated with effective and safe drug regimens, particularly in the setting of clinical manifestations of onchocerciasis. Although a 4-6-week course of doxycycline, a regimen known to be macrofilaricidal for O. volvulus23 but not L. loa24, is impractical at the community level, it could be used safely in this context. In summary, this TaNT strategy based on a novel and scalable point-of-contact tool that allows rapid identification (and exclusion from ivermectin-based treatment) of individuals at risk of Loa-related SAEs has enabled district-level community treatment of onchocerciasis. Though this TaNT strategy was motivated by the need to tackle hypoendemic onchocerciasis in Central Africa, it could also be considered for other foci co-endemic for onchocerciasis and loiasis. As many (but not all) meso-hyperendemic areas are already covered by CDTI, TaNT would target ivermectinnaïve individuals and systematic non-compliers. Supplementary Material Supplementary Appendix Funding and external contributions The study was funded by the Bill and Melinda Gates Foundation and in part by the Division of Intramural Research, NIAID, NIH and the USAID through the Higher Education Solutions Network partnership with UC Berkeley. Ivermectin was provided by the Mectizan Donation Program through the National Programme for Onchocerciasis Control of Cameroon. KLA-Tencor (Milpitas, CA, USA) supported production of LoaScope devices. The sponsors had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

vermectin was provided by the Mectizan Donation Program through the National Programme for Onchocerciasis Control of Cameroon. KLA-Tencor (Milpitas, CA, USA) supported production of LoaScope devices. The sponsors had no role in study design, data collection, data analysis, data interpretation, or writing of the report. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1612665..

Introduction Globally, maternal mortality ratios range from 3 to 1360 per 100,000 births; neonatal mortality rates from 0.95 to 40.6 per 1000; and stillbirths from 1.2 to 56.3 per 1000, with low- and middle- income countries experiencing rates an order of magnitude higher than high-income regions.1,2 While we have made progress in recent decades, there is substantial room for improvement.1,3-5 Despite a dramatic shift from home to facility-based births, birth attendants often do not perform practices known to reduce mortality and mortality rates have not decreased as expected.6 Research has shown that programs solely seeking to strengthen birth attendants’ training or to improve supply availability are insufficient to meaningfully improve patient care or outcomes.7 Conversely, interventions incorporating job aids and on-site coaching have proven effective in improving individual clinical practices, such as newborn resuscitation or active management of the third stage of labor, as well as outcomes.8-12 To bridge the gap between evidence and practice, the World Health Organization (WHO) created the Safe Childbirth Checklist, a practical tool to assist birth attendants in planning for and performing a more complete bundle of 28 essential birth practices.13,14 These key practices address the most common causes of avoidable mortality for women and newborns.15

practice, the World Health Organization (WHO) created the Safe Childbirth Checklist, a practical tool to assist birth attendants in planning for and performing a more complete bundle of 28 essential birth practices.13,14 These key practices address the most common causes of avoidable mortality for women and newborns.15 Studies have previously shown that, when well implemented at a small scale, the WHO Safe Childbirth Checklist improves facility-based birth attendants’ adherence to evidence-based care.16-18 We tested the BetterBirth program, a coaching-based implementation of the Checklist, in a large-scale, matched-pair, cluster-randomized controlled trial in Uttar Pradesh, India.19 We intended our intervention to support providers at multiple levels of the health system in using the Checklist appropriately, to identify gaps in facilities’ quality of care, and to activate resources (e.g. skills training and supply provision) within the existing healthcare system to address these gaps (Figure 1). We hypothesized that the intervention, implemented at the facility (cluster level), would improve quality of care in facility-based childbirth by increasing birth attendants’ adherence to Checklist practices. Further, we hypothesized that this increase in adherence to practices would lead to a decrease in our composite outcome, consisting of maternal mortality [day 0-7], maternal severe morbidity [day 0-7], early neonatal mortality [day 0-7], and stillbirth.

by increasing birth attendants’ adherence to Checklist practices. Further, we hypothesized that this increase in adherence to practices would lead to a decrease in our composite outcome, consisting of maternal mortality [day 0-7], maternal severe morbidity [day 0-7], early neonatal mortality [day 0-7], and stillbirth. Methods Trial Design We conducted a matched-pair, cluster-randomized controlled trial in government health facilities, which received either the BetterBirth program, a coaching-based implementation of the WHO Safe Childbirth Checklist (60 facilities; Figure 1) or the existing standard of care (60 facilities). We described the methodology of the BetterBirth Trial19, the BetterBirth Program/intervention20,21, and our Data Quality Assurance system22 elsewhere. Study Setting and Participants The most populous state in India (204 million, 77% rural),23 Uttar Pradesh is a high-priority region for national and international public-health organizations due to its persistently high neonatal mortality rate (25 per 1000) and maternal mortality ratio (258 per 100,000).24,25

Methods Trial Design We conducted a matched-pair, cluster-randomized controlled trial in government health facilities, which received either the BetterBirth program, a coaching-based implementation of the WHO Safe Childbirth Checklist (60 facilities; Figure 1) or the existing standard of care (60 facilities). We described the methodology of the BetterBirth Trial19, the BetterBirth Program/intervention20,21, and our Data Quality Assurance system22 elsewhere. Study Setting and Participants The most populous state in India (204 million, 77% rural),23 Uttar Pradesh is a high-priority region for national and international public-health organizations due to its persistently high neonatal mortality rate (25 per 1000) and maternal mortality ratio (258 per 100,000).24,25 The Government of Uttar Pradesh permitted the trial to proceed in 38 districts, where we identified 320 eligible facilities. We considered a facility eligible if it was designated as a Primary, Community, or First Referral Unit Health Center; had >=1000 deliveries annually; had >=3 birth attendants trained as auxiliary nurse midwives (or higher); had no other concurrent quality improvement or research programs; and had district and facility leadership willing to participate. The final study sample included 120 facilities across 24 districts (Supplement 1).

had >=1000 deliveries annually; had >=3 birth attendants trained as auxiliary nurse midwives (or higher); had no other concurrent quality improvement or research programs; and had district and facility leadership willing to participate. The final study sample included 120 facilities across 24 districts (Supplement 1). We matched facilities (i.e. clusters) on the following criteria prior to randomization: geographic zone, functional classification, distance to a district hospital, annual birth volume, and number of birth attendants (Supplement 1). We randomized facilities to study arm within the matched pair (Figure 2). After matching and randomization, we collected baseline practice-adherence data in 10 sites to confirm successful matching (Figure 2). Women registered for labor and delivery—excluding women who delivered outside the facility, women who were referred in from another facility, or women who were managed for abortion— were eligible for the study. At each intervention facility and its matched control site, we began enrolling patients 2 months after intervention initiation. Enrollment continued until the site’s target sample size was reached or for 24 weeks after intervention completion, whichever occurred first with a 12-week minimum follow-up.

le for the study. At each intervention facility and its matched control site, we began enrolling patients 2 months after intervention initiation. Enrollment continued until the site’s target sample size was reached or for 24 weeks after intervention completion, whichever occurred first with a 12-week minimum follow-up. Intervention We implemented the BetterBirth program following the Engage-Launch-Support model (Figure 1) piloted at non-study sites in Karnataka and Uttar Pradesh, India.18,21,26 Coaches (nurses) and Coach Team Leaders (physicians or public-health professionals), unaffiliated with facilities and comprehensively trained to apply a standard behavior-change framework, conducted site visits over the 8-month Support phase.20,21 We expected Coaches to conduct 43 day-long visits to each facility, beginning twice-weekly and tapering to monthly visits. Coach Team Leaders accompanied Coaches on alternating visits (23 total visits) (Figure 1). Each facility chose at least one staff member to serve as a Childbirth Quality Coordinator, a local champion for the use of the Checklist and continued coaching.

ch facility, beginning twice-weekly and tapering to monthly visits. Coach Team Leaders accompanied Coaches on alternating visits (23 total visits) (Figure 1). Each facility chose at least one staff member to serve as a Childbirth Quality Coordinator, a local champion for the use of the Checklist and continued coaching. Coaches motivated birth attendants to use the Checklist and to identify, understand, and resolve barriers to providing quality care.20,21 Coach Team Leaders supported facility leadership in fostering team communication and addressing gaps in care at facility and district levels by accessing resources through the established healthcare system. Cloud-based data collection enabled rapid feedback on a facility’s progress. We provided no clinical-skills training, financial support, or clinical supplies (except paper copies of the Checklist). Data Collection and Outcomes We measured a composite outcome of events occurring within the first 7-days postpartum, incorporating stillbirth; early neonatal death; maternal death; or self-reported maternal severe morbidity, including seizures, loss of consciousness for >1 hour, fever with foul-smelling vaginal discharge, hemorrhage, or stroke. We selected morbidity measures based on the WHO maternal near-miss approach using questions previously validated for self-report.27-30 We calculated an additional composite outcome consisting of mortality events only.

loss of consciousness for >1 hour, fever with foul-smelling vaginal discharge, hemorrhage, or stroke. We selected morbidity measures based on the WHO maternal near-miss approach using questions previously validated for self-report.27-30 We calculated an additional composite outcome consisting of mortality events only. Secondary maternal outcomes by 7-days postpartum included maternal mortality, maternal morbidity, inter-facility transfer (referral), cesarean section, hysterectomy, blood transfusion, and return to the facility for a health problem. Secondary newborn outcomes included stillbirth, early neonatal death, and inter-facility transfer. We assessed all outcomes from facility register information and by contacting women or close family members by telephone between 8 and 22 days post-partum. If we received no response by 22 days post-partum, a fieldworker conducted a home visit. Additionally, we selected a convenience sample of 15 matched pairs of facilities in which trained nurse-data-collectors directly observed birth attendants providing care over a 12-hour (daytime) shift at 2 months and 12 months after intervention initiation (4 months after coaching cessation). These independent observers measured practice adherence, including supply availability (Table 2). Intervention staff and independent observers were not present at the same facility simultaneously.

over a 12-hour (daytime) shift at 2 months and 12 months after intervention initiation (4 months after coaching cessation). These independent observers measured practice adherence, including supply availability (Table 2). Intervention staff and independent observers were not present at the same facility simultaneously. Due to the nature of the intervention, we were unable to blind any facility staff, most trial staff, or any investigators to the identity of intervention and control facilities. Call center staff, who collected the majority of outcome data, were blinded to facility assignment. Sample Size A priori, we hypothesized a 15% reduction in the primary composite outcome in the intervention arm. We estimated the intracluster (within facility) correlation (ICC) to be 0.01 and the matching effect to reduce the standard error by 45%, basing parameters on previous studies.31 We aimed to enroll 171,964 women (85,982 per arm) to detect a decrease from 60 events per 1000 births (control arm) to 51 events per 1000 births (intervention arm) with 80% power and alpha=0.05. Based on limited data available from Uttar Pradesh, the baseline rate of the primary composite outcome may have been as high as 120 events per 1000 births. The baseline rate used in calculations was purposively set lower than the estimated due to limited information as well as inclusion of community-based birth events in the available data, which may have elevated mortality rates.

e rate of the primary composite outcome may have been as high as 120 events per 1000 births. The baseline rate used in calculations was purposively set lower than the estimated due to limited information as well as inclusion of community-based birth events in the available data, which may have elevated mortality rates. In assessing practice adherence, we assumed an ICC (within facility) of 0.01 and a design effect of matching of 0.75. With 15 matched pairs, we had more than 80% power at alpha=0.05 to detect an absolute difference of 8.5% in any birth practice between study arms. Statistical Methods Using an intent-to-treat approach, we compared outcomes between study arms using a Rao- Scott chi-square test, accounting for the matched-pair, cluster design.32 The main outcome was the dichotomous composite outcome that was present if any of the 3 main outcomes occurred (maternal mortality, stillbirth/early neonatal mortality, or maternal severe morbidity). This variable was then used to estimate a composite relative risk.33,34 An additional secondary composite outcome included maternal and perinatal mortality only. In secondary analyses, each of the main outcomes were compared across arms; a Rao-Scott 3-degree-of-freedom test was used to assess the overall causal effect. No adjustment for multiplicity of testing was made.

relative risk.33,34 An additional secondary composite outcome included maternal and perinatal mortality only. In secondary analyses, each of the main outcomes were compared across arms; a Rao-Scott 3-degree-of-freedom test was used to assess the overall causal effect. No adjustment for multiplicity of testing was made. For the subset of directly observed births, we calculated adherence frequencies for each measured practice at 2 months and 12 months after intervention initiation. Further, we calculated an aggregate adherence score by summing the total number of 18 non-conditional practices performed in each delivery (Supplement 1). We generated the mean number of practices (presented as a fraction of 18) performed in each study arm and compared the differences at each time point, using a Rao-Scott chi-square test.32 For comparison of individual practices, we used a bias-corrected logistic regression approach that can handle zero cells and complete separation of points within strata and clusters.35 We conducted all statistical analyses using SAS v9.4 (SAS Institute, Cary, NC).

t each time point, using a Rao-Scott chi-square test.32 For comparison of individual practices, we used a bias-corrected logistic regression approach that can handle zero cells and complete separation of points within strata and clusters.35 We conducted all statistical analyses using SAS v9.4 (SAS Institute, Cary, NC). Ethical Compliance Each facility’s leadership provided facility-level consent for participation and permission for trial staff to collect de-identified data on every eligible woman from facility registers. Before patient discharge, we obtained verbal informed consent and contact information from each woman or her surrogate for follow-up. Data collectors reconfirmed verbal consent at start of the follow-up call or visit. In directly observed births, women or their surrogates provided written consent for observation. At trial initiation, birth attendants and facility staff verbally agreed to participate. Before an independent observer collected data, the birth attendant verbally reconfirmed agreement. Electronic data were de-identified and stored in a HIPAA-compliant database to ensure participant privacy.

Ethical Compliance Each facility’s leadership provided facility-level consent for participation and permission for trial staff to collect de-identified data on every eligible woman from facility registers. Before patient discharge, we obtained verbal informed consent and contact information from each woman or her surrogate for follow-up. Data collectors reconfirmed verbal consent at start of the follow-up call or visit. In directly observed births, women or their surrogates provided written consent for observation. At trial initiation, birth attendants and facility staff verbally agreed to participate. Before an independent observer collected data, the birth attendant verbally reconfirmed agreement. Electronic data were de-identified and stored in a HIPAA-compliant database to ensure participant privacy. The study protocol was approved by the following ethical review boards: Community Empowerment Lab (Ref no: 2014006), Jawaharlal Nehru Medical College (Ref no: MDC/IECHSR/2015-16/A-53), Harvard T.H. Chan School of Public Health (Protocol 21975-102), Population Services International (Protocol ID: 47.2012), World Health Organization (Protocol ID: RPC 501), and Indian Council of Medical Research. The protocol was reviewed and reapproved on an annual basis. A Data Safety Monitoring Board (DSMB) met every 6 months after enrollment initiation, including interim-analysis when 30% of data was collected (Supplement 1).

47.2012), World Health Organization (Protocol ID: RPC 501), and Indian Council of Medical Research. The protocol was reviewed and reapproved on an annual basis. A Data Safety Monitoring Board (DSMB) met every 6 months after enrollment initiation, including interim-analysis when 30% of data was collected (Supplement 1). Results Facility and Patient Characteristics All 120 matched and randomized facilities initiated the study. During data collection, 2 facilities closed for renovations, halting enrollment prematurely for those facilities and their matched pairs. Of the 163,939 women registered for labor and delivery (83,166 intervention; 80,773 control), 98.3% (161,107) were eligible for study inclusion and 97.9% (157,689) consented (Figure 2). We collected 7-day outcomes for all but 544 (0.3%) of the consented women. We found no significant differences between intervention and control arms in facility, maternal, or newborn characteristics (Table 1). The BetterBirth Program was successfully implemented in all 60 intervention facilities, achieving high fidelity to the expected number of coaching visits (average 42.1 visits of 43 expected), interactions with facility leadership (average 14.8 interactions of 11 expected), and facility data- sharing meetings (average 8.6 meetings of 11 expected).

successfully implemented in all 60 intervention facilities, achieving high fidelity to the expected number of coaching visits (average 42.1 visits of 43 expected), interactions with facility leadership (average 14.8 interactions of 11 expected), and facility data- sharing meetings (average 8.6 meetings of 11 expected). Adherence to Birth Practices After 2 months of twice-weekly coaching, birth attendants in intervention facilities (1259 observations) performed 73% of measured practices compared to 42% in control facility attendants (1304 observations) (p≤0.01) (Table 2). Birth attendants performed specific practices, such as blood pressure and temperature assessment, proper hand hygiene, and early newborn care, at significantly higher rates in the intervention arm. Supply availability was similar between study arms. In intervention sites, the Checklist was used at admission in 56.8% of observed births and in 74.3% of births within the first hour post-partum. Although adherence in intervention facilities remained significantly higher than control facilities, adherence in the intervention arm decreased to 62% of practices per childbirth at 12-months, four months after coaching ceased. For example, administration of oxytocin soon after delivery decreased by nearly one third (from 79.5% to 53.9%) between 2 and 12 months. Similarly, Checklist use declined in intervention sites at 12 months (17.4% at admission; 35.1% within 1 hour post-partum). In control sites, overall birth attendant adherence remained low at both 2 and 12 months (42% and 44%, respectively).

y decreased by nearly one third (from 79.5% to 53.9%) between 2 and 12 months. Similarly, Checklist use declined in intervention sites at 12 months (17.4% at admission; 35.1% within 1 hour post-partum). In control sites, overall birth attendant adherence remained low at both 2 and 12 months (42% and 44%, respectively). Mortality and Morbidity We found no significant difference between intervention and control facilities in our primary outcome (Intervention 15.1%, Control 15.3%, RR: 0.99, 95% CI: 0.83-1.18, p=0.90) or in any secondary outcomes (Table 3). We found no difference in the rates of follow-up care required for women or newborns, hysterectomy, blood transfusion, or inter-facility transfer (referral) for women or newborns. In stratified analyses, we observed no differences between arms based on phase of the intervention (intensive coaching, tapered coaching, and four months post- intervention), time of delivery, or in-facility mortality rate (data not shown).

blood transfusion, or inter-facility transfer (referral) for women or newborns. In stratified analyses, we observed no differences between arms based on phase of the intervention (intensive coaching, tapered coaching, and four months post- intervention), time of delivery, or in-facility mortality rate (data not shown). Discussion Implementation of the WHO Safe Childbirth Checklist and similarly constructed tools have suggested impact on quality of care, but have lacked rigorous data evaluating both adherence to essential birth practices and morbidity and mortality.11,12,16-18 In this large matched-pair, cluster-randomized controlled trial in Uttar Pradesh, India, we found that the BetterBirth Program—a coaching-based implementation of the WHO Safe Childbirth Checklist— demonstrated no impact on our primary composite maternal/perinatal health outcome (nor on any secondary health outcomes), despite significant improvement in birth attendant adherence to essential practices in intervention facilities.

terBirth Program—a coaching-based implementation of the WHO Safe Childbirth Checklist— demonstrated no impact on our primary composite maternal/perinatal health outcome (nor on any secondary health outcomes), despite significant improvement in birth attendant adherence to essential practices in intervention facilities. The majority of maternal and neonatal mortality happens around the time of birth and within the first 7 days postpartum;3,36 thus, interventions to improve early outcomes are desperately needed. The BetterBirth Program’s theory of change—that improving the quality of childbirth- related care provided in facilities would translate into improved patient outcomes—reflects basic assumptions underlying current childbirth work in global health. We found that the largely rural population living in this resource-limited setting had a perinatal mortality rate (47 per 1000) and a maternal morbidity rate (12%) much higher than anticipated;25 and event rates varied widely across facilities, with up to 10-fold differences observed (Supplement 1). Quality of care in control sites, as measured through birth attendant adherence to practices, was far lower than previously recognized.16,25,37-41 Overall, birth attendants in non-intervention facilities performed just 40% of measured essential practices in a typical birth, such as appropriate hand hygiene (used in >1% of deliveries) or administration of oxytocin within the first minute postpartum to reduce hemorrhage (used in >25%).

iously recognized.16,25,37-41 Overall, birth attendants in non-intervention facilities performed just 40% of measured essential practices in a typical birth, such as appropriate hand hygiene (used in >1% of deliveries) or administration of oxytocin within the first minute postpartum to reduce hemorrhage (used in >25%). We appeared to achieve the first step in the theory of change, demonstrating that a coaching- based implementation of the Checklist could produce broad-based improvement in the quality of care of facility-based childbirth. In intervention facilities, birth attendants substantially increased their performance of measured practices. At two months, intervention sites saw significant improvement in almost all of the practices that were rarely performed in control sites. Staff at intervention facilities had a corresponding increase in Checklist use during the coaching intervention. However, overall levels of adherence and Checklist use diminished after coaching ceased, and some practices never improved compared to controls. It is possible that Checklist use was not sustained due to lack of Checklist stock, staff belief that they knew the items on the Checklist, lack of enthusiasm, or other reasons. While we achieved relative success in improving the quality of care delivered in int

, and some practices never improved compared to controls. It is possible that Checklist use was not sustained due to lack of Checklist stock, staff belief that they knew the items on the Checklist, lack of enthusiasm, or other reasons. While we achieved relative success in improving the quality of care delivered in int A potential conclusion from our findings is that increasing adherence to these practices is not a worthwhile goal, as these practices did not lead to improved outcomes. We strongly believe this conclusion to be false. Each of the practices incorporated in the Checklist (and therefore in the BetterBirth Program) has its own evidence base, including effectiveness on improving maternal and/or neonatal outcomes.13,15 At least one practice (proper hand hygiene) has evidence of saving mothers and newborns stretching as far back as the 1840s.42

of the practices incorporated in the Checklist (and therefore in the BetterBirth Program) has its own evidence base, including effectiveness on improving maternal and/or neonatal outcomes.13,15 At least one practice (proper hand hygiene) has evidence of saving mothers and newborns stretching as far back as the 1840s.42 Several other factors may have affected the outcome. The measured levels of improvement in adherence to essential birth practices may not have reached sufficient levels to affect outcomes. For instance, hand hygiene reached only 35% adherence; although skin-to-skin initiation was 79%, maintenance at one hour was just 19%; and magnesium sulfate administration did not increase despite improved maternal blood pressure measurement. The measured levels of adherence may have been misleading, if staff practiced markedly differently when unobserved.43 However, we compared treatment effects between sites with and without observation, finding no differences. Persistent gaps in technical skills, complication management, quality as well as quantity of supplies and medicines, access to supportive management and systems level accountability—which were mostly unmeasured—could also have had a significant impact on the ability to improve health outcomes. Factors not targeted by the BetterBirth Program may have also limited the impact, including women’s underlying health and nutrition status, the quality of pre- and post-natal care, and the quality of referral care for those with more complex needs.

ave had a significant impact on the ability to improve health outcomes. Factors not targeted by the BetterBirth Program may have also limited the impact, including women’s underlying health and nutrition status, the quality of pre- and post-natal care, and the quality of referral care for those with more complex needs. In sum, we found that a coaching-based implementation of the WHO Safe Childbirth Checklist appeared to drive substantial improvements in the quality of facility-based childbirth care, but did not achieve impact on adverse maternal and perinatal health outcomes. High-quality research on large-scale childbirth improvement programs is feasible; such studies must continue to measure both processes and outcomes of care. Further investigation is required to understand and modify the complex interaction between quality of care, morbidity, and mortality.

nd perinatal health outcomes. High-quality research on large-scale childbirth improvement programs is feasible; such studies must continue to measure both processes and outcomes of care. Further investigation is required to understand and modify the complex interaction between quality of care, morbidity, and mortality. Supplementary Appendix Supplementary Material Acknowledgements We recognize the Governments of India and Uttar Pradesh for collaboration and support to conduct this trial in public health facilities. We thank the facility staff, women and newborns for their participation in the study. We are grateful to the members of the trials’ Scientific Advisory Committee who contributed crucial guidance to the development of this study protocol: Himanshu Bhushan, Zulfiqar Bhutta, Waldemar Carlo, Vinita Das, Amod Kumar, Matthews Mathai, Packirisamy Padmanbhan, Vinod Paul, and Rajiv Tandon. We also thank Jonathan Spector, Steve Ringer, Robyn Churchill, Gary King, Grace Galvin, Atul Kapoor, and the past and current members of the BetterBirth study teams in Boston and Uttar Pradesh. This trial is funded by the Bill & Melinda Gates Foundation, which reviewed the study design and sample size calculations. The funders did not have input on data collection, management, analysis, or interpretation of the data. Further, they did not have any authority over the writing of the reports or decision to submit findings for publication.

