Browse the corpus

What is already known on this topic? Treatment approaches for children with cleft palate differ broadly based on the cleft centre and cleft surgeon delivering care. The effect of these different approaches on clinical outcomes, and in particular the need for a secondary palate surgery, is unclear. What this study hopes to add? Among children with cleft palate, almost half will undergo secondary palate surgery. However, there is greater than four-fold variation among centres in their use of secondary surgery. These results suggest substantial opportunity for cleft centres and surgeons to improve outcomes for children with cleft palate by reducing variation in treatment approach. Introduction Cleft palate occurs in approximately 1 of every 1000 births, making it one of the world’s most common congenital anomalies.1–3 Children undergo cleft palate repair early in life, with the goals of closing the abnormal connection between the mouth and nose and obtaining normal speech. Failure to achieve either goal substantially impacts a child’s quality of life by leaving a fistula that allows liquids and food to exit their nose while eating or by making their speech unintelligible.4 Either complication may lead children to undergo secondary palate surgery. Secondary surgery requires a major operation, several weeks recovery and medical costs similar to the primary repair.5

e by leaving a fistula that allows liquids and food to exit their nose while eating or by making their speech unintelligible.4 Either complication may lead children to undergo secondary palate surgery. Secondary surgery requires a major operation, several weeks recovery and medical costs similar to the primary repair.5 Patients with cleft palate are principally cared for in cleft centres, which are multidisciplinary teams typically associated with a single hospital and multiple cleft surgeons. Cleft centres exhibit broad variation in both their treatment approach6 and their complication rates after palate repair. Fistula incidence ranges from 0% to 35%.7 Speech problems attributable to abnormal palate function range from 10% to 56%.4 8 9 This variation in both fistulae and speech problems may lead to differences in rates of secondary palate surgery. Understanding the variation in secondary surgery along with causative factors for this variation may impact how, where and by whom treatment is delivered. There are many controversial elements of cleft palate care, including use of postoperative antibiotics10 and the importance of hospital volume and provider volume.11 12 Timing of primary cleft palate closure is also controversial: some providers advocate performing repair before 9 months of age to improve speech outcomes,13–15 while others delay palate repair until 12 or 18 months of age when the palate has grown larger and the risk of harming future facial growth may be reduced.16 17

Timing of primary cleft palate closure is also controversial: some providers advocate performing repair before 9 months of age to improve speech outcomes,13–15 while others delay palate repair until 12 or 18 months of age when the palate has grown larger and the risk of harming future facial growth may be reduced.16 17 Previous comparisons of secondary palate surgery among cleft centres in England, Denmark, Sweden and the Netherlands identified significant differences in outcomes among centres, and these findings led to government-mandated changes in cleft surgery delivery.18 A study of four centres in the USA and Canada found the hazard of secondary palate surgery varied sixfold across teams, although these differences were not statistically significant (95% CI 0.76 to 45.65, p=0.057).19 Thus, it remains unknown whether variability in secondary palate surgery seen in Western Europe generalises to cleft centres around the globe. The present study is a large, retrospective analysis of children undergoing cleft palate repair at 43 free-standing children’s hospitals in the USA. We tested our hypotheses that (1) significant variation exists among hospitals in the use of secondary palate surgery and (2) modifiable hospital factors and surgeon factors at the time of primary palate repair, including age at primary palate repair, hospital volume and surgeon volume, use of postoperative antibiotics and duration of postoperative hospitalisation, are associated with subsequent secondary palate surgery.

alate surgery and (2) modifiable hospital factors and surgeon factors at the time of primary palate repair, including age at primary palate repair, hospital volume and surgeon volume, use of postoperative antibiotics and duration of postoperative hospitalisation, are associated with subsequent secondary palate surgery. Patients and methods Study design and data source We performed a retrospective cohort study of children with cleft lip and palate who underwent cleft palate repair at children’s hospitals in the USA contributing to the Pediatric Health Information System (PHIS). PHIS contains detailed administrative data for all inpatient admissions at participating hospitals. Extensive processes ensure data quality and reliability.20 PHIS contains a hospital-specific patient identifier that enables tracking a patient across all admissions at an individual hospital. Forty-three free-standing children’s hospitals participate in PHIS, accounting for 85% of children’s hospital admissions in the USA.21 The Cincinnati Children’s Hospital Medical Center Institutional Review Board reviewed this study and determined it was not human subjects research, as defined by the Common Rule (45CFR46.102[f]), because the dataset was deidentified.

Patients and methods Study design and data source We performed a retrospective cohort study of children with cleft lip and palate who underwent cleft palate repair at children’s hospitals in the USA contributing to the Pediatric Health Information System (PHIS). PHIS contains detailed administrative data for all inpatient admissions at participating hospitals. Extensive processes ensure data quality and reliability.20 PHIS contains a hospital-specific patient identifier that enables tracking a patient across all admissions at an individual hospital. Forty-three free-standing children’s hospitals participate in PHIS, accounting for 85% of children’s hospital admissions in the USA.21 The Cincinnati Children’s Hospital Medical Center Institutional Review Board reviewed this study and determined it was not human subjects research, as defined by the Common Rule (45CFR46.102[f]), because the dataset was deidentified. Study population Children younger than 2 years diagnosed with cleft lip and palate who were discharged between 1 January 1999 and 30 December 2013, after having undergone cleft palate repair were included. Patients with complex chronic conditions that might influence treatment of their cleft palate, including 22q11.2 deletion syndrome, were excluded using previously validated International Classification of Diseases, Ninth Revision (ICD-9) codes.22 Standard care for cleft palate repair includes palate repair before 2 years of age, so patients with initial palate repair after this age were excluded as they may possess additional medical conditions influencing palate repair timing. If a child underwent cleft palate repair at an individual hospital more than once during the study period, only the earliest repair was included; subsequent repairs were considered secondary palate surgeries.

fter this age were excluded as they may possess additional medical conditions influencing palate repair timing. If a child underwent cleft palate repair at an individual hospital more than once during the study period, only the earliest repair was included; subsequent repairs were considered secondary palate surgeries. Study definitions Study patients were identified using the ICD-9 codes indicating cleft lip and palate (749.20–749.25) in any discharge diagnosis field. Among these patients, admissions during which the patient underwent primary palate repair were determined by the presence of the ICD-9 code for correction of cleft palate (27.62). Any subsequent patient encounter at the same hospital that included an ICD-9 code for correction of cleft palate (27.62), revision of cleft palate repair (27.63) or other plastic repair of palate (27.69) was defined as a secondary palate surgery. No information was available on receipt of secondary palate surgery at hospitals other than the patient’s initial treating facility. Outcome The outcome variable was time from primary palate repair until the patient underwent secondary palate surgery. Patients not undergoing secondary surgery during the observation period were censored on 30 December 2013, the last date for which outcome status was available.

Study definitions Study patients were identified using the ICD-9 codes indicating cleft lip and palate (749.20–749.25) in any discharge diagnosis field. Among these patients, admissions during which the patient underwent primary palate repair were determined by the presence of the ICD-9 code for correction of cleft palate (27.62). Any subsequent patient encounter at the same hospital that included an ICD-9 code for correction of cleft palate (27.62), revision of cleft palate repair (27.63) or other plastic repair of palate (27.69) was defined as a secondary palate surgery. No information was available on receipt of secondary palate surgery at hospitals other than the patient’s initial treating facility. Outcome The outcome variable was time from primary palate repair until the patient underwent secondary palate surgery. Patients not undergoing secondary surgery during the observation period were censored on 30 December 2013, the last date for which outcome status was available. Covariates Covariates were defined using information available in PHIS from the patient’s primary palate repair. Demographic data included sex, race and median household income by postcode of residence. Race was included in the final model because prior research suggests variation among racial/ethnic groups in receipt of cleft palate surgery.23 24 Median household income by postcode of patient residence was obtained from 2010 US Census data and split into four categories based on US federal poverty guidelines for a family of four in 2010, as previously described.25

or research suggests variation among racial/ethnic groups in receipt of cleft palate surgery.23 24 Median household income by postcode of patient residence was obtained from 2010 US Census data and split into four categories based on US federal poverty guidelines for a family of four in 2010, as previously described.25 Additional covariates were specified a priori. Age at primary palate repair was categorised into three groups based on existing approaches to timing of primary palate repair: less than 9 months, 9–15 months and 16–24 months.13 26 27 Postoperative antibiotic use was defined as receipt of any antibiotic on the first and/or second day after primary palate repair and was dichotomised as present or absent. We dichotomised hospital length of stay as less than two nights and two or more nights to test the hypothesis that increased length of stay would reduce secondary surgery by enabling improved parent education on postsurgical care.

and/or second day after primary palate repair and was dichotomised as present or absent. We dichotomised hospital length of stay as less than two nights and two or more nights to test the hypothesis that increased length of stay would reduce secondary surgery by enabling improved parent education on postsurgical care. Surgeon and hospital procedure volume was determined for each patient on the day of the patient’s primary palate repair by counting all cleft palate repairs (ICD-9 codes 27.62 and 27.63) performed by that surgeon or hospital, respectively, in the prior 365 days. This approach decreases exposure misclassification compared with annual procedure volume for the same year the procedure was performed.28 Specifically, procedure volume was used as a measure of experience; any procedures performed after a specific patient’s procedure do not contribute to the surgeon’s or hospital’s experience at the time that patient’s procedure was performed. Thus, annual procedure volume for the year the procedure was performed misclassifies experience, and the approach used here reduces misclassification by determining volume using the 365 days prior to each patient’s procedure.

to the surgeon’s or hospital’s experience at the time that patient’s procedure was performed. Thus, annual procedure volume for the year the procedure was performed misclassifies experience, and the approach used here reduces misclassification by determining volume using the 365 days prior to each patient’s procedure. Procedure volume counts included all cleft palate repairs in patients younger than 4 years of age, regardless of cleft type or additional medical conditions, as all palate repairs were predicted to increase the surgical team’s experience with the procedure. For surgeons or hospitals who had reported in PHIS for fewer than 365 days at the time of a patient’s palate repair, procedure volume was calculated as the number of cleft palate repairs performed during their first 365 days of reporting. Procedure volume was categorised into tertiles.

’s experience with the procedure. For surgeons or hospitals who had reported in PHIS for fewer than 365 days at the time of a patient’s palate repair, procedure volume was calculated as the number of cleft palate repairs performed during their first 365 days of reporting. Procedure volume was categorised into tertiles. Statistical analyses We calculated descriptive statistics with means and percentages. We plotted a Kaplan-Meier time-to-event curve for all patients in the cohort.29 Prior to investigating variation in time-to-event among hospitals, we performed a power analysis to ensure we had a power of 0.8 to detect a 50% difference in time to secondary palate surgery among hospitals, with a type I error rate of 0.05, adjusted for pairwise comparisons.30 Assuming administrative censoring would occur and that secondary palate surgery would occur during the observation period for 47% of patients (from descriptive analysis of the data), we determined it was necessary to restrict this subgroup analysis to hospitals with at least 175 patients during the study period. We then plotted Kaplan-Meier time-to-event curves for hospitals meeting this criterion. We tested for variation in time-to-event among hospitals using a Log-rank test that stratified by patient gender, race and median household income by ZIP code of patient residence.

with at least 175 patients during the study period. We then plotted Kaplan-Meier time-to-event curves for hospitals meeting this criterion. We tested for variation in time-to-event among hospitals using a Log-rank test that stratified by patient gender, race and median household income by ZIP code of patient residence. We then fit a three-level mixed-effects parametric survival-time model with clustering of patients within surgeons and clustering of surgeons within hospitals. We assumed a Weibull distribution for this model, after confirming appropriateness of this assumption by visual inspection of log–log plots of survival. Random effects were assumed to have normal distributions with zero means. We tested for all interactions. After fitting the full model, we estimated the variability attributable to the surgeon and hospital. Sensitivity analyses were conducted to evaluate whether the choice of follow-up time imposed bias through right censoring. Sensitivity analyses included: (1) censoring children 2 years after their last encounter; (2) censoring children 4 years after their last encounter; (3) excluding children with less than 1 year of follow-up; and (4) excluding children with less than 4 years of follow-up. Results of all sensitivity analyses were nearly identical to those of our main analysis and are not presented. In a separate sensitivity analysis, we repeated our main analysis while adjusting for year of primary cleft palate repair; results were similar to our primary analysis and are included in online supplementary file 1.

of all sensitivity analyses were nearly identical to those of our main analysis and are not presented. In a separate sensitivity analysis, we repeated our main analysis while adjusting for year of primary cleft palate repair; results were similar to our primary analysis and are included in online supplementary file 1. 10.1136/bmjpo-2017-000063.supp1Supplementary file 1 supplementary file 2 Statistical analyses were performed using Stata V.13 and V.14 (StataCorp, College Station, Texas, USA). Statistical significance was set at p<0.05. Results A total of 5846 children underwent cleft palate repair before 2 years of age at PHIS hospitals. We excluded 907 children with additional complex chronic conditions.22 Our final cohort included 4939 children who underwent primary cleft palate repair between 1999 and 2013. Secondary palate surgery was performed in 1421 children (29%) during the observation period; 71% of the observations (n=3518) were right-censored. Characteristics of the study population are shown in table 1. Table 1 Characteristics of patients and the care delivered at their initial palate repair

Results A total of 5846 children underwent cleft palate repair before 2 years of age at PHIS hospitals. We excluded 907 children with additional complex chronic conditions.22 Our final cohort included 4939 children who underwent primary cleft palate repair between 1999 and 2013. Secondary palate surgery was performed in 1421 children (29%) during the observation period; 71% of the observations (n=3518) were right-censored. Characteristics of the study population are shown in table 1. Table 1 Characteristics of patients and the care delivered at their initial palate repair Characteristic No. (%) Total 4939 Sex Male 3144 (64) Female 1795 (36) Race White 3374 (68) Black 322 (7) Asian or Pacific Islander 342 (7) American Indian 66 (1) Other 589 (12) Not specified 246 (5) Median annual household income of ZIP code $33 525 or less (<1.5 FPL*) 1230 (25) $33 526–$44 700 (1.5–2 FPL) 1555 (32) $44 701–$67 050 (2–3 FPL) 1527 (31) $67 051 or more (>3 FPL) 483 (10) No data available 144 (3) Age at primary palate repair <9 months 1164 (24) 9–15 months 3186 (64) 16–24 months 589 (12) Postoperative antibiotic use None 2537 (51) Yes 2402 (49) Surgeon procedure volume (on day of repair) Low (<10 repairs in prior year) 1307 (27) Medium (10–25) 2343 (47) High (>25) 1289 (26) Hospital procedure volume (on day of repair) Low (<25 repairs in prior year) 1068 (22) Medium (25–50) 2568 (52) High (>50) 1303 (26) Length of stay after surgery ≤1 night 3034 (61) ≥2 nights 1905 (39) *FPL, US Federal Poverty Level for a family of four.

rior year) 1307 (27) Medium (10–25) 2343 (47) High (>25) 1289 (26) Hospital procedure volume (on day of repair) Low (<25 repairs in prior year) 1068 (22) Medium (25–50) 2568 (52) High (>50) 1303 (26) Length of stay after surgery ≤1 night 3034 (61) ≥2 nights 1905 (39) *FPL, US Federal Poverty Level for a family of four. Time to secondary surgery for the cohort is shown in figure 1A. At 3 years after primary palate repair, 18% of patients had undergone secondary palate surgery. At 5 years after primary palate repair, 25% of patients had undergone secondary palate surgery. At 10 years after primary palate repair, 44% of patients had undergone secondary palate surgery. Figure 1 Kaplan-Meier curves for time until secondary palate surgery. (A) Time to secondary surgery for all patients in the study. (B) Time to secondary surgery by hospital for hospitals with >175 patients undergoing palate repair during the observation period. At-risk table and censoring times for each hospital are shown in figure 3 online supplementary file 3. Log-rank test is stratified by patient gender, race and median household income for ZIP code of residence. 10.1136/bmjpo-2017-000063.supp3Supplementary file 3 supplementary file 1

Figure 1 Kaplan-Meier curves for time until secondary palate surgery. (A) Time to secondary surgery for all patients in the study. (B) Time to secondary surgery by hospital for hospitals with >175 patients undergoing palate repair during the observation period. At-risk table and censoring times for each hospital are shown in figure 3 online supplementary file 3. Log-rank test is stratified by patient gender, race and median household income for ZIP code of residence. 10.1136/bmjpo-2017-000063.supp3Supplementary file 3 supplementary file 1 Variation among hospitals Figure 1B displays the variability in time to secondary surgery among the nine hospitals with at least 175 patients. Significant differences in time to secondary palate surgery existed across hospitals (p<0.001, stratified Log-rank test). At 10 years after primary palate repair, the proportion of children who had undergone secondary palate surgery ranged from 9% at hospital H to 77% at hospital A.

ospitals with at least 175 patients. Significant differences in time to secondary palate surgery existed across hospitals (p<0.001, stratified Log-rank test). At 10 years after primary palate repair, the proportion of children who had undergone secondary palate surgery ranged from 9% at hospital H to 77% at hospital A. Hospital-specific and surgeon-specific predictors Using a mixed-effects time-to-event model, we investigated the patient and surgical factors associated with time to secondary surgery (table 2). This analysis included all PHIS hospitals. After adjustment for patient demographics (sex, race and household income level), postoperative antibiotic use, surgeon procedure volume, hospital procedure volume and length of stay after surgery were not associated with time to secondary surgery. Age at primary palate repair was associated with time to secondary palate surgery, with the hazard of secondary surgery increased for children who underwent primary repair before 9 months of age (p<0.001). Table 2 Adjusted HRs for secondary palate surgery

Hospital-specific and surgeon-specific predictors Using a mixed-effects time-to-event model, we investigated the patient and surgical factors associated with time to secondary surgery (table 2). This analysis included all PHIS hospitals. After adjustment for patient demographics (sex, race and household income level), postoperative antibiotic use, surgeon procedure volume, hospital procedure volume and length of stay after surgery were not associated with time to secondary surgery. Age at primary palate repair was associated with time to secondary palate surgery, with the hazard of secondary surgery increased for children who underwent primary repair before 9 months of age (p<0.001). Table 2 Adjusted HRs for secondary palate surgery Risk factor Secondary palate surgery* HR (95% CI) p Value Sex 0.70 Male 0.98 (0.87 to 1.10) Female Reference Race 0.10 White Reference Black 0.75 (0.58 to 0.97) Asian or Pacific Islander 1.01 (0.77 to 1.32) American Indian 1.09 (0.59 to 2.02) Other 0.97 (0.81 to 1.18) Not specified 1.28 (0.99 to 1.65) Median annual household income of ZIP code 0.16 $33 525 or less (<1.5 FPL†) Reference $33 526–$44 700 (1.5–2 FPL) 0.95 (0.82 to 1.10) $44 701–$67 050 (2–3 FPL) 0.84 (0.72 to 0.98) $67 051 or more (>3 FPL) 0.91 (0.74 to 1.14) Age at primary palate repair <0.001 <9 months‡ At baseline 6.74 (5.20 to 8.73) At 1 year after repair 4.70 (3.44 to 6.43) At 5 years after repair 1.11 (0.66 to 1.89) 9–15 months 1.15 (0.94 to 1.42) 16–24 months Reference Postoperative antibiotic use 0.06 None Reference Yes 0.86 (0.74 to 1.01) Surgeon procedure volume (on day of repair) 0.17 Low (<10 repairs in prior year) Reference Medium (10–25) 1.11 (0.94 to 1.31) High (>25) 1.25 (0.99 to 1.58) Hospital procedure volume (on day of repair) 0.14 Low (<25 repairs in prior year) Reference Medium (25–50) 0.94 (0.77 to 1.14) High (>50) 0.78 (0.60 to 1.02) Length of stay after surgery 0.33 ≤1 night 1.07 (0.94 to 1.22) ≥2 nights Reference Model assumes clustering of patients within surgeons and clustering of surgeons within hospitals; p<0.001 for likelihood-ratio tests of theta=0 for both surgeon and hospital.

ium (25–50) 0.94 (0.77 to 1.14) High (>50) 0.78 (0.60 to 1.02) Length of stay after surgery 0.33 ≤1 night 1.07 (0.94 to 1.22) ≥2 nights Reference Model assumes clustering of patients within surgeons and clustering of surgeons within hospitals; p<0.001 for likelihood-ratio tests of theta=0 for both surgeon and hospital. †FPL, US Federal Poverty Level for a family of four. ‡Age less than 9 months at primary repair is a time varying covariate, with baseline HR 6.74 (5.20−8.73) that decreases by 30% (26–34%) each subsequent year.

ium (25–50) 0.94 (0.77 to 1.14) High (>50) 0.78 (0.60 to 1.02) Length of stay after surgery 0.33 ≤1 night 1.07 (0.94 to 1.22) ≥2 nights Reference Model assumes clustering of patients within surgeons and clustering of surgeons within hospitals; p<0.001 for likelihood-ratio tests of theta=0 for both surgeon and hospital. †FPL, US Federal Poverty Level for a family of four. ‡Age less than 9 months at primary repair is a time varying covariate, with baseline HR 6.74 (5.20−8.73) that decreases by 30% (26–34%) each subsequent year. As shown in figure 2, for children when underwent primary palate repair before 9 months of age, the hazard of secondary surgery was greatest immediately following the primary repair and diminished as the time since primary repair increased. Immediately following primary repair, children who had repair before 9 months of age had a 6.74-fold increased hazard of secondary surgery (95% CI 5.20 to 8.73) compared with children who underwent repair at 16–24 months of age. For children who did not undergo secondary repair during the first year after palate repair, their hazard of secondary surgery diminished slightly, with a HR of 4.70 (95% CI 3.44 to 6.43). For children who reached the fifth anniversary of their palate repair without undergoing secondary surgery, their hazard of secondary palate surgery at any time in the future was similar to children who had primary repair at 9–24 months of age. Among children who did not undergo secondary surgery, the duration of follow-up was greatest for children undergoing repair before 9 months of age (p<0.001, online supplementary file 2).

ir hazard of secondary palate surgery at any time in the future was similar to children who had primary repair at 9–24 months of age. Among children who did not undergo secondary surgery, the duration of follow-up was greatest for children undergoing repair before 9 months of age (p<0.001, online supplementary file 2). 10.1136/bmjpo-2017-000063.supp2Supplementary file 2 supplementary file 3 Figure 2 Kaplan-Meier curve for time until secondary surgery based on age at primary palate repair. This figure demonstrates the time-dependent hazard of secondary surgery. For children who underwent primary palate repair before 9 months of age, the hazard of secondary surgery lies principally in the first 2 years after primary repair. From the mixed-effects model, we estimated the variation attributable to hospitals and surgeons. Between-hospital differences accounted for 59% of this variation (p<0.001), while between-surgeon differences accounted for 41% (p<0.001).

Figure 2 Kaplan-Meier curve for time until secondary surgery based on age at primary palate repair. This figure demonstrates the time-dependent hazard of secondary surgery. For children who underwent primary palate repair before 9 months of age, the hazard of secondary surgery lies principally in the first 2 years after primary repair. From the mixed-effects model, we estimated the variation attributable to hospitals and surgeons. Between-hospital differences accounted for 59% of this variation (p<0.001), while between-surgeon differences accounted for 41% (p<0.001). Discussion We found substantial variability in secondary surgery for children with cleft palate treated at children’s hospitals in the USA. Ten years after primary palate repair, 44% of children underwent secondary surgery, but this varied from 9% to 77% among hospitals. Performing primary palate repair before 9 months of age was associated with a significantly increased hazard of secondary surgery. After adjusting for patient demographics, procedure volume, antibiotic use and timing of primary palate repair, the remaining variability in outcome was attributable to both between-hospital and between-surgeon differences. These results suggest there is a substantial opportunity for hospitals and surgeons to reduce the need for secondary palate surgery in children with cleft palate by reducing variation in treatment approach.