Bill & Melinda Gates Foundation, which reviewed the study design and sample size calculations. The funders did not have input on data collection, management, analysis, or interpretation of the data. Further, they did not have any authority over the writing of the reports or decision to submit findings for publication. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1701075.. Figure 1 Intervention and implementation strategy for the BetterBirth program in Uttar Pradesh, India BetterBirth Trial CONSORT Diagram

Introduction Trachoma control programs have distributed more than 600 million doses of oral azithromycin in an effort to eliminate the ocular strains of chlamydia that cause the disease.12 Azithromycin has been effective against trachoma, although distributions have caused gastrointestinal side effects and selected for macrolide-resistant strains of Streptococcus pneumoniae and Escherichia coli.3-8 Investigators have also noted possible benefits against a number of infectious diseases including malaria, diarrhea, and pneumonia.9-14 A case-control study and community-randomized trial in a trachoma-endemic area of Ethiopia suggested that mass azithromycin may even reduce childhood mortality.15,16 Experts believed a mortality benefit possible, although likely smaller in magnitude than found in these studies.17 Here, we tested the hypothesis that biannual mass distributions of oral azithromycin can reduce mortality in children aged 1-59 months. The study was performed in 3 geographically distinct areas: Malawi in Southern Africa, Niger in West Africa, and Tanzania in East Africa. Azithromycin affects transmissible diseases, so treating one individual might influence others in the same community. Thus randomization and intervention were at the community level, and inferences of efficacy were made at the community level. As mortality is a relatively rare event even in these settings, a large study population was required. Hence we adopted a large simple trial paradigm with a straightforward intervention and primary outcome.18

tion and intervention were at the community level, and inferences of efficacy were made at the community level. As mortality is a relatively rare event even in these settings, a large study population was required. Hence we adopted a large simple trial paradigm with a straightforward intervention and primary outcome.18 Methods Eligibility MORDOR (Macrolides Oraux pour Réduire les Décès avec un Oeil sur la Résistance) was a community-randomized trial conducted in the Malawian district of Mangochi, the Nigerien districts of Boboye and Loga, and the Tanzanian districts of Kilosa and Gairo. The randomization unit was a health surveillance assistant area in Malawi, a grappe in Niger, and a hamlet in Tanzania. Communities with a population between 200 and 2000 inhabitants on the most recent census were eligible for enrollment (Supplementary Appendix). Enrollment was based on census information available prior to the study. Communities remained in the study even if the population drifted out of this range. All children aged 1-59 months (truncated to month) weighing at least 3,800 grams were eligible for treatment.

ere eligible for enrollment (Supplementary Appendix). Enrollment was based on census information available prior to the study. Communities remained in the study even if the population drifted out of this range. All children aged 1-59 months (truncated to month) weighing at least 3,800 grams were eligible for treatment. Randomization and masking Lists of communities from the most recent pre-trial census were submitted to the UCSF data coordinating center. For each country, communities were randomly assigned in equal proportions to 1 of 10 letters, with 5 letters coded for azithromycin and 5 for placebo (Statistical Analysis Plan, SAP). Randomization was generated in R (R Foundation for Statistical Computing, Vienna, Austria) using the sample command (TCP), with knowledge of the link between treatment letter and arm assignment limited (TCP, KJR, and personnel necessary for labeling and packaging). Centralized randomization and simultaneous assignment of communities facilitated complete allocation concealment. Participants, observers, investigators, and data-cleaning team members were masked to treatment arm. The placebo contained the vehicle of the oral azithromycin suspension and was identical in appearance with identical bottles and labels.

imultaneous assignment of communities facilitated complete allocation concealment. Participants, observers, investigators, and data-cleaning team members were masked to treatment arm. The placebo contained the vehicle of the oral azithromycin suspension and was identical in appearance with identical bottles and labels. Census A house-to-house census was performed during 5 prescribed 6-month periods, allowing a 2-month grace period for the initial census. At the initial census, all households in the community were entered into a custom-built mobile application (Conexus Inc., Los Gatos, CA), with the head of household name and GPS coordinates used to facilitate locating the household at the subsequent census. All children in the household aged 1-59 months were enumerated. Pregnant women and children under 1 month were also documented in anticipation of the following census. At follow-up censuses, the vital status (alive, dead, or unknown) and residence (living in community, moved outside community, or unknown) were recorded for children present in census records. New 0-59 month-old children and pregnant women were also entered. Communities were censused in the same general order throughout the study. Data were uploaded to the Salesforce Cloud Database Service (Salesforce.com, San Francisco, CA). Data cleaning was performed using Salesforce.com, Stata (Statacorp, College Station, TX), and R.

h-old children and pregnant women were also entered. Communities were censused in the same general order throughout the study. Data were uploaded to the Salesforce Cloud Database Service (Salesforce.com, San Francisco, CA). Data cleaning was performed using Salesforce.com, Stata (Statacorp, College Station, TX), and R. Intervention Each child aged 1-59 months at the census was offered a single, directly observed dose of oral azithromycin or placebo (both provided by Pfizer, Inc., New York, NY). A volume of suspension corresponding to at least 20 mg/kg was given by height-stick approximation according to the country’s trachoma program guidelines, or by weight for those unable to stand. Children known to be allergic to macrolides were not treated. Treatments were administered at the census and during additional visits in an attempt to achieve at least 80% coverage. Administration of study medication was documented for each child in the mobile application, and community coverage was calculated relative to the census. Guardians were instructed to contact a village representative for any adverse events noted after taking the study medication. This individual reported to the site study coordinator, who in turn reported to the Data Coordinating Center at UCSF.

e application, and community coverage was calculated relative to the census. Guardians were instructed to contact a village representative for any adverse events noted after taking the study medication. This individual reported to the site study coordinator, who in turn reported to the Data Coordinating Center at UCSF. Primary outcome The pre-specified primary outcome was the community-level, aggregate, 3- country mortality rate determined by biannual census. Each inter-census period was treated separately, with a mortality event counted only when a child was recorded as being alive and living in the household at the initial census, and recorded as having died while residing in the community at the subsequent census. By design, no attempt was made to track down a child’s status after movement outside the community. Person-time at risk was calculated as days between consecutive censuses, with children who moved, died, or had an unknown follow-up status contributing one half the inter-census period. All children documented as alive and living in the household at the initial census of each inter-census period were included in the analysis. No changes to trial outcomes were made after the trial had commenced. Subgroup analyses Mortality rates were assessed by country site and age group. An abbreviated version of the 2007 World Health Organization verbal autopsy questionnaire for children aged 4 weeks to 14 years was used to collect data for verbal autopsies.19 Causes of death were assigned using an algorithm based on a published verbal autopsy hierarchy.20

were assessed by country site and age group. An abbreviated version of the 2007 World Health Organization verbal autopsy questionnaire for children aged 4 weeks to 14 years was used to collect data for verbal autopsies.19 Causes of death were assigned using an algorithm based on a published verbal autopsy hierarchy.20 Sample size and statistical analysis plan We estimated that inclusion of 620 communities per country would provide at least 80% power to detect an overall reduction in all-cause mortality of 10%. Specifically, we assumed mortality rates of between 14 and 20 per 1000 child-years, average community sizes of 600 to 799 people (16.7% to 19.0% of which were children aged 1- 59 months), coefficients of variation of between 0.40 and 0.51, and loss to follow-up of 10% (SAP).

all reduction in all-cause mortality of 10%. Specifically, we assumed mortality rates of between 14 and 20 per 1000 child-years, average community sizes of 600 to 799 people (16.7% to 19.0% of which were children aged 1- 59 months), coefficients of variation of between 0.40 and 0.51, and loss to follow-up of 10% (SAP). The pre-specified primary analysis was negative binomial regression of the number of deaths per community, with treatment arm and country as predictors and total person-time at risk as an offset. All 3 country sites contributed to the primary outcome. Hypothesis testing was 2-sided, allowing a total alpha of 0.05 for the interim and final analyses. A P-value was determined by Monte Carlo permutation testing (10,000 replications). An interim efficacy analysis after the 12- month census was designed to spend 0.001 of the total alpha, reserving an alpha of 0.049 for the primary 24-month analysis. Community-level clustering was taken into account by the dispersion parameter in the negative binomial regression. Pre-specified subgroup analyses included negative binomial regression of community-level mortality rates by country, age group, and inter-census period (SAP). A sample of 250 verbal autopsies from each site were compared using the chi-squared statistic, with clustering taken into account by permutation at the community level. All statistical analyses were conducted in R.

omial regression of community-level mortality rates by country, age group, and inter-census period (SAP). A sample of 250 verbal autopsies from each site were compared using the chi-squared statistic, with clustering taken into account by permutation at the community level. All statistical analyses were conducted in R. Ethics Approval for the study was obtained from the ethical committees of the College of Medicine, University of Malawi, Blantyre, the Niger Ministry of Health, and the Tanzanian National Institute for Medical Research, as well as London School of Hygiene & Tropical Medicine, UCSF Committee for Human Research, Emory University, and Johns Hopkins University School of Medicine. Informed consent was obtained from the local Ministries of Health, village leaders, and guardians of children. No incentives were offered for participation, although all children in the Niger site were offered azithromycin at the conclusion of the study, and those in Malawi entered into the country’s trachoma program. The study was undertaken in accordance with the Declaration of Helsinki. A Data and Safety Monitoring Committee provided oversight. Members of the investigator steering committee (Supplementary Appendix) designed the trial, vouch for its adherence to the protocol, and attest to the accuracy and completeness of the data and analyses as specified in the protocol and SAP. The corresponding author wrote the initial draft, and all coauthors reviewed the manuscript and agreed to publication.

ommittee (Supplementary Appendix) designed the trial, vouch for its adherence to the protocol, and attest to the accuracy and completeness of the data and analyses as specified in the protocol and SAP. The corresponding author wrote the initial draft, and all coauthors reviewed the manuscript and agreed to publication. Results Participant flow As displayed in Figure 1, 1624 of 2260 communities in the chosen districts were eligible and randomized to either this study (1533) or parallel studies (91). In Niger, 1 community refused to participate, and 20 were excluded due to being misspelled duplicates, nonexistent at the time of census, or indistinguishable from larger neighboring communities. No randomization units were lost to follow-up after the initial census.

o either this study (1533) or parallel studies (91). In Niger, 1 community refused to participate, and 20 were excluded due to being misspelled duplicates, nonexistent at the time of census, or indistinguishable from larger neighboring communities. No randomization units were lost to follow-up after the initial census. Figure 1. Enrollment, Randomization, and Treatment Census periods started in December 2014, August 2015, February 2016, August 2016, and February 2017. Data collection was completed by July 2017 and the database closed for primary analysis on October 15, 2017. Baseline characteristics of the communities are displayed in Table 1. 323,302 person-years were monitored over the 5 census visits, including 111,559 person-years in Malawi, 145,597 person-years in Niger, and 66,146 person-years in Tanzania. To validate the census data collection, a subset of households was censused by an independent field team later during the census period. The majority of recensused children had been enumerated on the first census: 95% (257/271) for Malawi, 92% (286/310) for Niger, and 95% (4544/4791) for Tanzania. Coverage of the targeted population of children was 90.4% (standard deviation 10.1%) in the placebo arm and 90.3% (±10.6%) in the azithromycin arm: in Malawi, antibiotic coverage was 91.5% (±6.4%) in the placebo arm and 91.5% (±6.1%) in the azithromycin arm, in Niger 94.5% (±6.7%) in the placebo arm and 94.5% (±6.0%) in the azithromycin arm, and in Tanzania 86.1% (±12.3%) in the placebo arm and 85.5% (±13.6%) in the azithromycin arm (Supplementary Appendix).

arm: in Malawi, antibiotic coverage was 91.5% (±6.4%) in the placebo arm and 91.5% (±6.1%) in the azithromycin arm, in Niger 94.5% (±6.7%) in the placebo arm and 94.5% (±6.0%) in the azithromycin arm, and in Tanzania 86.1% (±12.3%) in the placebo arm and 85.5% (±13.6%) in the azithromycin arm (Supplementary Appendix). Table 1. Baseline characteristics in the azithromycin and placebo treated arms All Countries Malawi Niger Tanzania Azithromycin Placebo Azithromycin Placebo Azithromycin Placebo Azithromycin Placebo Communities 762 750 152 152 303 291 307 307 Children 97,047 93,191 39,386 39,534 40,345 35,747 17,316 17,910 Children/community (mean ± sd) 171 ± 126 169 ± 128 259 ± 121 260 ± 123 133 ± 93 123 ± 88 56 ± 29 58 ± 33 Male 50.7% 50.6% 50.2% 50.0% 51.2% 51.4% 50.5% 50.6% Age

Placebo Azithromycin Placebo Azithromycin Placebo Communities 762 750 152 152 303 291 307 307 Children 97,047 93,191 39,386 39,534 40,345 35,747 17,316 17,910 Children/community (mean ± sd) 171 ± 126 169 ± 128 259 ± 121 260 ± 123 133 ± 93 123 ± 88 56 ± 29 58 ± 33 Male 50.7% 50.6% 50.2% 50.0% 51.2% 51.4% 50.5% 50.6% Age 1-5 months 7.4% 7.4% 7.0% 6.9% 6.8% 6.9% 9.4% 9.2% 6-11 months 13.2% 13.2% 12.3% 12.3% 13.5% 13.6% 14.5% 14.5% 12-23 months 19.1% 19.2% 20.5% 20.2% 17.0% 16.8% 21.0% 21.8% 23-59 months 60.4% 60.2% 60.2% 60.6% 62.8% 62.7% 55.1% 54.5% Primary results The annual mortality rate for eligible children in the placebo-treated communities in the 3 countries combined was 16.5 per 1000 person-years (9.6 per 1000 person- years in Malawi, 27.5 per 1000 person-years in Niger, and 5.5 per 1000 person-years in Tanzania). A 12-month interim assessment for efficacy did not trigger the pre-specified early stopping rule (set at P<0.001). Over all 4 inter-census periods, community-level, intention-to- treat analysis revealed that the azithromycin-treated arm had 13.5% lower mortality overall (95% CI 6.7%—19.8%, P<0.001). The proportion of children whose census status was recorded as moved or unknown was not significantly different between the two arms (P=0.71 and P=0.36 respectively).

community-level, intention-to- treat analysis revealed that the azithromycin-treated arm had 13.5% lower mortality overall (95% CI 6.7%—19.8%, P<0.001). The proportion of children whose census status was recorded as moved or unknown was not significantly different between the two arms (P=0.71 and P=0.36 respectively). Subgroup results Mortality rates in the azithromycin-treated arm were 5.7% lower in Malawi (- 9.7%—18.9%, P=0.45), 18.1% lower in Niger (CI 10.0%—25.5%, P<0.001), and 3.4% lower in Tanzania (CI -21.2%—23.0%, P=0.77; Figure 2). Children in the 1-5 month old age group had the highest overall mortality and the largest observed reduction in mortality with azithromycin (24.9% reduction, 95% CI 10.6%—37.0%, P=0.001, Figure 3). The estimated efficacy of azithromycin increased with each inter-census period, going from 7.3% (CI -5.9% to 18.8%, P=0. 26) in the first period to 22.0% (CI 10.6% to 31.9%, P<0.001) in the last period (Figure 4). Efficacy was not significantly different by country site (P=0.17), age group (P =0.20), treatment period (P =0.09), or treatment coverage (P =0.34).

reased with each inter-census period, going from 7.3% (CI -5.9% to 18.8%, P=0. 26) in the first period to 22.0% (CI 10.6% to 31.9%, P<0.001) in the last period (Figure 4). Efficacy was not significantly different by country site (P=0.17), age group (P =0.20), treatment period (P =0.09), or treatment coverage (P =0.34). Figure 2. Efficacy of azithromycin overall and by country Figure 3. Efficacy of azithromycin by age Figure 4. Efficacy of azithromycin over time Serious adverse events and overall causes of death Not including the primary outcome of mortality, 20 hospitalizations or life-threatening illnesses occurred: 1 from Malawi, 3 from Niger, and 16 from Tanzania. 11 events were in the treated arm and 9 in the untreated arm. Medical review was unable to declare that any serious adverse event was probably caused by azithromycin. A random sample of 250 verbal autopsies from each of the 3 sites estimated that 41% of deaths were due to malaria, 18% diarrhea or dysentery, and 12% pneumonia (Supplementary Appendix). Cause of death was significantly different between the country sites (P<0.001), with relatively more deaths attributed to malaria in Niger and pneumonia in Tanzania.

utopsies from each of the 3 sites estimated that 41% of deaths were due to malaria, 18% diarrhea or dysentery, and 12% pneumonia (Supplementary Appendix). Cause of death was significantly different between the country sites (P<0.001), with relatively more deaths attributed to malaria in Niger and pneumonia in Tanzania. Discussion Biannual distribution of oral azithromycin to post-neonatal preschool children significantly reduced all-cause mortality by approximately 14%. A majority of the deaths and of the observed effect was seen in Niger, which had an 18% reduction. In subgroup analysis, only the Niger site revealed a statistically significant reduction of mortality. The overall 14% effect is less than that seen in a previous case-control study and community-randomized trial in Ethiopia, but is in line with the 18% effect that a group of experts had anticipated when polled before the study.15-17 Azithromycin was most effective in children aged 1-5 months, preventing 1 in 4 deaths. The United States Food and Drug Administration has not approved azithromycin for children in that age group, and the World Health Organization does not currently recommend including them in trachoma distributions.21 However, the Centers for Disease Control does recommend oral azithromycin for all ages for treatment and prophylaxis of pertussis.22 Any mass distribution below 1 month of age would need to consider the risk of inducing infantile hypertophic pyloric stenosis (IHPS).23-25

mend including them in trachoma distributions.21 However, the Centers for Disease Control does recommend oral azithromycin for all ages for treatment and prophylaxis of pertussis.22 Any mass distribution below 1 month of age would need to consider the risk of inducing infantile hypertophic pyloric stenosis (IHPS).23-25 This study did not investigate the mechanism by which azithromycin reduced mortality. Before the trial, experts thought a protective effect would most likely be due to reductions in respiratory infections, diarrhea, and malaria, in that order.17 Such a hypothesis seems reasonable, given azithromycin’s activity against bacterial pathogens of the lung and gastrointestinal tract, and the plasmodial apicoplast. Further study will be necessary to identify how azithromycin prevents mortality. Investigation is already underway. Smaller parallel trials with detailed microbiological and anthropometric assessments were conducted at each study site. Inference from these smaller trials will be directly applicable to the mortality result, because they drew communities at random from the same pool as the parent trial. Azithromycin has been linked to cardiac death in adults, although results are mixed and may not be relevant to children in this setting.26-29 This community-based trial, and even the more detailed parallel studies, did not have the capacity to monitor QT intervals as would be possible in a hospital-based setting.28

al. Azithromycin has been linked to cardiac death in adults, although results are mixed and may not be relevant to children in this setting.26-29 This community-based trial, and even the more detailed parallel studies, did not have the capacity to monitor QT intervals as would be possible in a hospital-based setting.28 Non-specific antibiotic use is discouraged due to concern over antibiotic resistance. Repeated mass azithromycin distributions for trachoma select for macrolide resistance in nasopharyngeal S. pneumoniae and rectal E. coli.6,7,30,31 Resistance emerging during mass azithromycin distributions could curb or even reverse any potential mortality benefit. We did not observe such a waning effect—in fact, the observed effect increased from 7% to 22% over the 4 biannual inter-census periods. Nonetheless, longer follow-up is warranted to determine whether the mortality effect observed in the present trial changes with subsequent rounds of treatment.

ntial mortality benefit. We did not observe such a waning effect—in fact, the observed effect increased from 7% to 22% over the 4 biannual inter-census periods. Nonetheless, longer follow-up is warranted to determine whether the mortality effect observed in the present trial changes with subsequent rounds of treatment. The study had several limitations. As a large simple trial, little information was collected on each child and community. Deaths were determined by consecutive censuses. Children who were born and died between censuses did not contribute to either death count or person-time at risk for the primary outcome. Secondary analyses may reveal whether these children were better off being in a treated community even if they themselves were not born in time for treatment. No effort was made to follow children after they had moved. Death rates may have differed in children who moved or had an unknown census status. As distributions were offered only biannually, a child’s first treatment might not be until 7 months of age. Supplementary treatments given to infants during a scheduled vaccination visit to a health clinic could potentially add benefit. While mortality is seasonal, communities were treated in a rolling fashion over each 6-month time period for logistical reasons. Secondary analyses may reveal whether the drug was particularly effective in certain seasons. Lastly, the study was performed in 3 geographically diverse sites in Africa, but the results may not be generalizable outside of these districts. In fact, subgroup analyses only confirmed a statistically significant reduction in 1 of the 3 sites.

ay reveal whether the drug was particularly effective in certain seasons. Lastly, the study was performed in 3 geographically diverse sites in Africa, but the results may not be generalizable outside of these districts. In fact, subgroup analyses only confirmed a statistically significant reduction in 1 of the 3 sites. Two doses of oral azithromycin per year significantly reduced childhood mortality across 3 geographically diverse settings in sub-Saharan Africa. The largest effect was found in Niger, which has one of the highest child mortality rates in the world. Identifying specific mechanisms for mortality reduction will require further investigation. Any policy recommending mass distribution of oral azithromycin for childhood mortality would need to consider not only cost, but also the potential for antibiotic resistance.32 Supplementary Material Supplementary Appendix Funding The Bill and Melinda Gates Foundation provided the funding for the trial (OP1032340). Pfizer Inc. (New York City) provided both the azithromycin and the placebo oral suspensions. The Salesforce Foundation provided user licenses to Salesforce.com and cloud storage. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1715474.

Introduction The Global Polio Eradication Initiative (GPEI) has relied on oral poliovirus vaccine (OPV) to bring polio to the brink of eradication. Just 22 cases of poliomyelitis caused by wild poliovirus (WPV) were reported in 2017 (all serotype 1) as of 9 March 2018. OPV is currently used in routine immunisation and mass campaigns among children <5 years old in over 150 countries globally to ensure high levels of population immunity. It is a live-attenuated vaccine (containing Sabin poliovirus strains) that is cheap, easy to administer and, unlike the parenteral inactivated poliovirus vaccine (IPV), replicates in the intestine to induce mucosal immunity that limits further infection and transmission. However, it is genetically unstable and can evolve during replication in the human intestine to regain the neurovirulence and replication characteristics of its parental wild-type strains1-3. In rare instances, it can cause vaccine-associated paralytic poliomyelitis ( ~1-2 per million vaccinated) or seed outbreaks of circulating vaccine-derived polioviruses (cVDPV) that cause poliomyelitis (~1 outbreak per 500 million vaccinated)4.

ulence and replication characteristics of its parental wild-type strains1-3. In rare instances, it can cause vaccine-associated paralytic poliomyelitis ( ~1-2 per million vaccinated) or seed outbreaks of circulating vaccine-derived polioviruses (cVDPV) that cause poliomyelitis (~1 outbreak per 500 million vaccinated)4. The last naturally occurring case of poliomyelitis caused by serotype 2 WPV was reported in 1999 in India.5,6 However, most (>90%) cVDPV poliomyelitis cases reported over the last decade have been caused by serotype 2 (cVDPV2), due in part to declining immunity following widespread use of serotypes 1 and 3 bivalent OPV (bOPV) in supplemental immunization activities (i.e. mass campaigns)7. WHO therefore recommended globally synchronised withdrawal of serotype-2 OPV (OPV2) during a two-week period in April 2016 to prevent further cVDPV2 emergences (replacing trivalent with bivalent OPV)8,9. To mitigate the risks associated with OPV2 withdrawal, WHO recommended at least one dose of (trivalent) IPV should be used in routine immunisation in all countries.

withdrawal of serotype-2 OPV (OPV2) during a two-week period in April 2016 to prevent further cVDPV2 emergences (replacing trivalent with bivalent OPV)8,9. To mitigate the risks associated with OPV2 withdrawal, WHO recommended at least one dose of (trivalent) IPV should be used in routine immunisation in all countries. A major risk associated with OPV2 withdrawal is occurrence of further cVDPV2 outbreaks, resulting from continued circulation of cVDPV2 seeded by trivalent OPV (tOPV) used prior to withdrawal or accidental use of tOPV after withdrawal. The risk of an outbreak occurring is likely to be highest during the first 12 months following OPV2 withdrawal10 . However, if cVDPV2 circulates later in time, the scale of an outbreak will be greater due to the accumulation of children un-immunised against serotype 2. Furthermore, any outbreak of cVDPV2 must be responded to with monovalent OPV2 (mOPV2) because of the superior mucosal immunity induced by this vaccine compared with IPV11. Use of mOPV2 could seed more cVDPV2, particularly as time increases since OPV2 withdrawal, risking escalating OPV2 usage and ‘cessation failure’.12

Furthermore, any outbreak of cVDPV2 must be responded to with monovalent OPV2 (mOPV2) because of the superior mucosal immunity induced by this vaccine compared with IPV11. Use of mOPV2 could seed more cVDPV2, particularly as time increases since OPV2 withdrawal, risking escalating OPV2 usage and ‘cessation failure’.12 The global withdrawal of OPV2 is therefore seen as a major test of the feasibility of eradication of all polioviruses as envisaged by the GPEI13. Serotype 2 vaccine poliovirus (i.e. Sabin-2 virus) detection has declined after OPV2 withdrawal14, although analysis of data in 17 countries found some unexpected detections during the first 8 months15. Here, we analyse the geographic distribution of Sabin-2 and cVDPV2 detected in stool and sewage samples collected from 112 countries over the first 15 months after OPV2 withdrawal.

s declined after OPV2 withdrawal14, although analysis of data in 17 countries found some unexpected detections during the first 8 months15. Here, we analyse the geographic distribution of Sabin-2 and cVDPV2 detected in stool and sewage samples collected from 112 countries over the first 15 months after OPV2 withdrawal. Methods Data Children aged 0-14 years with acute flaccid paralysis (AFP) are reported through a network of healthcare providers and data from clinical and epidemiological investigations recorded in the Polio Information System maintained by the GPEI.16 Two stool samples from each AFP case are analysed for the presence of wild, vaccine (Sabin) or vaccine-derived polioviruses using standard protocols17,18. Most (99%) AFP is not caused by poliovirus (non-polio AFP) and detection of Sabin poliovirus is usually a coincidental finding rather than an indicator of vaccine-associated paralytic poliomyelitis 19. We analysed epidemiological and laboratory data from AFP cases reported from 112 countries in the African, Eastern Mediterranean, Southeast Asian and European regions with stool samples collected between 1-Jan-2013 and 8-Aug-2017. VDPV2 are defined as vaccine-derived polioviruses that are at least 0.6% divergent from Sabin-2 in the VP1 region, and genetically linked isolates consistent with circulation are classified as cVDPV220.

n Mediterranean, Southeast Asian and European regions with stool samples collected between 1-Jan-2013 and 8-Aug-2017. VDPV2 are defined as vaccine-derived polioviruses that are at least 0.6% divergent from Sabin-2 in the VP1 region, and genetically linked isolates consistent with circulation are classified as cVDPV220. Environmental surveillance (ES), the systematic collection and testing of sewage samples for polioviruses, is performed in >30 countries to supplement AFP surveillance.21,22 We analysed ES data from four high-risk countries (Afghanistan, Pakistan, Nigeria and Kenya) collected between 1-Jan- 2013 and 8-Aug-2017. Samples were tested for polioviruses using standard WHO protocols23. The number and spatial distribution of collection sites is shown in Fig S1.

to supplement AFP surveillance.21,22 We analysed ES data from four high-risk countries (Afghanistan, Pakistan, Nigeria and Kenya) collected between 1-Jan- 2013 and 8-Aug-2017. Samples were tested for polioviruses using standard WHO protocols23. The number and spatial distribution of collection sites is shown in Fig S1. Statistical analysis Sabin-2 detection: We fit logistic regression models to data on the prevalence of Sabin-2 poliovirus isolated from non-polio AFP cases before OPV2 withdrawal. We assumed the odds of Sabin-2 detection declines as an exponential function of time since the last tOPV campaign in that province. It asymptotically approaches a low constant background level, resulting from either routine vaccination with tOPV or migration of recently vaccinated children from other provinces, estimated using a logit- link offset that is independent of time. Countries were grouped by region in the analysis except for the 3 wild poliovirus endemic countries and India, which were analysed at the national (Afghanistan) or sub-national level (India, Pakistan, Nigeria; Table S1). Fixed effects determining the rate of Sabin-2 decline and the background level were estimated for each population. Data were censored at the time of the next campaign or 6 months, whichever was sooner. We used the same approach for ES data, but with a mixed-effects model to account for repeated observations at each ES site and site-specific variation in sensitivity to detect poliovirus. The models were fitted under a Bayesian framework using the integrated nested Laplace approximation through the R-INLA package24 and the R programming language25. We used the fitted models to predict the prevalence of Sabin-2 detection after OPV2 withdrawal, accounting for the use of mOPV2 in subsequent outbreak response campaigns by assuming decline after mOPV2 occurs at the same rate as after tOPV (Supplementary Material).

the R-INLA package24 and the R programming language25. We used the fitted models to predict the prevalence of Sabin-2 detection after OPV2 withdrawal, accounting for the use of mOPV2 in subsequent outbreak response campaigns by assuming decline after mOPV2 occurs at the same rate as after tOPV (Supplementary Material). We also tested the correlation between serotype-2 population immunity in children <36 months old and the estimated time for Sabin-2 prevalence to reach background levels (+0.1%) following a tOPV campaign. Population immunity was estimated at subnational levels over 6-month time periods from vaccination histories of non-polio AFP cases (<36 months old) and estimates of OPV efficacy against serotype-2 poliomyelitis using a spatiotemporal random-effects model as previously described.26 cVDPV2 cases: We performed univariable and multivariable mixed-effects logistic regression to identify risk factors for whether a province reported cVDPV2 cases after OPV2 withdrawal. The most parsimonious yet adequate model was identified using the ‘widely applicable information criteria’ (WAIC) 27. Full details of the statistical methods and data sources are provided in the Supplementary Appendix. Data analysis was performed by the first author; the initial draft of the manuscript was written by the first and last authors; and all authors provided final approval for publication of the manuscript.

cVDPV2 cases: We performed univariable and multivariable mixed-effects logistic regression to identify risk factors for whether a province reported cVDPV2 cases after OPV2 withdrawal. The most parsimonious yet adequate model was identified using the ‘widely applicable information criteria’ (WAIC) 27. Full details of the statistical methods and data sources are provided in the Supplementary Appendix. Data analysis was performed by the first author; the initial draft of the manuscript was written by the first and last authors; and all authors provided final approval for publication of the manuscript. Results Sabin-2 poliovirus decline before OPV2 withdrawal The proportion of non-polio AFP cases positive for Sabin-2 in stool was highest within the first month following a tOPV campaign, and rapidly declined to a low background level (<3%) within 1-2 months of the campaign (Fig 1a). Background levels varied from 0.3%-2.2% and were highest in Nigeria. Decline was slower in the Horn of Africa, North Nigeria, West Africa, Afghanistan and Pakistan (outside of Punjab and Islamabad provinces) such that the odds ratios of detecting Sabin-2 thirty days after the last tOPV campaign (compared to the time of the campaign) were significantly larger in these populations (Table S2). Low serotype-2 immunity appeared to be a risk-factor for persistent detection, with the time for Sabin-2 detection to reach background levels increasing with lower immunity (Figure 1c, r2=0.34, p=0.037). Estimated serotype 2 population immunity increased in the majority of countries up until April 2016, from 81.5% on average in July-December 2015 to 88.4% January-April 2016 (Fig S2).

t detection, with the time for Sabin-2 detection to reach background levels increasing with lower immunity (Figure 1c, r2=0.34, p=0.037). Estimated serotype 2 population immunity increased in the majority of countries up until April 2016, from 81.5% on average in July-December 2015 to 88.4% January-April 2016 (Fig S2). Figure 1 The proportion of environmental samples positive for Sabin-2 virus also declined after tOPV campaigns, albeit at a significantly slower rate and to a higher background level than in stool samples (Fig 1b, Table S2). Sabin-2 poliovirus decline after OPV2 withdrawal The prevalence of Sabin-2 in stool from children with non-polio AFP rapidly declined in all countries following OPV2 withdrawal in April 2016, from 3.9% in March to 0.16% in June 2016 (chi-squared test, p<0.0001; Fig 2). The geographic distribution became more localised over time and in June 2016 the virus was only detected in children with non-polio AFP in Nigeria, Western Africa and the Horn of Africa (Fig 3, Sabin-2 detections by country are shown in Fig S3-6). These declines and persistence in areas with poorer serotype-2 immunity were consistent with the statistical model of Sabin-2 detection (Fig 2). However, four Sabin-2 detections occurred in Sudan, South Nigeria and Afghanistan between August and December 2016 that were not expected from the statistical model of Sabin-2 detection (Fig 2).

persistence in areas with poorer serotype-2 immunity were consistent with the statistical model of Sabin-2 detection (Fig 2). However, four Sabin-2 detections occurred in Sudan, South Nigeria and Afghanistan between August and December 2016 that were not expected from the statistical model of Sabin-2 detection (Fig 2). Figure 2 Figure 3 During 2016-17, the number of ES samples collected by month increased over time (Figure S7). The proportions of monthly samples that detected Sabin-2 in all four countries were relatively high (>=20%) prior to OPV2 withdrawal and declined following the last tOPV campaign in each country (Fig 2). The rate of decline was faster than expected from the statistical model of Sabin-2 detection in northern Nigeria and Pakistan (excluding Punjab and Islamabad provinces).

n-2 in all four countries were relatively high (>=20%) prior to OPV2 withdrawal and declined following the last tOPV campaign in each country (Fig 2). The rate of decline was faster than expected from the statistical model of Sabin-2 detection in northern Nigeria and Pakistan (excluding Punjab and Islamabad provinces). VDPV2 detection after OPV2 withdrawal Between April 2016 and 8 August 2017, thirty-six cVDPV2 poliomyelitis cases were reported (Fig 4) linked to five different emergences in four different countries: Nigeria, Pakistan, Democratic Republic of the Congo (DRC) (two separate emergences28) and Syrian Arab Republic. Each cVDPV2 outbreak was restricted to a single province, with the majority (27) of cases reported from the Syrian Arab Republic. In univariable analyses, low routine immunisation coverage, serotype-2 population immunity and population density were risk factors for cVDPV2 cases in a province (Table 1). There was no association between cVDPV2 cases and the number or timing of tOPV campaigns prior to OPV2 withdrawal. In the final multivariable model, provinces with poor routine immunisation coverage and low population immunity were more likely to report cVDPV2 cases, such that an absolute decrease of 10% resulted in an increased odds of reporting cVDPV2 cases in a province by a factor of 2.6 and 4.7, respectively (Table 1).

drawal. In the final multivariable model, provinces with poor routine immunisation coverage and low population immunity were more likely to report cVDPV2 cases, such that an absolute decrease of 10% resulted in an increased odds of reporting cVDPV2 cases in a province by a factor of 2.6 and 4.7, respectively (Table 1). Figure 4 Table 1 Univariable and multivariable mixed-effects logistic regression model results for risk factors associated with the reporting of one or more cVDPV2 cases in a province after OPV2 withdrawal (01 May 2016). Data are for all provinces in Nigeria, Pakistan, Syrian Arab Republic and Democratic Republic of the Congo as of 08 August 2017 Fixed Effect Data source Univariable Odds Ratio (95% Credible Interval) Multivariable Odds Ratio (95% Credible Interval) Routine immunisation coverage+ Non-polio AFP cases 12-23 months old* or P3 coverage29,‡ 1.69 (1.06 – 3.05) 2.59 (1.26 – 6.33) Serotype-2 population immunity in the first half of 2016+ Non-polio AFP cases <36 months old 2.83 (1.28 – 6.80) 4.65 (1.71 – 15.28) Number of tOPV campaigns during 6 months prior to OPV2 withdrawal Vaccination campaign calendar (Polio Information System) 1.05 (0.44 – 2.18) - Time (days) since the last tOPV campaign prior to April 2016 Vaccination campaign calendar (Polio Information System) 1.01 (0.81 – 1.27) - Population size (log10) Worldpop30 0.96 (0.10 – 9.10) 20 (0.49 – 1243) Population density (log10 people / km2) Worldpop30 and WHO geodata 0.18 (0.02 – 0.95) - + OR for a 10% absolute decrease * Pakistan and Syrian Arab Republic ‡ Nigeria and DRC

Fixed Effect Data source Univariable Odds Ratio (95% Credible Interval) Multivariable Odds Ratio (95% Credible Interval) Routine immunisation coverage+ Non-polio AFP cases 12-23 months old* or P3 coverage29,‡ 1.69 (1.06 – 3.05) 2.59 (1.26 – 6.33) Serotype-2 population immunity in the first half of 2016+ Non-polio AFP cases <36 months old 2.83 (1.28 – 6.80) 4.65 (1.71 – 15.28) Number of tOPV campaigns during 6 months prior to OPV2 withdrawal Vaccination campaign calendar (Polio Information System) 1.05 (0.44 – 2.18) - Time (days) since the last tOPV campaign prior to April 2016 Vaccination campaign calendar (Polio Information System) 1.01 (0.81 – 1.27) - Population size (log10) Worldpop30 0.96 (0.10 – 9.10) 20 (0.49 – 1243) Population density (log10 people / km2) Worldpop30 and WHO geodata 0.18 (0.02 – 0.95) - + OR for a 10% absolute decrease * Pakistan and Syrian Arab Republic ‡ Nigeria and DRC VDPV2 without evidence of circulation (ambiguous or ‘aVDPV2’) were isolated from 4 poliomyelitis cases and 18 environmental samples after OPV2 withdrawal; 14 of the latter occurred within 4 months of mOPV2 campaigns (Fig S8, Videos 1 and 2).

Fixed Effect Data source Univariable Odds Ratio (95% Credible Interval) Multivariable Odds Ratio (95% Credible Interval) Routine immunisation coverage+ Non-polio AFP cases 12-23 months old* or P3 coverage29,‡ 1.69 (1.06 – 3.05) 2.59 (1.26 – 6.33) Serotype-2 population immunity in the first half of 2016+ Non-polio AFP cases <36 months old 2.83 (1.28 – 6.80) 4.65 (1.71 – 15.28) Number of tOPV campaigns during 6 months prior to OPV2 withdrawal Vaccination campaign calendar (Polio Information System) 1.05 (0.44 – 2.18) - Time (days) since the last tOPV campaign prior to April 2016 Vaccination campaign calendar (Polio Information System) 1.01 (0.81 – 1.27) - Population size (log10) Worldpop30 0.96 (0.10 – 9.10) 20 (0.49 – 1243) Population density (log10 people / km2) Worldpop30 and WHO geodata 0.18 (0.02 – 0.95) - + OR for a 10% absolute decrease * Pakistan and Syrian Arab Republic ‡ Nigeria and DRC VDPV2 without evidence of circulation (ambiguous or ‘aVDPV2’) were isolated from 4 poliomyelitis cases and 18 environmental samples after OPV2 withdrawal; 14 of the latter occurred within 4 months of mOPV2 campaigns (Fig S8, Videos 1 and 2). Video 1 Video 2 Sabin-2 detection after mOPV2 campaigns mOPV2 campaigns were implemented in several northern states of Nigeria; adjacent areas of Niger, Chad and Cameroon; and Balochistan, Pakistan following detection of cVDPV2 in stool or sewage after OPV2 withdrawal (or shortly before in Borno state, Nigeria31); and additionally, in central Mozambique following detection of an aVDPV2 (Fig 3). All these campaigns resulted in subsequent Sabin-2 detection in stool from non-polio AFP cases, as expected from the statistical model of Sabin- 2 detection (Figs 2-3, Videos 1,2). The virus was also detected in a cluster of samples collected from North Nigeria in September 2016 (>1.5 months after a mOPV2 campaign), from southern Chad in April 2017 and from Punjab in Pakistan in June 2017, which were not expected from the statistical model of Sabin-2 detection (Fig 2). The virus was detected for longer following mOPV2 campaigns in ES samples than non-polio AFP stool (Videos 1,2), as predicted by the statistical model of Sabin-2 detection (Table S2). In general, Sabin-2 detections from ES samples in Nigeria occurred within the mOPV2 response areas (Fig 2, Video 2) but in Pakistan, the virus was also detected in ES samples collected from sites outside the response zone and across the border in Kandahar, Afghanistan (Fig 2, Video 1).

case surge, we performed phylogenetic analysis of these 14 genomes from 2018. A maximum likelihood phylogeny shows that the 2018 genomes fall within previously known Lassa virus diversity in Nigeria (Fig. 2A) and do not display substantial clustering by date of sampling, consistent with multiple zoonotic transmissions. Estimated dates for the branch points of closely related 2018 samples in this small dataset, which are in the range of years, do not support a surge in human-to-human transmission in 2018 (Fig. S2). We also identified several intra-host Single Nucleotide Variants at a minor allele frequency >5% in 5 of the 14 patient samples, indicating some virus evolution and de novo mutation within hosts. However, none of these variants were in coding regions and only 1 was shared between samples (Table S2).

of Sabin-2 detection (Table S2). In general, Sabin-2 detections from ES samples in Nigeria occurred within the mOPV2 response areas (Fig 2, Video 2) but in Pakistan, the virus was also detected in ES samples collected from sites outside the response zone and across the border in Kandahar, Afghanistan (Fig 2, Video 1). Outbreak response campaigns with mOPV2 occurred in late June and July 2017 in DRC and the Syrian Arab Republic, but as of 8 August 2017 non-polio AFP cases have not yet been reported with completed lab testing of stool from these areas. Discussion The success of global polio eradication depends not only on eradication of wild polioviruses, but of all live polioviruses including the attenuated oral vaccine strains. Although many wealthier countries have successfully switched from OPV to IPV in their routine schedules 32,33, the synchronized withdrawal of OPV2 in April 2016 in all OPV-using countries was a major test of the feasibility of poliovirus eradication. We show here that serotype-2 vaccine poliovirus disappeared rapidly following OPV2 withdrawal, but in a small number of high-risk locations it has persisted – as a result of mOPV2 use in response to VDPV outbreaks or unplanned administration of tOPV from old stocks14. We also show that variation in the rate of decline can in part be explained by differences in population immunity, which is likely to affect the duration of individual shedding and the extent of secondary transmission of vaccine poliovirus. This supports the targeted use of preventive campaigns in advance of OPV2

ks14. We also show that variation in the rate of decline can in part be explained by differences in population immunity, which is likely to affect the duration of individual shedding and the extent of secondary transmission of vaccine poliovirus. This supports the targeted use of preventive campaigns in advance of OPV2 Outbreaks of cVDPV2 were reported in Nigeria, Pakistan, Syrian Arab Republic and DRC in the period after OPV2 withdrawal. These outbreaks occurred in populations with low routine immunisation coverage and low population immunity against serotype-2 poliomyelitis, in agreement with analyses of VDPV2 emergences and spread in Nigeria.34 Whilst highlighting the challenges facing the programme, this clear association with known risk factors for poliovirus transmission and the absence of more widespread cVDPV2 outbreaks, offers support for the GPEI strategy of globally synchronized OPV withdrawal. GPEI currently recommends at least two high-quality immunisation campaigns with mOPV2 to respond to cVDPV2 outbreaks, given its superior ability to induce mucosal immunity compared with IPV35. There is concern that use of mOPV2 threatens eradication of this serotype of poliovirus, given the risk of creating further cVDPV2 in populations with limited routine IPV coverage and growing susceptibility to serotype 2 poliomyelitis36. Since OPV2 withdrawal, multiple campaigns with mOPV2 have been implemented in response to cVDPV2 and we show that decline of Sabin-2 after these campaigns has been rapid and in line with predictions from estimates before OPV2 withdrawal.

ted routine IPV coverage and growing susceptibility to serotype 2 poliomyelitis36. Since OPV2 withdrawal, multiple campaigns with mOPV2 have been implemented in response to cVDPV2 and we show that decline of Sabin-2 after these campaigns has been rapid and in line with predictions from estimates before OPV2 withdrawal. In terms of geographic spread, in Nigeria there were few detections in stool or the environment outside the response zone after multiple, large-scale mOPV2 campaigns in northern states. In contrast, Sabin-2 was frequently detected outside the response zone in Pakistan, which was initially quite small (600,000 mOPV2 doses across 2,700 km2) compared with Nigeria (2-50 million doses across 32,000- 661,000 km2). This may reflect differences in campaign quality, scale and/or population movement. Isolated aVDPV2s have been detected in sewage samples after mOPV2 campaigns in Nigeria and Pakistan but importantly there is no evidence to date that mOPV2 has led to emergent cVDPV2s with sustained transmission and associated cases of poliomyelitis. Indeed, genetic sequence analysis suggests that all but one cVDPV2 that we report here represent continued transmission of lineages that emerged before OPV2 withdrawal28.

ut importantly there is no evidence to date that mOPV2 has led to emergent cVDPV2s with sustained transmission and associated cases of poliomyelitis. Indeed, genetic sequence analysis suggests that all but one cVDPV2 that we report here represent continued transmission of lineages that emerged before OPV2 withdrawal28. Our study had several limitations. We do not report Sabin-2 isolations from the Americas or the Western Pacific region as these data were not available through the Polio Information System. The reporting rate of AFP varies across populations and our findings are more uncertain in areas with few AFP cases37. Environmental surveillance provides important additional data, showing more sustained Sabin-2 detection, consistent with the greater sensitivity of this surveillance method. Expansion of environmental surveillance is an important component of long term polio eradication strategy22. Lastly, we do not consider the effect of seasonality on Sabin-2 detection, which may affect the accuracy of our projections because of effects on virus survival and transmission.19

y of this surveillance method. Expansion of environmental surveillance is an important component of long term polio eradication strategy22. Lastly, we do not consider the effect of seasonality on Sabin-2 detection, which may affect the accuracy of our projections because of effects on virus survival and transmission.19 In summary, high population immunity at the time of OPV2 withdrawal facilitated rapid disappearance of Sabin-2 and restricted cVDPV2 to areas known to be at high risk of transmission. This is the first test of the GPEI strategy to eradicate all polioviruses including the live vaccine virus. Our findings offer support for the planned withdrawal of bOPV after eradication of wild-type polioviruses is confirmed, provided high immunity and effective surveillance is maintained in high- risk areas. Nonetheless, in 2017, the number of poliomyelitis cases associated with cVDPV2 (96) exceeded those caused by wild poliovirus (22) for the first time and outbreak response campaigns with mOPV2 are continuing in several countries. Timely control of these outbreaks in the context of a growing cohort of children without immunity to type-2 poliovirus is critical to the success of polio eradication. Supplementary Material Supplementary Appendix Box AbbreviationDefinitionGPEIGlobal Polio Eradication Initiative WPVWild poliovirus VDPV2Serotype-2 vaccine-derived polioviruses that are at least 0.6% divergent from Sabin-2 in the VP1 region20

In summary, high population immunity at the time of OPV2 withdrawal facilitated rapid disappearance of Sabin-2 and restricted cVDPV2 to areas known to be at high risk of transmission. This is the first test of the GPEI strategy to eradicate all polioviruses including the live vaccine virus. Our findings offer support for the planned withdrawal of bOPV after eradication of wild-type polioviruses is confirmed, provided high immunity and effective surveillance is maintained in high- risk areas. Nonetheless, in 2017, the number of poliomyelitis cases associated with cVDPV2 (96) exceeded those caused by wild poliovirus (22) for the first time and outbreak response campaigns with mOPV2 are continuing in several countries. Timely control of these outbreaks in the context of a growing cohort of children without immunity to type-2 poliovirus is critical to the success of polio eradication. Supplementary Material Supplementary Appendix Box AbbreviationDefinitionGPEIGlobal Polio Eradication Initiative WPVWild poliovirus VDPV2Serotype-2 vaccine-derived polioviruses that are at least 0.6% divergent from Sabin-2 in the VP1 region20 cVDPV2Circulating VDPV2; VDPV2 strain that is genetically linked to another VDPV2 strain indicating person-to-person transmission. A detailed definition is provided in GPEI guidelines20. By definition, cVDPV2 refers to an outbreak of VDPV2. aVDPV2Ambiguous VDPV2; unlinked VDPV2 isolates that are not from an immunodeficient patient OPVOral polio vaccine tOPVTrivalent OPV against serotypes 1, 2, 3 bOPVBivalent OPV against serotypes 1, 3 mOPV2Monovalent OPV against serotype 2

cVDPV2Circulating VDPV2; VDPV2 strain that is genetically linked to another VDPV2 strain indicating person-to-person transmission. A detailed definition is provided in GPEI guidelines20. By definition, cVDPV2 refers to an outbreak of VDPV2. aVDPV2Ambiguous VDPV2; unlinked VDPV2 isolates that are not from an immunodeficient patient OPVOral polio vaccine tOPVTrivalent OPV against serotypes 1, 2, 3 bOPVBivalent OPV against serotypes 1, 3 mOPV2Monovalent OPV against serotype 2 OPV2 withdrawalGlobal replacement of tOPV with bOPV within a two-week synchronised period in April 2016 and no further use of OPV2 except for mOPV2 use in outbreak response campaigns IPVInactivated polio vaccine AFPAcute flaccid paralysis ESEnvironmental surveillance (systematic testing of sewage for poliovirus) Acknowledgements We thank the 146 laboratories of the WHO Global Polio Laboratory Network (GPLN) operating at all three-levels, with and across varying capacities and technologies (Isolation, Identification and/or Sequencing), and in all six WHO regions: Global Specialized, Regional Reference and National and subnational Laboratories for providing the data used in this study.

Global Polio Laboratory Network (GPLN) operating at all three-levels, with and across varying capacities and technologies (Isolation, Identification and/or Sequencing), and in all six WHO regions: Global Specialized, Regional Reference and National and subnational Laboratories for providing the data used in this study. We are especially grateful to the GPLN laboratories that provided the sequencing data: [Victorian Infectious Diseases Reference Laboratory (VIDRL), Melbourne, Australia; Centers for Disease Control (CDC) and Prevention, Atlanta, USA; National Institute for Health and Welfare (THL), Helsinki, Finland; Regional Reference Laboratory for Polio, Accra, Ghana; National Institutes of Hygiene Rafael Rangel, Caracas, Bolivarian Republic of Venezuela; Guangdong Poliovirus Lab, Guangzhou, China; Enterovirus Research Centre of the Indian Council Of Medical Research (ERC), Mumbai, India; National Laboratory for Polio, Tel Hashomer, Israel; National Institute of Infectious Diseases, Musashimuraya-shi, Japan; Regional Reference Laboratory for Polio (IPVE), Moscow, Russian Federation; National Institute for Biological Standards and Control, Potters Bar, United Kingdom of Great Britain and Northern Ireland; Chinese Center for Disease Control and Prevention (China CDC), Beijing, China; Regional Reference Laboratory for Polio, Giza, Egypt; National Laboratory for Polio, Bandung, Indonesia; Istituto Superiore di Sanità, Rome, Italy; Institute for Medical Research, Kuala Lumpur, Malaysia; Netherlands National Institute for Public Health and the Environment, Bilthoven, Netherlands; National Institutes for Health (NIH), Islamabad, Pakistan; CNR des Enterovirus et Parechovirus, Lyon, France; Russian Subnational Laboratory for Polio, Khabarovsk, Russian Federation; Center for Health Protection, Hong Kong, SAR China, National Microbiology Laboratory, Winnipeg, Canada; Oswaldo Cruz Foundation, Rio de Janeiro, Brazil; Institut Pasteur, Paris, France; Instituto Nacional de Salud, Bogota D.C, Colombia; Administración Nacional de Laboratorios e Institutos de Salud “Dr. Carlos G.

ation; Center for Health Protection, Hong Kong, SAR China, National Microbiology Laboratory, Winnipeg, Canada; Oswaldo Cruz Foundation, Rio de Janeiro, Brazil; Institut Pasteur, Paris, France; Instituto Nacional de Salud, Bogota D.C, Colombia; Administración Nacional de Laboratorios e Institutos de Salud “Dr. Carlos G. Malbrán” (ANLIS), Malbran, Argentina; Institute of Environmental Science and Research, Upper Hutt, New Zealand, Singapore General Hospital, Singapore; National Institute for Communicable Diseases (NICD), Johannesburg, South Africa; Regional Reference Laboratory for Polio, Nonthaburi, Thailand; Regional Reference Laboratory for Polio, Tunis, Tunisia; National Laboratory for Polio, Solna, Sweden; Robert Koch Institute, Berlin, Germany; National Laboratory for Polio, Budapest, Hungary; National Laboratory for Polio, Oslo, Norway and the National Laboratory for Polio, Almaty, Kazakhstan]. In particular, we would like to thank the following individuals in the principal sequencing laboratories: Jane Iber (CDC, USA), Salmann Sharif (NIH, Pakistan), Uma Nalavade (ERC, India), Wayne Howard (NICD, South Africa), Shuangli Zhu (China CDC), and Liubov Koslovskaya (Institute of Poliomyelitis and Viral Encephalitis, Russia). Finally, we acknowledge the guidance and supervision of the coordinators of the WHO GPLN: Gloria Janneth Rey (PAHO); Sirima Pattamadilok (SEARO), Zhang Yan (WPRO), Humayun Asghar (EMRO), Evgeniy Gavrilin (EURO) and Hieronyma Nelisiwe Gumede-Moeletsi (AFRO). We would also like to thank Harvard Rue for implementing the logit offset link in the R-INLA package.

Finally, we acknowledge the guidance and supervision of the coordinators of the WHO GPLN: Gloria Janneth Rey (PAHO); Sirima Pattamadilok (SEARO), Zhang Yan (WPRO), Humayun Asghar (EMRO), Evgeniy Gavrilin (EURO) and Hieronyma Nelisiwe Gumede-Moeletsi (AFRO). We would also like to thank Harvard Rue for implementing the logit offset link in the R-INLA package. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1716677..

BACKGROUND Environmental and dietary regulation of fetal and infant growth remains inadequately understood. Observational studies have demonstrated associations of early-life exposures with later anthropometric outcomes1,2, but there is limited evidence of effects of prenatal nutritional interventions on childhood linear growth3. Small-for-gestational age (SGA) and postnatal linear growth faltering continue to be major public health problems in low- and middle-income countries4,5. Vitamin D may influence fetal and postnatal growth through effects on calcium absorption6, parathyroid hormone (PTH) expression7, phosphate metabolism8, growth plate function9,10, and possible regulation of the insulin-like growth factor axis11. Meta-analyses of observational studies12 and clinical trials13 have suggested a possible beneficial effect of vitamin D on fetal growth, but most previous trials had important methodological limitations13. In a previous small trial in Bangladesh, we found that early postnatal linear growth was accelerated in infants born to vitamin D-supplemented mothers14. In Bangladesh, 30% of newborns are SGA4 and 36% of children under 5-years of age are stunted (height-for-age z-score <-2)15. Vitamin D deficiency is common in Bangladeshi women of reproductive age16. In this Maternal Vitamin D for Infant Growth (MDIG) trial, we aimed to evaluate the dose-dependent effects of prenatal vitamin D supplementation, with and without postpartum supplementation, on infant growth and other maternal, newborn and infant outcomes in Dhaka, Bangladesh.

n Bangladeshi women of reproductive age16. In this Maternal Vitamin D for Infant Growth (MDIG) trial, we aimed to evaluate the dose-dependent effects of prenatal vitamin D supplementation, with and without postpartum supplementation, on infant growth and other maternal, newborn and infant outcomes in Dhaka, Bangladesh. METHODS Trial Design and Oversight MDIG was a randomized, double-blind (participants and study personnel), placebo-controlled, dose-ranging parallel five-arm trial of maternal vitamin D supplementation17. The protocol and statistical analysis plan are available at NEJM.org. The study was overseen by a trial steering committee and an independent data and safety monitoring board. The protocol was approved by research ethics committees at The Hospital for Sick Children (Toronto, Canada; REB1000039072) and icddr,b (PR-13055). All authors contributed to finalizing the manuscript and attest to the completeness and accuracy of analyses and adherence to the protocol. The trial funder had no role in trial design, data collection, analysis, or interpretation of the results. Participants Generally healthy women between 17 and 24 weeks of gestation were enrolled after providing written informed consent between March 2014 and September 2015 at the Maternal and Child Health Training Institute (MCHTI), a public hospital in Dhaka, Bangladesh. Inclusion and exclusion criteria are shown in Table S1 in the Supplementary Appendix.

men between 17 and 24 weeks of gestation were enrolled after providing written informed consent between March 2014 and September 2015 at the Maternal and Child Health Training Institute (MCHTI), a public hospital in Dhaka, Bangladesh. Inclusion and exclusion criteria are shown in Table S1 in the Supplementary Appendix. Interventions Participants were randomly allocated at enrolment to one of five groups: 0 IU/week vitamin D from enrolment until delivery and 0 IU/week from 1 to 26 weeks postpartum (0;0 or ‘placebo’ group); 4200 IU/week prenatal and 0 IU/week postpartum (4200;0); 16800 IU/week prenatal and 0 IU/week postpartum (16800;0); 28000 IU/ prenatal and 0 IU/week postpartum (28000;0); or, 28000 IU/week prenatal and postpartum (28000;28000). A computer-generated simple randomization scheme was created independently by the trial statistician (A.R.W.). The master list linking unique participant identifiers to supplementation groups was held by the supplement manufacturer and not accessed by any study personnel until final unmasking. Allocation concealment was ensured by using pre-labeled sequentially-numbered and otherwise identical supplement vials assigned to participants according to the allocation sequence. Oral vitamin D3 and placebo tablets were manufactured by Toronto Institute for Pharmaceutical Technology (Toronto, Canada). Vitamin D content of each batch of tablets was verified in product testing17. Tablets of varying doses were identical in appearance and taste. Tablets were routinely administered under direct observation by study personnel; however, up to 4 consecutive doses could be unobserved when participants were unavailable for scheduled visits. Missed doses were administered up to 7 days late. Calcium (500 mg/day), iron (66 mg/day), and folic acid (350 μg/day) were provided to all participants throughout the intervention phase17. A mid-trial audit of self-reported calcium and iron-folic acid adherence revealed that >85% participants reported >85% adherence. If a participant reported non-study vitamin D or calcium supplement use for >1 week, study supplements were suspended until non-study supplement use was discontinued. Supplementation was discontinued in participants with confirmed hypercalcemia (see definition below), fetal or infant death, or a new condition or medication that could alter vitamin D metabolism.