, the remaining variability in outcome was attributable to both between-hospital and between-surgeon differences. These results suggest there is a substantial opportunity for hospitals and surgeons to reduce the need for secondary palate surgery in children with cleft palate by reducing variation in treatment approach. These findings are consistent with and extend prior reports by confirming that broad differences in secondary surgery among cleft centres in Western Europe also occur in the USA. During the 1990s, investigators found differences in both clinical outcomes and use of secondary surgery among cleft centres in Western Europe.12 18 31–34 Cleft centres in the UK achieved the worst outcomes, and this led them to consolidate centres and impose statutory mandates for outcome reporting from each centre.35 Small intercentre studies in the USA found similar differences in clinical outcomes and use of secondary surgery across cleft centres.19 36–38 However, our study is the first to examine whether differences among smaller groups of North American cleft centres generalise to cleft teams in the USA. Although the best achievable rate of secondary surgery is not known and may vary among sites, the variation from 9% to 77% observed in this study is substantial. The literature does indicate that fistula rates below 10% are readily achievable, as are rates of secondary surgery for speech disorders below 20%.7 39 40 The results of this study suggest that many US hospitals achieve these results, and there is an opportunity to learn from these hospitals.

this study is substantial. The literature does indicate that fistula rates below 10% are readily achievable, as are rates of secondary surgery for speech disorders below 20%.7 39 40 The results of this study suggest that many US hospitals achieve these results, and there is an opportunity to learn from these hospitals. Age at cleft palate repair While there is universal agreement that ‘primary palatal repair should be done at the age that allows optimal speech development and facial growth’,27 there is little agreement about precisely what that age should be.13–17 27 Currently, the debate is between primary repair at 6 versus 12 months of age.41 Dorf and Curtin13 showed that children who undergo palate repair after 12 months of age exhibit more compensatory articulations, which require speech therapy to correct. This result has been confirmed by several others, yet none of these studies examined rates of secondary surgery for both speech and fistulae.15 26 Our results show the hazard of secondary surgery is initially 6.7-fold higher for children who have palate repair before 9 months of age and that this increased hazard persists for 5 years after surgery. We do not believe this increased hazard with early palate repair is due to underlying patient differences, as complicating medical or socioeconomic conditions lead to repair at older rather than younger ages in our experience. The Timing of Primary Surgery for Cleft Palate trial is an ongoing randomised control trial comparing palate repair at 6 and 12 months of age and results should be available by 2021. Until then, our study suggest that surgeons and hospitals should be cautious to recommend cleft palate repair before 9 months of age and do so only after reviewing their own results.

e trial is an ongoing randomised control trial comparing palate repair at 6 and 12 months of age and results should be available by 2021. Until then, our study suggest that surgeons and hospitals should be cautious to recommend cleft palate repair before 9 months of age and do so only after reviewing their own results. The volume–outcome relationship Our study failed to show an association between hospital volume and surgeon volume of cleft palate repairs and time until secondary palate surgery. A previous study of cleft palate surgery volume found surgeons who performed ≥3 palate repairs per year achieved superior speech outcomes to surgeons performing <3 repairs per year.11 While three repairs per year may be an important threshold, in the present study, only 6% of repairs were performed by surgeons below that threshold. We chose to set the threshold for low volume surgeons at <10 repairs per year and found that performing 10 or more palate repairs per year did not improve surgical outcomes. This is consistent with prior studies in paediatric surgery that suggested the effects of increased volume are dependent on the specific procedure, outcome and method of characterising volume.28

geons at <10 repairs per year and found that performing 10 or more palate repairs per year did not improve surgical outcomes. This is consistent with prior studies in paediatric surgery that suggested the effects of increased volume are dependent on the specific procedure, outcome and method of characterising volume.28 Limitations These data must be interpreted in the context of the study design. Misclassification of diagnosis or procedure could bias patient selection, although the direction of this bias is difficult to assess. Referral of more complex patients to specific hospitals or surgeons within the study cohort could explain between-hospital and between-surgeon variation, although that is unlikely given the non-overlapping referral patterns of most participating hospitals.25 Children may have had additional medical conditions that influenced timing of primary palate repair; we excluded children with complex chronic conditions,22 but we cannot eliminate the possibility of confounding by other medical conditions. We may have overestimated time to secondary surgery for censored individuals, as these children may have received secondary palate surgery at an institution other than their initial treating hospitals and this would not be in the dataset. Extent of variation among hospitals observed in figure 1b may not generalise to all children’s hospitals; the hospitals included in this figure were selected based on the large number of children undergoing palate repair at these hospitals during the observation period and may be outliers among all children’s hospitals.

xtent of variation among hospitals observed in figure 1b may not generalise to all children’s hospitals; the hospitals included in this figure were selected based on the large number of children undergoing palate repair at these hospitals during the observation period and may be outliers among all children’s hospitals. Conclusions This large retrospective multicentre study demonstrated substantial variation in the hazard of secondary palate surgery for children with cleft lip and palate. Performing primary palate repair before 9 months of age substantially increases the hazard of secondary surgery, and choosing to perform palate repair after this age may be one approach to lowering rates of secondary surgery. Additional research is needed to identify other factors contributing to variation in palate surgery outcomes, along with testing of evidence-based interventions to decrease rates of secondary surgery. At present, these results suggest that cleft centres may be able to improve outcomes for their patients by adopting treatment practices from the best-performing centres. Performing cleft palate repair before 9 months of age may increase the risk of secondary procedures, unless the surgeon and centre have demonstrated successful long-terms outcomes.

est that cleft centres may be able to improve outcomes for their patients by adopting treatment practices from the best-performing centres. Performing cleft palate repair before 9 months of age may increase the risk of secondary procedures, unless the surgeon and centre have demonstrated successful long-terms outcomes. Previous presentations: an earlier version of this paper was presented as a poster at the Annual Meeting of the Paediatric Academic Societies, San Diego, California, April 2015, and as an oral presentation at the Annual Meeting of the American Cleft Palate—Craniofacial Association, Palm Springs, California, April 2015. Other substantial contributions to the work: the authors would like to acknowledge the William Shaw for performing critical review of the manuscript. Contributors: TJS conceptualised and designed the study, carried out the initial analyses and interpretation of data, drafted the initial manuscript and approved the final manuscript as submitted. MH, ACC, HN and MTB designed the study, interpreted results of the analyses, reviewed and revised the manuscript and approved the final manuscript as submitted. Funding: TJS received support from the National Institute of Dental and Craniofacial Research of the National Institutes of Health (K23 DE025023). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. No other external funding was provided for this manuscript.

the National Institute of Dental and Craniofacial Research of the National Institutes of Health (K23 DE025023). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. No other external funding was provided for this manuscript. Competing interests: MTB reports personal fees and from the American Board of Pediatrics Research Advisory Committee, outside the submitted work; all other authors report no other relationships or activities that could appear to have influenced the submitted work. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Constipation and soiling are common in childhood, but little is known about factors that increase the risk of these problems persisting at primary school age. The majority of earlier studies examining risk factors for childhood constipation and soiling are cross-sectional, and many are based on small/clinic samples. No prospective cohort studies have examined the association between risk factors in early childhood and different trajectories of constipation and soiling at primary school age. What this study hopes to add? This study finds evidence that risk factors in early childhood are differentially associated with different trajectories of childhood constipation and soiling. Contrary to common beliefs of clinicians, we found that among children with soiling those with soiling alone (ie, without constipation) outnumbered those with associated constipation. Experiencing hard stools in early childhood is a risk factor for constipation; developmental delay is a risk factor for soiling alone and constipation with soiling.

Contrary to common beliefs of clinicians, we found that among children with soiling those with soiling alone (ie, without constipation) outnumbered those with associated constipation. Experiencing hard stools in early childhood is a risk factor for constipation; developmental delay is a risk factor for soiling alone and constipation with soiling. Introduction Bowel problems are common in childhood and have a considerable impact on quality of life.1 It is believed that 80% of faecal incontinence is due to overflow from chronic constipation, while 20% have no constipation (functional non-retentive faecal incontinence).2 The Rome-IV definition for functional constipation at developmental age ≥4 years requires at least two of six symptoms (two or fewer defecations in the toilet per week; at least one episode of faecal incontinence per week; history of retentive posturing/stool retention; history of painful or hard stools; a large faecal mass in the rectum; large diameter stools that can obstruct the toilet) present once a week or more for at least 1 month.3 Rome-IV diagnostic criteria are also available for functional constipation in children under 4 years.4 The diagnostic criteria for functional non-retentive faecal incontinence are inappropriate defecation; no medical condition for symptoms and no retention (criteria should be met for at least 1 month).3 Other clinical definitions are sometimes used.5 6 Epidemiological studies of the prevalence of constipation and soiling vary probably because of different definitions. A systematic review reported the median prevalence of constipation in children aged 0–18 years to be 8.9%, with similar prevalence in boys and girls;5 however, more recent findings suggest a higher proportion of constipation in girls.6 The prevalence of childhood soiling is between 1% and 4% and is consistently found to be two to four times more common in boys.7 8 A recent large cross-sectional study of children aged 5–13 years reported that 7.8% (9.8% boys, 5.8% girls) experienced faecal incontinence.9 Only one epidemiological survey, of children aged 10–16 years in Sri Lanka, differentiated between soiling with and without constipation and reported that 2.0% experienced faecal incontinence and 18% of those did not have constipation.10

years reported that 7.8% (9.8% boys, 5.8% girls) experienced faecal incontinence.9 Only one epidemiological survey, of children aged 10–16 years in Sri Lanka, differentiated between soiling with and without constipation and reported that 2.0% experienced faecal incontinence and 18% of those did not have constipation.10 Early identification of children at risk of constipation and soiling could lead to timely interventions to reduce the adverse impacts on quality of life and psychosocial development. Clinicians believe that pain of passing hard stools in infancy and early childhood is the principal contributing factor for acute childhood constipation,11 leading to chronic constipation which causes soiling.11 12 Hard stools lead to withholding and toileting refusal,13 retaining a stool mass and increasing the difficulty of evacuating. Breastfed infants produce softer stools,14 and those breastfed for <6 months may develop constipation more commonly.15 Other risk factors include lower levels of parental education,9 16 income9 and socioeconomic status,8 10 low birth weight and prematurity17 and developmental delay.18 Timing of toilet training has also been investigated but findings are inconsistent.7 19 20 Only one study specified whether constipation occurred with or without soiling.11 Very little is known about risk factors for soiling without constipation. Finally, most earlier studies of risk factors for constipation and soiling are cross-sectional which makes the timing of events more difficult to determine.

7 19 20 Only one study specified whether constipation occurred with or without soiling.11 Very little is known about risk factors for soiling without constipation. Finally, most earlier studies of risk factors for constipation and soiling are cross-sectional which makes the timing of events more difficult to determine. Although most children achieve bowel control by 3–5 years,19 21 there is recent evidence for different patterns of development of bowel control.22 These ‘developmental trajectories’ distinguish children with normative development (89.0%), delayed attainment (4.1%), persistent soiling (2.7%) and relapses in soiling (4.1%).22 Describing developmental trajectories of soiling alone does not allow the determination of whether soiling is occurring with or without constipation. The aims of this paper are twofold: first, we extend previous work using data from the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort to determine the degree of comorbidity between constipation and soiling in childhood, and second, we examine the association between risk factors in early childhood and trajectories of constipation and soiling at primary school age.

ing data from the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort to determine the degree of comorbidity between constipation and soiling in childhood, and second, we examine the association between risk factors in early childhood and trajectories of constipation and soiling at primary school age. Methods Participants The sample comprised participants from the ALSPAC. Detailed information is available at http://www.bristol.ac.uk/alspac, including a fully searchable data dictionary http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary. Pregnant women resident in the former Avon Health Authority in southwest England, having an estimated date of delivery between 1 April 1991 and 31 December 1992 were invited to take part, resulting in a cohort of 14 541 pregnancies.23 Of the 13 978 singletons/twins alive at 1 year, 24 participants withdrew consent, leaving a starting sample of 13 954.

rmer Avon Health Authority in southwest England, having an estimated date of delivery between 1 April 1991 and 31 December 1992 were invited to take part, resulting in a cohort of 14 541 pregnancies.23 Of the 13 978 singletons/twins alive at 1 year, 24 participants withdrew consent, leaving a starting sample of 13 954. Soiling and constipation in mid-childhood When children were 4½, 5½, 6½, 7½ and 9½ years, their parents were asked, ‘How often usually does your child dirty his/her pants during the day?’ with options: ‘never’, ‘occasional accident but less than once/week’, ‘about once/week’, ‘2–5 times/week’, ‘nearly every day’ and ‘more than once a day’. Responses were collapsed (‘never’ vs all other responses) to indicate presence of soiling at each age. Parents were also asked about their child’s constipation across a similar age range: ‘Has he/she had any constipation in the past 12 months?’ with options: ‘Yes, and saw a doctor’; ‘Yes, but did not see the doctor’ and ‘No, did not have’. Responses were collapsed (yes with/without consultation vs no) to indicate presence of constipation at each age.

heir child’s constipation across a similar age range: ‘Has he/she had any constipation in the past 12 months?’ with options: ‘Yes, and saw a doctor’; ‘Yes, but did not see the doctor’ and ‘No, did not have’. Responses were collapsed (yes with/without consultation vs no) to indicate presence of constipation at each age. Risk factors in early childhood Potential risk factors were identified from the literature. Data were obtained from questionnaires completed by mothers and included the child’s stool consistency at 2½ years, breast feeding during the first six months, family socioeconomic position assessed during the antenatal period (parental social class, maternal educational attainment) or at 33 months (material hardship, home ownership and car access), length of gestation and birth weight, developmental level24 and age at initiation of toilet training (assessed when the child was 6, 15 and 24 months).

sition assessed during the antenatal period (parental social class, maternal educational attainment) or at 33 months (material hardship, home ownership and car access), length of gestation and birth weight, developmental level24 and age at initiation of toilet training (assessed when the child was 6, 15 and 24 months). Statistical modelling We have previously used longitudinal Latent Class Analysis (LLCA) to derive developmental trajectories of soiling22 where we showed that patterns of soiling at 4–9 years could be adequately explained by a four-class solution. However, such a model ignores the comorbidity with constipation. Therefore, in the current study we (i) estimated a similar latent class model of constipation and (ii) combined this with soiling in a parallel model to describe changes in both problems (using Mplus V.7.1125). Further details regarding the estimation can be found in the online supplementary appendix. In addition, as various sample sizes have been used for the different analytical steps online supplementary figure 1 shows a flow chart describing how each sample was obtained. 10.1136/bmjpo-2017-000230.supp1Supplementary file 1

Statistical modelling We have previously used longitudinal Latent Class Analysis (LLCA) to derive developmental trajectories of soiling22 where we showed that patterns of soiling at 4–9 years could be adequately explained by a four-class solution. However, such a model ignores the comorbidity with constipation. Therefore, in the current study we (i) estimated a similar latent class model of constipation and (ii) combined this with soiling in a parallel model to describe changes in both problems (using Mplus V.7.1125). Further details regarding the estimation can be found in the online supplementary appendix. In addition, as various sample sizes have been used for the different analytical steps online supplementary figure 1 shows a flow chart describing how each sample was obtained. 10.1136/bmjpo-2017-000230.supp1Supplementary file 1 Once the optimal model had been established, we examined the association between the risks (above) and constipation/soiling class membership using multinomial logistic regression. Coefficients and SEs were estimated using the bias-adjusted three-step approach26 27 which has been shown to reduce the bias inherent in such regression models.26 For most risks, we report univariable estimates of association; however for toilet training initiation, we considered the potential confounding effects of gender, social class, early parenthood, mother’s education, housing adequacy, major financial difficulties, family size, social network/support, developmental delay and maternal depression.

t univariable estimates of association; however for toilet training initiation, we considered the potential confounding effects of gender, social class, early parenthood, mother’s education, housing adequacy, major financial difficulties, family size, social network/support, developmental delay and maternal depression. Results Table 1 shows rates of soiling and constipation at each age. While rates of reported constipation decrease steadily, there is a more complex pattern for soiling. Of those reporting constipation, the proportion who saw a doctor decreased from 19.7% to 11.6% (data not shown). Table 1 Prevalence of soiling and constipation in the Avon Longitudinal Study of Parents and Children study

Results Table 1 shows rates of soiling and constipation at each age. While rates of reported constipation decrease steadily, there is a more complex pattern for soiling. Of those reporting constipation, the proportion who saw a doctor decreased from 19.7% to 11.6% (data not shown). Table 1 Prevalence of soiling and constipation in the Avon Longitudinal Study of Parents and Children study 4½ years 5½ years 6½ years 7½ years 9½ years Subset of dataset containing at least one non-missing time points for both soiling and constipation (max n=10 450) No soiling 8720 (92.8%) 8284 (93.8%) 7752 (92.2%) 7617 (93.1%) 7288 (94.8%) Soiling 673 (7.2%) 548 (6.2%) 654 (7.8%) 565 (6.9%) 397 (5.2%) 4 years 9 months 5 years 9 months 6 years 9months 7 years 7 months 8 years 7 months 10 years 8 months No constipation 7947 (85.6%) 7622 (89.6%) 7580 (89.7%) 7346 (89.9%) 7062 (90.2%) 6631 (90.4%) Constipation 1342 (14.4%) 881 (10.4%) 873 (10.3%) 822 (10.1%) 765 (9.8%) 707 (9.6%) Subset of dataset containing at least three non-missing time points for both soiling and constipation (max n=8435) No soiling 7507 (92.8%) 7518 (94.0%) 7268 (92.2%) 7153 (93.2%) 6801 (94.8%) Soiling 585 (7.2%) 477 (6.0%) 611 (7.8%) 525 (6.8%) 372 (5.2%) 4 years 9 months 5 years 9 months 6 years 9 months 7 years 7 months 8 years 7 months 10 years 8 months No constipation 6893 (85.5%) 7014 (89.7%) 7086 (89.6%) 6894 (90.0%) 6549 (90.5%) 6194 (90.6%) Constipation 1173 (14.5%) 809 (10.3%) 825 (10.4%) 766 (10.0%) 689 (9.5%) 644 (9.4%) Sample with complete data on all eleven measures (n=4931) No soiling 4575 (92.8%) 4638 (94.1%) 4568 (92.6%) 4590 (93.1%) 4670 (94.7%) Soiling 356 (7.2%) 293 (5.9%) 363 (7.4%) 341 (6.9%) 261 (5.3%) 4 years 9 months 5 years 9 months 6 years 9 months 7 years 7 months 8 years 7 months 10 years 8 months No constipation 4244 (86.1%) 4425 (89.7%) 4402 (89.3%) 4426 (89.8%) 4454 (90.3%) 4458 (90.4%) Constipation 687 (13.9%) 506 (10.3%) 529 (10.7%) 505 (10.2%) 477 (9.7%) 473 (9.6%) Unconditional model for constipation There was good support for a four-class solution when considering the six repeated measurements of constipation. Focusing on the sample consisting of cases with at least three non-missing measures of constipation (n=8979), the following classification of children was obtained: normative (82%; very low probability of constipation throughout), early childhood occurrence (7%; children suffering from constipation until 6 years), late childhood occurrence (8% problem with constipating emerging after 6 years) and persistent (3%; high probability of problems throughout).

n of children was obtained: normative (82%; very low probability of constipation throughout), early childhood occurrence (7%; children suffering from constipation until 6 years), late childhood occurrence (8% problem with constipating emerging after 6 years) and persistent (3%; high probability of problems throughout). Parallel model for soiling and constipation The next step was to merge the two LLCA models into a single parallel model permitting an investigation into the longitudinal association between soiling and constipation (see table 2 and figure 1). Based on these results, we collapsed the 16 groups defined by such a two-way classification into four clinical relevant subgroups to facilitate further study. The four resulting groups were: normative (74.5% of total)—normative classes for both constipation and soiling; constipation alone (13.2%)—normative class for soiling and non-normative classes for constipation; soiling alone (7.5%)—normative class for constipation and non-normative classes for soiling; constipation with soiling (4.8%)—classified as non-normative in terms of the progression of both constipation and soiling during the period of childhood studied. Table 2 Associations between constipation and soiling: joint distribution of classes derived from parallel longitudinal Latent Class Analysis model (n=8435)

Parallel model for soiling and constipation The next step was to merge the two LLCA models into a single parallel model permitting an investigation into the longitudinal association between soiling and constipation (see table 2 and figure 1). Based on these results, we collapsed the 16 groups defined by such a two-way classification into four clinical relevant subgroups to facilitate further study. The four resulting groups were: normative (74.5% of total)—normative classes for both constipation and soiling; constipation alone (13.2%)—normative class for soiling and non-normative classes for constipation; soiling alone (7.5%)—normative class for constipation and non-normative classes for soiling; constipation with soiling (4.8%)—classified as non-normative in terms of the progression of both constipation and soiling during the period of childhood studied. Table 2 Associations between constipation and soiling: joint distribution of classes derived from parallel longitudinal Latent Class Analysis model (n=8435) Constipation classes Soiling classes (%) Normative (soiling) Delayed Relapse Persistent Total Normative (constipation) 74.5* 4.3† 2.2† 1.0† 82 Early childhood occurrence 5.5‡ 1.1§ <0.1§ 0.5§ 7.1 Late childhood occurrence 5.8‡ 0.6§ 0.9§ 0.3§ 7.6 Persistent 1.9‡ 0.4§ 0.5§ 0.5§ 3.3 Total 87.7 6.4 3.6 2.3 100 Data shown in the table are class proportions for the latent class patterns based on the estimated model. These results are for the sample of children with both constipation and soiling data for at least three time points.

occurrence 5.8‡ 0.6§ 0.9§ 0.3§ 7.6 Persistent 1.9‡ 0.4§ 0.5§ 0.5§ 3.3 Total 87.7 6.4 3.6 2.3 100 Data shown in the table are class proportions for the latent class patterns based on the estimated model. These results are for the sample of children with both constipation and soiling data for at least three time points. *Normative (74.5% of total)—normative classes for both constipation and soiling. †Soiling alone (7.5% of total)—normative class for constipation and non-normative classes for soiling. ‡Constipation alone (13.2% of total)—normative class for soiling and non-normative classes for constipation. §Constipation with soiling (4.8% of total)—non-normative classes for constipation and soiling. Figure 1 Trajectories for constipation (left) and soiling (right).