D or calcium supplement use for >1 week, study supplements were suspended until non-study supplement use was discontinued. Supplementation was discontinued in participants with confirmed hypercalcemia (see definition below), fetal or infant death, or a new condition or medication that could alter vitamin D metabolism. Assessments Study personnel contacted participants at weekly intervals from enrolment until 26 weeks postpartum, then every three months. Visits were conducted in the home or clinic, and included standardized questionnaires, point-of-care tests, anthropometry, and specimen collection (Table S2 in the Supplementary Appendix). Socioeconomic and household characteristics were collected at baseline. Weekly prenatal questionnaires included healthcare encounters and a clinical symptom checklist. Postnatal follow-ups included history of the infant’s health and feeding practices, and a basic physical exam of the infant. Maternal blood pressure was routinely measured at enrolment, 24 and 30 weeks gestation, and weekly from 36 weeks gestation to delivery. Study personnel tracked pregnancy outcomes, study physician encounters, hospitalizations and deaths. They attended all facility-based deliveries and home births when feasible. Participants were encouraged to seek medical attention from study physicians or to notify study personnel of any concerns. Free medical care was provided throughout the trial. Pregnancies were completed from June 2014 to February 2016; one-year postnatal visits were conducted from June 2015 to March 2017.

hen feasible. Participants were encouraged to seek medical attention from study physicians or to notify study personnel of any concerns. Free medical care was provided throughout the trial. Pregnancies were completed from June 2014 to February 2016; one-year postnatal visits were conducted from June 2015 to March 2017. Infant crown-to-heel length (to the last completed 0.1 cm), head circumference (HC), upper arm length (UAL), mid-upper arm circumference (MUAC), and rump-to-knee length (RKL) – all to the last completed 0.1 cm – and weight (to the nearest 5 g up to 10 kg, and nearest 10 g for >10 kg) were measured according to standardized procedures by trained personnel, as previously described17 and adapted from Intergrowth-21st protocols18. Length, weight, HC, UAL, and RKL were measured at birth. Length, weight, and HC were measured at a randomly timed visit during the first 2 months, and then at 3, 6, 9 and 12 months of age. MUAC, UAL, and RKL were measured at 3, 6, and 12 months. Each parameter was measured independently by two study personnel; paired measurements were compared and repeated if the difference exceeded 7 mm for length, 5 mm for HC, UAL, MUAC, or RKL, and 50 g for weight. Means of the final pair of values were used in analysis. Missing, outlying or implausible values were identified and interrogated (Method 1 in the Supplementary Appendix)19. There was high inter-rater reliability and few measurements were dropped due to implausibility or temporal inconsistencies (Table S3 in the Supplementary Appendix). Length, weight, weight-for-length (WFL), body mass index (BMI), HC, and MUAC were expressed as sex- and age- (or gestational age-) standardized z-scores using Intergrowth-21st standards for newborn size20, postnatal growth standards for preterm infants to 64 weeks post-menstrual age21 (weight, length, HC only), or World Health Organization (WHO) child growth standards22.

dex (BMI), HC, and MUAC were expressed as sex- and age- (or gestational age-) standardized z-scores using Intergrowth-21st standards for newborn size20, postnatal growth standards for preterm infants to 64 weeks post-menstrual age21 (weight, length, HC only), or World Health Organization (WHO) child growth standards22. Serum calcium (sCa) and urinary calcium:creatinine ratio (uCa:Cr) were measured by routine methods at icddr,b (Dhaka, Bangladesh). Serum 25-hydroxyvitamin D (25(OH)D) and intact parathyroid hormone (iPTH) measurements were performed by the Analytical Facility for Bioactive Molecules, The Hospital for Sick Children (Toronto, Canada)(Method 2 in the Supplementary Appendix). Biochemical screening for rickets was scheduled at 6 months of age (Method 3 in the Supplementary Appendix). Radiological confirmation of rickets was based on interpretations of wrist and/or knee radiographs by a pediatric radiologist (J.S.) blinded to clinical or laboratory data. Classification of infant neurological disabilities and congenital anomalies, and clustering of study physician-assigned diagnostic codes for clinical encounters and hospitalizations, were done by investigators (D.E.R., S.Z., S.K.M., R.W.), post-hoc but masked to treatment allocation.

ded to clinical or laboratory data. Classification of infant neurological disabilities and congenital anomalies, and clustering of study physician-assigned diagnostic codes for clinical encounters and hospitalizations, were done by investigators (D.E.R., S.Z., S.K.M., R.W.), post-hoc but masked to treatment allocation. Outcomes The primary outcome was length-for-age z-score (LAZ) at one year (364–420 days) of age. Secondary anthropometric outcomes included weight-for-age z-score, WFL z-score, BMI-for-age z-score, HC-for-age z-score, MUAC-for-age z-score, UAL, and RKL (Table S4 in Supplementary Appendix). Stunting was defined as LAZ<-2. For measurements within 48 hours of birth, SGA was weight-for-age<10th percentile (using Intergrowth-21st newborn standards20) and low birth weight (LBW) was <2500 g. Preterm birth was gestational age (GA) at birth <37 weeks based on last menstrual period and 2nd trimester ultrasound (Table S1 in the Supplementary Appendix). Vitamin D status was based on serum 25(OH)D23; deficiency was defined as 25(OH)D<30 nmol/L24. The C3-epimer fraction was included in sensitivity analyses (Method 2 in the Supplementary Appendix). The primary safety measure was maternal total sCa measured at baseline, 30 weeks of gestation, delivery, 3-months and 6-months postpartum, or when hospitalized due to illness if feasible. Possible hypercalcemia was any sCa>2.60 mmol/L and confirmed hypercalcemia (primary safety outcome) was defined when sCa>2.60 mmol/L on a repeat specimen or a single sCa>2.80 mmol/L. Secondary safety indicators included infant sCa at 3 and 6 months of age and maternal urinary Ca:Cr at delivery. Maternal possible hypercalciuria was a single uCa:Cr>1 mmol/mmol. Participants with uCa:Cr>1 on two consecutive specimens (confirmed hypercalciuria) and/or symptoms of renal colic underwent abdominal ultrasound for uro- or nephrolithiasis. Infant uCa:Cr was measured at 6 months of age. Secondary clinical outcomes included gestational hypertension, delivery characteristics, stillbirth, congenital anomalies, rickets, clinical encounters, hospitalization and death (Table S4 in the Supplementary Appendix).

nt abdominal ultrasound for uro- or nephrolithiasis. Infant uCa:Cr was measured at 6 months of age. Secondary clinical outcomes included gestational hypertension, delivery characteristics, stillbirth, congenital anomalies, rickets, clinical encounters, hospitalization and death (Table S4 in the Supplementary Appendix). Statistical Analysis The primary analysis was a complete-case intention-to-treat analysis. Analysis of variance (ANOVA) was performed to compare LAZ at one year of age across all groups. To estimate the effect of prenatal vitamin D (IU/week), five pairwise comparisons were conducted using t-tests: 4200;0 versus placebo, 16800;0 versus placebo, 16800;0 versus 4200;0, 28000;0 versus placebo, and 28000;0 versus 16800;0. Statistical significance was tested at an overall alpha=0.05 (two-sided), applying the Holm test for multiple comparisons25. Sample size determination conservatively assumed that if each between-group comparison had a two-sided alpha=0.01 and 90% power, 220 analyzable participants per group would enable detection of a between-group difference in LAZ of at least 0.4014. To accommodate 15% attrition, we aimed to enroll 260 pregnant women in each arm.

size determination conservatively assumed that if each between-group comparison had a two-sided alpha=0.01 and 90% power, 220 analyzable participants per group would enable detection of a between-group difference in LAZ of at least 0.4014. To accommodate 15% attrition, we aimed to enroll 260 pregnant women in each arm. The effect of postpartum vitamin D on LAZ at one year of age was assessed by the pairwise comparison of 28000;28000 versus 28000;0 (two-sided alpha=0.05) using a t-test. Secondary outcomes were compared across groups using ANOVA for continuous normally-distributed variables and Kruskal-Wallis tests for skewed distributions; Chi-square and Fischer’s exact tests were used for categorical variables. Zero-inflated negative binomial models were used to compare incidence rates of clinical encounters, hospitalizations, and other adverse events. Where a global test was significant at p<0.05, post-hoc pairwise comparisons were performed, applying the Holm test for multiplicity25. We conducted sensitivity and stratified analyses of the primary outcome, including multiple imputation by chain equations to account for missing length at one year of age (Method 4 in the Supplementary Appendix). Infant LAZ trajectories (and other anthropometric parameters) were estimated using restricted cubic regression spline models (Method 5 in the Supplementary Appendix). Per-protocol analyses were restricted to participants who consumed at least 90% of scheduled supplement doses and had no episodes of reported consumption of non-study vitamin D or calcium (Method 6 in the Supplementary Appendix). All analyses were performed using Stata version 13 (StataCorp, College Station, TX).

dix). Per-protocol analyses were restricted to participants who consumed at least 90% of scheduled supplement doses and had no episodes of reported consumption of non-study vitamin D or calcium (Method 6 in the Supplementary Appendix). All analyses were performed using Stata version 13 (StataCorp, College Station, TX). RESULTS Trial Population 1,300 pregnant women were enrolled and randomized (Figure 1). Baseline characteristics including vitamin D status were similar across groups (Table 1; Table S5 and S6 in the Supplementary Appendix). Overall, 64% of women were vitamin D deficient. Groups did not differ by breast-feeding patterns or reported infant supplement use (Tables S7 and S8 in the Supplementary Appendix). Participants in primary analyses had higher average asset indices than those excluded but were otherwise similar (Table S9 in the Supplementary Appendix). Across all groups, ≥90% of scheduled doses were received by >90% of women in the prenatal period and >80% of women in postpartum periods (Table S10 in the Supplementary Appendix). Figure 1: Trial flow diagram a Provisional screening refers to the initial eligibility assessment of pregnant women presenting for antenatal care at the Maternal and Child Health Training Institute; individual women may have been provisionally screened more than once during the trial enrolment period. b Detailed screening refers to the complete eligibility assessment supervised by a study physician. c Prenatal; postpartum vitamin D dose. d Participants not formally lost to follow-up but for whom data at the 1-year visit were not obtained.

Figure 1: Trial flow diagram a Provisional screening refers to the initial eligibility assessment of pregnant women presenting for antenatal care at the Maternal and Child Health Training Institute; individual women may have been provisionally screened more than once during the trial enrolment period. b Detailed screening refers to the complete eligibility assessment supervised by a study physician. c Prenatal; postpartum vitamin D dose. d Participants not formally lost to follow-up but for whom data at the 1-year visit were not obtained. Table 1 Maternal characteristics at enrolment, by supplementation group Prenatal; Postpartum Vitamin D Dose (IU/Week)

b Detailed screening refers to the complete eligibility assessment supervised by a study physician. c Prenatal; postpartum vitamin D dose. d Participants not formally lost to follow-up but for whom data at the 1-year visit were not obtained. Table 1 Maternal characteristics at enrolment, by supplementation group Prenatal; Postpartum Vitamin D Dose (IU/Week) 0; 0 4200; 0 16800; 0 28000; 0 28000; 28000 Participants, N 259a 260 259a 260 260 Age, median (min, max) 23 (18, 38) 22.5 (18, 40) 22 (18, 35) 22 (18, 38) 23 (18, 38) Gestational age (weeks), median (min, max) 20.4 (17, 24) 20.1 (17, 24) 20.3 (17, 24) 20.4 (17, 24) 20.1 (17, 24) Married, n (%)b 255 (99.2) 259 (100) 254 (100) 257 (100) 256 (100) Secondary schooling complete or higher, n (%) 52 (20.1) 70 (26.9) 51 (19.7) 58 (22.3) 55 (21.2) Occupation outside the home, n (%)b 17 (6.6) 19 (7.3) 15 (5.9) 16 (6.2) 14 (5.5) Asset index, median (min, max)c -0.1 (-4.5, 4.1) -0.2 (-3.2, 3.6) 0.0 (-4.5, 3.8) -0.2 (-3.5, 4.9) 0.2 (-3.7, 4.5) Gravidityd, median (min, max) 2 (1, 9) 2 (1, 6) 2 (1, 6) 2 (1, 7) 2 (1, 6) Parity, median (min, max) 2 (0, 6) 2 (0, 5) 2 (0, 5) 2 (0, 5) 2 (0, 4) Height (cm), mean ± SD 151.2 ± 5.4 150.9 ± 5.0 150.7 ± 5.5 150.2 ± 5.4 151.8 ± 5.5 Weight (kg), mean ± SD 54.5 ± 10.3 53.2 ± 10.1 53.8 ± 9.9 53.3 ± 9.1 55.2 ± 10.6 Serum 25(OH)D concentration (nmol/L), mean ± SDe 27.7 ± 13.8 27.4 ± 14.3 28.7 ± 14.0 27.0 ± 14.7 26.6 ± 13.2 a Participants found to be ineligible after randomization were excluded from analyses (one in each of the 0;0 and 16800;0 groups).

Weight (kg), mean ± SD 54.5 ± 10.3 53.2 ± 10.1 53.8 ± 9.9 53.3 ± 9.1 55.2 ± 10.6 Serum 25(OH)D concentration (nmol/L), mean ± SDe 27.7 ± 13.8 27.4 ± 14.3 28.7 ± 14.0 27.0 ± 14.7 26.6 ± 13.2 a Participants found to be ineligible after randomization were excluded from analyses (one in each of the 0;0 and 16800;0 groups). b N0; 0 = 257, N4200; 0 =259, N16800; 0 = 254, N28000; 0 = 257, N28000; 28000 = 256 c N0; 0 = 257, N4200; 0 = 258, N16800; 0 = 253, N28000; 0 = 256, N28000; 28000 = 256. Higher scores indicate greater household asset ownership relative to other participants. See Method 7 in the Supplementary Appendix for a description of the construction and interpretation of the asset index. d Number of pregnancies, including the current pregnancy. e N0; 0 = 253, N4200; 0 = 258, N16800; 0 = 258, N28000; 0 = 258, N28000; 28000 = 256. Vitamin D deficiency defined as 25(OH)D <30 nmol/L.

c N0; 0 = 257, N4200; 0 = 258, N16800; 0 = 253, N28000; 0 = 256, N28000; 28000 = 256. Higher scores indicate greater household asset ownership relative to other participants. See Method 7 in the Supplementary Appendix for a description of the construction and interpretation of the asset index. d Number of pregnancies, including the current pregnancy. e N0; 0 = 253, N4200; 0 = 258, N16800; 0 = 258, N28000; 0 = 258, N28000; 28000 = 256. Vitamin D deficiency defined as 25(OH)D <30 nmol/L. Infant Growth Outcomes Infant follow-up at 1 year of age was completed for 90% of pregnancies and 94% of infants alive at 1 year (Figure 1). Overall, mean LAZ at 1 year was -1.00 (SD 1.04) and the prevalence of stunting was 16%. Prenatal or postpartum maternal vitamin D supplementation did not increase or decrease infant length or other anthropometric outcomes by one year of age (Table 2; Figures S1-S5 and Tables S11 and S12 in the Supplementary Appendix). The lack of effect of prenatal vitamin D on length was evident from birth (Table 3) and was supported by sensitivity and stratified analyses (Tables S13-S22 in the Supplementary Appendix). Results of the multiple imputation analysis agreed closely with the complete case analysis (Table S18 in the Supplementary Appendix); therefore, the complete case analysis is shown in Table 2. Table 2 Anthropometric outcomes of infants at 1 year of age, by supplementation group Prenatal; Postpartum Vitamin D Dose (IU/Week)

Infant Growth Outcomes Infant follow-up at 1 year of age was completed for 90% of pregnancies and 94% of infants alive at 1 year (Figure 1). Overall, mean LAZ at 1 year was -1.00 (SD 1.04) and the prevalence of stunting was 16%. Prenatal or postpartum maternal vitamin D supplementation did not increase or decrease infant length or other anthropometric outcomes by one year of age (Table 2; Figures S1-S5 and Tables S11 and S12 in the Supplementary Appendix). The lack of effect of prenatal vitamin D on length was evident from birth (Table 3) and was supported by sensitivity and stratified analyses (Tables S13-S22 in the Supplementary Appendix). Results of the multiple imputation analysis agreed closely with the complete case analysis (Table S18 in the Supplementary Appendix); therefore, the complete case analysis is shown in Table 2. Table 2 Anthropometric outcomes of infants at 1 year of age, by supplementation group Prenatal; Postpartum Vitamin D Dose (IU/Week) 0; 0 4200; 0 16800; 0 28000; 0 28000; 28000 pa Age at measurement (days), median (min, max) 364 (364, 419) 365 (364, 415) 365 (364, 418) 365 (364, 419) 365 (364, 416) 0.70 Length N 229 237 237 230 231 - Length (cm), mean ± SD 72.62 ± 2.76 72.31 ± 2.84 72.56 ± 2.54 72.39 ± 2.80 72.67 ± 2.53 0.53 Length-for-age z-score, mean ± SD -0.93 ± 1.05 -1.11 ± 1.12 -0.97 ± 0.97 -1.06 ± 1.07 -0.94 ± 1.00 0.23 Stuntedb, n (%) 36 (15.7) 46 (19.4) 36 (15.2) 39 (17.0) 31 (13.4) 0.49 Other Anthropometric Indices Weight-for-age z-score, mean ± SDc -0.81 ± 1.12 -1.00 ± 1.14 -0.86 ± 1.09 -0.96 ± 1.09 -0.89 ± 1.04 0.34 Weight-for-length z-score, mean ± SDc -0.47 ± 1.07 -0.60 ± 1.03 -0.52 ± 1.08 -0.58 ± 1.01 -0.59 ± 1.01 0.62 Body mass index-for-age z-score, mean ± SDc -0.36 ± 1.05 -0.48 ± 1.00 -0.40 ± 1.07 -0.46 ± 0.99 -0.48 ± 1.00 0.66 Head circumference-for-age z-score, mean ± SDd -1.11 ± 0.99 -1.25 ± 0.96 -1.21 ± 1.05 -1.22 ± 0.92 -1.22 ± 0.89 0.61 Mid-upper arm circumference-for-age z-score, mean ± SDe -0.14 ± 0.97 -0.27 ± 0.92 -0.21 ± 0.93 -0.29 ± 0.88 -0.23 ± 0.86 0.42 Wastedf, n (%)c 14 (6.1) 22 (9.3) 18 (7.6) 18 (7.9) 21 (9.1) 0.72 a P-values for multiple group comparisons are from ANOVA or Kruskal-Wallis tests for continuous variables, and Chi-square tests for categorical variables.

e-for-age z-score, mean ± SDe -0.14 ± 0.97 -0.27 ± 0.92 -0.21 ± 0.93 -0.29 ± 0.88 -0.23 ± 0.86 0.42 Wastedf, n (%)c 14 (6.1) 22 (9.3) 18 (7.6) 18 (7.9) 21 (9.1) 0.72 a P-values for multiple group comparisons are from ANOVA or Kruskal-Wallis tests for continuous variables, and Chi-square tests for categorical variables. b Length-for-age z-score <-2. c N0; 0 = 228, N 4200; 0 = 236, N 16800; 0 = 237, N 28000; 0 =229, N 28000; 28000 = 231 d N0; 0 = 225, N 4200; 0 = 235, N 16800; 0 = 234, N 28000; 0 =229, N 28000; 28000 = 231 e Same sample size as length-for-age z-score. f Weight-for-length z-score <-2. Table 3 Delivery characteristics and pregnancy outcomes, by supplementation group Prenatal; Postpartum Vitamin D Dose (IU/Week) 0; 0 4200; 0 16800; 0 28000; 0 28000; 28000 pb Characteristic/Outcomea N = 259 N = 260 N = 259 N = 260 N = 260 Live birth, n (%) 247 (95.4) 254 (97.7) 252 (97.3) 252 (96.9) 249 (95.8) 0.53 Gestational age at birth (weeks), median (min, max) 39.1 (32, 43) 39.1 (34, 42) 39.0 (26, 43) 39.1 (29, 43) 39.1 (30, 42) 0.62 Preterm (<37 weeks), n (%) 24 (9.7) 21 (8.3) 31 (12.3) 26 (10.3) 22 (8.8) 0.60 Caesarean section, n (%) 121 (49.0) 143 (56.3) 131 (52.0) 127 (50.4) 132 (53.0) 0.54 Facility (hospital or clinic) deliveryc, n (%) 211 (85.4) 216 (85.0) 216 (85.7) 212 (84.1) 207 (83.1) 0.93 Female infant, n (%) 129 (52.2) 117 (46.1) 132 (52.4) 124 (49.2) 121 (48.6) 0.58 Maternal serum 25(OH)D concentration at/near delivery (nmol/L), mean ± SDd 25.4 ± 21.0 69.0 ± 19.4 99.7 ± 23.7 110.0 ± 27.6 112.2 ± 28.8 <0.001e Newborn anthropometryf, mean ± SD

211 (85.4) 216 (85.0) 216 (85.7) 212 (84.1) 207 (83.1) 0.93 Female infant, n (%) 129 (52.2) 117 (46.1) 132 (52.4) 124 (49.2) 121 (48.6) 0.58 Maternal serum 25(OH)D concentration at/near delivery (nmol/L), mean ± SDd 25.4 ± 21.0 69.0 ± 19.4 99.7 ± 23.7 110.0 ± 27.6 112.2 ± 28.8 <0.001e Newborn anthropometryf, mean ± SD Birth weight (kg)g 2.72 ± 0.36 2.70 ± 0.39 2.72 ± 0.35 2.67 ± 0.34 2.76 ± 0.35 0.25 Length at birth (cm)h 47.4 ± 2.1 47.5 ± 1.9 47.4 ± 1.9 47.2 ± 2.1 47.5 ± 2.0 0.74 Head circumference at birth (cm)i 33.0 ± 1.3 33.0 ± 1.3 33.0 ± 1.1 32.9 ± 1.2 33.0 ± 1.1 0.73 Gestational age/sex-standardized growth parameterf, mean ± SD Weight-for-age z-score at birthg -1.12 ± 0.83 -1.27 ± 0.89 -1.15 ± 0.90 -1.30 ± 0.82 -1.12 ± 0.85 0.16 Length-for-age z-score at birthh -0.83 ± 1.04 -0.95 ± 1.00 -0.90 ± 1.05 -1.00 ± 1.02 -0.88 ± 0.95 0.61 Head circumference-for-age z-score at birthi -0.58 ± 0.96 -0.66 ± 1.04 -0.57 ± 0.94 -0.72 ± 0.98 -0.58 ± 0.91 0.57 Low birth weightf,j, n (%)g 42 (25.3) 53 (31.0) 42 (25.0) 53 (32.9) 40 (23.7) 0.23 Small for gestational agef, k, n (%)g 72 (43.4) 88 (51.5) 77 (45.8) 84 (52.2) 76 (45.0) 0.38 a Except for maternal serum 25(OH)D concentration at/near delivery, all characteristics and outcomes presented are among live births only. b P-values from ANOVA or Kruskal-Wallis tests for continuous variables, and Chi-square or Fischer’s tests for categorical variables. c Two infants (in group 0; 0 and 16800; 0) were born at a location other than a hospital/clinic or home; all other deliveries were home births.

Weight-for-age z-score at birthg -1.12 ± 0.83 -1.27 ± 0.89 -1.15 ± 0.90 -1.30 ± 0.82 -1.12 ± 0.85 0.16 Length-for-age z-score at birthh -0.83 ± 1.04 -0.95 ± 1.00 -0.90 ± 1.05 -1.00 ± 1.02 -0.88 ± 0.95 0.61 Head circumference-for-age z-score at birthi -0.58 ± 0.96 -0.66 ± 1.04 -0.57 ± 0.94 -0.72 ± 0.98 -0.58 ± 0.91 0.57 Low birth weightf,j, n (%)g 42 (25.3) 53 (31.0) 42 (25.0) 53 (32.9) 40 (23.7) 0.23 Small for gestational agef, k, n (%)g 72 (43.4) 88 (51.5) 77 (45.8) 84 (52.2) 76 (45.0) 0.38 a Except for maternal serum 25(OH)D concentration at/near delivery, all characteristics and outcomes presented are among live births only. b P-values from ANOVA or Kruskal-Wallis tests for continuous variables, and Chi-square or Fischer’s tests for categorical variables. c Two infants (in group 0; 0 and 16800; 0) were born at a location other than a hospital/clinic or home; all other deliveries were home births. d N0; 0 = 125, N4200; 0 = 119, N16800; 0 = 127, N28000; 0 = 113, N28000; 28000 = 124. Median (IQR) gestational age (days) at timing of measurements was 275 (268-281) days. e Post-hoc pairwise comparisons using t-tests showed significant pairwise differences between all groups except both groups who received a prenatal dose of 28000 IU/week, after adjusting for multiple comparisons using the Holm test. f Limited to measurements obtained within 48 hours of birth. g N0; 0 = 166, N4200; 0 = 171, N16800; 0 = 168, N28000; 0 =161, N28000; 28000 = 169 h N0; 0 = 164, N4200; 0 = 169, N16800; 0 = 166, N28000; 0 = 160, N28000; 28000 = 165

e Post-hoc pairwise comparisons using t-tests showed significant pairwise differences between all groups except both groups who received a prenatal dose of 28000 IU/week, after adjusting for multiple comparisons using the Holm test. f Limited to measurements obtained within 48 hours of birth. g N0; 0 = 166, N4200; 0 = 171, N16800; 0 = 168, N28000; 0 =161, N28000; 28000 = 169 h N0; 0 = 164, N4200; 0 = 169, N16800; 0 = 166, N28000; 0 = 160, N28000; 28000 = 165 i N0; 0 = 167, N4200; 0 = 168, N16800; 0 = 169, N28000; 0 = 159, N28000; 28000 = 167 j Weight <2500 g. k Weight-for-age z-score below the 10th percentile, based on the Intergrowth-21st Neonatal Standards. Biochemical effects of supplementation Vitamin D had dose-dependent effects on maternal, cord blood and infant 25(OH)D (Table 3; Figure 2; Tables S23 and S24 in the Supplementary Appendix) and maternal iPTH at delivery; the 28000;28000 group continued to have significantly lower iPTH at 6-months postpartum versus other groups (Figure 2; Table S25 in the Supplementary Appendix).

pendent effects on maternal, cord blood and infant 25(OH)D (Table 3; Figure 2; Tables S23 and S24 in the Supplementary Appendix) and maternal iPTH at delivery; the 28000;28000 group continued to have significantly lower iPTH at 6-months postpartum versus other groups (Figure 2; Table S25 in the Supplementary Appendix). Figure 2: Study design and CONSORT diagram Maternal, venous cord, and infant 25-hydroxyvitamin D (25(OH)D) [nmol/L], and maternal iPTH [pmol/L] concentrations, by vitamin D supplementation group. (A) Mean maternal 25(OH)D at baseline (n=1285), delivery (n=656; delivery specimens were collected within -19 days to 11 days of delivery [median: 0 days]), 3 months postpartum (n=581), and 6 months postpartum (n=590). Bars denote 95% confidence intervals (CI); (B) Mean 25(OH)D concentrations in venous cord (n=507), infants at 3 months (n=345), infants at 6 months (n=254), and infants at 12 months (n=182). Bars denote 95% CI; (C) Geometric means of maternal iPTH concentrations [pmol/L] at baseline (n=608), delivery (n=551; delivery specimens were collected within -19 days to 11 days of delivery [median: 0 days]), and 6 months postpartum (n=588). Bars denote 95% CI.

s at 6 months (n=254), and infants at 12 months (n=182). Bars denote 95% CI; (C) Geometric means of maternal iPTH concentrations [pmol/L] at baseline (n=608), delivery (n=551; delivery specimens were collected within -19 days to 11 days of delivery [median: 0 days]), and 6 months postpartum (n=588). Bars denote 95% CI. Safety outcomes Episodes of confirmed hypercalcemia (all asymptomatic) occurred in 0 women in the prenatal period, 8 women (0.7%) postpartum (5 in the 28000;28000 group), and 6 infants (0.7%); however, the frequencies of maternal postpartum or infant confirmed or possible hypercalcemia did not differ significantly across groups (Table S25 in the Supplementary Appendix). Prenatal vitamin D supplementation modestly elevated maternal and fetal/infant mean sCa; differences were most pronounced for the 16800;0, 28000;0, and 28000;28000 versus 0;0 comparisons up to 3-months postpartum (Table S25, S26 in the Supplementary Appendix). At 6-months postpartum, groups were similar with respect to both maternal and infant sCa (Table S25 in the Supplementary Appendix), but the 28000;28000 group had higher maternal sCa versus 0;0 or 28000;0 in post-hoc comparisons (Table S26 in the Supplementary Appendix). Maternal uCa:Cr at delivery varied significantly across groups (lowest median value in the 0;0 group) and the risk of maternal possible hypercalciuria at delivery increased in a dose-dependent manner; however, only the 28000;28000 group differed significantly from 0;0 in pairwise comparisons after correcting for multiple testing (Table S25 in the Supplementary Appendix). There were two asymptomatic cases of maternal confirmed hypercalciuria, one each in the 0;0 and 28000;0 groups (Table S25 in the Supplementary Appendix), but no women with urinary tract stones (Table S27 in the Supplementary Appendix). None of the women with confirmed hypercalcemia or confirmed hypercalciuria experienced serious adverse events (hospitalizations or deaths). Two of the six infants with confirmed hypercalcemia had neonatal hospitalizations for acute illnesses, but these events were temporally and clinically unrelated to the later findings of asymptomatic hypercalcemia. There was one infant with confirmed hypercalciuria (4200;0 group) and no differences across groups in infant uCa:Cr at 6 months of age (Table S25 in the Supplementary Appendix).

alizations for acute illnesses, but these events were temporally and clinically unrelated to the later findings of asymptomatic hypercalcemia. There was one infant with confirmed hypercalciuria (4200;0 group) and no differences across groups in infant uCa:Cr at 6 months of age (Table S25 in the Supplementary Appendix). Secondary Clinical Outcomes Delivery characteristics, duration of gestation, preterm birth, SGA, and LBW were similar across groups (Table 3 and Table S28 in the Supplementary Appendix). There was no consistent evidence of beneficial or harmful effects of any vitamin D dose on maternal or infant morbidity (Tables S27, S29-S31 in the Supplementary Appendix), gestational hypertension (Tables S27 in the Supplementary Appendix) or maternal self-reported or caregiver-reported infant symptoms (Tables S32 and S33 in the Supplementary Appendix). Stillbirth (Table S28 in the Supplementary Appendix) and infant death rates (Table S27 in the Supplementary Appendix) did not differ significantly across groups. Of four infants with radiologically-confirmed rickets, three were in 0;0 and one was in the 4200;0 group (Table S27 in the Supplementary Appendix). DISCUSSION In a setting of widespread vitamin D deficiency and fetal-infant growth restriction, vitamin D supplementation from mid-pregnancy to delivery or 6-months postpartum is safe but does not influence offspring growth patterns, and has no discernible effects on numerous pregnancy or infant clinical outcomes.

Secondary Clinical Outcomes Delivery characteristics, duration of gestation, preterm birth, SGA, and LBW were similar across groups (Table 3 and Table S28 in the Supplementary Appendix). There was no consistent evidence of beneficial or harmful effects of any vitamin D dose on maternal or infant morbidity (Tables S27, S29-S31 in the Supplementary Appendix), gestational hypertension (Tables S27 in the Supplementary Appendix) or maternal self-reported or caregiver-reported infant symptoms (Tables S32 and S33 in the Supplementary Appendix). Stillbirth (Table S28 in the Supplementary Appendix) and infant death rates (Table S27 in the Supplementary Appendix) did not differ significantly across groups. Of four infants with radiologically-confirmed rickets, three were in 0;0 and one was in the 4200;0 group (Table S27 in the Supplementary Appendix). DISCUSSION In a setting of widespread vitamin D deficiency and fetal-infant growth restriction, vitamin D supplementation from mid-pregnancy to delivery or 6-months postpartum is safe but does not influence offspring growth patterns, and has no discernible effects on numerous pregnancy or infant clinical outcomes. These findings do not support the hypothesis that prenatal vitamin D status in the 2nd half of pregnancy is a determinant of newborn size, contrary to the conclusions of prior meta-analyses of observational studies12 and mostly small trials13, but consistent with higher-quality trials in settings with lower prevalence of vitamin D deficiency or fetal growth restriction26-29. Our earlier trial in Bangladesh14 and a study in the United Kingdom30 found that prenatal vitamin D increased infant linear growth. However, they were small studies (n<135) with postnatal growth as a post-hoc outcome, and the between-group differences may have been due to chance. The present findings are broadly consistent with a meta-analysis of six trials of prenatal daily multiple micronutrient supplements (200 to 400 IU vitamin D) in low- and middle-income countries, which showed no effect on height at 2-8.5 years of age3.

come, and the between-group differences may have been due to chance. The present findings are broadly consistent with a meta-analysis of six trials of prenatal daily multiple micronutrient supplements (200 to 400 IU vitamin D) in low- and middle-income countries, which showed no effect on height at 2-8.5 years of age3. There was no clear evidence of other health benefits of improved vitamin D status in the latter half of pregnancy or early infancy. In particular, purported effects of vitamin D on preterm birth31 were not substantiated, consistent with a recent meta-analysis13. Published prenatal vitamin D trials in high-income countries have been limited by the inclusion of few (if any) women with vitamin D deficiency and the lack of a placebo (no vitamin D) group26,27,32. MDIG had several advantages: most participants were vitamin D deficient at enrolment; true placebo control group; excellent adherence; robust dose-response effect on vitamin D status across a wide dose range up to the tolerable upper intake level (4000 IU/day); rigorous anthropometric and clinical data collection; and, high retention rates.

tages: most participants were vitamin D deficient at enrolment; true placebo control group; excellent adherence; robust dose-response effect on vitamin D status across a wide dose range up to the tolerable upper intake level (4000 IU/day); rigorous anthropometric and clinical data collection; and, high retention rates. The surprising lack of any demonstrable adverse effects of maternal vitamin D deficiency on infant outcomes underscored the uncertainties about vitamin D requirements in pregnancy and lactation. The dose equivalent to the Institute of Medicine recommended dietary allowance24 (4200 IU/week) was sufficient for eliminating maternal vitamin D deficiency (25(OH)D<30 nmol/L) without elevating 25(OH)D above a conservative long-term risk threshold (125 nmol/L)24. However, 16800 IU/week prevented maternal vitamin D deficiency at the <50 nmol/L cut-off, maximally suppressed maternal iPTH, and prevented cord 25(OH)D<30 nmol/L. Maternal postpartum supplementation (28000 IU/week) maintained infant 25(OH)D above 30 nmol/L up to 6 months of age, despite variability in feeding patterns. The occurrence of four cases of rickets in the placebo and 4200;0 groups was consistent with a plausible effect of 16800 IU/week or higher prenatally in the prevention of early rickets, but the incidence of x- ray-confirmed rickets was too low to be conclusive. Ongoing analyses of infant acute respiratory infections33 and other proposed follow-up studies of the MDIG cohort may provide additional insights into longer-term effects of maternal vitamin D supplementation on child health (e.g., asthma27,32, bone mass26).

nce of x- ray-confirmed rickets was too low to be conclusive. Ongoing analyses of infant acute respiratory infections33 and other proposed follow-up studies of the MDIG cohort may provide additional insights into longer-term effects of maternal vitamin D supplementation on child health (e.g., asthma27,32, bone mass26). Although active clinical surveillance in the MDIG cohort did not reveal effects of vitamin D on maternal health measures including gestational hypertension, earlier initiation of supplementation and a larger sample size may be required to assess effects on hypertensive disorders of pregnancy or gestational diabetes. Some effects of vitamin D may have been masked by concomitant supplementation with calcium or other untreated nutrient deficiencies. Also, linear growth faltering was milder in participants than in national surveys15 suggesting that participants had relatively better baseline health and access to care. For example, participants had high rates of facility deliveries (85%, versus 37% nationally) and C-sections (51%, which is typical of local health facilities but more than double the national average)15. Consistent with WHO recommendations34, MDIG trial findings do not support routine vitamin D supplementation in pregnancy or lactation to improve birth outcomes or infant growth, even in communities with endemic vitamin D deficiency and fetal-infant growth restriction. Supplementary Appendix Supplementary Material Acknowledgements Presented in part at the IUNS International Congress of Nutrition 2017, Buenos Aires, Argentina, October 17-20, 2017.

Although active clinical surveillance in the MDIG cohort did not reveal effects of vitamin D on maternal health measures including gestational hypertension, earlier initiation of supplementation and a larger sample size may be required to assess effects on hypertensive disorders of pregnancy or gestational diabetes. Some effects of vitamin D may have been masked by concomitant supplementation with calcium or other untreated nutrient deficiencies. Also, linear growth faltering was milder in participants than in national surveys15 suggesting that participants had relatively better baseline health and access to care. For example, participants had high rates of facility deliveries (85%, versus 37% nationally) and C-sections (51%, which is typical of local health facilities but more than double the national average)15. Consistent with WHO recommendations34, MDIG trial findings do not support routine vitamin D supplementation in pregnancy or lactation to improve birth outcomes or infant growth, even in communities with endemic vitamin D deficiency and fetal-infant growth restriction. Supplementary Appendix Supplementary Material Acknowledgements Presented in part at the IUNS International Congress of Nutrition 2017, Buenos Aires, Argentina, October 17-20, 2017. Disclosure forms provided by the authors are available with the full text of this article at NEJM.org.

Although active clinical surveillance in the MDIG cohort did not reveal effects of vitamin D on maternal health measures including gestational hypertension, earlier initiation of supplementation and a larger sample size may be required to assess effects on hypertensive disorders of pregnancy or gestational diabetes. Some effects of vitamin D may have been masked by concomitant supplementation with calcium or other untreated nutrient deficiencies. Also, linear growth faltering was milder in participants than in national surveys15 suggesting that participants had relatively better baseline health and access to care. For example, participants had high rates of facility deliveries (85%, versus 37% nationally) and C-sections (51%, which is typical of local health facilities but more than double the national average)15. Consistent with WHO recommendations34, MDIG trial findings do not support routine vitamin D supplementation in pregnancy or lactation to improve birth outcomes or infant growth, even in communities with endemic vitamin D deficiency and fetal-infant growth restriction. Supplementary Appendix Supplementary Material Acknowledgements Presented in part at the IUNS International Congress of Nutrition 2017, Buenos Aires, Argentina, October 17-20, 2017. Disclosure forms provided by the authors are available with the full text of this article at NEJM.org. We thank the staff of icddr,b (Dhaka, Bangladesh) who implemented the trial and collected data including Tahmeed Kashem, Rokshana Yazmin, Sanzida Afrin; and, Kazi Moksedur Rahman (Executive Director, Shimantik) for his collaboration; present and former staff at the Centre for Global Child Health, The Hospital for Sick Children (Toronto, Canada) including A.K. Onoyovwi, Nadine Francis, Brendon Pezzack, Michelle Dimitris, Elnathan Mesfin, Jo-Anna Baxter, and Ashley Motran; Hayley Craig-Barnes and Ashley St. Pierre of the Analytical Facility for Bioactive Molecules, The Hospital for Sick Children (Toronto, Canada) for assistance with 25- hydroxyvitamin D and parathyroid hormone measurements; David Hamer for serving as the external member of the trial steering committee; Frank Martinuzzi, Toronto Institute of Pharmaceutical Technology; members of the data and safety monitoring board: AKM Nurul Anwar (Chair), Mamunar Rashid, Choudhury Ali Kawser, Meerjady Sabrina Flora, Pradip K. Bardhan, Ahmed Shafiqur Rahman; and, staff at the Bill and Melinda Gates Foundation: Sindura Ganapathi, Jeff Murray, Sharon Bergquist, Kate Fay.

of Pharmaceutical Technology; members of the data and safety monitoring board: AKM Nurul Anwar (Chair), Mamunar Rashid, Choudhury Ali Kawser, Meerjady Sabrina Flora, Pradip K. Bardhan, Ahmed Shafiqur Rahman; and, staff at the Bill and Melinda Gates Foundation: Sindura Ganapathi, Jeff Murray, Sharon Bergquist, Kate Fay. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1800927..

Introduction Childhood diarrheal deaths are largely preventable. Unfortunately, the burden of diarrhea remains high and inadequately characterized due to the complex interplay that the environment, food, water, and sanitation have with poverty and deprivation.1 A significant proportion of cases can be prevented through rotavirus immunization,2,3 safe drinking-water,4 safely-managed sanitation and hygiene,5 and establishment of processes to eliminate exposure to contaminated food.6 Meanwhile, case management with oral rehydration salts (ORS),7,8 zinc supplementation,9,10 and antibiotics11 have the potential to prevent those with diarrhea from dying. Clear information on locations with the greatest diarrheal burden is required to accelerate progress and efficiently target intervention and treatment programs.

case management with oral rehydration salts (ORS),7,8 zinc supplementation,9,10 and antibiotics11 have the potential to prevent those with diarrhea from dying. Clear information on locations with the greatest diarrheal burden is required to accelerate progress and efficiently target intervention and treatment programs. The Global Burden of Diseases, Injuries, and Risk Factors Study 2016 (GBD 2016) estimates that 330,000 children under 5 – approximately 2 in 1,000 – died from diarrheal diseases in 2015 in Africa.3 Since 2000, the diarrhea mortality rate has decreased by 54% and the severe diarrhea incidence rate has decreased by 18%. However, in part due to population growth between 2000- 2015, the absolute number of severe diarrhea episodes has increased from 25,650,000 (95% Uncertainty Interval [UI], 23,162,000 - 28,084,000) to 29,761,000 (95% UI, 26,598,000 - 33,421,000).12 Because of this significant amenable mortality13 and the long-lasting negative health impacts on nutrition, growth, and development with recurrent diarrhea,14,15 further reduction of the global diarrhea burden remains a priority.

Interval [UI], 23,162,000 - 28,084,000) to 29,761,000 (95% UI, 26,598,000 - 33,421,000).12 Because of this significant amenable mortality13 and the long-lasting negative health impacts on nutrition, growth, and development with recurrent diarrhea,14,15 further reduction of the global diarrhea burden remains a priority. Initiatives such as the Global Action Plan for the Prevention and Control of Pneumonia and Diarrhea (GAPPD) establish ambitious goals to address the high diarrhea burden among children. These goals aim to reduce child mortality rates to below 1 in 1,000 persons and reduce severe diarrhea episodes to 75% of their 2010 values by 2025. Precision public health, the use of high resolution data to guide tailored interventions, is necessary to identify the most vulnerable populations and better target lifesaving preventive and treatment measures.16,17 No previous study has attempted a comprehensive, sub-national analysis of diarrhea burden across any large region of Africa, although there have been several focused analyses of spatial and spatio-temporal variation in diarrheal burden within selected countries.1,12,18–20 A history of mapping malaria burden 21,22 combined with recent work in mapping under-5 mortality rates, child growth failure,23 and educational attainment24 has demonstrated the utility of household surveys for identifying local patterns of health across the continent and thereby identifying the greatest opportunities for impact.

f mapping malaria burden 21,22 combined with recent work in mapping under-5 mortality rates, child growth failure,23 and educational attainment24 has demonstrated the utility of household surveys for identifying local patterns of health across the continent and thereby identifying the greatest opportunities for impact. Here we present the first comprehensive, systematic analysis of local variation in diarrheal morbidity and mortality in children under 5 across Africa during the Millennium Development Goal era (2000-2015). Using Bayesian model-based geostatistics, 51,355 geolocated point level survey clusters and 2,524 small geolocated polygons, and existing GBD 2016 methods, we produce yearly, 5-km2 gridded estimates of diarrhea prevalence, incidence by severity, and mortality for children under 5, from 2000 through 2015 across Africa.

Using Bayesian model-based geostatistics, 51,355 geolocated point level survey clusters and 2,524 small geolocated polygons, and existing GBD 2016 methods, we produce yearly, 5-km2 gridded estimates of diarrhea prevalence, incidence by severity, and mortality for children under 5, from 2000 through 2015 across Africa. Methods We compiled a database of 191 surveys from Africa that contained geocoded information corresponding to coordinates of 51,355 survey clusters and 2,524 subnational polygon boundaries. Survey clusters are the geographic unit in the sampling design from which households are randomly sampled—often a village, enumeration area, or census tract. For data that we could not match to specific survey clusters (e.g. GPS data was unavailable), we instead identified the smallest area/polygon and aggregated all observations within the unit to that level for modelling. Sources were excluded if they did not record period prevalence of diarrhea for every child in the home in the proceeding 2 – 4 weeks; if they did not include strata, primary sampling unit, and design weights for each participant; and if they did not include geographic information more specific than national (admin0) scale. We included datasets from the Demographic and Health Surveys (DHS), Multiple Indicator Cluster Surveys (MICS), as well as World Bank and country-specific surveys from 1998-2016.25-28 It was more probable that surveys not part of a larger series conducted independently would be excluded due to missing these criteria. Each source recorded period prevalence of diarrhea for every child in every home sampled over the preceding two to four weeks. Details on each data source for each country are provided in the Supplementary Appendix.

eys not part of a larger series conducted independently would be excluded due to missing these criteria. Each source recorded period prevalence of diarrhea for every child in every home sampled over the preceding two to four weeks. Details on each data source for each country are provided in the Supplementary Appendix. Prevalence data was adjusted for season and converted from period prevalence to point prevalence as described in the GBD 2016 study.29 The resulting adjusted point prevalence data was modeled directly in a Bayesian model-based geostatistical framework described in detail elsewhere.1,29 Briefly, a spatially and temporally explicit hierarchical logistic regression model was fit to point prevalence of diarrhea. In this model, points that are closer together in space and time – and which have similar covariate patterns – are expected to have similar diarrheal prevalence. To reflect the social, structural, and environmental factors that may influence diarrheal prevalence, we assembled a collection of 27 covariates (Table S3). Posterior distributions of all model parameters and hyperparameters were estimated using R-INLA.30,31 Due to the spatial resolution of the main covariates, all predictions were made at the 5-km2 scale. After fitting the geospatial model, 1,000 draws (samples) were taken from the joint posterior distribution of diarrheal prevalence. Each draw contains a single possible diarrheal prevalence value for each 5-km2 location for each modeled year.

olution of the main covariates, all predictions were made at the 5-km2 scale. After fitting the geospatial model, 1,000 draws (samples) were taken from the joint posterior distribution of diarrheal prevalence. Each draw contains a single possible diarrheal prevalence value for each 5-km2 location for each modeled year. The GBD 2016 study produced estimates of diarrhea prevalence, incidence, and mortality for every country in Africa for each year from 1990-2016.29 We combined our posterior distributions from above with the modeled results and diarrhea severity distributions from GBD 2016 in two stages. First, we maintained consistency with the GBD 2016 estimates by scaling our results such that these 5-km2 estimates of diarrhea prevalence – when aggregated and averaged to the national level by calculating a population-weighted mean – match the national level GBD estimates for each country and year. Second, we used the GBD 2016 ratios between incidence, prevalence, and mortality for every country-year to convert our prevalence estimates to corresponding estimates for mortality and incidence of severe diarrheal episodes. Draws of prevalence, incidence, and mortality were then summarized as mean estimates and Bayesian uncertainty intervals. Aggregated administrative subdivision estimates were also calculated at the draw level and then summarized as population-weighted means with uncertainty intervals. Annual case fatality ratios were calculated for each country by dividing the number of diarrhea deaths estimated by the GBD project by the corresponding number of incident cases.32 Model validation was conducted both in-sample and out-of-sample using several hold out methods. Additional details on model, estimation, and validation processes can be found in the Supplemental Appendix, sections 3.0 and 4.0.

diarrhea deaths estimated by the GBD project by the corresponding number of incident cases.32 Model validation was conducted both in-sample and out-of-sample using several hold out methods. Additional details on model, estimation, and validation processes can be found in the Supplemental Appendix, sections 3.0 and 4.0. This study complies with the Gather for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations (Table S1). All code used for these analyses will be available online upon publication at https://github.com. Given the continental scope and fine spatial scale of this work, additional results are provided in the Supplemental Appendix and will be made available upon publication on an online visualisation tool (http://vizhub.healthdata.org/lbd/diarrhea), which will be updated annually. Results Diarrheal mortality Our findings suggest an unequal distribution of diarrheal burden for children under 5 across Africa from 2000-2015. Locations in Nigeria and Chad have maintained high mortality rates through the study period; each country had several first administrative subdivision that exceed 6 deaths per 1,000 in 2015 (Figure 1). In 2015, the largest difference in within-country mortality rate observed was in Nigeria, with estimates ranging from Bayelsa at 1.6 (95% UI, 1.0 - 2.3) per 1,000 to Yobe at 9.5 (95% UI, 7.1 – 12.8) per 1,000 (Figure 1).

try had several first administrative subdivision that exceed 6 deaths per 1,000 in 2015 (Figure 1). In 2015, the largest difference in within-country mortality rate observed was in Nigeria, with estimates ranging from Bayelsa at 1.6 (95% UI, 1.0 - 2.3) per 1,000 to Yobe at 9.5 (95% UI, 7.1 – 12.8) per 1,000 (Figure 1). Figure 1: Diarrhea mortality rates in children under 5 in 2000 and 2015 Panels A and B show the estimated mean rate per 1,000 of mortality attributable to diarrhea in 2000. Panels C and D show the estimated mean rate per 1,000 of mortality attributable to diarrhea in 2015. Panels B and D display the rates at the 5-km2 scale at which the model is fit. Panels A and C display the rates aggregated up to first administrative subdivision using population weighting. The color scales for mortality are set to indicate the locations in which the mean mortality rate estimates have achieved the GAPPD goal of less than 1 in 1,000. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey.

tive subdivision using population weighting. The color scales for mortality are set to indicate the locations in which the mean mortality rate estimates have achieved the GAPPD goal of less than 1 in 1,000. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey. Our estimates reveal that the number of under 5 deaths due to diarrhea in Africa is highly geographically concentrated. Bauchi, Gombe, and Yobe, the first administrative subdivisions with the three worst mortality rates in Nigeria, account for 6% of all diarrhea death count in Africa while making up just 1% of the population at risk (with 9,928 [95% UI, 7,583 – 13,019], 4,778 [95% UI, 3,515 – 6,494] and 5,436 [95% UI, 4,055 – 7,322] deaths, respectively) (Figure 2). Nearly 50% of all childhood diarrhea deaths in Africa were estimated to occur in just 7.0% (55/782) of the first administrative subdivisions on the continent (35% of population, Figure 3). While the burden of diarrheal deaths continues to vary across the continent, diarrheal mortality rates have decreased in nearly all locations in Africa from 2000 to 2015, increasing in only certain parts of the Central African Republic, Gabon, Zimbabwe, Côte d'Ivoire, and Nigeria (Figure 1).

% of population, Figure 3). While the burden of diarrheal deaths continues to vary across the continent, diarrheal mortality rates have decreased in nearly all locations in Africa from 2000 to 2015, increasing in only certain parts of the Central African Republic, Gabon, Zimbabwe, Côte d'Ivoire, and Nigeria (Figure 1). Figure 2: Ten highest number and rates of diarrhea associated mortality by first administrative subdivision from 2000 to 2015 The left panel shows the 10 first administrative subdivisions with the most childhood death counts associated with diarrhea in 2000 and 2015. The right panel shows the 10 first administrative units with the highest mortality rates (per 1,000) associated with diarrhea in 2000 and 2015. Regions not in the top 10 in both 2000 and 2015 are listed below vertical ellipses with associated year-specific rank. The lines connecting regions are solid if rank increased from 2000 to 2015 and dashed if the rank decreased. Relative change in values is shown in the 2015 columns. SNNPR: Southern Nations, Nationalities, and People’s Region. Figure 3: Number of diarrheal deaths in children under 5 in 2000 and 2015 Panel A shows the estimated mean number of diarrheal death counts in 2000. Panel B shows the estimated mean number of diarrheal death counts in 2015. Both panels display diarrheal death counts aggregated up to the first administrative subdivision using population weighting. All color scales are on a log scale. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey.

an number of diarrheal death counts in 2015. Both panels display diarrheal death counts aggregated up to the first administrative subdivision using population weighting. All color scales are on a log scale. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey. Diarrheal incidence Nigeria contains the regions with the highest rates of severe diarrhea cases per 1,000 in 2015 (Yobe, Bauchi, and Gombe at 422 (95% UI, 315 - 569), 366 (95% UI, 280 - 480), and 349 (95% UI, 257 – 474, respectively; Figure 4). The burden of diarrheal incidence was also highly concentrated within parts of Ethiopia and the Democratic Republic of the Congo (DRC). In 2015, 9.4% (2,800,000 [95% UI, 2,390,000 - 3,300,000]) of all severe cases of diarrhea in Africa took place within just five first-level administrative units in these two countries: the Southern Nations, Nationalities, and People’s Region (SNNPR), Oromia, and Amhara in Ethiopia and the Orientale and Katanga regions of the DRC.

4% (2,800,000 [95% UI, 2,390,000 - 3,300,000]) of all severe cases of diarrhea in Africa took place within just five first-level administrative units in these two countries: the Southern Nations, Nationalities, and People’s Region (SNNPR), Oromia, and Amhara in Ethiopia and the Orientale and Katanga regions of the DRC. Figure 4: Severe diarrhea incidence rates in children under 5 in 2000 and 2015 in first administrative units Panels A and B show the estimated mean rate per 1,000 of severe diarrhea episodes in 2000. Panels C and D show the estimated mean rate per 1,000 of severe diarrhea episodes in 2015. Panels B and D display the rates at the 5-km2 scale at which the model is fit. Panels A and C display the rates aggregated up to the first administrative subdivision using population weighting. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey. Case fatality rates and avertable deaths In 2015, Lesotho (0.18% [95% UI, 0.12% - 0.25%]), Mali (0.17% [95% UI, 0.12% - 0.24%]), Sierra Leone (0.16% [95% UI, 0.11% - 0.23%]), Benin (0.16% [95% UI, 0.11% - 0.21%]), and Nigeria (0.16% [95% UI, 0.11% - 0.21%]) had the highest diarrheal case fatality rates in Africa (Figure 5 Panel A). Although the case fatality ratio in Benin increased from its estimated value in 2000 (0.15% [95% UI, 0.10% - 0.22%]), the remaining four countries listed above experienced relative improvements between 2000 and 2015.