†Soiling alone (7.5% of total)—normative class for constipation and non-normative classes for soiling. ‡Constipation alone (13.2% of total)—normative class for soiling and non-normative classes for constipation. §Constipation with soiling (4.8% of total)—non-normative classes for constipation and soiling. Figure 1 Trajectories for constipation (left) and soiling (right). Rates of associated symptoms within each class To gain further insight into this classification, we examined the extent to which the rates of daytime wetting, bed-wetting and stomach ache in mid-childhood, and infrequent bowel movements in early childhood differed across the four composite classes (online supplementary figure 2). First, the two classes involving soiling exhibited similar rates of both daytime wetting and bed-wetting, and furthermore, these rates were consistently higher than the constipation only and the normative class. Second, rates of stomach ache were lower for the soiling alone class compared with either class reporting constipation. Finally, in terms of the rates of infrequent bowel movements in early childhood, those classified as soiling alone were no different to the normative class for whom infrequent movements were rare, whereas both classes experiencing constipation were seen to have infrequent bowel movements which became a more common finding as the children became older.

infrequent bowel movements in early childhood, those classified as soiling alone were no different to the normative class for whom infrequent movements were rare, whereas both classes experiencing constipation were seen to have infrequent bowel movements which became a more common finding as the children became older. Risk factors for constipation/soiling class membership Table 3 shows the association between the risk factors and ORs for membership of the classes of ‘constipation alone’, ‘soiling alone’ and ‘constipation with soiling’ with reference to the normative class. For completeness, we append the results for the separate models of soiling and constipation (online supplementary table 3). Table 3 Association between risk factors and the parallel classes of soiling and constipation (n≤8435 depending on risk factor)

Risk factors for constipation/soiling class membership Table 3 shows the association between the risk factors and ORs for membership of the classes of ‘constipation alone’, ‘soiling alone’ and ‘constipation with soiling’ with reference to the normative class. For completeness, we append the results for the separate models of soiling and constipation (online supplementary table 3). Table 3 Association between risk factors and the parallel classes of soiling and constipation (n≤8435 depending on risk factor) n (%) Constipation alone Soiling alone Constipation with soiling Omnibus P value OR (95% CI) OR (95% CI) OR (95% CI) Sex Female (ref) 4082 (52%) <0.001 Male 4353 (48%) 0.63 (0.53 to 0.75) 1.78 (1.39 to 2.27) 1.38 (1.06 to 1.79) Hard stools regularity at 2½ years Never hard (ref) 1414 (18%) <0.001 Sometimes 4312 (54%) 2.08 (1.56 to 2.77) 1.12 (0.82 to 1.52) 1.22 (0.84 to 1.78) Usually hard 2235 (28%) 1.90 (1.40 to 2.60) 0.68 (0.46 to 0.99) 1.26 (0.85 to 1.89) Breast feeding Child never breast fed (ref) 1695 (21%) 0.309 Child breast fed for <6 months 3579 (45%) 1.25 (0.99 to 1.58) 1.34 (0.96 to 1.86) 1.08 (0.77 to 1.52) Child breastfed for at least 6 months 2677 (34%) 1.24 (0.97 to 1.59) 1.22 (0.86 to 1.73) 0.94 (0.65 to 1.35) Parental social class Professional, managerial or skilled 6579 (85%) 0.736 Partly skilled/unskilled 1178 (15%) 0.95 (0.74 to 1.20) 1.16 (0.85 to 1.60) 0.98 (0.67 to 1.42) Maternal educational attainment A level/degree (ref) 3399 (41%) 0.358 O level 2916 (35%) 0.92 (0.76 to 1.12) 0.82 (0.62 to 1.08) 1.16 (0.87 to 1.54) Vocational or none 1909 (23%) 0.82 (0.66 to 1.03) 0.94 (0.70 to 1.26) 0.92 (0.65 to 1.30) Material hardship at 33 months*

to 1.20) 1.16 (0.85 to 1.60) 0.98 (0.67 to 1.42) Maternal educational attainment A level/degree (ref) 3399 (41%) 0.358 O level 2916 (35%) 0.92 (0.76 to 1.12) 0.82 (0.62 to 1.08) 1.16 (0.87 to 1.54) Vocational or none 1909 (23%) 0.82 (0.66 to 1.03) 0.94 (0.70 to 1.26) 0.92 (0.65 to 1.30) Material hardship at 33 months* No material hardship (score<5) (ref.) 5733 (73%) 0.026 Material hardship (score≥5) 2084 (27%) 1.00 (0.82 to 1.22) 1.06 (0.81 to 1.40) 1.52 (1.15 to 2.00) Home ownership at 33 months Home owned/ mortgaged (ref.) 6524 (85%) 0.401 Privately rented 312 (4.1%) 0.73 (0.44 to 1.20) 1.12 (0.62 to 2.02) 1.26 (0.69 to 2.29) Subsidised rented 832 (11%) 0.74 (0.55 to 1.00) 1.01 (0.68 to 1.48) 0.92 (0.60 to 1.43) Car access at 33 months Yes (ref.) 7230 (93%) 0.018 No 581 (7%) 0.82 (0.57 to 1.18) 1.53 (1.02 to 2.27) 1.49 (0.97 to 2.29) Gestational age at delivery ≥37 weeks (ref) 8000 (95%) 0.334 <37 weeks 435 (5%) 0.76 (0.50 to 1.16) 1.28 (0.80 to 2.05) 1.12 (0.66 to 1.93) Birth weight ≥2500 g (ref) 7964 (96%) 0.488 <2500 g 372 (4%) 1.29 (0.88 to 1.88) 1.32 (0.79 to 2.21) 1.05 (0.56 to 1.97) Developmental level at 18 m†

2 to 2.27) 1.49 (0.97 to 2.29) Gestational age at delivery ≥37 weeks (ref) 8000 (95%) 0.334 <37 weeks 435 (5%) 0.76 (0.50 to 1.16) 1.28 (0.80 to 2.05) 1.12 (0.66 to 1.93) Birth weight ≥2500 g (ref) 7964 (96%) 0.488 <2500 g 372 (4%) 1.29 (0.88 to 1.88) 1.32 (0.79 to 2.21) 1.05 (0.56 to 1.97) Developmental level at 18 m† Per 1 SD reduction in development 7931 0.95 (0.87 to 1.05) 1.44 (1.28 to 1.62) 1.31 (1.13 to 1.51) <0.001 Age at initiation of toilet training Unadjusted model <0.001 Before 6 months 168 (2%) 1.20 (0.67 to 2.15) 1.50 (0.69 to 3.28) 1.73 (0.78 to 3.81) Between 6 and 15 months 1056 (14%) 1.22 (0.95 to 1.56) 1.09 (0.74 to 1.59) 0.81 (0.49 to 1.32) Between 15 and 24 months (ref) 3841 (50%) After 24 months 2650 (34%) 0.91 (0.75 to 1.12) 1.47 (1.13 to 1.91) 1.56 (1.17 to 2.07) *Material hardship was assessed using the set of questions: ‘How difficult at the moment do you find it to afford these items? Food, clothing, heating, rent, items for child’: each on a four-point scale from very difficult through to not difficult. Responses were summed and a binary variable was derived to indicate the top 20% of the sample.

assessed using the set of questions: ‘How difficult at the moment do you find it to afford these items? Food, clothing, heating, rent, items for child’: each on a four-point scale from very difficult through to not difficult. Responses were summed and a binary variable was derived to indicate the top 20% of the sample. †Developmental level was assessed using a questionnaire developed by Avon Longitudinal Study of Parents and Children including items from the Denver Developmental Screening Test24 and comprising four domains of development (fine motor, gross motor, communication and social skills). We used a total development score derived from the sum of the scores on each domain. Scores on each domain were adjusted for age in weeks and standardised (using a linear regression model and extracting the residuals) and reversed where appropriate so that high values on all scores reflected a lower level of development (increase in the odds of membership to the latent classes per 1 SD increase in developmental level score).

ere adjusted for age in weeks and standardised (using a linear regression model and extracting the residuals) and reversed where appropriate so that high values on all scores reflected a lower level of development (increase in the odds of membership to the latent classes per 1 SD increase in developmental level score). There was strong evidence that boys were at greater odds of being a member of the ‘soiling alone’ and ‘constipation with soiling’ classes; however, at lower odds of experiencing constipation alone. We also found strong evidence that the presence of hard stools (sometimes or usually) at 2½ years was associated with increased odds of membership to the ‘constipation alone’ class; however, there was little evidence of hard stool consistency being associated with constipation with soiling. There was also a suggestion that hard stools might be protective against the development of ‘soiling without constipation’.

rs was associated with increased odds of membership to the ‘constipation alone’ class; however, there was little evidence of hard stool consistency being associated with constipation with soiling. There was also a suggestion that hard stools might be protective against the development of ‘soiling without constipation’. We found little evidence of an association between breastfeeding duration and later problems with constipation and/or soiling, and we found no association between gestational age or birth weight and constipation and soiling classes. For socioeconomic position, the pattern of results was inconsistent. We found no evidence that parental social class, maternal educational attainment and home ownership were associated with constipation and soiling. Material hardship, however, was associated with constipation with soiling and lack of car access was associated with soiling alone. Lower developmental level at 18 months was associated with ‘soiling alone’ and ‘constipation and soiling’. However, there was no association between developmental level ‘constipation alone’. Finally, a later age at initiation of toilet training was associated with ‘soiling alone’ and ‘constipation with soiling’. Estimates were attenuated following adjustment for the confounders: ‘soiling alone’ (adjusted OR 1.30 (0.97 to 1.75)), ‘constipation with soiling’ (adjusted OR 1.27 (0.92 to 1.75)). The most influential confounders were developmental delay, major financial difficulties, sex (male) (membership of ‘soiling alone’ class) and maternal depression (‘constipation with soiling’ class).

unders: ‘soiling alone’ (adjusted OR 1.30 (0.97 to 1.75)), ‘constipation with soiling’ (adjusted OR 1.27 (0.92 to 1.75)). The most influential confounders were developmental delay, major financial difficulties, sex (male) (membership of ‘soiling alone’ class) and maternal depression (‘constipation with soiling’ class). Discussion We found that the variability in longitudinal data on childhood constipation and soiling years in a large UK birth cohort could each be adequately explained by four latent classes. On cross-classifying these groupings we identified four clinically relevant longitudinal classes. Three quarters of children were members of the ‘normative’ class with a very low probability of constipation or soiling across childhood. Two classes comprised children with constipation alone (13.2% of children) and soiling alone (7.5%), and finally, a class of children who had constipation with soiling (4.8%). Among children with soiling, around 60% had soiling alone and around 40% had soiling with constipation. The prevalence of constipation (18%) observed is higher than the median prevalence (8.9%) reported in a systematic review of children aged 0–18 years.5 Boys were more likely than girls to experience soiling as reported elsewhere.5 7

ith soiling, around 60% had soiling alone and around 40% had soiling with constipation. The prevalence of constipation (18%) observed is higher than the median prevalence (8.9%) reported in a systematic review of children aged 0–18 years.5 Boys were more likely than girls to experience soiling as reported elsewhere.5 7 We found that hard stools in early childhood were associated with an increased odds of ‘constipation alone’ at school age, as well as a reduced risk for ‘soiling alone’. Children with developmental delay had more ‘soiling alone’ and ‘constipation with soiling’, but not ‘constipation alone’. We found limited evidence for socioeconomic disadvantage and no evidence that a shorter duration of breast feeding, shorter gestation, lower birth weight or the timing of toilet training were risk factors for constipation or soiling at school age. A major strength is the availability of repeated measures of constipation and soiling across childhood in a large, representative cohort. Using these data, we extended our previous work by modelling constipation and soiling in parallel. The resultant latent classes allowed us to estimate, the prevalence of soiling with or without constipation and for constipation alone across childhood and the differences in early risk factors.

ge, representative cohort. Using these data, we extended our previous work by modelling constipation and soiling in parallel. The resultant latent classes allowed us to estimate, the prevalence of soiling with or without constipation and for constipation alone across childhood and the differences in early risk factors. A potential limitation is the use of maternal report measures of constipation and soiling. Parents were asked to report whether their child had suffered from constipation in the past 12 months and whether they saw a doctor. Previous studies report that only a small proportion of children see a doctor for soiling problems,8 perhaps because parents are unaware that medical advice and treatment is available. Also, parents of children who soil may be unaware that their child is constipated. In addition, the questions relating to soiling did not elicit information about quantity so did not distinguish between leakage, normal bowel movement and staining. Some parents added addenda to the questionnaire noting, for instance, active diarrhoea and poor wiping and did not categorise their child as soiling, so anecdotal evidence indicates that parents are able to distinguish between true soiling and other occurrences.

uish between leakage, normal bowel movement and staining. Some parents added addenda to the questionnaire noting, for instance, active diarrhoea and poor wiping and did not categorise their child as soiling, so anecdotal evidence indicates that parents are able to distinguish between true soiling and other occurrences. Parents in the present study were not asked about the duration of constipation or soiling at each time point; however, the repeated measurements of these conditions suggest that constipation and soiling are persisting problems. Information on frequency of soiling was omitted from the latent class models because frequent soiling was rare. Our aim was to describe the trajectories of constipation and soiling in the community and not to focus solely on children whose bowel problems meet the current established diagnostic criteria. Finally, there was no information available on underlying organic causes of constipation and soiling (eg, Hirschsprung’s disease or anorectal malformations), but the majority of cases are functional. Children with soiling and/or constipation have relatives with these problems,7 28 29 information we did not have.

teria. Finally, there was no information available on underlying organic causes of constipation and soiling (eg, Hirschsprung’s disease or anorectal malformations), but the majority of cases are functional. Children with soiling and/or constipation have relatives with these problems,7 28 29 information we did not have. Clinicians believe that most cases of childhood soiling are from chronic constipation; however, we found that among soilers, constipation with soiling (39%) was less common than soiling alone (61%). Other estimates for constipation with soiling2 28 are from clinic samples. Children with soiling present less often to a clinician, perhaps because they feel ashamed or they believe it is due to laziness.8Compared with constipation with soiling, which comes with abdominal pain and infrequent stools, soiling may have no other symptoms.2 In agreement with earlier studies,11 12 we found that hard stools in early childhood were strongly associated with later constipation. Early hard stools were, however, not associated with soiling alone or constipation with soiling, suggesting that soiling is a primary continence issue and not secondary to constipation.

In agreement with earlier studies,11 12 we found that hard stools in early childhood were strongly associated with later constipation. Early hard stools were, however, not associated with soiling alone or constipation with soiling, suggesting that soiling is a primary continence issue and not secondary to constipation. There is evidence from earlier studies that constipation and soiling are more common among children from lower socioeconomic background.8–10 16 We found little evidence that socioeconomic factors are associated with childhood constipation and soiling. Developmental delay was associated with any soiling but not with constipation alone. Constipation alone may be more strongly related to family history, early experiences of painful defecation and defecation anxiety while soiling is related to delays in reaching social developmental milestones. We found no evidence that length of gestation or birth weight was associated with constipation or soiling at school age consistent with an earlier study.19 After adjusting for confounders, earlier or later initiation of toilet training had no significant effect on constipation or soiling similar to previous studies19 20 and in contrast to a large cohort study.7

on or birth weight was associated with constipation or soiling at school age consistent with an earlier study.19 After adjusting for confounders, earlier or later initiation of toilet training had no significant effect on constipation or soiling similar to previous studies19 20 and in contrast to a large cohort study.7 This study finds evidence that experiencing hard stools in early childhood is a risk factor for later problems with constipation at school age. Early identification of children at risk of constipation is important because more than a third of cases become chronic and require secondary care.30 Early diagnosis and treatment will reduce the risk of constipation persisting to school age, resulting in poor quality of life. The burden of constipation is large—13.2% of primary school-age children. Extra healthcare costs are substantial as 5.9% of the ALSPAC cohort 4–9 years of age saw a doctor at least once (data not shown). Many would have been referred to secondary care and tertiary paediatric gastroenterology and paediatric surgery services.

urden of constipation is large—13.2% of primary school-age children. Extra healthcare costs are substantial as 5.9% of the ALSPAC cohort 4–9 years of age saw a doctor at least once (data not shown). Many would have been referred to secondary care and tertiary paediatric gastroenterology and paediatric surgery services. Further population-based studies are needed, and if non-retentive soiling rates are confirmed, parents should be encouraged to seek help for soiling from clinicians who recognise the problem. As with daytime wetting the mainstay of treatment for soiling (alone and the soiling aspect of soiling with constipation) will be behavioural, for example, supporting children to attempt to empty their bowels into the toilet at least once per day. This form of therapy requires practitioners to be skilled at providing behavioural interventions concurrently with medications for children who are constipated. Supplementary Material Reviewer comments Author's manuscript The authors are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. Contributors: Secured funding: DT and CJ. Study design: JH, DT, AvG and CJ. Literature search: CJ. Data analysis: JH and MG. Data interpretation and writing:JH, MG, DT, AvG and CJ.

Supplementary Material Reviewer comments Author's manuscript The authors are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. Contributors: Secured funding: DT and CJ. Study design: JH, DT, AvG and CJ. Literature search: CJ. Data analysis: JH and MG. Data interpretation and writing:JH, MG, DT, AvG and CJ. Funding: The MRC and Wellcome Trust (grant ref: 102215/2/13/2) and University of Bristol provide core support for ALSPAC. This publication is the work of the authors who will serve as guarantors for the contents of this paper. This work was supported by the MRC (MR/L007231/1) and NHS Greater Glasgow & Clyde Health Board (Stephen McLeod, Head of Specialist Children’s Services). Competing interests: None declared. Ethics approval: ALSPAC Law and Ethics committee and local research ethics committees. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Disabled children are admitted to hospital more often than other children. Research suggests that disabled children’s experience as inpatients is not always optimal and identifies communication between staff, children and families as a key issue. Improving children’s experience of healthcare is a priority for the National Health Service. What this study hopes to add? We describe development of a novel training using the intervention mapping approach. The training comprises videos of parent carers discussing hospital experiences, interactive tasks, small group discussion, personal reflection and intention planning. The training was feasible and is documented in a manual to enable replication.

What this study hopes to add? We describe development of a novel training using the intervention mapping approach. The training comprises videos of parent carers discussing hospital experiences, interactive tasks, small group discussion, personal reflection and intention planning. The training was feasible and is documented in a manual to enable replication. Introduction Disabled children should be consulted about their care and decisions that affect them, a right asserted by The UN Conventions on the Rights of the Child1 and the Rights of Persons with Disabilities.2 The UK is a signatory to both conventions and National Health Service (NHS) policy affirms that children have the right to be treated with respect and enabled to make decisions about their healthcare.3 Despite legislation and policy at international and national level to promote equitable healthcare, the practical reality is that poorer experiences are still reported for disabled children.4 This is particularly evident with regards to disabled children and their involvement in decision-making processes for their own healthcare and well-being. The NHS has committed to improve patient experience of care through the NHS Outcomes Framework and identified responsiveness to inpatients’ needs as an indicator to evaluate improvement.5

vident with regards to disabled children and their involvement in decision-making processes for their own healthcare and well-being. The NHS has committed to improve patient experience of care through the NHS Outcomes Framework and identified responsiveness to inpatients’ needs as an indicator to evaluate improvement.5 Disabled children are more often admitted to hospital than other children.6 7 Communication in hospital can be particularly challenging for children with learning disabilities,8 and those who use augmentative and alternative communication.9 It has been reported that parents often feel unable to leave such children because of concerns about communication.10–13 When communication is poor, children and parent carers may not understand their choices and have inadequate opportunity to engage in decision-making.

use augmentative and alternative communication.9 It has been reported that parents often feel unable to leave such children because of concerns about communication.10–13 When communication is poor, children and parent carers may not understand their choices and have inadequate opportunity to engage in decision-making. Synthesis of qualitative research on the experience of disabled children as inpatients suggested that communication mediates many aspects of their experience, but is often inadequate in practice.14 Communication was found to be an overarching theme, with impact on other factors and influenced the hospital experience as a whole. Good communication can help to alleviate adverse emotional states, contribute to a more positive perception of the environment and improve confidence in staff. Further qualitative research with parents and ward staff identified key barriers and facilitators to good communication. Barriers included time pressures and the low priority given to communication; facilitators were making time to build a rapport with a child, previous experience of working with children and a family-centred outlook.15 The professional participants in our earlier qualitative study clearly expressed awareness and personal frustration with failing to meet children’s communication needs. Some parents felt that their knowledge and experience of their child was not always considered or valued, which has been highlighted in previous studies.16 Our findings were similar to those of Oulton et al8 which emphasised that training in caring for children with learning disabilities was a key factor in taking an individualised approach to inpatient care.14

experience of their child was not always considered or valued, which has been highlighted in previous studies.16 Our findings were similar to those of Oulton et al8 which emphasised that training in caring for children with learning disabilities was a key factor in taking an individualised approach to inpatient care.14 The UK Department of Health Education Outcomes Framework includes five domains, one of which is ‘NHS Values and Behaviours’ and refers to healthcare staff developing values and behaviours, through training, to enhance the quality of the patient experience.17 This is a mandate for the provision of relevant training for the existing workforce. This paper describes research to develop a training package for staff in partnership with parent carers and ward staff called ‘Improving Inpatient Experiences for Disabled Children’, to test the feasibility of delivery in a ward setting and to gauge the utility of such an intervention to staff.

The UK Department of Health Education Outcomes Framework includes five domains, one of which is ‘NHS Values and Behaviours’ and refers to healthcare staff developing values and behaviours, through training, to enhance the quality of the patient experience.17 This is a mandate for the provision of relevant training for the existing workforce. This paper describes research to develop a training package for staff in partnership with parent carers and ward staff called ‘Improving Inpatient Experiences for Disabled Children’, to test the feasibility of delivery in a ward setting and to gauge the utility of such an intervention to staff. Methods Stakeholder involvement The research had a strong ethos of public involvement. We report this involvement using GRIPP2 guidance.18 Six parents of children with neurodisability from the PenCRU Family Faculty collaborated at various times to develop the training (www.pencru.org/getinvolved/ourfamilyfaculty). Financial acknowledgement of their time and travel expenses were reimbursed. Parent carers suggested the topic, helped design the training, suggested and facilitated invitation to the hospital Patient and Carer Experience Group, recorded their experiences for the video content and participated in meetings to reflect on the training with the facilitator. Parents are also involved in sharing the findings. The involvement of parent carers profoundly influenced the content of the training to deliver family experiences and messages. We did not have resources to plan for the meaningful involvement of young people; involving disabled young people would be desirable in future work. There was little ethnic diversity among the parents we work with; accounting for cultural differences might add another dimension to the training. Paediatricians and nurses were represented on the team and other clinicians were consulted about the design of the intervention.

ng people would be desirable in future work. There was little ethnic diversity among the parents we work with; accounting for cultural differences might add another dimension to the training. Paediatricians and nurses were represented on the team and other clinicians were consulted about the design of the intervention. Ethics statement The Royal Devon and Exeter NHS Foundation Trust Research and Development Office approved the study. The Health Research Authority does not require ethics approval for studies involving NHS staff as research participants by virtue of their professional role (www.hra.nhs.uk/resources/before-you-apply/research-requiring-nhs-rd-review-but-not-ethical-review). Theoretical underpinning The process of developing the training was informed by intervention mapping.19 Intervention mapping includes six iterative steps for developing and evaluating health interventions (table 1). This study describes the use of the first four steps to design the intervention. We took account of recommendations from the National Institute for Health and Care Excellence (NICE) guidance on behaviour change20 and the Template for Intervention Description and Replication.21 Table 1 Intervention mapping framework Products Tasks Step 1 Needs assessment Establish a stakeholder group Conduct needs assessment Assess capacity Determine programme outcomes Step 2 Programme objective matrices State expected changes in behaviour and environment State performance objectives Specify modifiable determinants Create a logic model of change

Table 1 Intervention mapping framework Products Tasks Step 1 Needs assessment Establish a stakeholder group Conduct needs assessment Assess capacity Determine programme outcomes Step 2 Programme objective matrices State expected changes in behaviour and environment State performance objectives Specify modifiable determinants Create a logic model of change Step 3 Theory-based methods and practical strategies Review programme ideas with representative participants Identify relevant theories Choose programme methods Select or design strategies appropriate to change objectives Step 4 Programme Consult intended participants and implementers Create programme scope, sequence and resources list Develop design documents Review available programme materials Draft programme materials Pretest programme materials with target group and implementers. Produce materials and protocols Step 5 Adoption and Implementation Plan Identify potential adopters and users Specify adoption, implementation and sustainability performance objectives Specify determinants and create a matrix of change objectives Select methods and strategies Design interventions for adoption and implementation Step 6 Evaluation plan Develop evaluation model Develop indicators and measures Specify evaluations designs Write an evaluation plan

Specify adoption, implementation and sustainability performance objectives Specify determinants and create a matrix of change objectives Select methods and strategies Design interventions for adoption and implementation Step 6 Evaluation plan Develop evaluation model Develop indicators and measures Specify evaluations designs Write an evaluation plan Step 1: Needs assessment The first step in intervention mapping involves assessing the need for an intervention. The research was initiated by a parent approaching the researchers following a difficult inpatient experience with their disabled child. A structured review confirmed that the inpatient experience of disabled children is not always optimal and that communication is a key determinant of inpatient experience.14 A qualitative study was undertaken to explore the experiences of families of disabled children and ward staff and to gain a fuller understanding about the concerns, skills and resources influencing communication. Fifteen parents and 25 ward staff took part in semistructured interviews or focus groups. Difficulty was experienced in recruiting children and evaluating their experiences, despite considerable efforts. Thematic analysis of the interviews and focus groups identified barriers and facilitators to effective communication on children’s wards.15

ward staff took part in semistructured interviews or focus groups. Difficulty was experienced in recruiting children and evaluating their experiences, despite considerable efforts. Thematic analysis of the interviews and focus groups identified barriers and facilitators to effective communication on children’s wards.15 There was no formal-specific training in the hospital focusing on communication or disability. A consultation meeting with staff representative of those requiring training informed the needs assessment to take account of the social and environmental context of the intervention. This explored interest in the training topic, practical considerations that might influence participation. Step 2: Identifying outcomes and change objectives The second step involved considering the objectives of training and creating a logic model of change (figure 1). The ultimate outcome is to improve the inpatient experience of disabled children through enhanced ward staff communication with disabled children and their parent carers. However, this outcome is dependent on many intermediary variables. The more proximal objectives involve who and what needs to change in order to achieve this. Figure 1 Logic model of the training intervention and outcomes. NHS, National Health Service.