6% [95% UI, 0.11% - 0.21%]) had the highest diarrheal case fatality rates in Africa (Figure 5 Panel A). Although the case fatality ratio in Benin increased from its estimated value in 2000 (0.15% [95% UI, 0.10% - 0.22%]), the remaining four countries listed above experienced relative improvements between 2000 and 2015. Figure 5: Diarrhea CFR between 2000 and 2015 and deaths averted Panel A shows each country’s diarrheal CFR value in 2000 and in 2015. Panel B shows “Scenario 1,” the estimated number of deaths averted had all countries with the highest 50% CFRs in 2015 achieved the median CFR in 2015. Panel C shows, “Scenario 2,” the estimated number of deaths averted had the countries with the worst change in CFR between 2000-2015 achieved the median CFR change during that time period. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey.

dian CFR in 2015. Panel C shows, “Scenario 2,” the estimated number of deaths averted had the countries with the worst change in CFR between 2000-2015 achieved the median CFR change during that time period. Pixels with fewer than ten people per 1-km2 and classified as “barren or sparsely vegetated” are colored in grey. The median country-level case fatality ratio in 2015 was 0.0498%. Had all countries with case fatality ratios worse than the median 2015 value achieved the median 2015 case fatality ratio value (“Scenario 1”), an estimated 251,202 deaths (95% UI 220,859 – 283,164) could have been averted across the continent (Figure 5, Panel B). Approximately 41% of these averted deaths (103,161 [95% UI, 83,765 – 126,065]) would have occurred in Nigeria. Specifically, Bauchi (9,709 deaths averted [95% UI, 7,415 – 12,731]), Kano (8,212 deaths averted [95% UI, 5,987 – 11,228]), and Jigawa (7,232 deaths averted [95% UI, 5,307 – 9,703]) would have seen the most lives saved of any African first administrative subdivisions. Similarly, the median reduction in case fatality ratio from 2000 to 2015 was 51.4%, between that of Cameroon (51.0% [95% UI, 26.2% - 70.5%]) and Kenya (51.8% [95% UI, 42.8% - 59.7%]). If the countries below the median change during this period had reduced their case fatality ratio by this median value (“Scenario 2”), approximately 49,059 (95% UI, 41,888 – 56,314) lives could have been saved in 2015 (Figure 5, Panel C).

meroon (51.0% [95% UI, 26.2% - 70.5%]) and Kenya (51.8% [95% UI, 42.8% - 59.7%]). If the countries below the median change during this period had reduced their case fatality ratio by this median value (“Scenario 2”), approximately 49,059 (95% UI, 41,888 – 56,314) lives could have been saved in 2015 (Figure 5, Panel C). Data Validation Validation of model fit and model specification were performed using two instances of 5-fold cross validation. Folds were spatially selected using a quad-tree algorithm or by second administrative unit, such that data near each other were selected for the same fold. This provided a more stringent test for our spatially correlated model, and more closely resembled the spatially patchy nature of data sparsity in the input data. Out of sample statistics such as root mean squared error (RMSE), correlation, and coverage were generated on the data held out of the model and subsequently summarized by aggregating to administrative areas. Across the continent at the first administrative unit level, we have an RMSE of .01039 for 2000 and 0.00962 in 2015, while the correlation for these years were 0.86 and 0.95 respectively. Additional statistics on model validity can be found in the supplementary materials.

marized by aggregating to administrative areas. Across the continent at the first administrative unit level, we have an RMSE of .01039 for 2000 and 0.00962 in 2015, while the correlation for these years were 0.86 and 0.95 respectively. Additional statistics on model validity can be found in the supplementary materials. Discussion Our modeled maps demonstrate substantial local variation in both incidence and mortality associated with diarrhea in children under 5 in Africa over the last 15 years. The rates of decline in incidence and mortality have varied both between and within countries at every level of spatial aggregation considered. Some countries appear to have significantly reduced their diarrhea burden uniformly, while others are behind on their progress countrywide. Additionally, these high-resolution subnational estimates identify a third group of countries whose progress has varied subnationally. By providing estimates of current rates and counts of severe incidence and mortality, we identify locations most in need of interventions to reduce diarrhea burden.

rogress countrywide. Additionally, these high-resolution subnational estimates identify a third group of countries whose progress has varied subnationally. By providing estimates of current rates and counts of severe incidence and mortality, we identify locations most in need of interventions to reduce diarrhea burden. Over half of all diarrheal deaths in Africa occur in about 7% of the first administrative subdivisions, which contain 35% of Africa’s population. These highly populated locations with high mortality rates - many of which are in Nigeria, Ethiopia, and Niger - are places where targeted interventions to improve mortality rate, even modestly, could avert many deaths. Conversely, in-depth evaluation of the factors contributing to success in countries like Ethiopia, where case fatality rate declined by over 60% from 2000 to 2015, could reveal important strategies for reducing case fatality in other areas. As noted in the work by Troeger et al.,1 Ethiopia has shown significant improvements in child nutrition over the last 15 years. That, combined with an expanded use in oral rehydration therapy, appears to account for much of the reduction of mortality in that country.

trategies for reducing case fatality in other areas. As noted in the work by Troeger et al.,1 Ethiopia has shown significant improvements in child nutrition over the last 15 years. That, combined with an expanded use in oral rehydration therapy, appears to account for much of the reduction of mortality in that country. The relative intractability of diarrhea incidence compared to diarrhea mortality, as shown in the present analysis and elsewhere,1 may suggest that growing access to timely and appropriate treatment, better nutritional status, and fewer comorbidities are contributing factors to reducing diarrhea mortality. A variety of interventions – including programs to promote immunization, hygiene, breastfeeding, oral rehydration therapy, and zinc supplementation – have been effectively employed on a small scale to combat diarrheal disease and death.1,33 Targeting the worst regions of those countries with the highest case fatality ratio, such as those in Lesotho and Mali, is likely to have a substantially larger impact than untargeted approaches. Though the introduction of the rotavirus vaccine into Africa is relatively recent and coverage is still incomplete, the GBD study found that rotavirus vaccine coverage was negatively correlated with all-cause diarrhea. There was however a significant range of estimated percent attributable fractions for rotavirus across Africa (6.5%-64.2% in 2016), and as the vaccine becomes more established this warrants further investigation.

ete, the GBD study found that rotavirus vaccine coverage was negatively correlated with all-cause diarrhea. There was however a significant range of estimated percent attributable fractions for rotavirus across Africa (6.5%-64.2% in 2016), and as the vaccine becomes more established this warrants further investigation. Local estimates of diarrheal burden can be used to prioritize improved access to safe water and sanitation, which varies greatly between dense and sparse populations;34,35 childhood growth monitoring, which has improved in most regions of Africa but not universally;36,37 delivery and uptake of vaccines, including the rotavirus vaccine;38 and access to diarrheal care and prevention interventions for marginalized populations that live in remote regions or areas of conflict.39,40 Nepal, for example, outpaced its neighboring countries in reducing diarrhea case fatality rates in part by implementing a district-level community intervention program.41 Additionally, Brazil successfully used targeted interventions in the 1980s, when it drastically reduced infant mortality due to diarrheal diseases through policy efforts aimed at the northeast of the country, a poorer region with the country’s highest burden.42

menting a district-level community intervention program.41 Additionally, Brazil successfully used targeted interventions in the 1980s, when it drastically reduced infant mortality due to diarrheal diseases through policy efforts aimed at the northeast of the country, a poorer region with the country’s highest burden.42 As with any work of this scope, our results are subject to several limitations. First, in order to produce continent-wide estimates, we combine data from a broad range of sources which require making assumptions about their utility and consistency. For example, while diarrheal prevalence was assessed with the same, standard question across heath surveys, they rely on self-reported stooling patterns, and as such are subject to recall and reporting bias. Additionally, conversions from prevalence to incidence leverage the GBD modeled estimates and the diarrhea severity distribution. Incorporating etiology-specific estimates of diarrheal incidence and severity would likely enhance the accuracy of the conversion. Similar to the GBD study which parses all-cause diarrhea into percent attributable fractions for multiple etiologies,1,29 we are working towards etiology-specific maps of mortality and morbidity for Africa. Currently neither approach uses information on bloody stools to assist in either severity splits or etiology-splits as that information was not included in all surveys. While the conversion from incidence to mortality leverages various data sources1,12,29 and allows for variation in case-fatality ratio by country, year, sex, and age, it does not currently allow for variation in case-fatality ratios by diarrheal etiology and does not incorporate the effects of comorbidities. Our geospatial approach naturally borrows strength from neighboring areas, and as such may smooth over extremely focal epidemics, such as those frequently associated with cholera. Finally, there is significant evidence of difference in risk within the 0-5 age group. Due to the nature of the data and methods we utilize, we are currently unable to parse mortality and morbidity estimates into finer age groups.

smooth over extremely focal epidemics, such as those frequently associated with cholera. Finally, there is significant evidence of difference in risk within the 0-5 age group. Due to the nature of the data and methods we utilize, we are currently unable to parse mortality and morbidity estimates into finer age groups. This work provides a foundation for several important directions for future research. First, accounting for etiological distributions within prevalence, incidence, and death associated with diarrhea will provide increased capacity to create targeted intervention strategies (e.g. rotavirus vaccine coverage needs). Second, the approaches outlined in this work are directly applicable to other continents where similar data sources are available. Expanding estimates out of Africa to all low- and middle- income countries will be the next step towards the ultimate goal of globally mapping diarrheal morbidity and mortality. Third, as this statistical modeling approach deliberately values predictive performance over interpretability of the relationships between covariates and diarrhea, a parallel effort is underway to build spatio-temporal models more capable of causal inference to assess the impact of interventions such as vaccination and improvements in water, sanitation, and hygiene. These sorts of associations will be very important to explore in identifying the root causes of entrenched disease burden at the local- level. Future analyses will leverage these estimates to explore the extent to which high diarrheal burden in a subnational location reveals deeper patterns of eco-social inequity within countries. This work clearly demonstrates the marked local variability in childhood morbidity and mortality due to diarrhea across Africa. For every country in Africa, these estimates can be used to identify the optimal regions to more precisely target interventions. These estimated deaths are largely preventable at the population and clinical levels. Our work can help accelerate the already impressive reduction in childhood diarrhea deaths across the continent.

ry country in Africa, these estimates can be used to identify the optimal regions to more precisely target interventions. These estimated deaths are largely preventable at the population and clinical levels. Our work can help accelerate the already impressive reduction in childhood diarrhea deaths across the continent. Supplementary Appendix Supplementary Material Author Contributions S.I.H. conceived and R.C.R. designed and implemented the study and wrote the first draft of the paper. All other authors contributed to subsequent revisions. G.M.G. and S.J.S. obtained, extracted, processed, and geopositioned the data. N.G., D.C.C., C.T., A.D., S.J.S., and I.M.D. wrote the computer code and designed and carried out statistical analysis with input from R.C.R. and S.I.H. R.C.R., D.C.C., S.J.S., I.D.L., and L.E. prepared tables and figures. B.F.B. and P.C.R. managed the project. R.C.R., C.J.L.M., and S.I.H. hold responsibility for the final decision to submit. All authors provided intellectual inputs into aspects of this study. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1716766..

Mycobacterium tuberculosis killed more people than any other pathogen in 2016, when over 10 million active cases were estimated, and 1.7 million patients died.1 In 2014, the World Health Organization (WHO) set a target to ‘END TB’ by 2035, acknowledging that success depends on the development of better preventative, diagnostic and therapeutic interventions. The global emergence of antimicrobial resistance poses a major challenge. Despite a call for universal access to drug susceptibility testing to direct individualised therapies, high costs and skills shortages mean it is unavailable in many countries with greatest need. Consequently, only 22% of an estimated 600,000 patients requiring treatment for multidrug-resistant tuberculosis were diagnosed and treated in 2016,1 facilitating the onward transmission of multidrug-resistant strains.2 The Xpert MTB/RIF (Cepheid, Sunnyvale, California, USA) assay has partially eased the global diagnostic need. It uses polymerase chain reaction technology to identify both M. tuberculosis complex and mutations in the rpoB gene (predictive of multidrug resistance) directly from clinical samples.3 However, as it targets only a few potential resistance-conferring mutations, antimicrobial susceptibility cannot be reliably inferred from a negative result.4 To direct individualised therapies, a diagnostic assay is needed to determine which drugs to give, in addition to which to avoid.

in 2018 (Fig. S2). We also identified several intra-host Single Nucleotide Variants at a minor allele frequency >5% in 5 of the 14 patient samples, indicating some virus evolution and de novo mutation within hosts. However, none of these variants were in coding regions and only 1 was shared between samples (Table S2). Figure 2. Distribution of Lassa virus genetic diversity in Nigeria. a) Maximum likelihood phylogenetic tree of the S segment of the Lassa virus genome. The tree incorporates the 77 new sequences presented here alongside 193 previously published sequences from Nigeria and the Mano River Union (in gray). The 77 new samples are coloured by geographic region in which the patient resides. Samples from 2018 are in bold. b) Map of Nigeria highlighting the states from which the 77 new sequences originate and the number of samples from each state. Colours are the same as in A. Kogi state, at the intersection of the 2 rivers, is shown in striped purple reflecting the clustering of the single sequenced sample from this state with others from the southwest region in A. The location of Irrua Specialist Teaching Hospital is marked in yellow.

irectly from clinical samples.3 However, as it targets only a few potential resistance-conferring mutations, antimicrobial susceptibility cannot be reliably inferred from a negative result.4 To direct individualised therapies, a diagnostic assay is needed to determine which drugs to give, in addition to which to avoid. Advances in whole-genome sequencing mean it is now the most promising solution to the need for universal drug susceptibility testing. It is faster, more scalable, and likely to become cheaper than phenotypic testing.5 As the number of genomic sites whole-genome sequencing covers are virtually unrestricted, it should be possible to infer M. tuberculosis antimicrobial susceptibility from the absence of resistance-conferring mutations.6 Here we assess how well this performs for first-line anti- tuberculosis drugs, considering WHO target product profiles for new molecular assays,7 and whether whole-genome sequencing can be used to accurately direct anti-tuberculosis therapy. Methods Sample selection Collections of M. tuberculosis complex isolates unenriched for resistance and largely sequenced prospectively for routine diagnostic reasons, or for disease surveillance, were included from Germany, Italy, the Netherlands and the UK. Collections enriched for antimicrobial resistance, were included from across six continents (Table1, Supplement S1). Analyses of both the unenriched and complete collection were planned.

pectively for routine diagnostic reasons, or for disease surveillance, were included from Germany, Italy, the Netherlands and the UK. Collections enriched for antimicrobial resistance, were included from across six continents (Table1, Supplement S1). Analyses of both the unenriched and complete collection were planned. Table 1 Number of isolates by country and drug resistance profile Country of sample origin Time period of isolation Enriched for resistance Susceptible to all 4 drugs Susceptible to 3 drugs, with missing pyrazinamide result Isoniazid resistant, rifampicin susceptible Isoniazid susceptible, rifampicin resistant Isoniazid resistant, rifampicin resistant Other pattern Total Australia 2006-2016 Yes 0 0 4 0 38 0 42 Belgium 2007-2015 Yes 121 0 2 0 97 14 234 Canada 2003-2014 Yes 11 1,118 164 14 24 12 1343 China 2009-2012 Yes 0 44 0 0 236 0 280 Germany 1998-2015 No 248 0 9 1 13 2 273 Italy 2008-2016 Yes and No* 82 1 9 0 132 2 226 Netherlands 1993-2016 No 420 42 24 1 149 31 667 Pakistan 2014-2015 Yes 47 5 11 6 345 1 415 Peru 1997-2009 Yes 24 12 49 18 199 13 315 Russia 2008-2010 Yes 282 0 116 15 407 22 842 Serbia 2008-2014 Yes 0 0 0 0 105 0 105 South Africa 2012-2014 Yes 593 11 37 69 151 130 991 Spain 2013-2015 Yes 45 3 5 2 8 1 64 Swaziland 2009-2010 Yes 2 130 14 4 116 7 273 Thailand 1998-2013 Yes 0 53 7 4 188 0 252 UK 2009-2017 Yes and No* 3,036 82 167 6 442 154 3,887 Total 4911 1501 618 140 2650 389 10209 * More than one collection was derived from Italy and the UK, some enriched and some not enriched for resistance. See supplement for details.

Table 1 Number of isolates by country and drug resistance profile Country of sample origin Time period of isolation Enriched for resistance Susceptible to all 4 drugs Susceptible to 3 drugs, with missing pyrazinamide result Isoniazid resistant, rifampicin susceptible Isoniazid susceptible, rifampicin resistant Isoniazid resistant, rifampicin resistant Other pattern Total Australia 2006-2016 Yes 0 0 4 0 38 0 42 Belgium 2007-2015 Yes 121 0 2 0 97 14 234 Canada 2003-2014 Yes 11 1,118 164 14 24 12 1343 China 2009-2012 Yes 0 44 0 0 236 0 280 Germany 1998-2015 No 248 0 9 1 13 2 273 Italy 2008-2016 Yes and No* 82 1 9 0 132 2 226 Netherlands 1993-2016 No 420 42 24 1 149 31 667 Pakistan 2014-2015 Yes 47 5 11 6 345 1 415 Peru 1997-2009 Yes 24 12 49 18 199 13 315 Russia 2008-2010 Yes 282 0 116 15 407 22 842 Serbia 2008-2014 Yes 0 0 0 0 105 0 105 South Africa 2012-2014 Yes 593 11 37 69 151 130 991 Spain 2013-2015 Yes 45 3 5 2 8 1 64 Swaziland 2009-2010 Yes 2 130 14 4 116 7 273 Thailand 1998-2013 Yes 0 53 7 4 188 0 252 UK 2009-2017 Yes and No* 3,036 82 167 6 442 154 3,887 Total 4911 1501 618 140 2650 389 10209 * More than one collection was derived from Italy and the UK, some enriched and some not enriched for resistance. See supplement for details. Sequencing Isolates were sequenced on Illumina platforms and reads processed by the Public Health England bioinformatics pipeline at Genomics England,8 as described.6 Reads were mapped to the pan- susceptible M. tuberculosis reference genome (Genbank NC_000962.2) using Stampy (v.1.0.17)9, with repetitive regions masked. SAMtools mpileup10 (v.0.1.18) made variant-calls based on a minimum depth of 5X and at least one read on each strand. Mixed-calls were assigned where minority alleles composed >10% of read depth. Insertions and deletions were determined using Cortex (v.1.0.5.21).11

ng Stampy (v.1.0.17)9, with repetitive regions masked. SAMtools mpileup10 (v.0.1.18) made variant-calls based on a minimum depth of 5X and at least one read on each strand. Mixed-calls were assigned where minority alleles composed >10% of read depth. Insertions and deletions were determined using Cortex (v.1.0.5.21).11 Drug susceptibility testing and prediction Phenotypic drug susceptibility testing was performed locally using MGIT 960 (Becton Dickinson, New Jersey, USA), 7H10 or Löwenstein-Jensen agar, or by microscopic-observation drug- susceptibility (MODS), with method-specific critical concentrations for isoniazid (MGIT 0.1-0.2μg/mL; Agar 0.2μg/mL; MODS 0.4μg/mL), rifampicin (MGIT 1.0μg/mL; 40μg/mL Agar), ethambutol (MGIT 5.0μg/mL; Agar 0.2μg/mL), and pyrazinamide (100μg/mL). Not all laboratories routinely tested all agents (S1). Genotypic predictions were based on mutations in, or upstream of, genes associated with resistance to isoniazid (ahpC, inhA, fabG1, katG), rifampicin (rpoB), ethambutol (embA, embB, embC), and pyrazinamide (pncA).6 A knowledgebase of mutations predicting antimicrobial resistance, or not, was informed by (i) the molecular targets of WHO-recommended line-probe assays (MTBDRplus, MTBDRsl v1.0, HAIN Lifesciences, Germany), (ii) a systematic literature review,12 (iii) the CDC, Atlanta, USA, panel and (iv) two recent studies, with no isolates in common with this study (S2),6,13 of which one became available after this study commenced.13

ular targets of WHO-recommended line-probe assays (MTBDRplus, MTBDRsl v1.0, HAIN Lifesciences, Germany), (ii) a systematic literature review,12 (iii) the CDC, Atlanta, USA, panel and (iv) two recent studies, with no isolates in common with this study (S2),6,13 of which one became available after this study commenced.13 Isolates containing resistance-mutations were predicted phenotypically resistant, whereas isolates containing only wild-type sequence, phylogenetic mutations,6 or mutations considered consistent with susceptibility, were predicted susceptible. Predictions were withheld for isolates containing mutations affecting target genes but of unknown association, or where no nucleotide-call could be determined at a resistance-associated site. In these circumstances, the genotype was reported ‘unknown’ or ‘failed’, respectively. Using phenotypic results as a gold-standard, sensitivity, specificity, negative and positive predictive value were calculated for the correct assignment of susceptibility or resistance. Primary analyses excluded phenotypes without a prediction. Laboratory error was assumed where three or more phenotypes were discordant with an isolate’s genotype, or where susceptible phenotypes were recorded despite the presence of high-level resistance katG S315T mutations for isoniazid, or rpoB S450L mutations for rifampicin.14 Such isolates were excluded from further analysis. Analysis was performed using STATA (Texas, USA, v13.1). No institutional review board approval was required except in Thailand, it was granted through Mahidol University (Si029/2557).

Laboratory error was assumed where three or more phenotypes were discordant with an isolate’s genotype, or where susceptible phenotypes were recorded despite the presence of high-level resistance katG S315T mutations for isoniazid, or rpoB S450L mutations for rifampicin.14 Such isolates were excluded from further analysis. Analysis was performed using STATA (Texas, USA, v13.1). No institutional review board approval was required except in Thailand, it was granted through Mahidol University (Si029/2557). The study was first designed by TMW,TEAP,DWC, with subsequent contributions from others (supplement). Data were gathered at participating centres. Initial analysis was performed by TMW,TEAP,ASW,ZI,MH,SL,DW,PF,PM with later input from others (supplement). TMW wrote the first draft. TMW vouches for the analysis and had full access to the data; all authors agreed to publication. Results 10,290 isolates were available for the study. 81 (0.8%) were excluded due to likely laboratory error. 10,209 isolates remained, for which full first-line phenotypic profiles were available for 7,516 (73.6%), and partial profiles for the remainder. 4,911 (48.1%) isolates were phenotypically susceptible to all drugs (Table 1).

isolates were available for the study. 81 (0.8%) were excluded due to likely laboratory error. 10,209 isolates remained, for which full first-line phenotypic profiles were available for 7,516 (73.6%), and partial profiles for the remainder. 4,911 (48.1%) isolates were phenotypically susceptible to all drugs (Table 1). For each isolate, the complete sequence of nine genes and their promoter regions was interrogated to make genotypic predictions of each available phenotypic result. Predictions could be made for 8,405/8,976 (93.6%) resistant and 26,879/28,746 (93.5%) susceptible phenotypes. The remainder contained uncharacterised mutations, or missing key nucleotide calls. For isoniazid and rifampicin, ethambutol and pyrazinamide, sensitivity (proportion of resistant phenotypes predicted resistant) was 97.1%, 97.5%, 94.6% and 91.3%, and specificity (proportion of susceptible phenotypes predicted susceptible) was 99.0%, 98.8%, 93.6% and 96.8%, respectively. By comparison, an in-silico prediction of the results that would have been obtained from WHO-recommended molecular assays (Xpert MTB/RIF, MTBDRplus, MTBDRsl v1.0) had a significantly lower sensitivity than whole-genome sequencing for isoniazid, rifampicin and ethambutol (p<0.001), but greater specificity for isoniazid and ethambutol (p<0.001) (Table 2a,b). Table 2 Prediction of individual drug phenotypes Resistant phenotype, n (%) Susceptible phenotype, n (%)

For each isolate, the complete sequence of nine genes and their promoter regions was interrogated to make genotypic predictions of each available phenotypic result. Predictions could be made for 8,405/8,976 (93.6%) resistant and 26,879/28,746 (93.5%) susceptible phenotypes. The remainder contained uncharacterised mutations, or missing key nucleotide calls. For isoniazid and rifampicin, ethambutol and pyrazinamide, sensitivity (proportion of resistant phenotypes predicted resistant) was 97.1%, 97.5%, 94.6% and 91.3%, and specificity (proportion of susceptible phenotypes predicted susceptible) was 99.0%, 98.8%, 93.6% and 96.8%, respectively. By comparison, an in-silico prediction of the results that would have been obtained from WHO-recommended molecular assays (Xpert MTB/RIF, MTBDRplus, MTBDRsl v1.0) had a significantly lower sensitivity than whole-genome sequencing for isoniazid, rifampicin and ethambutol (p<0.001), but greater specificity for isoniazid and ethambutol (p<0.001) (Table 2a,b). Table 2 Prediction of individual drug phenotypes Resistant phenotype, n (%) Susceptible phenotype, n (%) R S U F Total R S U F Total Sensitivity of predictions, %(95% CI) Specificity of predictions, % (95% CI) PPV, % (95% CI) NPV, % (95% CI) Sensitivity (all*), % Specificity (all*), % No genotypic prediction made, % Resistance prevalence (all), % (a) All isolates Isoniazid 3067 90 93 44 3294 65 6313 215 117 6710 97.1 (96.5-97.7) 99.0 (98.7-99.2) 97.9 (97.4-98.4) 98.6 (98.3-98.9) 93.1 94.1 4.7 32.9 Rifampicin 2743 69 7 84 2903 85 6763 232 147 7227 97.5 (96.9-98.1) 98.8 (98.5-99.0) 97.0 (96.3-97.6) 99.0 (98.7-99.2) 94.5 93.6 4.6 28.7 Ethambutol 1410 81 94 55 1640 468 6835 781 70 8154 94.6 (93.3-95.7) 93.6 (93.0-94.1) 75.1 (73.0-77.0) 98.8 (98.5-99.1) 86.0 83.8 10.2 16.7 Pyrazinamide 863 82 117 77 1139 204 6146 197 108 6655 91.3 (89.3-93.0) 96.8 (96.3-97.2) 80.9 (78.4-83.2) 98.7 (98.4-99.0) 75.8 92.4 6.4 14.6 (b) In silico prediction of performance of MTB/RIF Xpert and HAIN MTBDRplus/MTBDRsl line-probe assays for all isolates Isoniazid 2886 355 53 3294 27 6675 8 6710 89.0 (87.9-90.1)† 99.6 (99.4-99.7)† 99.1 (98.7-99.4)† 95.0 (94.4-95.5)† 0.6 32.9 Rifampicin 2669 143 91 2903 129 6826 272 7227 94.9 (94.0-95.7)†98.1 (97.8-98.4)‡ 95.4 (94.5-96.1)‡ 97.9 (97.6-98.3)† 3.6 28.7 Ethambutol 961 641 38 1640 241 7895 18 8154 60.0 (57.5-62.4)† 97.0 (96.6-97.4)† 80.0 (77.6-82.2)‡ 92.5 (91.9-93.0)† 0.6 16.7 Pyrazinamide

99.4-99.7)† 99.1 (98.7-99.4)† 95.0 (94.4-95.5)† 0.6 32.9 Rifampicin 2669 143 91 2903 129 6826 272 7227 94.9 (94.0-95.7)†98.1 (97.8-98.4)‡ 95.4 (94.5-96.1)‡ 97.9 (97.6-98.3)† 3.6 28.7 Ethambutol 961 641 38 1640 241 7895 18 8154 60.0 (57.5-62.4)† 97.0 (96.6-97.4)† 80.0 (77.6-82.2)‡ 92.5 (91.9-93.0)† 0.6 16.7 Pyrazinamide (c) Collections from Germany, Italy, the Netherlands and the UK, unenriched for resistance Isoniazid 314 8 9 4 335 15 3770 104 90 3979 97.5 (95.2-98.9) 99.6 (99.3-99.8)† 95.4 (92.6-97.4)‡ 99.8 (99.6-99.9)† 93.7 94.7 4.8 7.8 Rifampicin 126 0 0 9 135 31 3958 103 116 4208 100.0 (97.1-100.0) 99.2 (98.9-99.5)§ 80.3 (73.2-86.2)† 100.0 (99.9-100.0)† 93.3 94.1 5.2 3.1 Ethambutol 72 1 0 0 73 47 3711 458 36 4252 98.6 (92.6-100.0) 98.7 (98.3-99.1)† 60.5 (51.1-69.3)† 100.0 (99.8-100.0)† 98.6 87.3 11.4 1.7 Pyrazinamide 109 6 4 6 125 30 4003 14 58 4105 94.8 (89.0-98.1) 99.3 (98.9-99.5)† 78.4 (70.6-84.9) 99.9 (99.7-99.9)† 87.2 97.5 1.9 3.0 (d) In silico prediction of performance of MTB/RIF Xpert and HAIN MTBDRplus/MTBDRsl line-probe assays for collections unenriched for resistance Isoniazid 295 36 4 335 10 3965 4 3979 89.1 (85.3-92.3)† 99.7 (99.5-99.9) 96.7 (94.1-98.4) 99.1 (98.8-99.4)† 0.2 Rifampicin 114 11 10 135 22 3957 229 4208 91.2 (84.8-95.6)† 99.4 (99.2-99.7) 83.8 (76.5-89.6) 99.7 (99.5-99.9)† 5.5 Ethambutol 57 16 0 73 29 4220 3 4252 78.1 (66.9-86.9)† 99.3 (99.0-99.5)§ 66.3 (55.3-76.1) 99.6 (99.4-99.8)† 0.1 Pyrazinamide

samples from each state. Colours are the same as in A. Kogi state, at the intersection of the 2 rivers, is shown in striped purple reflecting the clustering of the single sequenced sample from this state with others from the southwest region in A. The location of Irrua Specialist Teaching Hospital is marked in yellow. Genomic epidemiology of Lassa virus in Nigeria We next assessed these genomes in the context of the recent history of Lassa virus diversity in Nigeria, to determine whether the larger picture showed patterns that could help explain the recent surge. To do so, we extended our dataset to include 63 new Lassa virus genomes from RT-qPCR-positive patient samples collected at ISTH between August 2015 and November 2016 (BioProject accession PRJNA436552; Table S3). The patients resided in 11 states, with most (68%) coming from Edo and Ondo. This combined dataset considerably expands and updates previous phylogenetic trees of Lassa virus in Nigeria.