Step 2: Identifying outcomes and change objectives The second step involved considering the objectives of training and creating a logic model of change (figure 1). The ultimate outcome is to improve the inpatient experience of disabled children through enhanced ward staff communication with disabled children and their parent carers. However, this outcome is dependent on many intermediary variables. The more proximal objectives involve who and what needs to change in order to achieve this. Figure 1 Logic model of the training intervention and outcomes. NHS, National Health Service. The intended participants were staff who interact with children admitted to the ward. The child’s journey through the ward was considered, including all interactions beyond medical and nursing care; for example, from first booking in with the ward administrator, meal times and meeting ward housekeeping staff. The target behaviour change was that ward staff prioritise and practice good communication with disabled children. Based on our needs assessment and consultation with parents in our advisory group four key practices were identified:Ask the parent or carer for advice about how to communicate with their child. Identify how a child communicates yes and no. Communicate directly with a child when appropriate. Feel comfortable to admit when you don’t know the best way to communicate.

The intended participants were staff who interact with children admitted to the ward. The child’s journey through the ward was considered, including all interactions beyond medical and nursing care; for example, from first booking in with the ward administrator, meal times and meeting ward housekeeping staff. The target behaviour change was that ward staff prioritise and practice good communication with disabled children. Based on our needs assessment and consultation with parents in our advisory group four key practices were identified:Ask the parent or carer for advice about how to communicate with their child. Identify how a child communicates yes and no. Communicate directly with a child when appropriate. Feel comfortable to admit when you don’t know the best way to communicate. It was evident that for the intervention to succeed, organisational and cultural changes were required to support individual behaviour change. A key contextual factor was for managers to recognise and prioritise the need for training and enable staff attendance.

Feel comfortable to admit when you don’t know the best way to communicate. It was evident that for the intervention to succeed, organisational and cultural changes were required to support individual behaviour change. A key contextual factor was for managers to recognise and prioritise the need for training and enable staff attendance. Step 3: theory-based methods and practical strategies The third stage of intervention mapping is to select practical methods and strategies consistent with behaviour change theories. With reference to the NICE guidance, the theories selected as being most appropriate were the Theory of Planned Behaviour22 and Bandura’s construct of self-efficacy.23 Azjen’s theory states that intention is the main determinant of action, this is predicted by attitude, subjective norm and perceived behavioural control. In Bandura’s construct, self-efficacy is the belief in one’s ability to succeed in specific situations. Table 2 lists the learning objectives for ward staff and how each links with behaviour change concepts and the content of the training. Table 2 Behaviour change concepts mapped to training content Learning objectives Behaviour change concepts Training content To understand the impact of communication behaviours on disabled children Outcome expectancies and positive attitude Personal and moral norms Parent video’s describing their child’s experience and how this could have been improved Inclusion of a positive experience Handouts including research findings To be motivated to change behaviour Personal relevance Self-efficacy

Learning objectives Behaviour change concepts Training content To understand the impact of communication behaviours on disabled children Outcome expectancies and positive attitude Personal and moral norms Parent video’s describing their child’s experience and how this could have been improved Inclusion of a positive experience Handouts including research findings To be motivated to change behaviour Personal relevance Self-efficacy Interactive tasks appropriate to role, small group discussion of personal experience Parent videos To develop empathy Personal and moral norms Opportunity for personal and small group reflection Practical exercises Parent videos To feel capable of behaviour change Self-efficacy Prompt/cue Four key practices, reinforced throughout training Basic awareness of some communication aids Signposting to local resources and policies Poster of four key practices displayed on ward To make a commitment to change Intention formation Concrete plans Opportunity to document how the training will change personal practice To feel supported by the organisation in changing behaviour Knowing and utilising existing processes and service models ‘Local slot’: highlighting local policies and useful resources

Poster of four key practices displayed on ward To make a commitment to change Intention formation Concrete plans Opportunity to document how the training will change personal practice To feel supported by the organisation in changing behaviour Knowing and utilising existing processes and service models ‘Local slot’: highlighting local policies and useful resources Common principles of adult learning were applied, for example, the concept of experiential learning and learning as a social activity. Continuous Professional Development tends to be more effective when time is allocated to reflect on learning and where organisational support is provided to facilitate change.24 This supports the importance of the secondary stream of the intervention; cultural and organisational change to support and maintain individual behaviour change. Step 4: Programme development This step produces the content and delivery of the intervention. Our consultation with ward staff suggested (1) it should not last more than 1 hour; (2) it should not be ‘mandatory training’ as this may mean some people attend reluctantly and (3) a face-to-face group session was preferred to online learning.

evelopment This step produces the content and delivery of the intervention. Our consultation with ward staff suggested (1) it should not last more than 1 hour; (2) it should not be ‘mandatory training’ as this may mean some people attend reluctantly and (3) a face-to-face group session was preferred to online learning. It was agreed with the advisory group that in order for the training to be sustainable and reproducible, videos of parents would be used to deliver key messages rather than parent facilitators. The facilitators would be local staff; in this instance the sessions were delivered by a paediatric registrar and a paediatric nurse with experience of working with disabled children. It was perceived that a nurse cofacilitator made the training more accessible to nurses who comprise the majority of ward staff. Personal and external factors influencing running sessions successfully were identified; this informed development of delivery strategies for the training (table 3). Advertising strategies were considered to ensure that all ward staff were made aware of the training. Face-to-face recruitment on the ward promoted discussion of the training and peer support to attend. Discussion with ward managers from different disciplines supported staff in attending. For example, an agreement was reached for nursing staff to use their half-hour lunch break and be given an additional half hour to enable them to attend the full-hour session. Table 3 Strategies for delivery objectives

Personal and external factors influencing running sessions successfully were identified; this informed development of delivery strategies for the training (table 3). Advertising strategies were considered to ensure that all ward staff were made aware of the training. Face-to-face recruitment on the ward promoted discussion of the training and peer support to attend. Discussion with ward managers from different disciplines supported staff in attending. For example, an agreement was reached for nursing staff to use their half-hour lunch break and be given an additional half hour to enable them to attend the full-hour session. Table 3 Strategies for delivery objectives Delivery objectives Personal factors External factors Strategies Raised awareness at organisational level with ‘buy in’ to cultural change Perceived importance of the need for training Competing interests and priorities Meeting with the hospital Patient Carer Experience Group Highlighting statutory requirements Reviewing policies and strategies at children’s ward business meeting Everyone working on the ward are able to attend training Knowledge of training sessions Allocation of time to attend (behavioural control) Accessibility of site Duration of training and timing in day Meeting with senior staff from all disciplines to agree permission to attend Identifying a suitable site Delivering the training at acceptable timings and duration to allow equity of access Everyone working on the ward attends training Confidence to attend Motivation to attend Modelling by peers who have attended (subjective norms)

Meeting with senior staff from all disciplines to agree permission to attend Identifying a suitable site Delivering the training at acceptable timings and duration to allow equity of access Everyone working on the ward attends training Confidence to attend Motivation to attend Modelling by peers who have attended (subjective norms) Visiting the ward and speaking to staff about the training Signing up peers to attend together Providing lunch as a motivator Identifying key figures and encouraging them to attend The training included a warm up activity to engage participants, parent videos, interactive tasks and small group discussion. Time was allocated at the end for personal reflection, to document a commitment to behaviour change and to provide feedback. Participants were asked for numerical scores for training elements and to identify two positive points, two areas for improvement and how the session was likely to change their practice. After each session the facilitators reviewed the feedback and proposed changes together, this was also discussed with the advisory group.

icipants were asked for numerical scores for training elements and to identify two positive points, two areas for improvement and how the session was likely to change their practice. After each session the facilitators reviewed the feedback and proposed changes together, this was also discussed with the advisory group. Results Eighty participants attended one of four sessions at the Royal Devon and Exeter NHS Foundation Trust, which is a UK University Hospital. All sessions were attended by various staff (table 4). Allied health professionals included physiotherapists, play therapists, dieticians, a pharmacist and a speech and language therapist. Non-clinical staff included housekeeping and catering staff, administrators, teachers and chaplain. A notable feedback comment was that ‘it felt very powerful to have such a cross section of staff and hierarchy all working at the same task’. Table 4 Roles of participants attending training

Results Eighty participants attended one of four sessions at the Royal Devon and Exeter NHS Foundation Trust, which is a UK University Hospital. All sessions were attended by various staff (table 4). Allied health professionals included physiotherapists, play therapists, dieticians, a pharmacist and a speech and language therapist. Non-clinical staff included housekeeping and catering staff, administrators, teachers and chaplain. A notable feedback comment was that ‘it felt very powerful to have such a cross section of staff and hierarchy all working at the same task’. Table 4 Roles of participants attending training Profession Number of participants Roles represented Medical 34 Seven consultants, 20 junior doctors, seven medical students Nursing 25 Four senior nurses, four specialty (epilepsy, oncology), 10 ward nurses, five nursing students, two nursing assistants Allied health professionals 9 Three physiotherapists, two play therapists, two dieticians, one pharmacist, one speech and language therapist Ward- non-clinical 12 Six housekeeping and catering staff, two ward administrators, 1 chaplain and three teachers The first session was attended by 26 people, including some unbooked participants. Subsequent sessions were strictly limited to 20. One session had fewer numbers at short notice due to unexpected clinical workload and staff capacity. Participant feedback was very positive, with high satisfaction scores for all areas. Comments in response to asking for positive points indicated that the videos and practical strategies designed to develop empathy and understand the impact of communication behaviours on disabled children had been successful. Comments in response to the question ‘How will you change your practice?’ indicated that participants were willing to make a commitment to change, felt capable of change and were assimilating key messages (box 1).Box 1 Examples of feedback Positive points ‘Easy understanding by watching interviews’

had been successful. Comments in response to the question ‘How will you change your practice?’ indicated that participants were willing to make a commitment to change, felt capable of change and were assimilating key messages (box 1).Box 1 Examples of feedback Positive points ‘Easy understanding by watching interviews’ ‘Really improved the level of understanding of how children can feel in some situations’ ‘Sharing experiences was helpful’ ‘Practical exercises were enlightening’ Areas for improvement ‘More discussion’ ‘Focus less on problems’ ‘Some slides text heavy’ ‘Room too hot and crowded’ How will you change your practice? ‘Considering the patient as an individual’ ‘Feel more confident to just ask’ ‘Take more time to think’ ‘More techniques on how to communicate’ Feedback on areas for improvement included allowing more time for discussion and highlighting parents’ positive and negative experiences. This resulted in the inclusion of a positive parent experience video and distributing background information prior to the session rather than during it, to allow more discussion time. One suggestion was for prompts and reminders to be displayed on the ward. We worked with young people with learning disability at a local further education college to create a poster that provides ‘4 Top Tips’ to improve communication with disabled children.

session rather than during it, to allow more discussion time. One suggestion was for prompts and reminders to be displayed on the ward. We worked with young people with learning disability at a local further education college to create a poster that provides ‘4 Top Tips’ to improve communication with disabled children. Discussion The training was well received in the context of a university hospital paediatric ward. The number of staff attending and the breadth of roles represented reflect successful recruitment strategies and a high level of interest in improving ward communication with disabled children. Strategies to raise awareness with ‘buy in’ to cultural change at management level were an essential part of the intervention. The commitment to wider cultural change was underpinned by the involvement of stakeholder groups and promotion at ward and hospital level meetings.

ng ward communication with disabled children. Strategies to raise awareness with ‘buy in’ to cultural change at management level were an essential part of the intervention. The commitment to wider cultural change was underpinned by the involvement of stakeholder groups and promotion at ward and hospital level meetings. The iterative approach of intervention mapping provided a structure to consider the complexity of personal and external determinants that influence behaviour at an individual and organisational level. Consultation provided guiding principles for an acceptable delivery model for training; feedback after sessions led to more focused content and staff feedback indicates provisional willingness and motivation to change behaviour. The intervention mapping process was useful as an approach, however, we found applying it in practice to be time-consuming as it required granular levels of analysis in steps 2 and 3 to define objectives. We followed the approach insofar as it was possible within the limitations of time and resources available. An alternative approach proposing six steps in intervention development,25 may be more practical and be sufficient for the needs of future research when the high degree of rigour required by intervention mapping is not feasible.

approach insofar as it was possible within the limitations of time and resources available. An alternative approach proposing six steps in intervention development,25 may be more practical and be sufficient for the needs of future research when the high degree of rigour required by intervention mapping is not feasible. A high level of commitment and enthusiasm was required by the paediatric registrar to encourage management support for staff participation. Strategic recruitment of influential staff, seeking validation from credible advocates and peer encouragement through word of mouth were vital to recruitment. This approach relied heavily on understanding the social dynamics of the ward. The commitment to cultural change in the hospital benefited from high-level support from the Parent and Carer Experience Group and seizing opportunities to raise the training at ward meetings. While it is difficult to identify the impact of these individual and local contextual mediators, we suggest that the professional leading these initiatives needs to know the key people in their own organisation.

from the Parent and Carer Experience Group and seizing opportunities to raise the training at ward meetings. While it is difficult to identify the impact of these individual and local contextual mediators, we suggest that the professional leading these initiatives needs to know the key people in their own organisation. Despite the relative success of delivering this initiative in one hospital, we accept the need for replication in other hospitals, and formal evaluation of the effectiveness of the intervention to actually improve children’s experience of care as inpatients. The transferability of the intervention to other hospitals will help to identify which intervention ingredients can influence organisational and cultural change on a broader scale. The hospital ward is a complex environment with a host of interacting variables. The premise of the logic model underpinning the intervention is that delivering training will lead to communication behaviour change in staff and that this will improve the disabled children’s experience of care as inpatients. This ultimate outcome could be measured using questionnaires developed to measure children’s experience as inpatients.26 27 However, before proceeding to such a study, we need to refine the intervention content and be confident in the delivery strategies that it is feasible for the training to be delivered in hospital children’s wards.

come could be measured using questionnaires developed to measure children’s experience as inpatients.26 27 However, before proceeding to such a study, we need to refine the intervention content and be confident in the delivery strategies that it is feasible for the training to be delivered in hospital children’s wards. Our training was designed to specifically address challenges to communication that arise on a paediatric ward. The group learning resources offered by disability matters are a potentially more comprehensive means to address the learning needs of professionals working with disabled children and young people across different settings.28 A Danish study incorporated communication skills into clinical practice using face-to-face methods in a 3-day course. We suggest that this time commitment is unlikely to be realistic in many acute settings.29 A strength of this training package is that it was designed purposefully to be short, sustainable and deliverable with minimal resources.

d communication skills into clinical practice using face-to-face methods in a 3-day course. We suggest that this time commitment is unlikely to be realistic in many acute settings.29 A strength of this training package is that it was designed purposefully to be short, sustainable and deliverable with minimal resources. A report by the Care Quality Commission (CQC) in England highlighted inequalities of inpatient experience for children with physical, learning or mental health needs.30 There was a call for action by representatives of the CQC, NHS England, professional bodies (RCPCH) and third sector advocacy groups (Young Minds). The importance of good communication was emphasised. These inequalities have persisted despite international and national legislation and ground level hospital policies and procedures. This agenda can be lost during the day-to-day ward pressures, but this is not acceptable. This intervention provides a low cost, fixed time, minimal resource option to raise awareness at organisational level and provides staff with training that is based on parental and child experiences, to motivate behaviour change. We have shown that it is feasible to deliver and that it was well received by staff. Early feedback was encouraging and indicated a real desire among staff to improve their communication skills. The next stage will be to test the transferability to other settings and to formally evaluate its impact. We believe that there is much potential for this intervention to improve the hospital inpatient experience of disabled children.

raging and indicated a real desire among staff to improve their communication skills. The next stage will be to test the transferability to other settings and to formally evaluate its impact. We believe that there is much potential for this intervention to improve the hospital inpatient experience of disabled children. The authors thank all parents involved in developing the intervention. We are particularly grateful to Bel McDonald, Julia Melluish and Mike Hurley who were interviewed for the videos, Jon Toomey for producing the films and Sheri Ostler for enabling access to the Parent and Carer Experience group. Thank you also to staff of Bramble Ward, Royal Devon and Exeter NHS Foundation Trust that participated in the study, and to the Exeter College students that produced the poster with Katharine Fitzpatrick our Family Involvement Coordinator. We are also grateful to Sharon Blake for helping to finalise the manual and videos. Contributors: All authors made substantive intellectual contributions to the study. RG and CM drafted the manuscript, all coauthors reviewed and revising it critically for important intellectual content and approved the final version. Funding: The study was part funded by a small grant from the Royal Devon & Exeter Foundation NHS Trust. PenCRU receives financial support from the charity Cerebra and the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care of the South West Peninsula (PenCLAHRC).

Funding: The study was part funded by a small grant from the Royal Devon & Exeter Foundation NHS Trust. PenCRU receives financial support from the charity Cerebra and the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care of the South West Peninsula (PenCLAHRC). Disclaimer: The views expressed in this publication are those of the authors and do not necessarily represent the views of Cerebra, the NHS, the NIHR or the Department of Health. Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known? A quarter of global neonatal deaths are due to birth asphyxia. The majority of these deaths occur in low-resource settings and are preventable. Neonatal resuscitation training (NRT) of birth attendants using mannequins result in improved knowledge and skills needed for resuscitation. Translation of NRT into improved neonatal outcomes and the effect estimates of improvements need to be re-evaluated and updated. What this study adds? This meta-analysis assessed the impact of NRT on stillbirths, 1-day neonatal mortality, 7-day neonatal mortality, 28-day neonatal mortality and perinatal mortality. NRT resulted in significant reduction in stillbirths and early neonatal mortality. However, continuum of care is needed for mortality reduction from day 7 to 28. Future studies also need to establish the best combination of settings, trainee characteristics and training frequency to sustain the existing effect on perinatal mortality reduction. Introduction Approximately a quarter of f million neonatal deaths worldwide are as a result of birth asphyxia.1 A large majority of these deaths occur in low-resource settings and are preventable. Approximately 5%–10% of newborns require some support to adapt to the extrauterine environment and to establish regular respiration.1 2 Simple resuscitative measures are often enough to resuscitate newborns that may even appear to be lifeless at birth. Studies have shown that essential newborn care has been effective in reducing stillbirths (SB).3

rns require some support to adapt to the extrauterine environment and to establish regular respiration.1 2 Simple resuscitative measures are often enough to resuscitate newborns that may even appear to be lifeless at birth. Studies have shown that essential newborn care has been effective in reducing stillbirths (SB).3 In developing countries, measures to improve resuscitative efforts through training of basic steps of neonatal resuscitation are expected to reduce birth asphyxia and neonatal mortality. Numerous studies have suggested that imparting neonatal resuscitation training (NRT) to healthcare providers involved in delivery process and handling of newborns has the potential to save newborn lives in low-income and middle-income settings4–10 Improvements in knowledge and skills of trainees following training programme in resource-limited settings have been reviewed. However, the impact on perinatal mortality outcomes has not been updated in last 5 years.9 The effect estimates of mortality reduction as a result of training of healthcare providers involved in delivery process and handling of newborns needs to be updated to inform hospital administrators and policy-makers the importance of investing in NRT to sustain and improve neonatal survival. A previous systematic review and meta-analysis11 assessed knowledge, skills, neonatal morbidity, neonatal mortality in first 7 days after birth and from day 8 to 28. However, it did not include outcomes of stlillbirth, 1-day neonatal mortality or perinatal mortality which has been included in our review.

atal survival. A previous systematic review and meta-analysis11 assessed knowledge, skills, neonatal morbidity, neonatal mortality in first 7 days after birth and from day 8 to 28. However, it did not include outcomes of stlillbirth, 1-day neonatal mortality or perinatal mortality which has been included in our review. The objective of this review is to assess the impact of NRT programme in reducing stillbirths, 1-day neonatal mortality, 7-day neonatal mortality, 28-day neonatal mortality and perinatal mortality. Materials and methods Inclusion criteria Types of studies We included relevant randomised, quasi-randomised controlled trials, interrupted time series studies and before–after studies regardless of language or publication status. Types of participants (population) trained We considered studies where NRT was provided to healthcare providers (including neonatologists, physicians, nurses, interns, midwives, traditional/community birth attendants, auxillary nurse midwives, village health workers, paramedics) involved in delivery process and handling of newborns in a community (home-based, rural and village clusters) or a hospital (including district hospitals, health centres, dispensaries, teaching/university hospitals, regional hospital, delivery/health centres, local hospitals and tertiary care hospital) setting.

medics) involved in delivery process and handling of newborns in a community (home-based, rural and village clusters) or a hospital (including district hospitals, health centres, dispensaries, teaching/university hospitals, regional hospital, delivery/health centres, local hospitals and tertiary care hospital) setting. Types of interventions and comparison Studies in which any NRT was compared with a control group (that received no NRT) or compared with data before the study (pre-NRT vs post-NRT) were included. For this purpose, we considered any NRT programme of healthcare professionals, including the American Academy of Pediatrics’ (AAP) Neonatal Resuscitation Program (NRP), Helping Babies Breathe (HBB) or any other training programme that had NRP or HBB as a clearly mentioned component of training methodology. Types of outcomes measures We included following outcomes in the review:Stillbirths: defined as number of deaths prior to complete expulsion or extraction of products of conception from its mother. Fresh stillbirth: clinically defined as those deaths with no signs of life at any time after birth and without any signs of maceration. 1-day neonatal mortality: defined as number of deaths in first 24 hours of life 7-day neonatal mortality: defined as number of deaths in first 7 days of life Perinatal mortality: defined as number of stillbirths and deaths in the first week of life. 28-day neonatal mortality: defined as number of deaths in the first 28 days of life.