3979 89.1 (85.3-92.3)† 99.7 (99.5-99.9) 96.7 (94.1-98.4) 99.1 (98.8-99.4)† 0.2 Rifampicin 114 11 10 135 22 3957 229 4208 91.2 (84.8-95.6)† 99.4 (99.2-99.7) 83.8 (76.5-89.6) 99.7 (99.5-99.9)† 5.5 Ethambutol 57 16 0 73 29 4220 3 4252 78.1 (66.9-86.9)† 99.3 (99.0-99.5)§ 66.3 (55.3-76.1) 99.6 (99.4-99.8)† 0.1 Pyrazinamide PPV = Positive Predictive Value; NPV = Negative Predictive Value; R=resistant; S=susceptible; U=mutation of unknown association present; F=genotypic prediction failed due to missing data around a genomic resistance locus; All % results based on R/S genotypic predictions only, excluding U and F except where * for which denominator includes R, S, U and F. †p≤0.001 , ‡p≤0.01, and §p≤0.05 comparing sensitivity, specificity, NPV and PPV for each drug for (b) and (c) against (a), and comparing (d) against (c); p>0.05 for all results not marked †, ‡ or §. In silico predictions of resistance for Xpert and HAIN assays were based on the presence of non-wild type sequence within the genomic regions interrogated by these assays. 'F' was reported in the presence of minority alleles at relevant sites, just as for WGS predictions. The negative predictive value (proportion of concordant susceptible predictions) was over 98.5% for all four drugs. Although dependent on prevalence, this also varied with isolates’ background phenotypic profiles. For example, at 20% prevalence of pyrazinamide resistance, the expected negative predictive value for pyrazinamide was 93.6% and 99.0% for isolates susceptible and resistant to the other three drugs, respectively (Table 3, S3).

ugh dependent on prevalence, this also varied with isolates’ background phenotypic profiles. For example, at 20% prevalence of pyrazinamide resistance, the expected negative predictive value for pyrazinamide was 93.6% and 99.0% for isolates susceptible and resistant to the other three drugs, respectively (Table 3, S3). Phenotypic profiles R S U F Total R S U F Total Prevalence of resistance among each of the listed drug profiles, % Sensitivity, % Specificity, % PPV, % NPV,% Expected NPV at given prevalence of resistance based on simulations, % (95% CI)* Calculated NPV at 20% prevalence of resistance, % (see table S3) Calculated NPV at 40% prevalence of resistance, % (see table S3) Isoniazid -SSS 391 30 18 12 451 21 4,653 133 104 4,911 8.4 93 100 95 99.4 99.3-100 98.2 95.4 -RSS 459 21 20 6 506 7 85 5 1 98 83.8 96 92 98 80.2 83.5-100 98.8 96.9 -RRS 424 3 13 4 444 2 2 2 0 6 98.7 99 50 100 40.0 73.7-85.6 99.6 99.1 -SRS 24 4 1 0 29 0 10 1 0 11 72.5 86 100 100 71.4 90.5-95.6 96.6 91.3 -SSR 24 1 2 1 28 0 95 6 3 104 21.2 96 100 100 99.0 98.5-99.7 99 97.4 -RRR 662 3 11 4 680 0 0 0 0 0 100.0 100 . 100 0.0 73.7-85.6 n/a n/a -RSR 217 3 5 5 230 0 3 0 0 3 98.7 99 100 100 50.0 73.7-85.6 99.7 99.1 -SRR 13 0 0 2 15 0 0 0 0 0 100.0 100 . 100 . 73.7-85.6 n/a n/a

10 1 0 11 72.5 86 100 100 71.4 90.5-95.6 96.6 91.3 -SSR 24 1 2 1 28 0 95 6 3 104 21.2 96 100 100 99.0 98.5-99.7 99 97.4 -RRR 662 3 11 4 680 0 0 0 0 0 100.0 100 . 100 0.0 73.7-85.6 n/a n/a -RSR 217 3 5 5 230 0 3 0 0 3 98.7 99 100 100 50.0 73.7-85.6 99.7 99.1 -SRR 13 0 0 2 15 0 0 0 0 0 100.0 100 . 100 . 73.7-85.6 n/a n/a Rifampicin S-SS 74 16 0 8 98 30 4,632 126 123 4,911 2.0 82 99 71 99.7 99.3-100 95.7 89.3 S-RS 6 0 0 0 6 1 9 1 0 11 35.3 100 90 86 100.0 97.8-99.5 100 100 S-SR 1 2 0 0 3 0 100 3 1 104 2.8 33 100 100 98.0 99.3-100 85.7 69.2 S-RR 0 0 0 0 0 0 0 0 0 0 . . . . . . n/a n/a R-SS 464 20 1 21 506 18 424 3 6 451 52.9 96 96 96 95.5 95.8-98.6 98.9 97.2 R-RS 424 7 2 11 444 4 25 0 0 29 93.9 98 86 99 78.1 76.2-86.6 99.5 98.8 R-SR 218 4 0 8 230 7 20 0 1 28 89.1 98 74 97 83.3 77.9-87.9 99.4 98.4 R-RR 665 2 0 13 680 10 3 0 2 15 97.8 100 23 99 60.0 76.2-86.6 99.7 99.1 Ethambutol SS-S 1 9 1 0 11 4 4,399 472 36 4,911 0.2 10 100 20 99.8 98.8-99.9 81.6 62.5 RS-S 21 5 3 0 29 31 376 40 4 451 6.0 81 92 40 98.7 98.8-99.9 95.1 87.8 SR-S 4 2 0 0 6 1 93 3 1 98 5.8 67 99 80 97.9 98.8-99.9 92.2 81.7 RR-S 375 20 30 19 444 203 241 48 14 506 46.7 95 54 65 92.3 93.4-96.7 97.7 94.1 SS-R 0 0 0 0 0 1 81 22 0 104 0.0 . 99 0 100.0 98.8-99.9 n/a n/a RS-R 12 2 1 0 15 7 20 1 0 28 34.9 86 74 63 90.9 95.7-98.1 95.4 88.6 SR-R 0 0 0 0 0 0 3 0 0 3 0.0 . 100 . 100.0 98.8-99.9 n/a n/a RR-R 625 9 26 20 680 150 50 25 5 230 74.7 99 25 81 84.7 82.0-88.2 98.6 96.4

41 48 14 506 46.7 95 54 65 92.3 93.4-96.7 97.7 94.1 SS-R 0 0 0 0 0 1 81 22 0 104 0.0 . 99 0 100.0 98.8-99.9 n/a n/a RS-R 12 2 1 0 15 7 20 1 0 28 34.9 86 74 63 90.9 95.7-98.1 95.4 88.6 SR-R 0 0 0 0 0 0 3 0 0 3 0.0 . 100 . 100.0 98.8-99.9 n/a n/a RR-R 625 9 26 20 680 150 50 25 5 230 74.7 99 25 81 84.7 82.0-88.2 98.6 96.4 Pyrazinamide SSS- 74 28 0 2 104 12 4,826 13 60 4,911 2.1 73 100 86 99.4 98.6-99.6 93.6 84.5 RSS- 13 8 4 3 28 5 431 2 13 451 5.8 62 99 72 98.2 98.6-99.6 91.2 79.6 RRS- 166 25 22 17 230 49 374 68 15 506 31.3 87 88 77 93.7 95.5-97.7 96.4 91 SRS- 0 3 0 0 3 0 97 0 1 98 3.0 0 100 . 97.0 98.6-99.6 80 60 RRR- 532 15 83 50 680 107 216 105 16 444 60.5 97 67 83 93.5 87.3-91.0 99 97.3 SRR- 0 0 0 0 0 0 6 0 0 6 0.0 . 100 . 100.0 98.6-99.6 n/a n/a RSR- 10 2 1 2 15 0 28 0 1 29 34.1 83 100 100 93.3 95.0-97.3 96 90 SSR- 0 0 0 0 0 0 11 0 0 11 0.0 . 100 . 100.0 98.6-99.6 n/a n/a

3 0 0 3 0 97 0 1 98 3.0 0 100 . 97.0 98.6-99.6 80 60 RRR- 532 15 83 50 680 107 216 105 16 444 60.5 97 67 83 93.5 87.3-91.0 99 97.3 SRR- 0 0 0 0 0 0 6 0 0 6 0.0 . 100 . 100.0 98.6-99.6 n/a n/a RSR- 10 2 1 2 15 0 28 0 1 29 34.1 83 100 100 93.3 95.0-97.3 96 90 SSR- 0 0 0 0 0 0 11 0 0 11 0.0 . 100 . 100.0 98.6-99.6 n/a n/a Phenotypic profiles are listed in the following order: Isoniazid, Rifampicin, Ethambutol, Pyrazinamide. '-' under 'Phenotypic profiles' marks the drug phenotype being assessed. PPV = Positive Predictive Value; NPV = Negative Predictive Value; R=resistant; S=susceptible; U=mutation of unknown association present; F=genotypic prediction failed due to missing data around a genomic resistance locus; All % results based on R/S genotypic predictions only, excluding U and F. Expected NPV was calculated as follows: specificity x (1-prevlence) / (specificity x (1-prevlence)+(1-sensitivity) x prevalence). * indicates that for prevalence <10% or >90%, simulated values are given for 10% and 90% respectively as simulations were not performed below or above these values. As some collections included clustered isolates, the analysis was repeated after randomly selecting one representative among genomically indistinguishable isolates, and again from isolates within five single nucleotide polymorphisms of another. No significant change in sensitivity or specificity was observed for any drugs (p>0.1, S4).

ons included clustered isolates, the analysis was repeated after randomly selecting one representative among genomically indistinguishable isolates, and again from isolates within five single nucleotide polymorphisms of another. No significant change in sensitivity or specificity was observed for any drugs (p>0.1, S4). To reflect the emerging practice of routinely sequencing isolates for clinical care, the analysis was repeated for the subset of 4,397 isolates from German, Italian, Dutch and UK collections that were not enriched for resistance. Among these isolates, 335 (7.6%) were isoniazid resistant and 125 (2.8%) multidrug-resistant. For each drug, specificity and negative predictive values increased, whilst positive predictive values (the proportion of concordant resistant predictions) decreased relative to the overall results. There was no significant change in sensitivity (Table 2c). Predicting complete phenotypic profiles For DNA sequencing to help individualise therapy, a minimum requirement is that all first-line antimicrobial phenotypes are predicted. Phenotypic profiles were thus predicted for 7,516 isolates with phenotypic data available for all first-line drugs (S1&6). ‘Unknown’ or ‘failed’ was reported for at least one drug for 1,651 (22.0%) profiles. 5,865 (78.0%) were predicted completely, of which 5,250 (89.5%) were predicted correctly (S5). Among the 5,865 profiles, 4,007 were phenotypically pan-susceptible, of which 3952 (98.6%) were predicted correctly (Table 4).

drugs (S1&6). ‘Unknown’ or ‘failed’ was reported for at least one drug for 1,651 (22.0%) profiles. 5,865 (78.0%) were predicted completely, of which 5,250 (89.5%) were predicted correctly (S5). Among the 5,865 profiles, 4,007 were phenotypically pan-susceptible, of which 3952 (98.6%) were predicted correctly (Table 4). Table 4 Genotypic drug profile predictions of pan-susceptibility Prediction Genotypic drug profile Number predicted to have drug profile Number predicted to have drug profile that are phenotypic ally pansusceptible (%) Sensitivity % Specificity % PPV % NPV % Predictions made % Inh Rif Emb Pza (a) Predicted pan-susceptible S S S S 4,037 3952 (97.9) (b) Predicted pansusceptible after inferring that 'U' mutations are consistent with susceptibility in this context S S S U 11 11 (100) S S U S 410 399 (97.3) S S U U 2 2 (100) S U S S 93 88 (94.6) S U U S 29 29 (100) Total 4,582 4481 (97.8) (c) Predicted to have some phenotypic resistance R S R or S 397 18 (4.5) S At least one R, no U or F 158 36 (22.8) R R R or S 1273 1 (0.1) Total 1828 55 (3.0) 95.4 98.6 97.0 97.9 78.0 94.6 98.8 97.0 97.8 85.1 No prediction made (drug profile prediction incomplete) U S or U 150 126 (84.0) At least one F, no R 280 240 (85.7) At least one R and U, no F 499 6 (1.2) At least one R and F, no U 159 3 (1.9) At least one R, U, and F 18 0 (0.0) Total 1106 375 (33.9)

R R R or S 1273 1 (0.1) Total 1828 55 (3.0) 95.4 98.6 97.0 97.9 78.0 94.6 98.8 97.0 97.8 85.1 No prediction made (drug profile prediction incomplete) U S or U 150 126 (84.0) At least one F, no R 280 240 (85.7) At least one R and U, no F 499 6 (1.2) At least one R and F, no U 159 3 (1.9) At least one R, U, and F 18 0 (0.0) Total 1106 375 (33.9) As the proportion of incompletely predicted profiles was substantial (22.0%), we assessed whether pan-susceptibility could be accurately predicted for some of these isolates anyway. Because isoniazid susceptibility predicts susceptibility to other first-line drugs,15 we maximised confidence in isoniazid predictions by conditioning predictions on the absence of ‘unknown’ mutations in isoniazid- related genes. ‘Unknown’ mutations relevant to other drugs were permitted. Doing this, pan- susceptibility was correctly predicted for 4,481/4,582 (97.8%) isolates, including 545/1,651 (33.0%) previously incompletely predicted profiles (Table 4). Among the collections unenriched for resistance, 3439/3450 (99.7%) profiles were thereby correctly predicted pan-susceptible (S7).

s were permitted. Doing this, pan- susceptibility was correctly predicted for 4,481/4,582 (97.8%) isolates, including 545/1,651 (33.0%) previously incompletely predicted profiles (Table 4). Among the collections unenriched for resistance, 3439/3450 (99.7%) profiles were thereby correctly predicted pan-susceptible (S7). To simulate how this approach would perform in settings with differing burdens of antimicrobial resistance, we assessed the decline in negative predictive value with increasing prevalence of resistance to individual drugs, and with prevalence of any resistance within drug profiles. We randomly sub-sampled 1,000 isolates to represent every 1% increment in antimicrobial-resistance prevalence between 10%-90%, repeating this 1,000 times for each drug and for complete drug profiles. Negative predictive value declined further for ethambutol and pyrazinamide than for complete drug profiles, but declined least for isoniazid and rifampicin. Below 47.0% prevalence of resistance to any drug, the simulated negative predictive value remained above 95% for 97.5% of drug profiles (Figure 1). Figure 1 Simulated negative predictive values for individual drugs and complete drug profiles Negative predictive vales shown for individual drugs and complete drug profiles, according to simulated prevalence of resistance to each drug, or within each drug profile (‘any resistance’). For each percentage prevalence between 10% and 90%, 1,000 isolates were randomly selected, 1,000 times. Lines indicate the median with shaded areas showing the 95% confidence intervals.

complete drug profiles, according to simulated prevalence of resistance to each drug, or within each drug profile (‘any resistance’). For each percentage prevalence between 10% and 90%, 1,000 isolates were randomly selected, 1,000 times. Lines indicate the median with shaded areas showing the 95% confidence intervals. Discrepancy analyses In Australia, eleven ethambutol susceptible isolates containing embB mutations were re- phenotyped. Three repeat assays failed, but seven of the remaining eight yielded, now consistent, resistant phenotypes. In Peru, 10 of 16 repeated assays remained phenotypically susceptible by MODS despite fabG1 C-15T or G-17T mutations. In isolates from the Netherlands, six resistant phenotypes predicted susceptible were identified as clerical errors, and three susceptible phenotypes predicted resistant tested phenotypically resistant by alternative phenotypic assays (S8). Although additional re- phenotyping was not possible, we conducted a ‘per mutation’ analysis to further assess discrepancies.

t phenotypes predicted susceptible were identified as clerical errors, and three susceptible phenotypes predicted resistant tested phenotypically resistant by alternative phenotypic assays (S8). Although additional re- phenotyping was not possible, we conducted a ‘per mutation’ analysis to further assess discrepancies. Of the 322 resistant phenotypes predicted susceptible, 290 (90.1%) had no mutations affecting targeted genes, and 32 (9.9%) had one or more of 15 mutations per isolate, each previously characterised as consistent with antimicrobial susceptibility. Supporting this, across all isolates in which these 15 mutations occurred as the sole mutation, they correctly predicted isoniazid susceptibility in 286/293 (97.6%) isolates and ethambutol susceptibility in 95/119 (79.8%) isolates. The one mutation relevant to pyrazinamide was seen in two isolates, both of which were phenotypically resistant. None of these mutations were relevant to rifampicin (S9).

ole mutation, they correctly predicted isoniazid susceptibility in 286/293 (97.6%) isolates and ethambutol susceptibility in 95/119 (79.8%) isolates. The one mutation relevant to pyrazinamide was seen in two isolates, both of which were phenotypically resistant. None of these mutations were relevant to rifampicin (S9). Among 822 susceptible phenotypes predicted resistant, 145 different resistance-conferring mutations were found. Of these, 142 (97.9%) featured as the only resistance-conferring mutation in at least one isolate in the dataset, allowing assessment of individual predictive performance. They correctly predicted resistance to isoniazid in 308/371 (83.0%) isolates, rifampicin in 548/627 (87.4%) isolates, ethambutol in 1280/1743 (73.4%) isolates, and pyrazinamide in 459/663 (69.2%) isolates (S9). 14 of 17 (82.3%) mutations leading to rifampicin resistance predictions in phenotypically susceptible isolates were in the genetic region targeted by Xpert MTB/RIF and MTBDRplus. Laboratory sample mislabelling probably also contributed discrepant results. This was estimated for each collection from the proportion of isolates excluded because of katG S315T or rpoB S450L mutations and susceptible phenotypes, the collection’s discrepancy rate, and the prevalence of antimicrobial resistance (S10). Overall, about 43% of isoniazid, and 12% of rifampicin discrepancies were thereby attributable to mislabelling.

tion from the proportion of isolates excluded because of katG S315T or rpoB S450L mutations and susceptible phenotypes, the collection’s discrepancy rate, and the prevalence of antimicrobial resistance (S10). Overall, about 43% of isoniazid, and 12% of rifampicin discrepancies were thereby attributable to mislabelling. Discussion This analysis of over 10,000 M. tuberculosis isolates collected from 16 countries across six continents, and representing all major lineages, demonstrates that whole-genome sequencing can now characterise susceptible first-line anti-tuberculosis drug profiles sufficiently accurately for clinical use. The importance of this is twofold: First, it demonstrates that the genomic approach can be used to tailor individual treatment regimens. Extended to all drugs, individualised therapies promise to improve cure rates over those achieved by semi-empiric regimens directed by more limited diagnostic tests.1 Second, it is now possible to reduce the phenotypic workload where routine whole-genome sequencing is performed.

to tailor individual treatment regimens. Extended to all drugs, individualised therapies promise to improve cure rates over those achieved by semi-empiric regimens directed by more limited diagnostic tests.1 Second, it is now possible to reduce the phenotypic workload where routine whole-genome sequencing is performed. The WHO’s target product profiles for new molecular assays for M. tuberculosis require over 90% and 95% sensitivity and specificity, respectively.7 Overall, both these targets were met for all drugs with the exception of specificity for ethambutol (93.6%). This is no surprise as phenotyping is an imperfect gold standard, in particular for isolates with embB mutations.6,13,16 For the collections unenriched for resistance, all drugs did however meet these targets, as did the predictions of pan- susceptibility in all collections. Only categorical agreement was assessed for complete drug profile predictions because of the number of permutations. These met the external quality assurance criteria (>80% concordance) for the European TB reference laboratory network.17

t these targets, as did the predictions of pan- susceptibility in all collections. Only categorical agreement was assessed for complete drug profile predictions because of the number of permutations. These met the external quality assurance criteria (>80% concordance) for the European TB reference laboratory network.17 There are three reasons why pan-susceptibility predictions were particularly accurate. First, the knowledgebase included both resistance-associated genomic mutations, and mutations compatible with phenotypic susceptibility. Second, anti-tuberculosis drug susceptibility phenotypes are not independent of one another, allowing the use of isoniazid susceptibility to predict susceptibility to other drugs. Third, no predictions were attempted for isolates containing genomic variation of unknown association in genes affecting isoniazid. This maximised confidence in isoniazid predictions that were made. Consequently, the prediction of drug profiles performed better than the per-drug analysis for ethambutol and pyrazinamide, and although there was a slight corresponding decline in performance for isoniazid and rifampicin, simulations showed that the prevalence of resistance would have to exceed that seen in most of the worst affected countries in the world before these predictions no longer satisfied the WHO targets.1

tol and pyrazinamide, and although there was a slight corresponding decline in performance for isoniazid and rifampicin, simulations showed that the prevalence of resistance would have to exceed that seen in most of the worst affected countries in the world before these predictions no longer satisfied the WHO targets.1 Our findings showed substantial improvements over the in-silico predictions for the sensitivity of WHO-recommended PCR-based assays because whole-genome sequencing is able to identify many more mutations. These additional mutations were however simultaneously responsible for the losses in specificity, largely because of the number of mutations for which a minority of isolates did not manifest a resistant phenotype. A typical example of such is the rpoB I491F mutation which frequently gives a susceptible rifampicin result in liquid culture but has been linked to treatment failure.4,18,19

or the losses in specificity, largely because of the number of mutations for which a minority of isolates did not manifest a resistant phenotype. A typical example of such is the rpoB I491F mutation which frequently gives a susceptible rifampicin result in liquid culture but has been linked to treatment failure.4,18,19 The broader discrepancy analysis highlighted the same phenomenon. Whilst the predictive performance of individual mutations, whether probed by WHO-recommended assays or not, was good, each mutation has an error rate, occasionally leading to an unexpected phenotype in a minority of isolates. This is most likely where a mutation elevates the minimum drug concentration required to inhibit bacterial growth to close to the concentration above which an isolate is considered resistant. Canonical ethambutol mutations are a classic example,20 but there are many others including the mutations missed by the MODS assay in Peru.16,21,22 Such phenomena are thus likely to explain the majority of isolates that were predicted resistant, yet were phenotypically susceptible. They are also the most likely reason why predicting pan-susceptible drug profiles was more accurate than predicting profiles apparently resistant to one or more drugs.

in Peru.16,21,22 Such phenomena are thus likely to explain the majority of isolates that were predicted resistant, yet were phenotypically susceptible. They are also the most likely reason why predicting pan-susceptible drug profiles was more accurate than predicting profiles apparently resistant to one or more drugs. One study limitation is that the scale and cost of repeat sequencing and phenotyping of isolates meant that we could not definitively resolve most discrepancies. This was most concerning for phenotypically resistant isolates predicted susceptible. For these, possible explanations include phenotypic error, resistant minority bacterial populations undetected by sequencing, mechanisms of resistance linked to genes we did not interrogate, or laboratory labelling error.

screpancies. This was most concerning for phenotypically resistant isolates predicted susceptible. For these, possible explanations include phenotypic error, resistant minority bacterial populations undetected by sequencing, mechanisms of resistance linked to genes we did not interrogate, or laboratory labelling error. More work remains to be done before predictions can be extended to second and third-line drugs, and to newer compounds. However, following external review, Public Health England has already decided to stop phenotyping isolates predicted pan-susceptible to first-line drugs (personal communication, Derrick Crook, Director, National Infection Service). Similar moves are expected in the Netherlands (Dick van Soolingen, Rijksinstituut voor Volksgezondheid en Milieu) and New York (Kimberlee Musser, Wadsworth Center, New York State Department of Health). For low and middle- income countries without easy access to phenotyping, there is now the prospect that emerging mobile sequencing platforms could be used to implement sequence-directed therapies, a potential solution to the call for universal susceptibility testing. Portable platform sequencing directly from spiked-samples has been achieved, although real-world systematic evaluation is still required.23

he prospect that emerging mobile sequencing platforms could be used to implement sequence-directed therapies, a potential solution to the call for universal susceptibility testing. Portable platform sequencing directly from spiked-samples has been achieved, although real-world systematic evaluation is still required.23 Should whole-genome sequencing perform as well for second and third-line drugs as for first- line, a clinical trial could be needed to assess the performance of individualised over standardized treatment regimens in countries with a high drug-resistant disease burden.24 Individualised therapies would be expected to reduce the amplification of resistance (to other drugs) in individual patients, side- effects, likelihood of onward transmission, and to exert a weaker selection pressure on strains at a population level, which is key where empiric regimens have been targeted on the basis of very narrow data on antimicrobial susceptibility.4 Welcome public health benefits could result from monitoring transmission using the very same sequences.2 The current investment in whole-genome sequencing in high-income countries is likely to help accelerate implementation in lower-income, higher-burden countries where the potential benefit is greatest.25 These data demonstrate how our understanding of the molecular determinants of resistance to first-line anti-tuberculosis drugs is now sufficiently good to start using DNA sequencing to guide therapy. Similar performance must now be replicated for the remaining drugs.

countries where the potential benefit is greatest.25 These data demonstrate how our understanding of the molecular determinants of resistance to first-line anti-tuberculosis drugs is now sufficiently good to start using DNA sequencing to guide therapy. Similar performance must now be replicated for the remaining drugs. Supplementary Appendix Supplementary Material Authors contributing are (in alphabetical order): Caroline Allix-Béguec, Irena Arandjelovic, Patrick Beckert, Lijun Bi, Maryline Bonnet, Phelim Bradley, Andrea M Cabibbe, Irving Cancino-Muñoz, Mark J Caulfield, Angkana Chaiprasert, Daniela Cirillo, David Clifton, Iñaki Comas, Derrick W Crook, Maria Rosaria De Filippo, Han de Neeling, Roland Diel, Francis A Drobniewski, Kiatichai Faksri, Maha R Farhat, Joy Fleming, Philip Fowler, Tom A Fowler, Qian Gao, Jennifer Gardy, Deborah Gascoyne-Binzi, Ana Gibertoni Cruz, Ana Gil-Brusola, Tanya Golubchik, Ximena Gonzalo, Louis Grandjean, Jennifer L Guthrie, Guangxue He, Sarah Hoosdally, Martin Hunt, Zamin Iqbal, Nazir Ismail, James Johnston, Faisal Masood Khanzada, Chiea Chuen Khor, Thomas A Kohl, Clare Kong, Sam Lipworth, Qingyun Liu, Gugu Maphalala, Elena Martinez, Vanessa Mathys, Matthias Merker, Paolo Miotto, Nerges Mistry, David Moore, Megan Murray, Stefan Niemann, Rick Twee-Hee Ong, Tim E A Peto, James E Posey, Therdsak Prammananan, Alexander Pym, Camilla Rodrigues, Mabel Rodrigues, Timothy Rodwell, Gian Maria Rossolini, Elisabeth Sánchez Padilla, Marco Schito, Xin Shen, Jay Shendure, Vitali Sintchenko, Alex Sloutsky, E Grace Smith, Matthew Snyder, Karine Soetaert, Angela M Starks, Philip Supply, Prapat Suriyapol, Sabira Tahseen, Patrick Tang, Yik-Ying Teo, Thuong Nguyen Thuy Thuong, Guy Thwaites, Enrico Tortoli, Shaheed Vally Omar, Dick van Soolingen, A Sarah Walker, Timothy M Walker, Mark Wilcox, Daniel J Wilson, David Wyllie, Yang Yang, Hongtai Zhang, Yanlin Zhao, Baoli Zhu.

, Angela M Starks, Philip Supply, Prapat Suriyapol, Sabira Tahseen, Patrick Tang, Yik-Ying Teo, Thuong Nguyen Thuy Thuong, Guy Thwaites, Enrico Tortoli, Shaheed Vally Omar, Dick van Soolingen, A Sarah Walker, Timothy M Walker, Mark Wilcox, Daniel J Wilson, David Wyllie, Yang Yang, Hongtai Zhang, Yanlin Zhao, Baoli Zhu. Author contributions: CAB, IA, PBe, LB, MB, AMC, AC, DMC, IC, MJC, RD, FAD, KF, MRF, JF, PF, TAF, QG, JGa, DGB, AGB, TG, XG, LG, JLG, GH, NI, JJ, CK, FMK, CCK, TAK, QL, GM, EM, ICM, VM, MMe, MMu, DM, HDN, SN, RTHO, TP, ESP, GMR, MR, TR, AS, VS, EGS, JS, KS, MSc, MSn, PhS, PrS, XS, PT, YYT, ST, ET, SVO, DvS, MW, HZ, YZ, and BZ contributed towards data acquisition (including whole-genome sequencing and phenotypic drug susceptibility testing); DMC, NI, DM, SN, CR, EGS, PhS, ST, DC, DWC, AGC, SH, PM, NM, TEAP, JEP, AP, AMS, TNTT, GT, ASW, TMW, YY, AMC, HDN, MRF, and PF contributed towards study design; DMC, DC, PM, TEAP, ASW, TMW, YY, AMC, PF, PBr, MRDF, MH, ZI, SL, DJW, and DW contributed towards data analysis; TEAP, ASW and TMW wrote the manuscript; all authors contributed feedback on the manuscript. We would like to thank Stéphanie Duthoy, Carina Hahn, Alamdar Hussain, Yannick Laurent, Mathilde Mairey, Vanessa Mohr and Mahmood Qadir and for technical assistance.

Author contributions: CAB, IA, PBe, LB, MB, AMC, AC, DMC, IC, MJC, RD, FAD, KF, MRF, JF, PF, TAF, QG, JGa, DGB, AGB, TG, XG, LG, JLG, GH, NI, JJ, CK, FMK, CCK, TAK, QL, GM, EM, ICM, VM, MMe, MMu, DM, HDN, SN, RTHO, TP, ESP, GMR, MR, TR, AS, VS, EGS, JS, KS, MSc, MSn, PhS, PrS, XS, PT, YYT, ST, ET, SVO, DvS, MW, HZ, YZ, and BZ contributed towards data acquisition (including whole-genome sequencing and phenotypic drug susceptibility testing); DMC, NI, DM, SN, CR, EGS, PhS, ST, DC, DWC, AGC, SH, PM, NM, TEAP, JEP, AP, AMS, TNTT, GT, ASW, TMW, YY, AMC, HDN, MRF, and PF contributed towards study design; DMC, DC, PM, TEAP, ASW, TMW, YY, AMC, PF, PBr, MRDF, MH, ZI, SL, DJW, and DW contributed towards data analysis; TEAP, ASW and TMW wrote the manuscript; all authors contributed feedback on the manuscript. We would like to thank Stéphanie Duthoy, Carina Hahn, Alamdar Hussain, Yannick Laurent, Mathilde Mairey, Vanessa Mohr and Mahmood Qadir and for technical assistance. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, the Department of Health or Public Health England. Use of trade names is for identification only and does not constitute endorsement by the U.S. Department of Health and Human Services, the U.S. Public Health Service, or the CDC. The findings and conclusions expressed by authors contributing to this journal do not necessarily reflect the official opinion of the Centers for Disease Control and Prevention, or the authors’ affiliated institutions.

ment by the U.S. Department of Health and Human Services, the U.S. Public Health Service, or the CDC. The findings and conclusions expressed by authors contributing to this journal do not necessarily reflect the official opinion of the Centers for Disease Control and Prevention, or the authors’ affiliated institutions. Institutional funding acknowledgements: The UK element of this research was made possible in part by the 100,000 Genomes Project which is managed by Genomics England Limited (a wholly owned company of the Department of Health). The 100,000 Genomes Project is funded by the National Institute for Health Research and NHS England. The Wellcome Trust, the Medical Research Council and Public Health England have also funded research infrastructure. The 100,000 Genomes Project uses data and samples collected by the National Health Service as part of care and support of patients. This work was also supported by Wellcome Trust/Newton Fund-MRC Collaborative Award [200205/Z/15/Z]; and Bill & Melinda Gates Foundation [OPP1133541] and by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre (BRC) and National Institute for Health Research Health Protection Research Unit (NIHR HPRU) in Healthcare Associated Infections and Antimicrobial Resistance at University of Oxford in partnership with Public Health England (PHE). This work was also facilitated by the NIHR Biomedical Research Centre at Barts and by the NIHR Biomedical Research Centre at Imperial.

rch Health Protection Research Unit (NIHR HPRU) in Healthcare Associated Infections and Antimicrobial Resistance at University of Oxford in partnership with Public Health England (PHE). This work was also facilitated by the NIHR Biomedical Research Centre at Barts and by the NIHR Biomedical Research Centre at Imperial. National Science and Technology Key Program of China (2014ZX10003002), National Basic Research program of China (2014CB744403). BCCDC Foundation for Population and Public Health in Canada. Borstel has been supported by a grant from the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement 278864 in the framework of the Patho-NGen- Trace project, and by the German Center for Infection Research (DZIF). Leibniz Science Campus Evolutionary Medicine of the Lung (EvoLUNG). Belgian Reference Centre for Tuberculosis & Mycobacteria from Bacterial Diseases Service is partially supported by the Belgian Ministry of Social Affairs through a fund within the Health Insurance System. Genoscreen was supported by the French governmental Program ‘Investing for the Future’ (Equipex LIGAN platform) and by a grant from the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement 278864 in the framework of the PathoNGenTrace project. EU FP& EUROGEN and PANNET grants.

System. Genoscreen was supported by the French governmental Program ‘Investing for the Future’ (Equipex LIGAN platform) and by a grant from the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement 278864 in the framework of the PathoNGenTrace project. EU FP& EUROGEN and PANNET grants. Individual funding acknowledgements: Angkana Chaiprasert (research grant Siriraj Research grant No. R015833003, Drug Resistant Tuberculosis Fund and Chalermprakiat grant, Faculty of Medicine Siriraj Hospital, Mahidol University). Iñaki Comas (MINECO research grant SAF2016-77346-R and the European Research Council (ERC) (638553-TB-ACCELERATE) (to IC)); Daniel Wilson and Zamin Iqbal are Sir Henry Dale Fellows, jointly funded by the Wellcome Trust and the Royal Society (grant nos. 101237/Z/13/Z and 102541/A/13/Z respectively)); Timothy Walker is an NIHR Academic Clinical Lecturer; Derrick Crook, Tim Peto and Mark Caulfield are NIHR Senior Investigators. Rick Twee-Hee Ong and Yik-Ying Teo (National University of Singapore Yong Loo Lin School of Medicine Aspiration Fund (NUHSRO/2014/069/AF- New Idea/04)). Francis A Drobniewski was supported by EU FP7 European Union Framework Programme 7 (Grant number 201483; TB-EUROGEN) and TB-PANNET (Grant FP7-223681). This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1800474..

a distinct sublineage (Fig. 2B). This pattern of distinct regional lineages, each internally diverse, indicates that Lassa virus has remained stably separated in the rodent populations of these regions; for example, the most recent common ancestor of lineage II occurred around 235 years ago (95% CI: 187-283; Fig. S2). The observed clustering aligns with the courses of the Niger and Benue rivers in Nigeria (Fig. 2B), suggesting that these major rivers present natural barriers to Mastomys rodents. This pattern further supports a key role for the rodent reservoir, and not humans, in the ongoing transmission of Lassa virus. Together with the long branch lengths of these groups – suggestive of extensive, uncaptured Lassa virus diversity in these regions – these results indicate sequestering of the rodent population and their associated Lassa virus lineages in these regions.

T-qPCR-positive patient samples collected at ISTH between August 2015 and November 2016 (BioProject accession PRJNA436552; Table S3). The patients resided in 11 states, with most (68%) coming from Edo and Ondo. This combined dataset considerably expands and updates previous phylogenetic trees of Lassa virus in Nigeria. Samples from 2015-2018 cluster geographically on the phylogenetic tree. All eleven samples sequenced here from northern Nigeria fall into lineage III (Fig. 2B), increasing our sampling of this lineage more than threefold. These samples confirm the high genetic diversity of this lineage and make clear that it is a regionally defined variant of Lassa virus. Our dataset further identifies a separation in lineage II between samples from southwestern and eastern states, with samples from the eastern states of Ebonyi, Taraba and Anambra forming a distinct sublineage (Fig. 2B). This pattern of distinct regional lineages, each internally diverse, indicates that Lassa virus has remained stably separated in the rodent populations of these regions; for example, the most recent common ancestor of lineage II occurred around 235 years ago (95% CI: 187-283; Fig. S2).

ognormal distribution, and a Bayesian SkyGrid coalescent tree prior. All of the Bayesian analyses were run for 200 million MCMC steps, sampling parameters and trees every 5,000 generations. Maximum-clade credibility trees summarizing all MCMC samples were generated using TreeAnnotator v1.8.4 with a burn-in rate of 10%. RESULTS Lassa fever case burden at ISTH in 2018 The ISTH Lassa ward, with 16 beds, is the largest Lassa fever facility in Nigeria and a major diagnostic referral center, receiving suspected Lassa fever patient samples from across the country. From January 1 to March 13, 2018, ISTH tested over 1500 clinically suspected Lassa fever cases, of which 368 were RT-qPCR-positive for Lassa virus (Fig. 1A & 1B). This number, which represents the majority of confirmed cases in Nigeria during this period, is markedly higher than that observed in previous years (Fig. 1A). There is a wide distribution of ages (Fig. S1A) and geographic source of confirmed cases (Fig. S1B), as previously observed for Lassa fever28. We did observe an approximate 2:1 male-to-female ratio among confirmed cases, in contrast to previous conclusions that Lassa fever does not exhibit sex disparity11, though it would be difficult to determine whether this reflects a true difference, given the sampling bias inherent in clinical surveillance. Patients included healthcare workers, farmers, lawyers and students, demonstrating the broad reach of the 2018 surge.

, including fever, headache, malaise and general weakness, often indistinguishable from malaria or common viral diseases. Case fatality rates, though challenging to determine, are estimated at 15-20% among hospitalized cases11, though a recent study estimated case fatality rates in Nigeria during 2015-2016 to be 60%30. Table 1 Demographic data and symptoms as reported for 14 patients whose virus was sequenced at ACEGID in 2018. ID Age/Sex State Symptom onset Sample Collection Symptoms Outcome Genbank # 0026 32y M Edo 2017-12-29 2018-01-07 Fever, headache, weakness Recovered MH157043, MH157046 0097 44y M Ondo 2018-01-08 2018-01-15 Fever, abdominal pain, sore throat, weakness Recovered MH157049, MH157035 0541 18y F Edo 2018-01-30 2018-02-01 Fever, headache, abdominal pain Recovered MH157048, MH157044 0611 41y F Ebonyi 2018-02-02 Fever, headache, unspecified bleeding MH157039 0664 20y F Ondo 2018-02-04 Fever, abdominal pain MH157053, MH157028 0959 32y M Edo 2018-02-03 2018-02-12 Fever, vomiting, diarrhea, haematuria, weakness Died MH157042, MH157032 0998 32y M Edo 2018-02-05 2018-02-13 Fever, abdominal pain, sore throat, cough, weakness Recovered MH157030 1024 25y M Edo 2018-02-01 2018-02-14 Fever, headache, cough, general body pain, weakness Recovered MH157047, MH157037 1079 43y M Ondo 2018-02-07 2018-02-15 Fever, headache, abdominal pain, vomiting, diarrhea, bleeding, sore throat, weakness Recovered MH157029, MH157038 1177 33y M Edo 2018-02-04 2018-02-18 Fever, weakness, abdominal pain, sore throat, haematemesis Died MH157036, MH157034 1375 48y M Ondo 2018-02-16 2018-02-23 Fever, abdominal pain, headache, sore throat, vomiting, diarrhea, weakness Died MH157033, MH157045 1381 30y F Kogi 2018-02-08 2018-02-23 Fever, abdominal pain, headache, sore throat, diarrhea, haematemesis Recovered MH157040, MH157041 1392 14y F Edo 2018-02-16 2018-02-24 Fever, vomiting, cough, haematuria Recovered MH157051, MH157052 1643 27y M Edo 2018-02-25 2018-03-05 Fever, headache, sore throat Recovered MH157031, MH157050 To look for evidence of a novel viral genetic variant or sustained human-to-human transmission driving the 2018 case surge, we performed phylogenetic analysis of these 14 genomes from 2018. A maximum likelihood phylogeny shows that the 2018 genomes fall within previously known Lassa virus diversity in Nigeria (Fig. 2A) and do not display substantial clustering by date of sampling, consistent with multiple zoonotic transmissions.

nclusions that Lassa fever does not exhibit sex disparity11, though it would be difficult to determine whether this reflects a true difference, given the sampling bias inherent in clinical surveillance. Patients included healthcare workers, farmers, lawyers and students, demonstrating the broad reach of the 2018 surge. Figure 1: Incidence of Lassa virus in Nigeria in recent years. a) Number of clinically suspected Lassa fever cases (blue) and RT-qPCR-positive cases (orange) tested at ISTH monthly from January 2012 to February 2018. Counts are those reported by ISTH. Gray shading denotes dry season months in Nigeria, when Lassa cases are typically highest. b) Samples processed at ISTH from January 1 to March 13, 2018. Outcome data, where available, are up to date as of March 22. Lassa virus sequencing of patient samples from 2018 surge To investigate the viral population underpinning this surge, we performed unbiased sequencing and assembled Lassa virus genomes on a subset of RT-qPCR-positive patient samples (Fig. 1B). We obtained complete or high-quality partial Lassa virus genomes from 14 out of 26 RTqPCR- positive patient samples. Table S1 summarizes sequence and assembly quality metrics for these samples. The mean unambiguous assembly length of these genomes was 9,039 bases (4,450-10,610) and mean coverage depth was 193x (1-1,834). 12 samples did not readily produce high-quality Lassa virus genomes. We did not find evidence consistent with other pathogenic viral infections in any of the samples from 2018, with the depth of sequencing available.

mbly length of these genomes was 9,039 bases (4,450-10,610) and mean coverage depth was 193x (1-1,834). 12 samples did not readily produce high-quality Lassa virus genomes. We did not find evidence consistent with other pathogenic viral infections in any of the samples from 2018, with the depth of sequencing available. The 14 patients from whom we assembled Lassa virus genomes were reflective of the demographic characteristics of the larger cohort, including age (Fig. S1A), sex (Table 1) and geographic distribution (Fig. S1B). Clinically, the picture is of a nonspecific febrile illness that sometimes develops into a bleeding diathesis. Hemorrhage was documented in 2 of the 3 patients who died and in at least 3 of the 9 who recovered, suggesting a range of disease severity29. This is broadly consistent with clinical descriptions of Lassa fever: patients typically present with nonspecific symptoms, including fever, headache, malaise and general weakness, often indistinguishable from malaria or common viral diseases. Case fatality rates, though challenging to determine, are estimated at 15-20% among hospitalized cases11, though a recent study estimated case fatality rates in Nigeria during 2015-2016 to be 60%30.

INTRODUCTION Lassa fever is a viral hemorrhagic disease endemic to parts of Western Africa that causes over 300,000 cases and 3,000 fatalities per year1. It has been recognized by the World Health Organization (WHO) and the Coalition for Epidemic Preparedness Innovations (CEPI) as a significant threat to global health and in need of urgent R&D attention2-4. Despite the burden of Lassa virus, there is currently no approved vaccine, and the only available pharmacologic therapy is early intravenous administration of the antiviral ribavirin5-7. In early 2018 there was a marked increase in Lassa fever cases in Nigeria: by early March, Nigeria had more confirmed cases (394) than in any previous year. Confirmed cases were observed in 19 Nigerian states, with an estimated case fatality rate of approximately 25%8. The factors underlying this increase were not known, raising concern among public health officials that something had fundamentally changed about this endemic disease.

cases (394) than in any previous year. Confirmed cases were observed in 19 Nigerian states, with an estimated case fatality rate of approximately 25%8. The factors underlying this increase were not known, raising concern among public health officials that something had fundamentally changed about this endemic disease. In a presumed Lassa fever outbreak, genomic analysis of contemporaneous Lassa virus in samples from infected patients can complement conventional epidemiological data by determining whether changes to intrinsic properties of the virus explain the increase in cases. In particular, viral genomic analysis can rapidly assess whether a novel variant or specific viral lineage, or a change in viral transmission route is associated with the case surge. Most human Lassa virus infections result from contact with infected Mastomys natalensis (the major natural reservoir9) or their excreta, but human-to-human transmission has been documented in hospital settings and is a focus of public health monitoring10,11. Previous retrospective investigation of the genomic epidemiology of Lassa virus in Nigeria between 2008 and 2014 showed extensive genetic diversity across the region and provided support for predominantly reservoir-to-human transmission12. Subsequent studies have extended the known genetic diversity of Lassa virus, of which there are at least four firmly established lineages13, as well as its geographic range in Western Africa14,15. Against this backdrop, genomic analysis of Lassa virus during the 2018 can quickly establish changes in the viral genome associated with period of increased Lassa fever cases.

iversity of Lassa virus, of which there are at least four firmly established lineages13, as well as its geographic range in Western Africa14,15. Against this backdrop, genomic analysis of Lassa virus during the 2018 can quickly establish changes in the viral genome associated with period of increased Lassa fever cases. Here we report near real-time genome analysis of Lassa virus from patients from January to March 2018, undertaken at the African Center of Excellence for Genomics of Infectious Disease (ACEGID), at Redeemer’s University in Nigeria. These data provide important genomic context to the recent Lassa fever surge and further resolve the geographic structure of the endemic Lassa virus population across Nigeria.

to March 2018, undertaken at the African Center of Excellence for Genomics of Infectious Disease (ACEGID), at Redeemer’s University in Nigeria. These data provide important genomic context to the recent Lassa fever surge and further resolve the geographic structure of the endemic Lassa virus population across Nigeria. METHODS Patient sample collection We obtained patient samples through a study evaluated and approved by Institutional Review Boards (IRBs) at Irrua Specialist Teaching Hospital (ISTH, Irrua, Nigeria), Redeemer’s University (Ede, Osun State, Nigeria), and Harvard University (Cambridge, Massachusetts). Study staff obtained informed consent from participants enrolled in the research study at ISTH. In addition, some samples were included under a waiver of consent to facilitate rapid public health response as the research involved minimal risk to the subjects. Samples from suspected Lassa fever cases were tested for Lassa virus by RT-qPCR (reverse transcriptase - quantitative polymerase chain reaction) at the clinical diagnostics laboratory at ISTH. We de-identified samples and obtained demographic and clinical data in line with ethical approvals. We prepared a subset of samples with positive Lassa virus RT-qPCR diagnosis, spanning the time frame of the surge, for sequencing.

- quantitative polymerase chain reaction) at the clinical diagnostics laboratory at ISTH. We de-identified samples and obtained demographic and clinical data in line with ethical approvals. We prepared a subset of samples with positive Lassa virus RT-qPCR diagnosis, spanning the time frame of the surge, for sequencing. Viral sequencing We extracted RNA from patient plasma using the QiAmp viral RNA mini kit (Qiagen) or Pathogen RNA/DNA kit (MagMax) according to the manufacturer’s instructions. We removed contaminating DNA by DNase treatment, synthesized cDNA, and prepared sequencing libraries using the Nextera XT kit (Illumina) as previously described16. We constructed sequencing libraries directly from clinical samples without culture or other intervention. We extracted, prepared, and sequenced samples from 2018 at ACEGID, Redeemer’s University, Ede, Osun State, Nigeria, and those from prior to 2018 at ACEGID or the Broad Institute, Cambridge, MA, USA. We additionally performed replicate sequencing of samples from 2018 at the Broad Institute for intra-host variant detection. We sequenced all samples using Illumina MiSeq and HiSeq 2500 machines with 100 nucleotide paired-end reads.

ria, and those from prior to 2018 at ACEGID or the Broad Institute, Cambridge, MA, USA. We additionally performed replicate sequencing of samples from 2018 at the Broad Institute for intra-host variant detection. We sequenced all samples using Illumina MiSeq and HiSeq 2500 machines with 100 nucleotide paired-end reads. Genomic data analysis We analyzed sequencing data using our publicly available software viral-ngs v1.19.217,18 implemented on the DNAnexus cloud-based platform. Briefly, we demultiplexed individual libraries, removed reads mapping to the human genome and to other known technical contaminants (e.g. sequencing adapters), and filtered the remaining reads against previously published Lassa virus genomes. We performed de novo assembly using Trinity19 and scaffolded contigs against one of three Lassa virus reference genomes (KM821997-8, GU481072-3, KM821772-3), representing the major viral lineages (II, III and IV). We used Kraken v0.10.620 in viral-ngs to identify other viral taxa present in the samples. To do so, we first built a database that encompassed the known diversity of all viruses that infect humans (similar to that described elsewhere21, but without insect species). We searched for viral species detected in the samples with a read count at least 1.5x greater than that of any viral taxon identified in negative control samples and manually investigated any potential hits. We detected intra-host variants in samples from 2018 using V-Phaser 222 implemented in viral-ngs v1.19.2 using default parameters. To do so, we leveraged data from independently prepared replicate sequencing libraries for 13 of the 14 samples.

ed in negative control samples and manually investigated any potential hits. We detected intra-host variants in samples from 2018 using V-Phaser 222 implemented in viral-ngs v1.19.2 using default parameters. To do so, we leveraged data from independently prepared replicate sequencing libraries for 13 of the 14 samples. In order to construct the phylogenetic tree of Lassa virus, we performed a multiple sequence alignment of our new genomes with a set of 193 previously published Lassa virus genomes from Nigeria, Sierra Leone, Liberia, and Côte d’Ivoire12. We performed codon-based multiple sequence alignments of the NP and GPC sequences using MAFFT23. We estimated maximum likelihood phylogenies of concatenated alignments of NP and GPC using IQ-TREE v1.5.524,25 using a GTR substitution model and ultrafast bootstrapping. To create time-aware phylogenies for the Nigerian lineage II sequences, we then performed Bayesian phylogenetic analyses using the program BEAST v1.8.426, incorporating the collection date for each sequence. We included GPC and NP lineage II alignments as separate partitions. We used a model consisting of an SRD06 codon-aware nucleotide substitution model27, an uncorrelated relaxed clock with a lognormal distribution, and a Bayesian SkyGrid coalescent tree prior. All of the Bayesian analyses were run for 200 million MCMC steps, sampling parameters and trees every 5,000 generations. Maximum-clade credibility trees summarizing all MCMC samples were generated using TreeAnnotator v1.8.4 with a burn-in rate of 10%.

rvoir, and not humans, in the ongoing transmission of Lassa virus. Together with the long branch lengths of these groups – suggestive of extensive, uncaptured Lassa virus diversity in these regions – these results indicate sequestering of the rodent population and their associated Lassa virus lineages in these regions. DISCUSSION We undertook genome sequencing of Lassa virus from patient samples to assess whether intrinsic properties of the viral genomes contributed to the recent increase in Lassa fever cases in Nigeria. In our initial dataset of 14 genomes from 2018, we observe no evidence that either a particular viral variant or extensive human-to-human transmission drove the surge. Lassa virus genomes both from 2018 and from 2015-16 were broadly distributed across different Lassa virus lineages, suggesting that no single variant was associated with the recent increase in Lassa fever. Furthermore, we do not observe phylogenetic clustering of Lassa virus genomes from samples collected close in time, as would be expected if this surge were driven by humanto- human transmission. The absence of these patterns supports the assertion that Lassa virus transmission in 2018 was sustained by multiple distinct cross-species transmission events, consistent with previous observations12,13. These findings suggest future studies of the 2018 increase in cases prioritize investigating changes in the rodent reservoir population as well as the role of heightened surveillance and clinical awareness31.

s sustained by multiple distinct cross-species transmission events, consistent with previous observations12,13. These findings suggest future studies of the 2018 increase in cases prioritize investigating changes in the rodent reservoir population as well as the role of heightened surveillance and clinical awareness31. The data reported here also improve our understanding of Lassa virus genetic diversity across Nigeria, revealing clear geographic population structure and extensive diversity in regions that have previously been poorly sampled. Intriguingly, we see substantial genetic divergence between regions demarcated by two major rivers, suggesting the importance of established, local rodent populations in sustaining Lassa virus transmission13. Together, these results reaffirm the need for widespread geographic sampling of Lassa virus in Nigeria, including more extensive sampling from the rodent reservoir, in order to better understand its genetic diversity. A comprehensive knowledge of this diversity is critical for development of urgently needed Lassa fever diagnostics and vaccines2,3.

irm the need for widespread geographic sampling of Lassa virus in Nigeria, including more extensive sampling from the rodent reservoir, in order to better understand its genetic diversity. A comprehensive knowledge of this diversity is critical for development of urgently needed Lassa fever diagnostics and vaccines2,3. The 2018 Lassa fever cases in this study were sequenced locally in Nigeria, leveraging longterm investments to establish local, responsive genomics laboratory capacity. These data were then rapidly shared with key public health organisations, who recognized the value of genomic data to inform case tracking and management. Continued development of local genomics capacity and growth of these collaborations will facilitate a more agile and integrated approach to outbreaks. We envision a model for genomics-informed outbreak investigation in which locally generated sequence data is rapidly integrated with traditional epidemiological data to refine response strategies.

cal genomics capacity and growth of these collaborations will facilitate a more agile and integrated approach to outbreaks. We envision a model for genomics-informed outbreak investigation in which locally generated sequence data is rapidly integrated with traditional epidemiological data to refine response strategies. Supplementary Materials Supplementary Materials ACKNOWLEDGEMENTS This project has been funded in part with Federal funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services, under Grant Numbers U19AI110818 and R01AI114855 to the Broad Institute (P.C.S.), Grant Numbers U19AI115589 and R44AI115754 to Tulane University (R.F.G.), and Grant Number U19AI135995 to The Scripps Research Institute (K.G.A.). Support was also received from the National Human Genome Research Institute, National Institutes of Health, Department of Health and Human Services, under Grant Numbers U01HG007480 and U54HG007480 to Redeemer’s University Nigeria (C.T.H.). This content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This project has also been supported by World Bank project ACE019 to Redeemer’s University Nigeria (C.T.H.), the Henry M Jackson Foundation to Redeemer’s University Nigeria (C.T.H.) and the Broad Institute (P.C.S.), and the Bill and Melinda Gates Foundation to Harvard University (P.C.S.). K.J.S. is supported by a fellowship from the Human Frontier Science Program (LT000553/2016); K.G.B. is supported by a Shope Fellowship from the American Society of Tropical Medicine and Hygiene; K.G.A. is a Pew Biomedical Scholar supported by NIH NCATS CTSA UL1TR001114; P.C.S. is an Investigator supported by the Howard Hughes Medical Institute. All genomic data has been publicly released at NCBI under BioProject PRJNA436552. All 64 genomes reported from 2015-2016 are available in GenBank under accessions MH053463-MH053590 All 14 genomes reported from 2018 are available in GenBank under accessions MH157028- MH157053.

ted by the Howard Hughes Medical Institute. All genomic data has been publicly released at NCBI under BioProject PRJNA436552. All 64 genomes reported from 2015-2016 are available in GenBank under accessions MH053463-MH053590 All 14 genomes reported from 2018 are available in GenBank under accessions MH157028- MH157053. This is an Author Final Manuscript, which is the version after external peer review and before publication in the Journal. The publisher’s version of record, which includes all New England Journal of Medicine editing and enhancements, is available at 10.1056/NEJMoa1804498..