1-day neonatal mortality: defined as number of deaths in first 24 hours of life 7-day neonatal mortality: defined as number of deaths in first 7 days of life Perinatal mortality: defined as number of stillbirths and deaths in the first week of life. 28-day neonatal mortality: defined as number of deaths in the first 28 days of life. Search strategy We searched following electronic databases from inception to July 2016: MEDLINE (PubMed), The Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library); Education Resources Information Centre (ERIC), Web of Science, Science Citation Index and Scientific Electronic Library Online. The search strategies for PubMed and CENTRAL can be found in supplementary files S1 and S2 respectively. We also searched for ongoing trials at www.clinicaltrials.gov and www.controlled-trials.com. We searched published abstracts of conferences and examined bibliographies of retrieved articles for additional studies. We contacted and requested experts and authors in this field to provide possible unpublished work. Study selection and data extraction Screening of studies Two reviewers (MNK and AB) independently examined studies identified by literature search; discarded articles that did not fulfil the inclusion criteria and assessed full texts of all relevant articles for inclusion. A third reviewer (AP) resolved disagreement among the primary reviewers.

xtraction Screening of studies Two reviewers (MNK and AB) independently examined studies identified by literature search; discarded articles that did not fulfil the inclusion criteria and assessed full texts of all relevant articles for inclusion. A third reviewer (AP) resolved disagreement among the primary reviewers. Data extraction and management For all studies that fulfilled the inclusion criteria, two reviewers (KK, SB) extracted data (table 1 and 2). Third review author (AP) cross-checked the data and resolved discrepancies. For studies where required data was lacking or could not be calculated, we requested the corresponding author for details. Table 1 Characteristic of included studies Sr. No. Author Country Study design Study period Funding 1 Bang et al20 India RCT 36 months (1995–1998) Ford Foundation USA The John D & Catherin T MacArthur Foundation USA 2 Ariawan et al* 8 Indonesia Pre–Post training NR  NR 3 Carlo et al17** Argentina, Democratic Republic of Congo, Guatemala, India, Pakistan and Zambia Pre–Post training and RCT 42 months (ENC: Mar 2005 and Feb 2007; NRP: Jul 2006–Aug 2008) NICHD, Global Network for Women’s and Children’s Health Research Bill & Melinda Gates Foundation 4 Carlo et al 18 Argentina, Democratic Republic of Congo, Guatemala, India, Pakistan and Zambia Pre–Post training and RCT 42 months (ENC: Mar 2005 and Feb 2007; NRP: Jul 2006–Aug 2008) NICHD, Global Network for Women’s and Children’s Health Research, Bill & Melinda Gates Foundation

2 Ariawan et al* 8 Indonesia Pre–Post training NR  NR 3 Carlo et al17** Argentina, Democratic Republic of Congo, Guatemala, India, Pakistan and Zambia Pre–Post training and RCT 42 months (ENC: Mar 2005 and Feb 2007; NRP: Jul 2006–Aug 2008) NICHD, Global Network for Women’s and Children’s Health Research Bill & Melinda Gates Foundation 4 Carlo et al 18 Argentina, Democratic Republic of Congo, Guatemala, India, Pakistan and Zambia Pre–Post training and RCT 42 months (ENC: Mar 2005 and Feb 2007; NRP: Jul 2006–Aug 2008) NICHD, Global Network for Women’s and Children’s Health Research, Bill & Melinda Gates Foundation 5 Gill et al21 Zambia Prospective, cluster randomised and controlled effectiveness study 30 months (Jun 2006–Nov 2008) Boston University and The Office of Health and Nutrition of The United State Agency for International Development AAP Unicef 6 Zhu et al26 China Perspective study, pre–post training (traditional resuscitation vs NRPG) 24 months (1993–1995)  NR 7 Deorari et al24 India Pre–post training ( NR Laerdal Foundation Norway 8 Jeffery et al28 Macedonia Pre–Post training 60 months (1997–2001) International Project Unit, Ministry of Health, Macedonia IDA Credit, World Bank 9 Vakrilova et al30 Bulgeria Pre–Post training ( 48 months (2000–2003)  NR 10 O’Hare et al25 Uganda Pre–Post training (historic group vs NRP pilot) 1 month (Dec 2001–Jan 2002) Child Advocacy International 11 Opiyo et al19 Kenya Pre–Post training NR Laerdal Foundation for Acute Medicine Wellcome Trust Senior Research Fellowship Award

9 Vakrilova et al30 Bulgeria Pre–Post training ( 48 months (2000–2003)  NR 10 O’Hare et al25 Uganda Pre–Post training (historic group vs NRP pilot) 1 month (Dec 2001–Jan 2002) Child Advocacy International 11 Opiyo et al19 Kenya Pre–Post training NR Laerdal Foundation for Acute Medicine Wellcome Trust Senior Research Fellowship Award 12 Boo31 Malaysia Pre–Post training, prospective observational study 100 months (Sep 1996–Dec 2004) Perinatal Society of Malaysia 13 Sorensen et al29 Tanzania Prospective study, Pre–Post training 14 weeks (Jul 2008–Nov 2008) Danish Society of Obstetrics and Gynecology 14 Hole et al32 Malawi, Africa Pre–Post training 30 months (Jun 2007–Dec 2009) Stanford University School of Medicines, Medical Scholars Research Program Department of Community Relations at Lucil Packard Children’s Hospital 15 Msemo et al22 Tanzania Pre–Post training 30 months (2009–2013) AAP Laerdal Foundation for Acute Medicine 16 Goudar et al23 India Pre–Post training (pretraining vs post HBB) 12 months (Oct 2009–Sep 2010) AAP Global Implementation Task Force HBB Program, Laerdal Foundation for Acute Medicine, Stavanger Norway 17 Vossius et al77 Tanzania Pre–Post training (pretraining vs post HBB) 24 months (Feb 2010–Jan 2012) Laerdal Foundation for Acute Medicine and Municipality of Stavanger Norway Research Department of HLH, Tanzania 18 Ashish et al*** Nepal Pre–Post training (pretraining vs post HBB) 15 months (Jul 2012–Sep 2013) Laerdal Foundation for Acute Medicine Swedish Society of Medicine

17 Vossius et al77 Tanzania Pre–Post training (pretraining vs post HBB) 24 months (Feb 2010–Jan 2012) Laerdal Foundation for Acute Medicine and Municipality of Stavanger Norway Research Department of HLH, Tanzania 18 Ashish et al*** Nepal Pre–Post training (pretraining vs post HBB) 15 months (Jul 2012–Sep 2013) Laerdal Foundation for Acute Medicine Swedish Society of Medicine 19 Bellad et al27 Kenya, India (Belgaum, Nagpur) Pre–Post training (pretraining vs post HBB) 24 months (Nov 2011–Oct 2013) NORAD Laerdal Foundation and NICHD 20 Patel et al*** India (Nagpur) Pre–Post training (pre-training vs post HBB) 24 months (Nov 2011–Oct 2013) NORAD Laerdal Foundation and NICHD *Data for this study has been taken from Lee et al8. **Data for very low birth weight (<1500 g). ***Unpublished data obtained via personal communication with the author AAP, American Academy of Pediatrics; ENC, essential newborn care; HBB, helping babies breathe; NICHD, National Institute of Child and Human Development; NR, not reported; NRPG, Neonatal Resuscitation Program Guidelines; RCT, randomised control trial. Assessment of risk of bias in included studies Two authors (SB, KK) independently assessed risk of bias for each study using criteria suggested by Cochrane Effective Practice and Organization of Care (EPOC)12 and using criteria outlined in Chapter 8 of Cochrane Handbook for Systematic Reviews of Interventions.13 Disagreements were resolved by discussion with the third reviewer (MNK).

) independently assessed risk of bias for each study using criteria suggested by Cochrane Effective Practice and Organization of Care (EPOC)12 and using criteria outlined in Chapter 8 of Cochrane Handbook for Systematic Reviews of Interventions.13 Disagreements were resolved by discussion with the third reviewer (MNK). Data analysis Measures of treatment effect We conducted meta-analysis and reported pooled statistics as risk ratios (RR) with 95% confidence interval (CIs) for dichotomous data. We followed recommendations of the Cochrane Handbook for Systematic Reviews of Interventions Sections 9.2 and 9.4 for measuring the effects.13 Assessment of heterogeneity We assessed heterogeneity amongst studies by inspecting forest plots for the overlap of confidence intervals, analysed statistical heterogeneity through Χ2 test (P value >0.10) and quantified through I2 statistics(Chapter 9.5 of Cochrane Handbook for Systematic Reviews).13 We regarded heterogeneity as substantial if in the Χ2 test for heterogeneity there was either I2>50%, or P value <0.10. We interpreted I2 values between 0% and 40% as possibly unimportant, 30% and 60% as possibly significant, 50% and 90% as possibly substantial and 75% and 100% as possibly considerable. Assessment of reporting bias We used funnel plots for assessment of publication bias if ten or more studies were included in a meta-analysis.

Assessment of heterogeneity We assessed heterogeneity amongst studies by inspecting forest plots for the overlap of confidence intervals, analysed statistical heterogeneity through Χ2 test (P value >0.10) and quantified through I2 statistics(Chapter 9.5 of Cochrane Handbook for Systematic Reviews).13 We regarded heterogeneity as substantial if in the Χ2 test for heterogeneity there was either I2>50%, or P value <0.10. We interpreted I2 values between 0% and 40% as possibly unimportant, 30% and 60% as possibly significant, 50% and 90% as possibly substantial and 75% and 100% as possibly considerable. Assessment of reporting bias We used funnel plots for assessment of publication bias if ten or more studies were included in a meta-analysis. Data synthesis and analysis We analysed the data using Review Manager V.5.3 software.14 We conducted meta-analyses for individual studies and reported pooled statistics as relative risk (RR) between experimental and control groups with 95% CI. We explored possible clinical and methodological reasons for heterogeneity, and in the presence of significant heterogeneity, we carried out sensitivity analysis and employed inverse-variance method with Random-effects model. We did not pool randomised and non-randomised (pre–post NRT) studies in the same meta-analysis.

e explored possible clinical and methodological reasons for heterogeneity, and in the presence of significant heterogeneity, we carried out sensitivity analysis and employed inverse-variance method with Random-effects model. We did not pool randomised and non-randomised (pre–post NRT) studies in the same meta-analysis. Summary of findings table We created ‘summary of findings’ (SoF) table using five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of a body of evidence. We used methods and recommendations described in Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions13 using GRADEpro software.15 GRADE working Group grades of evidence were used in the SoF.16 Results Search results We identified 148 records through database searching and 11 records through other sources. After initial screening on the basis of title and abstract, we assessed 47 full-text articles for eligibility and finally included 20 articles in the meta-analysis. The screening details are presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram (figure 1). Figure 1 Flow diagram of the study selection process. NRP, Neonatal Resuscitation Program.

Results Search results We identified 148 records through database searching and 11 records through other sources. After initial screening on the basis of title and abstract, we assessed 47 full-text articles for eligibility and finally included 20 articles in the meta-analysis. The screening details are presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram (figure 1). Figure 1 Flow diagram of the study selection process. NRP, Neonatal Resuscitation Program. Included studies Amongst included studies, two randomised trials addressed the efficacy of NRT in improving neonatal and perinatal outcomes, whereas 18 were pre–post studies. A full description of each study is included in table 1 and 2. All studies were from low-income and middle-income countries. Four studies were done in community setting, whereas 16 studies were carried in hospital setting. Table 2 Characteristic of included studies (training and outcomes) Sr. No. Author Training No. of births A: control/pre B: intervention/post Outcomes Criteria for delivery outcomes A: inclusion B: exclusion Duration Training setting Type Trainers Trainees Assessment 1 Bang et al20 NR Community (86 villages) A package of home-based neonatal care, health education includingENC Suction, stimulation Artificial respiration by mouth to mask and tube and mask NR Community birth attendants Village health workers NR A: 1159 B: 1005 1. SB 2. NMR: day 7 3. Perinatal mortality A: NR B: NR 2 Ariawan et al* 8 NR Community NRT includingUse of tube mask Refresher training at 3, 6 and 9 months, use of video Post resuscitation care

Suction, stimulation Artificial respiration by mouth to mask and tube and mask NR Community birth attendants Village health workers NR A: 1159 B: 1005 1. SB 2. NMR: day 7 3. Perinatal mortality A: NR B: NR 2 Ariawan et al* 8 NR Community NRT includingUse of tube mask Refresher training at 3, 6 and 9 months, use of video Post resuscitation care NR Midwives NR A: 9816 B: 16 053 1. SB 2. NMR: day 28 A: NR B: NR 3 Carlo et al17** 3 days Rural communities (7 sites in 6 countries for ENC; 88 for NRP) ENC sensitisation followed by in-depth NRT includingInitial resuscitation steps BMV AAP-trained trainer Research staff, either a physician or nurse Community birth attendants NR A: 359 B: 273 1. SB 2. FSB 3. NMR: day 7 4. PNMR A: BW <1500 g B: NR 4 Carlo et al18 3 days Rural communities (7 sites in six countries for ENC; 88 for NRP) ENC sensitisation followed by in-depth NRT includingInitial resuscitation steps BMV AAP-trained trainer Research staff, either a physician or nurse Community birth attendants NR A: 35 017 B: 29 715 1. SB 2. FSB 3. NMR: day 1 4. NMR: day 7 5. PNMR A: BW >1500 g B: NR 5 Gill et al21 2 weeks Community (rural district setting) NRT modified from AAP/AHA includingInitial steps PPV Use of manikins to demonstrate and practice skills NR 60 Community birth attendants/TBAs One to one skills assessment A: 1536 B: 1961 1. SB 2. NMR: day 7 3. NMR: day 28 4. PNMR A: NR B: NR 6 Zhu et al26 NR Hospital (1 hospital) NRPG curriculum established from AAP and AHA includingSuction BMV or ET ventilation Intubation

Community birth attendants NR A: 35 017 B: 29 715 1. SB 2. FSB 3. NMR: day 1 4. NMR: day 7 5. PNMR A: BW >1500 g B: NR 5 Gill et al21 2 weeks Community (rural district setting) NRT modified from AAP/AHA includingInitial steps PPV Use of manikins to demonstrate and practice skills NR 60 Community birth attendants/TBAs One to one skills assessment A: 1536 B: 1961 1. SB 2. NMR: day 7 3. NMR: day 28 4. PNMR A: NR B: NR 6 Zhu et al26 NR Hospital (1 hospital) NRPG curriculum established from AAP and AHA includingSuction BMV or ET ventilation Intubation NR Hospital birth attendants NR A: 1722 B: 4751 1. NMR: day 1 2. NMR: day 7 A: NR B: NR 7 Deorari et al24 NR Hospital (14 teaching hospitals) AAP/AHA-modified NRT withToT approach 2 Faculty member trainer per facility Hospital-based birth attendants No skills assessment A: 7070 B: 25 713 1. NMR: day 28 A: NR B: NR 8 Jeffery et al28 9 weeks Hospital (3 tertiary care, 13 district hospitals) A package of perinatal practices with NRT Australian-trained Macedonian teachers (doctors and nurses) Doctors and nurses MCQ, SAQ and OSCE (practical test) A: 69 840 B: 45 458 1. SB 2. NMR: day 7 3. PNMR A: NR B: NR 9 Vakrilova et al30 NR Hospital (delivery rooms of city hospitals) French–Bulgarian Program on NRT NR Neonatologist Obstetrician Midwives NR A: 67 948 B: 67 647 1. NMR: day 7 A: NR B: NR 10 O’Hare et al25 10 days training (5 days classroom+5 days delivery suite) Hospital (1 teaching hospital) NRT includingAirway management BMV Cardiac massage Use of manikins to demonstrate and practice skills

2 Faculty member trainer per facility Hospital-based birth attendants No skills assessment A: 7070 B: 25 713 1. NMR: day 28 A: NR B: NR 8 Jeffery et al28 9 weeks Hospital (3 tertiary care, 13 district hospitals) A package of perinatal practices with NRT Australian-trained Macedonian teachers (doctors and nurses) Doctors and nurses MCQ, SAQ and OSCE (practical test) A: 69 840 B: 45 458 1. SB 2. NMR: day 7 3. PNMR A: NR B: NR 9 Vakrilova et al30 NR Hospital (delivery rooms of city hospitals) French–Bulgarian Program on NRT NR Neonatologist Obstetrician Midwives NR A: 67 948 B: 67 647 1. NMR: day 7 A: NR B: NR 10 O’Hare et al25 10 days training (5 days classroom+5 days delivery suite) Hospital (1 teaching hospital) NRT includingAirway management BMV Cardiac massage Use of manikins to demonstrate and practice skills NR 5 members of nursing staff NR A: 1296 B: 1046 1. SB A: NR B: NR 11 Opiyo et al19 1 day Hospital (1 maternity hospital) NRT includingInitial steps BMV (use of bag valve mask device) CC Use of manikins to demonstrate and practice skills Instructor completed Kenya Resuscitation Council Advanced Life Support Generic Instructor Course Nurse/midwives MCQ and formal test scenario evaluating skills A: 4084 B: 4302 1. SB 2. NMR: day 28 A: NR B: NR 12 Boo31 NR Hospital AAP-NRT tailored to local needs includingInitial steps BMV CC ET ToT approach, a national-level training programme 37 Core instructors Doctors and nurses 14 575 Doctors Nurses Medical assistants Medical students

Instructor completed Kenya Resuscitation Council Advanced Life Support Generic Instructor Course Nurse/midwives MCQ and formal test scenario evaluating skills A: 4084 B: 4302 1. SB 2. NMR: day 28 A: NR B: NR 12 Boo31 NR Hospital AAP-NRT tailored to local needs includingInitial steps BMV CC ET ToT approach, a national-level training programme 37 Core instructors Doctors and nurses 14 575 Doctors Nurses Medical assistants Medical students Written and practical test A: 541 721 B: 465 140 1. SB 2. NMR: day 28 3. PNMR A: NR B: NR 13 Sorensen et al29 2 days Hospital (1 referral hospital) ALSO a widespread EmONCUse of manikins to demonstrate and practice skills NR High-level and mid-level staff involved in delivery NR A: 577 B: 565 1. SB A: BW >1000 g B: Missing data 14 Hole et al32 1 day Hospital (1 university hospital and 1 referral hospital) AAP modified NRT to includeInitial steps BMV CC and special consideration Use of manikins to demonstrate and practice skills Paediatrics residents from Stanford University Physician Clinical officers Midwives Survey covering knowledge, skills and attitude A: 3449 B: 3515 1. NMR: day 28 A: NR B: NR 15 Msemo et al22 1 day Hospital (3 referral hospitals, 4 regional hospitals and 1 district hospital) HBB training includingStimulation Suctioning Face and mask ventilation ToT approach Use of simulators for hands on practice FBOS training—reported by 1 site

Survey covering knowledge, skills and attitude A: 3449 B: 3515 1. NMR: day 28 A: NR B: NR 15 Msemo et al22 1 day Hospital (3 referral hospitals, 4 regional hospitals and 1 district hospital) HBB training includingStimulation Suctioning Face and mask ventilation ToT approach Use of simulators for hands on practice FBOS training—reported by 1 site 40 Trainers Hospital birth attendants Practical test A: 8124 B: 78 500 1. SB 2. FSB 3. NMR: day 1 A: BW >750 g for live birth BW >1000 g for FSB 16 Goudar et al23 1 day Hospital (primary health centres and rural and urban hospitals) HBB–AAP-based NRTInitial steps Stimulation Suctioning BMV ToT model Paired teaching Use of manikins to demonstrate and practice skills 18 Master trainers trained by AAP Physicians and nurses 599 Birth attendants Written and verbal MCQ, BMV by demonstration—OSCE A: 4187 B: 5411 1. SB 2. FSB 3. NMR: day 28 A: GA >28 wks B: NR 17 Vossius et al77 1 day Hospital (1 tertiary hospital) HBB–AAP-based NRT includingBNC and resuscitation Simulation-based training using manikins ToT approach 40 Master trainers Hospital-based birth attendants Knowledge and technical skills A: 4876 B: 4734 1. FSB 2. NMR: day 7 A: NR B: NR 18 Ashish et al*** 2 days Hospital (1 tertiary hospital) HBB–AAP-based NRT with QIC; train the trainer model, paired teachingSkills and practice ToT model Use of manikins to demonstrate and practice skills NR Obstetricians Anaesthesiologist Medical doctors Students Nurse/midwives

40 Master trainers Hospital-based birth attendants Knowledge and technical skills A: 4876 B: 4734 1. FSB 2. NMR: day 7 A: NR B: NR 18 Ashish et al*** 2 days Hospital (1 tertiary hospital) HBB–AAP-based NRT with QIC; train the trainer model, paired teachingSkills and practice ToT model Use of manikins to demonstrate and practice skills NR Obstetricians Anaesthesiologist Medical doctors Students Nurse/midwives NR A: 9588 B: 15 520 1. SB 2. FSB 3. NMR: day 1 4. PNMR A: GA >22 wks B: NR 19 Bellad et al27 3 days Hospital (39 primary, 21 secondary and 11 tertiary facilities) HBB–AAP-based NRT includingInitial steps Stimulation, suctioning BMV Refresher training QI activities ToT model Paired teaching Use of manikins to demonstrate and practice skills Neonatologists Paediatricians Obstetricians Nurses Hospital-based birth attendantsPaediatricians Obstetricians Physicians Residents Nursing staff Medical assistants MCQ, OSCE for skills assessment A: 15 232 B: 15 985 1. FSB 2. NMR: day 1 3. NMR: day 7 4. NMR: day 28 5. PNMR A: BW >1500 g B: BW unknown,<1500, >5500 and MSB 20 Patel et al*** 3 days Hospital (2 primary, 4 secondary HTML validation and 7 tertiary facilities) HBB–AAP-based NRT includingInitial steps Stimulation, suctioning BMV Refresher training and QI activities ToT model Paired teaching Use of manikins to demonstrate and practice skills Neonatologists Paediatricians Obstetricians Nurses eHospital-based birth attendantsPaediatricians Obstetricians Physicians Residents Nursing staff Medical assistants MCQ, OSCE for skills assessment A: 38 078 B: 40 870 1. SB 2. FSB 3. NMR:  day 1 4. NMR: day 7 6. PNMR A: GA >20 wks B: NR *Data for this study has been taken from Lee et al8.

Use of manikins to demonstrate and practice skills Neonatologists Paediatricians Obstetricians Nurses eHospital-based birth attendantsPaediatricians Obstetricians Physicians Residents Nursing staff Medical assistants MCQ, OSCE for skills assessment A: 38 078 B: 40 870 1. SB 2. FSB 3. NMR:  day 1 4. NMR: day 7 6. PNMR A: GA >20 wks B: NR *Data for this study has been taken from Lee et al8. **Data for very low-birth weight (<1500 g). ***Unpublished data obtained via personal communication with the author AAP, American Academy of Pediatrics; AHA, American Heart Association; ALSO, Advanced Life Support in Obstetrics; BMV, bag and mask ventilation; BW, birth weight; CC, chest compression; EmONC, Emergency Obstetrics & Neonatal Care; ENC, essential newborn care; ET, endotracheal tube; FBOS, frequent brief onsite simulation; FSB, fresh stillbirth; GA, gestational age, HBB, helping babies breathe; MCQ, multiple choice questions; NICHD, National Institute of Child and Human Development; NMR, neonatal mortality rate; NORAD, Norwegian Agency for Development Cooperation; NR, not reported; NRPG, Neonatal Resuscitation Program Guidelines; NRT, neonatal resuscitation training; OSCE, objective structured clinical evaluation; PNMR, perinatal mortality rate; PPV, positive pressure ventilation; QI, quality improvement; QIC, quality improvement cycle; RCT, randomised control trial; SAQ, short answer questions; SB, all stillbirth; TBA, traditional birth attendants; ToT, training of trainer; wks, weeks.

objective structured clinical evaluation; PNMR, perinatal mortality rate; PPV, positive pressure ventilation; QI, quality improvement; QIC, quality improvement cycle; RCT, randomised control trial; SAQ, short answer questions; SB, all stillbirth; TBA, traditional birth attendants; ToT, training of trainer; wks, weeks. Carlo et al17 18 assessed baseline perinatal outcomes, then imparted Essential Newborn Care (ENC) training to all which also included basic steps of NRT. They then randomised all clusters that had received ENC training into two groups. One group received an in-depth NRT while the other group did not (control group). For this study we evaluated the pre-ENC outcome of all clusters and compared them to outcomes of those clusters that received ENC +post ENC in-depth NRT. We therefore did not include this study in the NRT versus control analysis because the control group had also received NRT as a part of ENC training. The study from Kenya had a complex design of randomisation of health workers to two groups—early training (phase I) or late training (phase II) and did not include a control group without training.19 Therefore, we analysed this study as before–after study where the rate of stillbirths prior to any training were compared with the rate of stillbirths after all phases of training.

f health workers to two groups—early training (phase I) or late training (phase II) and did not include a control group without training.19 Therefore, we analysed this study as before–after study where the rate of stillbirths prior to any training were compared with the rate of stillbirths after all phases of training. Participants of the NRT programme differed across studies and included village health workers, community birth attendants,17 18 20 community birth attendants/traditional birth attendants,21 hospital-based birth attendants,19 22–26 or hospital-based birth attendants including high-level and mid-level staff/specialists.27–34 Different types of training employed by studies included AAP, HBB or NRP curricula23 24 27 31 32 34 35 AAP/American Heart Association (AHA),21 24 26 basic neonatal resuscitation and ENC,17–19 25 home-based neonatal care, basic training with mouth to mask or tube and mask resuscitation,35 Advanced Life Support in Obstetrics (ALSO),29 Bulgarian program on NRT.30 The duration of NRT also differed acrossstudies. We also included two unpublished trials after permission from authors (tables 1 and 2). Excluded studies Studies that included interventions that did not qualify as NRT were excluded from the review. These included trainings in safe birthing techniques,36 Emergency Obstetric and Neonatal Care (EmONC),37 38 ENC,39–41promotion of antenatal care and maternal health education,42and newborn care intervention package.43

studies Studies that included interventions that did not qualify as NRT were excluded from the review. These included trainings in safe birthing techniques,36 Emergency Obstetric and Neonatal Care (EmONC),37 38 ENC,39–41promotion of antenatal care and maternal health education,42and newborn care intervention package.43 Other interventions that did not qualify as NRT44–50 or included interventions like neonatal intensive care unit/special neonatal care unit training51 52 were also excluded. Studies in which desired outcomes (fetal and neonatal outcome) were not assessed,53–58 or only trainees/training outcomes were assessed,59–73 were also excluded from the analysis. Some studies that were subgroups of larger studies like Ersdal et al.74 75 (subgroup of Msemo et al22), Matendo et al76(subgroup of Carlo et al18), Matendo et al76 and Vossius et al77 (subgroup of Msemo et al22) were also not included. However, Vossius et al77 was included in the analysis for outcomes where data from22 Msemo et al22 were not available. Risk of bias in included studies has been depicted in table 3. Table 3 Risk of bias assessment across studies

Some studies that were subgroups of larger studies like Ersdal et al.74 75 (subgroup of Msemo et al22), Matendo et al76(subgroup of Carlo et al18), Matendo et al76 and Vossius et al77 (subgroup of Msemo et al22) were also not included. However, Vossius et al77 was included in the analysis for outcomes where data from22 Msemo et al22 were not available. Risk of bias in included studies has been depicted in table 3. Table 3 Risk of bias assessment across studies Bang et al20 Carlo et al17 Carlo et al18 Gill et al21 Zhu et al26 Deorari et al24 Jeffery et al28 O’Hare et al25 Opiyo et al19 Boo31 Sorensen et al29 Hole et al32 Msemo et al22 Goudar et al23 Vossius et al77 Ashish et al (Unpublished data) Bellard et al Patel et al (Unpublished data) Adequate sequence generation? High risk Low risk Allocation concealment? High risk Low risk Incomplete outcome data addressed? High risk Low risk Low risk Low risk Unclear risk Unclear risk Unclear risk Low risk Unclear risk Low risk Low risk High risk Unclear risk Unclear risk Low risk Low risk Low risk Low risk Free of selective reporting? Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Low risk Free of other bias? Unclear risk Low risk Low risk Low risk Low risk Low risk Low risk Unclear risk Unclear risk Uncleat risk Low risk Unclear risk Low risk Unclear risk High risk Low risk High risk Unclear risk Baseline outcomes similar? Low risk Low risk Unclear risk Unclear risk Unclear risk Unclear risk unclear risk Uncleat risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Free of contamination? Low risk Low risk Low risk Low risk Unclear risk Low risk Low risk High risk Low risk High risk Low risk Low risk High risk Low risk Low risk Low risk Baseline characteristics similar? Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Unclear risk Low risk Low risk Unclear risk Unclear risk Low risk Unclear risk High risk Low risk Low risk Effects of interventions Neonatal and perinatal outcomes were reported in majority of included studies. The overall analysis showed a trend towards reduction in neonatal deaths, early neonatal deaths, perinatal deaths and stillbirths with NRT; most of which are statistically significant.

isk High risk Low risk Low risk Effects of interventions Neonatal and perinatal outcomes were reported in majority of included studies. The overall analysis showed a trend towards reduction in neonatal deaths, early neonatal deaths, perinatal deaths and stillbirths with NRT; most of which are statistically significant. NRT verses control The meta-analysis for NRT verses control shows that NRT decreases the risk of all stillbirths by 21% (RR 0.79, 95% CI 0.44 to 1.41; participants=5661; studies=2; I2=67%) (figure 2), 7-day neonatal deaths by 47% (RR 0.53, 95% CI 0.38 to 0.73; participants=5518; studies=2; I2=0%) (figure 3), 28-day neonatal deaths by 50% (RR 0.50, 95% CI 0.37 to 0.68; participants=5442; studies=2; I2=0%) (figure 4), and perinatal deaths by 37% (RR 0.63, 95% CI 0.42 to 0.94; participants=5584; studies=2; I2=68%)(figure 5). The effect was significant for ay 7-day neonatal mortality, 28-day neonatal mortality and perinatal mortality. Significant heterogeneity was observed in analysis of total stillbirths and perinatal mortality. Figure 2 Forest plot comparing all SB between the NRT and the control groups. NRT, neonatal resuscitation training; SB, stillbirths. Figure 3 Forest plot comparing 7-day neonatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training. Figure 4 Forest plot comparing 28-day neonatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training. Figure 5 Forest plot comparing perinatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training.

Figure 3 Forest plot comparing 7-day neonatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training. Figure 4 Forest plot comparing 28-day neonatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training. Figure 5 Forest plot comparing perinatal mortality between the NRT and the control groups. NRT, neonatal resuscitation training. The grade of quality of evidence for the meta-analysis of the trials was moderate to high (table 4). Table 4 Summary of findings for NRT versus control groups Outcomes Anticipated absolute effects (95% CI) – risk with no NRP Anticipated absolute effects (95% CI) – risk with NRP Relative effect (95% CI) No of participants (studies) Quality of the evidence (GRADE) All stillbirth 29 per 1000 23 per 1000 (13 to 41) RR 0.79 (0.44 to 1.41) 5661 (2 RCTs) ⨁◯◯◯ Very low*† Fresh stillbirth Outcome not reported Outcome not reported Outcome not reported Outcome not reported ⨁◯◯◯ Very low‡ 1-day neonatal mortality Outcome not reported Outcome not reported Outcome not reported Outcome not reported ⨁◯◯◯ Very low‡ 7-day neonatal mortality 39 per 1000 20 per 1000 (15 to 28) RR 0.53 (0.38 to 0.73) 5518 (2 RCTs) ⨁⨁⨁⨁ High 28-day neonatal mortality 49 per 1000 24 per 1000 (18 to 33) RR 0.50 (0.37 to 0.68) 5442 (2 RCTs) ⨁⨁⨁⨁ High Perinatal mortality 68 per 1000 43 per 1000 (29 to 64) RR 0.63 (0.42 to 0.94) 5584 (2 RCTs) ⨁⨁⨁◯ Moderate§ *I2 is 67% and the two trials were inconsistent in the direction of effect. Quality of evidence downgraded by two for inconsistency and imprecision (figure 2).

to 33) RR 0.50 (0.37 to 0.68) 5442 (2 RCTs) ⨁⨁⨁⨁ High Perinatal mortality 68 per 1000 43 per 1000 (29 to 64) RR 0.63 (0.42 to 0.94) 5584 (2 RCTs) ⨁⨁⨁◯ Moderate§ *I2 is 67% and the two trials were inconsistent in the direction of effect. Quality of evidence downgraded by two for inconsistency and imprecision (figure 2). †The 95% CI of the pooled estimate includes null effect. Quality of evidence downgraded by one for imprecision (figure 2). ‡No evidence to support or refute. §Though I2 is 68%, the 95% CI of the pooled estimate does not include the null effect. Quality of evidence downgraded by one for inconsistency (figure 5). NRT, neonatal resuscitation training; RCTs, randomised controlled trial; RR, risk ratio.

†The 95% CI of the pooled estimate includes null effect. Quality of evidence downgraded by one for imprecision (figure 2). ‡No evidence to support or refute. §Though I2 is 68%, the 95% CI of the pooled estimate does not include the null effect. Quality of evidence downgraded by one for inconsistency (figure 5). NRT, neonatal resuscitation training; RCTs, randomised controlled trial; RR, risk ratio. Post-NRT verses pre-NRT The meta-analysis of post-NRT verses pre-NRT shows that post-NRT decreases the risk of all stillbirths by 12% (RR 0.88, 95% CI 0.83 to 0.94; participants=1 425 540; studies=12; I2=47%, figure 6), fresh stillbirths by 26% (RR 0.74, 95% CI 0.61 to 0.90; participants=296 819; studies=8; I2=84%, figure 7), 1-day neonatal mortality by 42% (RR 0.58, 95% CI 0.42 to 0.82; participants=280 080; studies=6; I2=89%, figure 8), 7-day neonatal mortality by 18% (RR 0.82, 95% CI 0.73 to 0.93; participants=360 383; studies=7; I2=71%, figure 9), 28-day neonatal mortality by 14% (RR 0.86, 95% CI 0.65 to 1.13; participants=1 116 463; studies=7; I2=95%, figure 10) and perinatal mortality by 18% (RR 0.82, 95% CI 0.74 to 0.91; participants=1 243 802; studies=6; I2=90%, figure 11). The changes were significant in all the outcomes; except 28-day neonatal mortality. Heterogeneity was significant in all outcomes except all stillbirths. We created a funnel plot for all stillbirths, which showed asymmetry, thereby indicating a publication bias (figure 12). Figure 6 Forest plot comparing all SB between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training; SB, stillbirths.

cribed.13 The total RNA concentration (ng/µL) and ratios (260/280 and 260/230) were measured using a NanoDrop ND-100 UV-vis spectrophotometer (NanoDrop Technologies, Delaware, USA) and the RNA integrity was assessed using the Agilent 2100 BioAnalyzer (Agilent Technologies, Edinburgh, UK) before and after concentration. RNA (2 µg) was reverse transcribed into cDNA using the Superscript double-stranded cDNA synthesis kit (Invitrogen, Paisley, UK) according to the manufacturer’s instructions. The double-stranded cDNA was purified using a GeneChip sample clean-up module (Invitrogen). The purified cDNA was then biotin labelled with the ENZO BioArray high-yield RNA transcript labelling kit (Affymetrix, High Wycombe, UK) and cleaned with a cRNA clean-up module (Invitrogen). Aliquots of labelled cRNA (20 µg) were fragmented at 94°C for 35 min and then hybridised to a Human Genome U133A GeneChip array for 16 hours rotating at 60 rpm at 45°C in a GeneChip Hybridization Oven 640 (Affymetrix). Each chip was washed and stained on a GeneChip Fluidics Station 450 (Affymetrix) and scanned using a GeneChip Scanner 3000 (Affymetrix) employing standard recommended protocols (Affymetrix).

Post-NRT verses pre-NRT The meta-analysis of post-NRT verses pre-NRT shows that post-NRT decreases the risk of all stillbirths by 12% (RR 0.88, 95% CI 0.83 to 0.94; participants=1 425 540; studies=12; I2=47%, figure 6), fresh stillbirths by 26% (RR 0.74, 95% CI 0.61 to 0.90; participants=296 819; studies=8; I2=84%, figure 7), 1-day neonatal mortality by 42% (RR 0.58, 95% CI 0.42 to 0.82; participants=280 080; studies=6; I2=89%, figure 8), 7-day neonatal mortality by 18% (RR 0.82, 95% CI 0.73 to 0.93; participants=360 383; studies=7; I2=71%, figure 9), 28-day neonatal mortality by 14% (RR 0.86, 95% CI 0.65 to 1.13; participants=1 116 463; studies=7; I2=95%, figure 10) and perinatal mortality by 18% (RR 0.82, 95% CI 0.74 to 0.91; participants=1 243 802; studies=6; I2=90%, figure 11). The changes were significant in all the outcomes; except 28-day neonatal mortality. Heterogeneity was significant in all outcomes except all stillbirths. We created a funnel plot for all stillbirths, which showed asymmetry, thereby indicating a publication bias (figure 12). Figure 6 Forest plot comparing all SB between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training; SB, stillbirths. Figure 7 Forest plot comparing fresh SB between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training; SB, stillbirths. Figure 8 Forest plot comparing 1-day neonatal mortality between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training.

Figure 6 Forest plot comparing all SB between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training; SB, stillbirths. Figure 7 Forest plot comparing fresh SB between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training; SB, stillbirths. Figure 8 Forest plot comparing 1-day neonatal mortality between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training. Figure 9 Forest plot comparing 7-day neonatal mortality between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training. Figure 10 Forest plot comparing 28-day neonatal mortality between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training. Figure 11 Forest plot comparing perinatal m between the post-NRT and the pre-NRT groups. NRT, neonatal resuscitation training. Figure 12 Funnel plot of comparison: Post-NRT verses pPre-NRT for all SB. NRT, neonatal resuscitation training; RR, risk ratio; SB, stillbirths. The quality of evidence for NRT verses control was very low for SB and 1-day neonatal mortality, high for 7-day and 28-day neonatal mortality and moderate for perinatal mortality (table 4). The quality of evidence for post-NRT verses pre-NRT was very low for all our outcomes (table 5). Table 5 Summary of findings for Post-NRT versus Pre-NRT groups

The quality of evidence for NRT verses control was very low for SB and 1-day neonatal mortality, high for 7-day and 28-day neonatal mortality and moderate for perinatal mortality (table 4). The quality of evidence for post-NRT verses pre-NRT was very low for all our outcomes (table 5). Table 5 Summary of findings for Post-NRT versus Pre-NRT groups Outcomes Anticipated absolute effects (95% CI) Risk with pre-NRP Anticipated absolute effects (95% CI) Risk with post-NRP Relative effect (95% CI) No of participants (studies) Quality of the evidence (GRADE) All stillbirths 8 per 1000 7 per 1000 (7 to 8) RR 0.88 (0.83 to 0.94) 1 425 540 (12 observational studies) ⨁◯◯◯ Very low*†‡ Fresh stillbirths 15 per 1000 11 per 1000 (9 to 13) RR 0.74 (0.61 to 0.90) 296 819 (8 observational studies) ⨁◯◯◯ Very low*†§ 1-day neonatal mortality 8 per 1000 5 per 1000 (4 to 7) RR 0.58 (0.42 to 0.82) 280 080 (6 observational studies) ⨁◯◯◯ Very low *¶ 7-day neonatal mortality 13 per 1000 11 per 1000 (9 to 12) RR 0.82 (0.73 to 0.93) 360 383 (7 observational studies) ⨁◯◯◯ Very low *† ** 28-day neonatal mortality 8 per 1000 7 per 1000 (5 to 9) RR 0.86 (0.65 to 1.13) 1 116 463 (7 observational studies) ⨁◯◯◯ Very low * †† Perinatal mortality 14 per 1000 12 per 1000 (10 to 13) RR 0.82 (0.74 to 0.91) 1 243 802 (6 observational studies) ⨁◯◯◯ Very low * §§ ¶¶ *Pre–post studies. Quality of evidence downgraded by one for risk of bias (table 1 and 2). † Studies differ in the settings, type of NRP, duration and type trainees. Quality of evidence downgraded by one for indirectness (table 1 and 2).

Outcomes Anticipated absolute effects (95% CI) Risk with pre-NRP Anticipated absolute effects (95% CI) Risk with post-NRP Relative effect (95% CI) No of participants (studies) Quality of the evidence (GRADE) All stillbirths 8 per 1000 7 per 1000 (7 to 8) RR 0.88 (0.83 to 0.94) 1 425 540 (12 observational studies) ⨁◯◯◯ Very low*†‡ Fresh stillbirths 15 per 1000 11 per 1000 (9 to 13) RR 0.74 (0.61 to 0.90) 296 819 (8 observational studies) ⨁◯◯◯ Very low*†§ 1-day neonatal mortality 8 per 1000 5 per 1000 (4 to 7) RR 0.58 (0.42 to 0.82) 280 080 (6 observational studies) ⨁◯◯◯ Very low *¶ 7-day neonatal mortality 13 per 1000 11 per 1000 (9 to 12) RR 0.82 (0.73 to 0.93) 360 383 (7 observational studies) ⨁◯◯◯ Very low *† ** 28-day neonatal mortality 8 per 1000 7 per 1000 (5 to 9) RR 0.86 (0.65 to 1.13) 1 116 463 (7 observational studies) ⨁◯◯◯ Very low * †† Perinatal mortality 14 per 1000 12 per 1000 (10 to 13) RR 0.82 (0.74 to 0.91) 1 243 802 (6 observational studies) ⨁◯◯◯ Very low * §§ ¶¶ *Pre–post studies. Quality of evidence downgraded by one for risk of bias (table 1 and 2). † Studies differ in the settings, type of NRP, duration and type trainees. Quality of evidence downgraded by one for indirectness (table 1 and 2). ‡ Publication bias detected in the funnel plot. Quality of evidence downgraded by one for publication bias (figure 12). § Although I2 is 84%, the effect estimates of all included studies do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 7).

† Studies differ in the settings, type of NRP, duration and type trainees. Quality of evidence downgraded by one for indirectness (table 1 and 2). ‡ Publication bias detected in the funnel plot. Quality of evidence downgraded by one for publication bias (figure 12). § Although I2 is 84%, the effect estimates of all included studies do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 7). ¶ Although I2 is 89%, the effect estimates of all the included studies (except Bellard et al.) do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 8). ** Although I2 is 71%, the effect estimates of all the included studies (except Bellard et al.) do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 9). †† I2 is 95% and the effect estimates cross the life of no effect. Quality of evidence downgraded by two for inconsistency and imprecision (figure 10). ‡‡The effect estimate crosses the line of no effect. Quality of evidence downgraded by one for imprecision (figure 10). §§ Although I2 is 90%, the effect estimates of all the included studies do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 11). ¶¶ Studies differ in setting, type of NRP and trainees. Quality of evidence downgraded by one for indirectness (table 1 and 2). NRP, Neonatal Resuscitation Program; NRT, neonatal resuscitation trainings; RR, risk ratio; SB, stillbirths.

§§ Although I2 is 90%, the effect estimates of all the included studies do not differ in the direction of effect. Quality of effect downgraded by one for inconsistency (figure 11). ¶¶ Studies differ in setting, type of NRP and trainees. Quality of evidence downgraded by one for indirectness (table 1 and 2). NRP, Neonatal Resuscitation Program; NRT, neonatal resuscitation trainings; RR, risk ratio; SB, stillbirths. Discussion This meta-analysis assessed the impact of any NRT programme either by itself or as a part of newborn care package on rates of stillbirths, perinatal mortality, all-cause neonatal mortality on day-1, up till day-7 and till 28th day after birth. We did not evaluate intrapartum-related neonatal deaths or asphyxia/cause-specific neonatal mortality. Mortality in neonates <7 days of life is a proxy measure for intrapartum-related deaths.43 78 Meta-analysis of before–after studies showed a significant reduction in all stillbirths by 12% (12 studies) and of FSB by 26% (8 studies). The reduction in fresh stillbirths can be attributed to NRT that helps in resuscitating neonates that appear lifeless at birth.17 18 Of 12 studies, seven studies reported a significant and one study reported a non-significant reduction in fresh stillbirths. However, a non-significant increase in risk of stillbirths was reported in three African studies which blunted the impact of NRT on reduction of stillbirths.

s that appear lifeless at birth.17 18 Of 12 studies, seven studies reported a significant and one study reported a non-significant reduction in fresh stillbirths. However, a non-significant increase in risk of stillbirths was reported in three African studies which blunted the impact of NRT on reduction of stillbirths. There was reduction in 1-day mortality of 42% (6 studies) and that of 7-day mortality was 18%. All studies included in the analysis (figures 8 and 9) showed a reduction with an exception of one study.27 Failure to observe reduction in mortality in Bellad et al could be due to two reasons. First, NRT was provided in diverse health systems within a short period of time. Second, mortality was not assessed in facilities where training was imparted but was measured in the population. The meta-analysis showed a non-significant reduction of 14% in 28-day mortality. Of the seven included studies only two studies reported a significant reduction in mortality. Resuscitation at delivery helps to reduce neonatal mortality in the first hour of birth when the neonate is at the highest risk of intrapartum-related deaths3 and the impact diminishes subsequently. For reduction of 28-day neonatal mortality, post-resuscitation specialised care for survivors is required and only NRT is unlikely to have the desired impact on 28-day neonatal mortality.79 80

st hour of birth when the neonate is at the highest risk of intrapartum-related deaths3 and the impact diminishes subsequently. For reduction of 28-day neonatal mortality, post-resuscitation specialised care for survivors is required and only NRT is unlikely to have the desired impact on 28-day neonatal mortality.79 80 Trials that randomise facilities to NRT versus controls (where NRT is not a standard practice) would be ideal to assess the reduction in neonatal mortality. Trials are also likely to result in higher impact as compared with before–after studies as other changes at health facilities or in communities during the time period of before–after studies can confound the results. Because NRT is a standard practice and randomising individuals or clusters to no resuscitation training is unethical, there were only two trials available for the meta-analysis.20 21 They showed a reduction of 7-day neonatal mortality and 28-day mortality by 47% (figure 3) and 50% (figure 4), respectively. The perinatal mortality reduced by 37% (figure 5) with no significant reduction in SB rates.

resuscitation training is unethical, there were only two trials available for the meta-analysis.20 21 They showed a reduction of 7-day neonatal mortality and 28-day mortality by 47% (figure 3) and 50% (figure 4), respectively. The perinatal mortality reduced by 37% (figure 5) with no significant reduction in SB rates. Previously, an expert panel published a systematic review for community-based studies and conducted a meta-analysis that evaluated whether NRT reduced all-cause neonatal mortality in th first 7 days of life. They reported a 38% reduction in mortality which is larger than the 18% (7 studies) reduction observed in the current meta-analysis. Our meta-analysis included community-based studies that resulted in a smaller effect size. Community-based studies (trials or before–after) report a smaller reduction effect on any day neonatal mortality.8 17 18 47 The reduction in effect size of neonatal mortality in these studies can arise due to several reasons. All births in the intervention community may not be attended by birth attendants trained in neonatal resuscitation, especially if it is a home delivery.81 82 Second, women may decide to deliver at facilities or homes outside communities where NRT has been imparted. Finally, assessing mortality outcomes in the community can be challenging. Another meta-analysis11 was published in Cochrane which evaluated outcomes such as knowledge, skills, neonatal morbidity, neonatal mortality in first 7 days after birth and from day 8 to 28. This analysis did not include stillbirths, 1-day neonatal mortality or perinatal mortality that was included in the current meta-analysis.

eta-analysis11 was published in Cochrane which evaluated outcomes such as knowledge, skills, neonatal morbidity, neonatal mortality in first 7 days after birth and from day 8 to 28. This analysis did not include stillbirths, 1-day neonatal mortality or perinatal mortality that was included in the current meta-analysis. The current meta-analysis consists largely of before–after studies with lack of concurrent control group that limits isolation of effect of resuscitation training alone from other changes at health facilities or in communities during the time period. Other limitation is lack of consistency of settings, duration of training, varying study designs and lack of consistent outcomes which contributed to substantial heterogeneity. Lack of subgroup analysis of type of health facilities may be perceived as a limitation. An improvement in mortality would be maximised in low-resource settings with poor quality of care. However, it is presumed that there is regular training of health workers in basic resuscitation skills in higher levels of care that would translate to higher quality of care. Our recent study83 84 that evaluated the knowledge and skills of trainees trained in HBB included 384 tertiary-level facilities in India. Only 3% of physicians and 5% of nurses were able to pass the pre-training bag and mask resuscitation skill assessment.84 Therefore, in the absence of reporting of pre-training skills of health workers in low-resource or high-resource settings or any indicator of quality of care, it would be erroneous to conduct a subgroup analysis based merely on resource settings and mostly will not change the results or the main message of this meta-analysis. We emphasise that despite the heterogeneity in settings, type of training, type of trainees, type of trainers and the duration of training, this study showed an improvement in mortality at and soon after birth.

merely on resource settings and mostly will not change the results or the main message of this meta-analysis. We emphasise that despite the heterogeneity in settings, type of training, type of trainees, type of trainers and the duration of training, this study showed an improvement in mortality at and soon after birth. To conclude, NRT resulted in reduction in stillbirths and improved survival of newborns. The impact on survival of newborns can be further improved by providing a continuum of care beyond 7 days which is not addressed by NRT alone. The meta-analysis performed showed beneficial effect of NRT in improving neonatal and perinatal outcomes. The models of training were not consistent across studies, with variations in training, trainee and setting. Generalisation of results of the pooled analysis to many currently available programme may not be appropriate. There was evidence of heterogeneity across studies in our meta-analyses; however, overall there is consistency in the direction of effect. This review identified several important limitations of the current evidence from included studies. Due to inadequate information about the methodology followed and variety of resuscitation programmes in included studies, the quality of the evidence was downgraded for risk of bias and indirectness resulting in inability to adequately assess the effects of this intervention. Conclusions Implications for practice This review shows that the implementation of NRT improves neonatal and perinatal outcomes.

This review identified several important limitations of the current evidence from included studies. Due to inadequate information about the methodology followed and variety of resuscitation programmes in included studies, the quality of the evidence was downgraded for risk of bias and indirectness resulting in inability to adequately assess the effects of this intervention. Conclusions Implications for practice This review shows that the implementation of NRT improves neonatal and perinatal outcomes. Implications for research Further good quality, multicentric randomised controlled trials addressing the role of NRT for improving neonatal and perinatal outcomes may be warranted. Impact of NRT on improving neonatal and perinatal outcomes as well as the best combination of settings and type of trainee should be established in future trials. More studies need to be done to assess the frequency with which NRT needs to be conducted to sustain the existing effect on perinatal mortality reduction. The authors wish to acknowledge Richard Kirubhakaran (Research Scientist, Cochrane South Asia, Prof B V Moses Centre for Evidence-Informed Healthcare & Health Policy, Christian Medical College, Vellore) for his inputs on meta-analysis and Lauren Arlington, Partner Healthcare, for her help in getting the full text of the articles required for this review.

ard Kirubhakaran (Research Scientist, Cochrane South Asia, Prof B V Moses Centre for Evidence-Informed Healthcare & Health Policy, Christian Medical College, Vellore) for his inputs on meta-analysis and Lauren Arlington, Partner Healthcare, for her help in getting the full text of the articles required for this review. Contributors: AP: conception of the work, design of the work, manuscript drafting with final approval of the version to be published. MNK: developed and run the search strategy, screened and selected studies, and did meta-analysis, GRADE assessment and manuscript drafting. KK and SB: involved in preparation of characteristic of studies table, data acquisition and manuscript drafting. AB: screening and selection of studies, data acquisition and manuscript drafting. Funding: This work was supported by Lata Medical Research Foundation, Nagpur, India (Grant no: LMRF/GRP02/072016). Competing interests: The authors AP and AB were investigators in two of the studies (Bellad et al and Patel et al) included in the meta-analysis. There were no other competing interest. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Invasive pneumococcal disease (IPD) caused by Streptococcus pneumoniae is a leading cause of pneumonia, meningitis and septicaemia worldwide. Globally, IPD is reported to cause about 11% (0.8 million) of all deaths in children less than 5 years of age annually. The overall burden of IPD is increased 40-fold in HIV-infected compared with HIV-uninfected children. However, mechanisms involved in host response during IPD are not yet fully understood. What this study hopes to add? We demonstrate for the first time differences in transcriptional profiles between HIV-infected and HIV-uninfected children with pneumococcal meningitis (as a homogeneous disease entity of invasive pneumococcal disease and healthy controls). We demonstrate increased expression in cases of genes regulating the innate immune response, leucocyte migration, glucose homeostasis and endothelial cell migration. Introduction Streptococcus pneumoniae infection is a leading cause of pneumonia, meningitis and septicaemia worldwide, and results in approximately 1 million deaths in children under the age of 5 years annually.1 The overall burden of invasive pneumococcal disease (IPD) is increased 40-fold in HIV-infected compared with HIV-uninfected children.2 Pneumococcal meningitis is a life-threatening disease with poor prognosis associated with neurologic complications and a high case-fatality ratio in African children, which is further increased by HIV coinfection.3

e pneumococcal disease (IPD) is increased 40-fold in HIV-infected compared with HIV-uninfected children.2 Pneumococcal meningitis is a life-threatening disease with poor prognosis associated with neurologic complications and a high case-fatality ratio in African children, which is further increased by HIV coinfection.3 Gene expression profiling during sepsis provides new insights into the host response to invasive bacterial disease. Several sepsis studies have demonstrated upregulation of pathogen recognition receptors and signal transduction pathways, and a downregulation of zinc homeostasis.4–6 The intricate host inflammatory response is associated with neuronal and vascular injury, even after cerebrospinal fluid (CSF) sterilisation with antibiotics. Adjunctive new therapies for bacterial meningitis have to date not shown any conclusive benefit, prompting the need for an improved understanding of key mechanisms that might reveal potential new therapeutic targets.7 Chronic coinfection may impact gene expression, even among asymptomatic patients. HIV is a major risk factor for IPD, characterised by waning immunity and dysregulated physiology among infected individuals, greatly escalating disease susceptibility and mortality outcomes,8 9 thereby influencing the host’s gene expression. We examine differences in gene expression using blood samples from children with pneumococcal meningitis and matched healthy controls, and also compare transcriptional profiles between children with pneumococcal meningitis with and without underlying HIV infection.

es,8 9 thereby influencing the host’s gene expression. We examine differences in gene expression using blood samples from children with pneumococcal meningitis and matched healthy controls, and also compare transcriptional profiles between children with pneumococcal meningitis with and without underlying HIV infection. Materials and methods Patients and controls Children (n=377) were recruited as part of a larger study investigating host determinants of susceptibility to IPD conducted at Queen Elizabeth Central Hospital, Blantyre, Malawi, between April 2004 and October 2006.10 Ethical approval was granted from the College of Medicine Research Committee, Malawi and the Liverpool School of Tropical Medicine Local Research Ethics Committee. Written informed consent was given prior to recruitment.

ucted at Queen Elizabeth Central Hospital, Blantyre, Malawi, between April 2004 and October 2006.10 Ethical approval was granted from the College of Medicine Research Committee, Malawi and the Liverpool School of Tropical Medicine Local Research Ethics Committee. Written informed consent was given prior to recruitment. We excluded children (n=135) infected with other commonly prevalent microbes (Salmonella typhimurium, Escherichia coli, Haemophilus influenzae b, Neisseria meningitidis, Staphylococcus aureus, Streptococcus pyogenes) identified by positive laboratory culture of blood, CSF or lung aspirate. Cases for the microarray analysis were HIV treatment-naive children with confirmed pneumococcal meningitis defined as abnormal CSF white cell count >10×109/L plus one or more of the following: CSF culture positive for pneumococci, CSF gram stain consistent with pneumococci, CSF positive for pneumococcal polysaccharide antigen or pneumococcal DNA. Cases for the real-time quantitative PCR (RT-qPCR) were children with confirmed IPD, which was defined as follows: pneumococcal pneumonia (n=40) or pneumococcal meningitis (n=189), confirmed by either culture, antigen test or PCR. Controls were healthy afebrile, malaria aparasitaemic children from the same villages as the index cases, and were as closely age matched as possible. Microarray analysis was conducted on 15 samples (12 from cases with pneumococcal meningitis and 3 from controls), and RT-PCR was conducted on 242 samples (229 from cases with IPD and 13 from controls), which included all those used for microarray analysis.

s the index cases, and were as closely age matched as possible. Microarray analysis was conducted on 15 samples (12 from cases with pneumococcal meningitis and 3 from controls), and RT-PCR was conducted on 242 samples (229 from cases with IPD and 13 from controls), which included all those used for microarray analysis. There were no known coinfections apart from HIV, and we did not test for other viruses in the CSF, like cytomegalovirus (CMV) or Epstein Barr Virus (EBV), which have been reported to be associated with increased mortality in Malawian adults with bacterial meningitis.11 12 Samples Whole blood was drawn at admission from consecutive children with IPD. The methodology for downstream transcriptome analysis from small blood samples has been previously described.13 RNA extraction, quantification and hybridisation Total RNA was extracted from whole blood using an optimised method for the PAXgene blood RNA kit (Qiagen, West Sussex, UK), as previously described.13 The total RNA concentration (ng/µL) and ratios (260/280 and 260/230) were measured using a NanoDrop ND-100 UV-vis spectrophotometer (NanoDrop Technologies, Delaware, USA) and the RNA integrity was assessed using the Agilent 2100 BioAnalyzer (Agilent Technologies, Edinburgh, UK) before and after concentration.

to a Human Genome U133A GeneChip array for 16 hours rotating at 60 rpm at 45°C in a GeneChip Hybridization Oven 640 (Affymetrix). Each chip was washed and stained on a GeneChip Fluidics Station 450 (Affymetrix) and scanned using a GeneChip Scanner 3000 (Affymetrix) employing standard recommended protocols (Affymetrix). Microarray data analysis Microarray experiment data were analysed using R and Bioconductor packages.14 Briefly, the human genome HGU133A array (Affymetrix) scans output was preprocessed using the affy package.15 The limma package was used to evaluate differential expressed genes.16 17 Gene annotations were performed using hgu133a.db, KEGG.db database packages.18–21 Genes were considered differentially expressed if they had a Benjamini and Hochberg (BH)-adjusted p value <1.5e-3 and >±2-fold change in gene expression. Bonferroni p value adjustments were also performed for comparison. Canonical pathways and functional networks that involve the differently expressed genes play were determined using the Ingenuity Pathway Analysis (IPA) catalogue of well-characterised metabolic and cell signalling cascades. Expression data can be accessed using accession number GSE47172 at the NCBI GEO database. Reverse transcription for qPCR RNA samples were DNAse (Ambion, Warrington, UK) treated to remove any contaminating genomic DNA. RNA (1 µg) was reverse transcribed with SuperScript II RNase H- Reverse Transcriptase and oligo (dT)12–18 (0.5 µg/µL) following the manufacturer’s guidelines. The cDNA was stored at −40°C until required.

nscription for qPCR RNA samples were DNAse (Ambion, Warrington, UK) treated to remove any contaminating genomic DNA. RNA (1 µg) was reverse transcribed with SuperScript II RNase H- Reverse Transcriptase and oligo (dT)12–18 (0.5 µg/µL) following the manufacturer’s guidelines. The cDNA was stored at −40°C until required. RT-qPCR measurement of target genes The Human Universal ProbeLibrary (UPL, Roche, Switzerland) employing proprietary locked nucleic acid analogues was used to design qPCR assays to measure expression levels in genes of interest. Using the Roche Online Assay Design Centre, specific primers and an associated probe were selected for the reference and target transcripts. Gene expression was determined using RT-qPCR on a Roche LightCycler 480 (online supplementary table 2). 10.1136/bmjpo-2017-000092.supp1Supplementary file 1 The following 34 genes were identified from literature and prioritised for RT-qPCR differential expression analysis between cases and controls: ACSL1, ANXA3, ATP, BAG1, BPGM, C3AR1, CA4, CD177, CD55, CD59, CEACAM, CFLAR, FOLR3, GNAI3, GNLY, GYG1, IL1R2, IL1RN, ITGAM, KLRF1, LCK, LCN2, LTF, MAPK14, MMP9, NUMB, OLAH, PSEN1, RETN, S100A12, SAMSN, SERPINA1, SUB1 and VNN1. We used a previously described RT-qPCR data normalisation method.22

tween cases and controls: ACSL1, ANXA3, ATP, BAG1, BPGM, C3AR1, CA4, CD177, CD55, CD59, CEACAM, CFLAR, FOLR3, GNAI3, GNLY, GYG1, IL1R2, IL1RN, ITGAM, KLRF1, LCK, LCN2, LTF, MAPK14, MMP9, NUMB, OLAH, PSEN1, RETN, S100A12, SAMSN, SERPINA1, SUB1 and VNN1. We used a previously described RT-qPCR data normalisation method.22 Statistical analysis of genes prioritised for RT-qPCR differential expression analysis First, we derived the relative gene expression in cases compared with controls for the 34 genes under assessment. We used a previously described RT-qPCR data normalisation method.22 Briefly, the amounts of target genes expressed in a sample were normalised to the average of the three endogenous controls. This is given by ΔCq, where ΔCq is determined by subtracting the average endogenous gene Cq value from the average target gene Cq value: Cq target gene – Cq average endogenous gene = ΔCq. The calculation of relative expression, ΔΔCq, involves subtraction of ΔCq value for the controls from the ΔCq value for the cases: ΔCq target gene(case) – ΔCq target gene(control) = ΔΔCq. 2-ΔΔCq is the relative expression of the target gene in cases compared with controls. Next, we examined statistically significant differences in relative gene expression between cases and controls using the Welch two-sample t-test implemented in the R package. We generate boxplots to visualise the mean and media relative expression in cases and controls separately, and the Welch two-sample t-test p value to show statistically significant differences in relative gene expression between these two groups.

controls using the Welch two-sample t-test implemented in the R package. We generate boxplots to visualise the mean and media relative expression in cases and controls separately, and the Welch two-sample t-test p value to show statistically significant differences in relative gene expression between these two groups. Results Transcriptional profiles among the cases and controls In the microarray discovery cohort, there were 12 children with pneumococcal meningitis (six male, six female, median age 1.1 years) and 3 controls (two male, one female, median age 7 years). The breakdown was as follows: HIV-infected survivors (n=3), HIV-infected non-survivors (n=3), HIV-uninfected survivors (n=3) and HIV-uninfected non-survivors (n=3) (online supplementary table 1). The RT-PCR validation cohort had a median age of 3.09 years. 10.1136/bmjpo-2017-000092.supp2Supplementary file 2

Results Transcriptional profiles among the cases and controls In the microarray discovery cohort, there were 12 children with pneumococcal meningitis (six male, six female, median age 1.1 years) and 3 controls (two male, one female, median age 7 years). The breakdown was as follows: HIV-infected survivors (n=3), HIV-infected non-survivors (n=3), HIV-uninfected survivors (n=3) and HIV-uninfected non-survivors (n=3) (online supplementary table 1). The RT-PCR validation cohort had a median age of 3.09 years. 10.1136/bmjpo-2017-000092.supp2Supplementary file 2 We examined whether global transcriptional profiles of peripheral blood from children with pneumococcal meningitis (n=12) were distinct from those of healthy controls (n=3) randomly selected from a larger data set by microarray expression profile analysis.10 In general, there was a marked distinction in differential expression between cases and controls (online supplementary figure 1). We identified 10 significantly differentially expressed genes (BH-adjusted p value <1.5e-3 and >±2-fold change) (figure 1). We observed significant upregulation of the following: S100 calcium-binding protein A12 (S100A12); vanin-1 (VNN1); arginase, liver (ARG1); matrix metallopeptidase 9 (gelatinase B, 92 kDa gelatinase, 92 kDa type IV collagenase) (MMP9); annexin A3 (ANXA3); interleukin 1 receptor, type II (IL1R2); CD177 molecule (CD177); S100 calcium-binding protein A9 (S100A9), cytoskeleton-associated protein 4 (CKAP4); and glycogenin 1 (GYG1). 10.1136/bmjpo-2017-000092.supp3Supplementary file 3

We examined whether global transcriptional profiles of peripheral blood from children with pneumococcal meningitis (n=12) were distinct from those of healthy controls (n=3) randomly selected from a larger data set by microarray expression profile analysis.10 In general, there was a marked distinction in differential expression between cases and controls (online supplementary figure 1). We identified 10 significantly differentially expressed genes (BH-adjusted p value <1.5e-3 and >±2-fold change) (figure 1). We observed significant upregulation of the following: S100 calcium-binding protein A12 (S100A12); vanin-1 (VNN1); arginase, liver (ARG1); matrix metallopeptidase 9 (gelatinase B, 92 kDa gelatinase, 92 kDa type IV collagenase) (MMP9); annexin A3 (ANXA3); interleukin 1 receptor, type II (IL1R2); CD177 molecule (CD177); S100 calcium-binding protein A9 (S100A9), cytoskeleton-associated protein 4 (CKAP4); and glycogenin 1 (GYG1). 10.1136/bmjpo-2017-000092.supp3Supplementary file 3 Figure 1 Distribution plot of the differentially expressed genes. The significantly differentially expressed genes are shown in red colour. The significance threshold (p<1.5e-3) is indicated by a dashed red line, and a fold change threshold of more than 2 is shown by the dashed vertical lines. The green line shows p<0.05. (A) Shows results for unadjusted p values, (B) results for BH-adjusted p values and (C) stringent Bonferroni-adjusted p values, which represent an overcorrection. BH, Benjamini and Hochberg.

a dashed red line, and a fold change threshold of more than 2 is shown by the dashed vertical lines. The green line shows p<0.05. (A) Shows results for unadjusted p values, (B) results for BH-adjusted p values and (C) stringent Bonferroni-adjusted p values, which represent an overcorrection. BH, Benjamini and Hochberg. Differential expression, underlying HIV infection and disease outcomes In the microarray discovery cohort, we did not find any significantly differentially expressed genes for comparisons between HIV-infected cases and HIV-uninfected cases or between survivors and non-survivors (online supplementary figure 2). 10.1136/bmjpo-2017-000092.supp4Supplementary file 4 RT-qPCR validation RT-qPCR validation was performed for a set of 34 prioritised genes selected from literature for cases (n=229) and controls (n=13). The RT-PCR results are in agreement with our microarray analysis findings. All the genes are significantly differentially expressed between the two groups with the exception of guanine nucleotide-binding protein subunit alpha-13 (GNA13), protein numb (NUMB) and presenilin-1 (PSEN1) (figure 2). There was no significantly increased gene expression in HIV-infected compared with HIV-uninfected cases (online supplementary figure 3). Interestingly, there was significant upregulation of folate receptor 3 (FOLR3), NUMB and S100A12 in survivors compared with non-survivors (online supplementary figure 4). There was wide variability in relative gene expression in cases, but not controls. 10.1136/bmjpo-2017-000092.supp5Supplementary file 5

RT-qPCR validation RT-qPCR validation was performed for a set of 34 prioritised genes selected from literature for cases (n=229) and controls (n=13). The RT-PCR results are in agreement with our microarray analysis findings. All the genes are significantly differentially expressed between the two groups with the exception of guanine nucleotide-binding protein subunit alpha-13 (GNA13), protein numb (NUMB) and presenilin-1 (PSEN1) (figure 2). There was no significantly increased gene expression in HIV-infected compared with HIV-uninfected cases (online supplementary figure 3). Interestingly, there was significant upregulation of folate receptor 3 (FOLR3), NUMB and S100A12 in survivors compared with non-survivors (online supplementary figure 4). There was wide variability in relative gene expression in cases, but not controls. 10.1136/bmjpo-2017-000092.supp5Supplementary file 5 10.1136/bmjpo-2017-000092.supp6Supplementary file 6 Figure 2 Validation of RNA transcription profile differential expression using real-time quantitative PCR. Relative gene expression in cases compared with controls for 34 genes assessed. The black line shows the boxplot median. The red dot shows the mean, and the Welch two-sample t-test p value is shown on the top right corner.

Validation of RNA transcription profile differential expression using real-time quantitative PCR. Relative gene expression in cases compared with controls for 34 genes assessed. The black line shows the boxplot median. The red dot shows the mean, and the Welch two-sample t-test p value is shown on the top right corner. Pathway analysis Networks were reconstructed using IPA software for the genes differentially expressed between cases and controls in the microarray experiment, and the combined microarray and RT-qPCR experiments (figure 3A,B). For the genes defined by the microarray experiments, the networks were predominantly related to granulocyte function, and including antimicrobial and endothelial activation responses (figure 3A). The merged set demonstrated wider networks involving immune cell activation as well as leucocyte migration and adhesion (figure 3B). The canonical pathways mapped by genes defined in the microarray experiment demonstrated greatest changes in the arginine and granulocyte pathways. Canonical pathways in the combined microarray and RT-qPCR experiments demonstrated greatest change in leucocytes, and especially neutrophil activation and migration, Notch and glucocorticoid receptor signalling pathways (online supplementary tables 3 and 4).

ed greatest changes in the arginine and granulocyte pathways. Canonical pathways in the combined microarray and RT-qPCR experiments demonstrated greatest change in leucocytes, and especially neutrophil activation and migration, Notch and glucocorticoid receptor signalling pathways (online supplementary tables 3 and 4). Figure 3 The gene network for significantly differentially expressed genes in the cases. Networks reconstructed to support the most direct connectivity between genes differentially expressed between cases and controls. IPA (Qiagen) software's Core analysis application has been used to perform an automatic graphical reconstruction of the network via utilisation of IPA's Knowledge Base database of protein interactions. The meaning of links and shapes is explained in the inserted legend. Functional connections are presented in blue, information connections to the associated categories in pink and grey. (A) Network reconstructed only for the DE genes identified by microarray analysis. (B) Network reconstructed for the merged data sets of DE genes identified via microarray (red) and PCR analysis (green). White blocks correspond to nodes added by IPA's network editor automatically to ensure network connectivity. DE, differential expression; IPA, Ingenuity Pathway Analysis.

fied by microarray analysis. (B) Network reconstructed for the merged data sets of DE genes identified via microarray (red) and PCR analysis (green). White blocks correspond to nodes added by IPA's network editor automatically to ensure network connectivity. DE, differential expression; IPA, Ingenuity Pathway Analysis. Discussion and conclusion We have shown significant differences in RNA transcriptional profiles in children with pneumococcal meningitis compared with controls. Children with pneumococcal meningitis demonstrated increased expression in genes involved in the inflammatory response, and glucose and L-arginine metabolic pathways. Dysregulation of these pathways can lead to an impaired adaptive host response to pneumococcal infection, thereby contributing to the increased morbidity and mortality. Our findings in the initial cohort of pneumococcal meningitis were validated in the larger cohort of all children with IPD, which includes presentations with pneumonia as well as meningitis. These findings add validity to the initial results from the microarray experiments, and suggest that the host response observed is systemic, and not simply localised to the process of blood–brain barrier disruption per se. We observe wide variability in relative gene expression in cases, which perhaps reflects differences in disease onset and robustness of host response within this group.

experiments, and suggest that the host response observed is systemic, and not simply localised to the process of blood–brain barrier disruption per se. We observe wide variability in relative gene expression in cases, which perhaps reflects differences in disease onset and robustness of host response within this group. Elevation of S100A12, VNN, ARG1, MMP9, ANXA3, IL1R2, CD177, S100A9, CKAP4 and GYG1 in pneumococcal meningitis supports previous findings that have highlighted the important roles of some of these genes during host pathogen response. These genes have important interconnected functions for cellular interactions and response to infection (figure 3A and online supplementary file 7), and play crucial roles during host immune response; cell regulatory processes such as apoptosis and differentiation; and metabolic processes such as amino acid and glucose metabolism. 10.1136/bmjpo-2017-000092.supp7Supplementary file 7

Elevation of S100A12, VNN, ARG1, MMP9, ANXA3, IL1R2, CD177, S100A9, CKAP4 and GYG1 in pneumococcal meningitis supports previous findings that have highlighted the important roles of some of these genes during host pathogen response. These genes have important interconnected functions for cellular interactions and response to infection (figure 3A and online supplementary file 7), and play crucial roles during host immune response; cell regulatory processes such as apoptosis and differentiation; and metabolic processes such as amino acid and glucose metabolism. 10.1136/bmjpo-2017-000092.supp7Supplementary file 7 Vanin-1 (VNN1) protein is expressed by human phagocytes, and involved in leucocyte adhesion and migration.23 The VNN1 knockout mice model has provided clear evidence that VNN1 modulates redox and immune pathways.24 Exposure of human mononuclear cells to oxidative stress results in upregulation of human VNN1 and downregulation of peroxisome proliferator-activated receptor (PPAR).25 IL1R2 and IL1RN were upregulated in cases, which is consistent with our previous report of the IL-1Ra single-nucleotide polymorphism rs4251961 playing a key role in the pathophysiology of IPD and in other human infections.10 Recent reports also demonstrate IL1R2 expression upregulation in sepsis, being more pronounced in Gram-negative than Gram-positive infections.26

is consistent with our previous report of the IL-1Ra single-nucleotide polymorphism rs4251961 playing a key role in the pathophysiology of IPD and in other human infections.10 Recent reports also demonstrate IL1R2 expression upregulation in sepsis, being more pronounced in Gram-negative than Gram-positive infections.26 During infection, host production of the cytokine nitric oxide (NO) after non-opsonic phagocytosis exerts microbicidal effects.27 28 ARG1 expression is inducible in the lungs of mice in response to pneumococcal infection.29 Phagocytosis of pneumococci by macrophages also results in increased production of nitric oxide synthase 2-dependent production of NO and reactive nitrogen species.30 Increased plasma arginase activity depletes L-arginine concentrations, the substrate for NO synthesis, leading to vascular dysfunction during severe sepsis and supressed NO-mediated microbicidal effects.31 Increased ARG1 activity may also be a bacterial survival strategy to escape the NO-dependent host antimicrobial immune response.30

arginase activity depletes L-arginine concentrations, the substrate for NO synthesis, leading to vascular dysfunction during severe sepsis and supressed NO-mediated microbicidal effects.31 Increased ARG1 activity may also be a bacterial survival strategy to escape the NO-dependent host antimicrobial immune response.30 Neutrophil-specific glycoprotein CD177 is expressed on a subset of human neutrophils, and is involved in neutrophil transendothelial migration. A previous microarray study of purified neutrophils from patients with septic shock revealed CD177 mRNA has the highest differential expression between cases and controls.32 Consistent with our data, the study also demonstrated increased expression of ARG1, ANXA3, CKAP4, IL1R2, MMP9 and VNN1. ANXA3 promotes endothelial cell junction integrity in animal models, and endothelial cell motility in vitro. ANXA3 is upregulated following neuronal injury, which may explain the finding in pneumococcal meningitis.33

our data, the study also demonstrated increased expression of ARG1, ANXA3, CKAP4, IL1R2, MMP9 and VNN1. ANXA3 promotes endothelial cell junction integrity in animal models, and endothelial cell motility in vitro. ANXA3 is upregulated following neuronal injury, which may explain the finding in pneumococcal meningitis.33 S100A12 plays a prominent role in the regulation of proinflammatory processes and immune response by recruiting leucocytes, promoting cytokine and chemokine production, and regulating leucocyte adhesion and migration.34 The S100A8/A9 heterodimer is expressed by myeloid cells, especially neutrophils, and has a protective effect in the host response to pneumococcal infection by increasing circulating neutrophils through increased granulocyte colony-stimulating factor production.35 It is an antimicrobial peptide, but plays an important role in leucocyte migration.36 During infection, S100 proteins stimulate the proinflammatory immune response through interaction with the immunoglobulin family transmembrane pattern recognition receptors: receptor for advanced glycation end-products and Toll-like receptor 4. This leads to nuclear factor kappa B-mediated proinflammatory response with production of proinflammatory cytokines. This inflammatory response in turn leads to increased expression of S100 proteins, and the start of a positive feedback loop. Although as antimicrobial proteins they protect against infection, they can also have a negative detrimental effect on the host by amplifying the destructive proinflammatory responses. The increased expression in survivors may be explained by this positive feedback loop.37

ns, and the start of a positive feedback loop. Although as antimicrobial proteins they protect against infection, they can also have a negative detrimental effect on the host by amplifying the destructive proinflammatory responses. The increased expression in survivors may be explained by this positive feedback loop.37 Glucocorticoids also play a significant role in immune response regulation by supressing immune and inflammatory responses, and modulating cytokines that promote host immune responses.38 We speculate that these responses are important in regulating immune responses to avoid host tissue damage. NUMB negatively regulates Notch, which in turn attenuates proinflammatory cytokines and increases anti-inflammatory cytokines, which may explain the increase in NUMB expression in survivors.39 40

une responses.38 We speculate that these responses are important in regulating immune responses to avoid host tissue damage. NUMB negatively regulates Notch, which in turn attenuates proinflammatory cytokines and increases anti-inflammatory cytokines, which may explain the increase in NUMB expression in survivors.39 40 Our results are consistent with previous studies on transcription profiling in other bacterial diseases, suggesting that some of the mechanisms are not specific to IPD. Although it could be argued that the lack of age matching in cases versus controls could cause differential expression due to maturation of the immune system in older children, the consistency of our results with other studies makes this unlikely.41 42 A limitation of our study is that the microarray analysis was performed using the Affymetrix Human Genome U133A array, and sample number was constrained at the time by cost. The microarray data set was not large enough to allow all possible multifactorial models of comorbidity and disease outcomes to be exhaustively examined, but sought to validate our findings using RT-qPCR in a larger cohort of children. Further evaluation is in a larger cohort using whole genome sequencing would provide more insights on the mechanisms of host response.

low all possible multifactorial models of comorbidity and disease outcomes to be exhaustively examined, but sought to validate our findings using RT-qPCR in a larger cohort of children. Further evaluation is in a larger cohort using whole genome sequencing would provide more insights on the mechanisms of host response. In conclusion, this comparative study of gene expression provides mechanistic insight for IPD in children, and demonstrates significant and widespread immune activation, with oxidative stress, recruitment of neutrophils, leucocyte adhesion and migration, activation of antimicrobial peptides and preservation of endothelial cell junction integrity. 10.1136/bmjpo-2017-000092.supp8Supplementary file 8 The IPD Study Group (C Antonio, M Chinamale, L Jere, D Mnapo, V Munthali, FNyalo, J Simwinga, M Kaole, DL Banda, A Manyika, K Phiri) recruited patients. We thank the children included in this study, and their parents and guardians for giving consent for them to participate in the study. We also extend thanks to the nursing and medical staff at the Malawi-Liverpool-Wellcome Trust Clinical Research Programme (MLW), Research Ward, for their contribution to this study. Contributors: FM, PJRD and EDC conceived and designed the study. LM and EDC collected the clinical samples. FM performed experiments. BWK, FM, OV, KN, PJRD and EDC performed the analysis. BWK, FM, PJRD and EDC wrote the first draft of the manuscript. BWK, FM, LM, KN, OV, MEM, EMM, PJRD and EDC contributed to writing the manuscript.

FM, PJRD and EDC conceived and designed the study. LM and EDC collected the clinical samples. FM performed experiments. BWK, FM, OV, KN, PJRD and EDC performed the analysis. BWK, FM, PJRD and EDC wrote the first draft of the manuscript. BWK, FM, LM, KN, OV, MEM, EMM, PJRD and EDC contributed to writing the manuscript. Funding: EDC was supported by a Wellcome Trust Career Development Grant (068026). BWK was supported by a Wellcome Trust Major Overseas programme award (084679/Z/08/Z). Disclaimer: The funding bodies had no role in collection, analysis or interpretation of data or in writing the manuscript. Competing interests: None declared. Patient consent: Obtained. Ethics approval: College of Medicine Research Committee (COMREC), Malawi. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Transcription data are openly available.

What is already known on this topic? Patients with cardiac disease transitioning from the paediatric to the adult centre need to take on increasing ownership for their condition. New technologies can provide unprecedented insight into the human body, but the narrative element remains a neglected dimension of data. What this study hopes to add? We present a first-person composite narrative from the perspective of adolescents with congenital heart disease (CHD). An immersive artistic workshop can allow adolescents with CHD to express imagery relating to their individuality, incorporating elements of their medical history. An interdisciplinary framework including a narrative element could contribute towards teasing out divergences between patients’ health status and their expectations.

An immersive artistic workshop can allow adolescents with CHD to express imagery relating to their individuality, incorporating elements of their medical history. An interdisciplinary framework including a narrative element could contribute towards teasing out divergences between patients’ health status and their expectations. Introduction Unearthing narratives and ‘honouring the stories of illness’1 are essential for developing a holistic approach to medicine. As discussed in the literature on co-creation, the management (from the carer’s perspective) and self-management (from the patient’s perspective) of a condition rely on practical and moral choices that are profoundly unique,2 and it has been advocated that a narrative approach could be illuminating with regards to adopting technological innovations to improve patients’ care. An interdisciplinary framework including the narrative element could thus potentially allow improvements, such as addressing lack of awareness or excessive anxieties and teasing out divergences between the patients’ health status and their expectations. Furthermore, while new technologies can dramatically improve our insight into the human body, with the most sophisticated imaging techniques or new technologies such as three-dimensional (3D) printing, they cannot exhaust its meaning.3 In other words, how is this medical reality reconciled with the experiential component? The narrative element has indeed been suggested as an essential yet neglected dimension of data.4 And its inclusion, particularly based on an artistic participatory approach, could prove beneficial to identify both collective and unique responses in a population of interest.5

reconciled with the experiential component? The narrative element has indeed been suggested as an essential yet neglected dimension of data.4 And its inclusion, particularly based on an artistic participatory approach, could prove beneficial to identify both collective and unique responses in a population of interest.5 More specifically, in the context of adolescent patients growing up with congenital heart disease (CHD), we have previously discussed how an immersive artistic workshop was conducive to generating imagery allowing young people with CHD to express the uniqueness of their condition, and that this process can give them the opportunity to explore their individuality within a group sharing a similar medical condition and life experience, incorporating elements of their medical history.6 Here, we discuss how a collective narrative can be developed, ensuing from this kind of creative activity, and what its value might be for patients and health professionals.

lore their individuality within a group sharing a similar medical condition and life experience, incorporating elements of their medical history.6 Here, we discuss how a collective narrative can be developed, ensuing from this kind of creative activity, and what its value might be for patients and health professionals. Materials and methods Participants Five young people with CHD (age 17–18 years, two men) were involved in the workshop process. All participants were under follow-up in a specialist cardiac transition clinic at a tertiary paediatric hospital and were invited to take part in the workshops on the basis of a primary diagnosis of CHD and availability of cardiovascular MRI data in order to create a 3D rendering of their heart (see Workshop process section). Their primary diagnoses included tetralogy of Fallot, total anomalous pulmonary venous drainage and transposition of the great arteries. Participants provided written assent/consent to be photographed during the workshop, for the conversations to be recorded, and for their creative outputs to be shared with the group and more widely. All patients were in the final 2 years of secondary education and were in the process of applying for university. They were accompanied in the process by an artist with long-standing experience in participatory practices, an adolescent clinical nurse specialist, a biomedical engineer and a health psychologist.

widely. All patients were in the final 2 years of secondary education and were in the process of applying for university. They were accompanied in the process by an artist with long-standing experience in participatory practices, an adolescent clinical nurse specialist, a biomedical engineer and a health psychologist. Workshop process All participants took part in two sessions, which were run 2.5 months apart. The first session explored the uniqueness of CHD with both artistic media and 3D printed models, while the second was more overtly focused on the heart as an organ. As part of the process, during the first session (described in detail in Layton et al6, see Supplementary File), participants were led through activities including a blind self-portrait drawing exercise, a blind self-portrait sculpting exercise, a creative writing activity and a body map exercise. These activities enabled them to develop language and imagery personal to them and their perception of themselves, without explicitly focusing on their heart and their condition.

luding a blind self-portrait drawing exercise, a blind self-portrait sculpting exercise, a creative writing activity and a body map exercise. These activities enabled them to develop language and imagery personal to them and their perception of themselves, without explicitly focusing on their heart and their condition. During the second session, the group re-engaged with one another and the facilitators. During an opening exercise, participants revised the imagery that was developed during the first workshop. They then undertook an embossing activity, whereby they were given a small A6 size metal plate (either aluminium or copper, allowing for participants’ preferences, see figure 1A) with a velvet flocked anatomical heart screen-printed on it; participants were asked to incorporate the imagery that had emerged from the first workshop, in particular the body mapping exercise, to create unique embossed pieces. Figure 1 Participants completed an embossing activity using small A6 copper or aluminium plates with an anatomical drawing of the heart printed onto A. The embossed pieces incorporated elements of each participant’s personal imagery that was explored during the workshop. An example of embossed piece is shown in B.

During the second session, the group re-engaged with one another and the facilitators. During an opening exercise, participants revised the imagery that was developed during the first workshop. They then undertook an embossing activity, whereby they were given a small A6 size metal plate (either aluminium or copper, allowing for participants’ preferences, see figure 1A) with a velvet flocked anatomical heart screen-printed on it; participants were asked to incorporate the imagery that had emerged from the first workshop, in particular the body mapping exercise, to create unique embossed pieces. Figure 1 Participants completed an embossing activity using small A6 copper or aluminium plates with an anatomical drawing of the heart printed onto A. The embossed pieces incorporated elements of each participant’s personal imagery that was explored during the workshop. An example of embossed piece is shown in B. A short creative writing exercise ensued, to further develop imagery relating to the self. Overall, across the two sessions, participants were asked to elaborate on how they saw themselves if they were: an animal, a colour, a weather/element, a building, a vegetable, a book/film, a childhood toy, a flower and a piece of clothing. This activity allows participants to brainstorm introspective images and they were encouraged by the artist facilitating the workshop not just to identify an image for each theme but to elaborate on the reason why they associated that specific image with themselves.

m, a childhood toy, a flower and a piece of clothing. This activity allows participants to brainstorm introspective images and they were encouraged by the artist facilitating the workshop not just to identify an image for each theme but to elaborate on the reason why they associated that specific image with themselves. The final activity focused explicitly on the heart. Participants were given an A4 size printout of their own hearts. These were obtained by three-dimensionally reconstructing the cardiovascular MRI data for each patient, namely the whole heart sequence, according to validated methodology,7 and then printing the 3D rendering in two dimensions. An example is provided in figure 2. First participants were asked to outline their own heart using tracing paper and connect with its lines and forms. The outlines were then photocopied on thicker A4 sketching sheets and, as a final activity, participants created their heart design, having access to a range of paints and pastels, which enabled them to incorporate colour and text in the design if they so wished. Figure 2 Participants worked on the outline of their own heart, which was reconstructed from three-dimensional (3D) data as part of their routine cardiovascular MRI scan. An example of 3D reconstruction is shown here. These patient-specific heart renderings were given to participants to trace in A4 size. Throughout the process, participants were encouraged to speak about their images and drawings. The workshop was recorded for later analysis and reflection.

Figure 2 Participants worked on the outline of their own heart, which was reconstructed from three-dimensional (3D) data as part of their routine cardiovascular MRI scan. An example of 3D reconstruction is shown here. These patient-specific heart renderings were given to participants to trace in A4 size. Throughout the process, participants were encouraged to speak about their images and drawings. The workshop was recorded for later analysis and reflection. Interpretation of findings: first-person narrative A composite first-person narrative approach was chosen to handle the emerged phenomenological descriptions.8 The composite first-person narrative is a reflective story and results in a representation of the phenomenon amalgamating the voices of multiple participants. Initially, the narrative was independently developed by different authors (artist, engineer and psychologist), and in order to do so, they revisited audio recordings and notes from the workshops, as well as—importantly—visually assessing the artistic output of the workshop. The latter chiefly included the heart designs realised at the end of the second session, but took into account all the imagery that had emerged and had been discussed throughout the process. Prior to approaching the writing, the authors dwelled on the materials for approximately 18 months. A composite approach incorporating narratives unearthed through formative research allows the researchers to use factually realistic details and shape a unified story.9 Three authors (GB, SL and JW) developed a narrative independently and differences in the approach, tone and key elements to be included were then discussed prior to creating a merged version, which resulted in the final ‘composite’. This was then shared with the fourth author (L-KL) to further check the truthfulness of the representation. As such, there was not a dominant writer but the approach was considered as a group authorship. The final version of the composite narrative (presented in the Results section of this paper) was also shared via email with the workshop participants. They were invited to comment on whether elements of it reflected their own individual narrative and the feedback that we received from them indicated that this was indeed the case.

rsion of the composite narrative (presented in the Results section of this paper) was also shared via email with the workshop participants. They were invited to comment on whether elements of it reflected their own individual narrative and the feedback that we received from them indicated that this was indeed the case. Results Participants engaged well in the workshop process overall. Two participants were taken aback by the final activity of working on their own heart printouts, but were guided by the artist through the activity and the resulting designs were rich in imagery. Importantly, all young people were willing to discuss elements of their artworks and elaborate on them. This sometimes happened not during a group reflection but rather during one of the activities, with the embossing exercise, for example, acting as a displacement activity allowing for participants to start elaborating on their imagery while slowly working on the metal plate. The group bonded well and the resulting element of peer support facilitated conversations around the creative outputs, which were indeed rich and detailed. The outcomes of the workshop process allowed the research team to develop the composite first-person narrative of adolescents with complex CHD, which is presented below. Is it red? That is what I thought it looked like when I was little.

Results Participants engaged well in the workshop process overall. Two participants were taken aback by the final activity of working on their own heart printouts, but were guided by the artist through the activity and the resulting designs were rich in imagery. Importantly, all young people were willing to discuss elements of their artworks and elaborate on them. This sometimes happened not during a group reflection but rather during one of the activities, with the embossing exercise, for example, acting as a displacement activity allowing for participants to start elaborating on their imagery while slowly working on the metal plate. The group bonded well and the resulting element of peer support facilitated conversations around the creative outputs, which were indeed rich and detailed. The outcomes of the workshop process allowed the research team to develop the composite first-person narrative of adolescents with complex CHD, which is presented below. Is it red? That is what I thought it looked like when I was little. My whole life has been defined by my heart—the good and the not so good. I knew that something was wrong with my heart, that it needed fixing, even when I was little. I remember people saying that there was a hole in it (a strange concept to grasp) and I imaged that they closed it by putting a plaster over it, and they also put some bandage around the aorta, as it too needed mending. Now I understand that it is much more complicated than that. It is something like a Rubick’s cube, a puzzle; something challenging, layered and complex, that maybe eventually cannot be done. People have tried to fix it but it cannot be fixed—it is unique and special but bits are in the wrong place. Sometimes I wonder: “Does it actually look like a heart?” because some of its parts are missing.

like a Rubick’s cube, a puzzle; something challenging, layered and complex, that maybe eventually cannot be done. People have tried to fix it but it cannot be fixed—it is unique and special but bits are in the wrong place. Sometimes I wonder: “Does it actually look like a heart?” because some of its parts are missing. And over the years, my heart survived the trauma of being repaired. Today, it bears scars, which are a testimony to this trauma, but it survived, and I am so proud of it. Almost like a soldier, who has been wounded during the wars, and lived on, so now, together with the scars, it is decorated with medals, like a war hero. The medals represent its resilience, its strength and its achievements. It is my condition that defines who I am and fuels my drive and my determination to succeed, to grab life whenever I can. I have scars, my soldier has scars, each a wound that tells a tale of a different fight but each victory necessary for me to be here. The soldier is camouflaged to protect and conceal, returning each time from the battlefield, bloodied and sore, but the wounds of war heal, leaving a scar and a medal to cherish—a symbol of the bravery.

soldier has scars, each a wound that tells a tale of a different fight but each victory necessary for me to be here. The soldier is camouflaged to protect and conceal, returning each time from the battlefield, bloodied and sore, but the wounds of war heal, leaving a scar and a medal to cherish—a symbol of the bravery. I know that the scars are always there, and they are black, like black lines, black marks. So the heart is red—and black. And then, sometimes I feel the colours start to change, gradually veering from cold to warmer and warmer shades. It’s still cold, but it’s warming up. And the warmer part is at its core. Colour has always been important, vital even. As a child I remember hearing about ‘blue blood’ and ‘red blood’, or people commenting on my blue lips and nails—not really blue, more like a purplish tinge. Little did I realise then that the colour I was indicated to others how well my heart was functioning. What colour am I today? If I was a colour it would be purple, that is how I see myself. The degree of purple tells me (and others) if I need to rest and how well I am feeling. But my colour changes, oscillating between blue and red, that process of freezing and warming like a frozen snowflake melting and being heated up by a burning fire. That is like my life—good days, bad days, days when I can do things and days when I can’t. And that is it—the puzzle and unpredictability of my life.

ling. But my colour changes, oscillating between blue and red, that process of freezing and warming like a frozen snowflake melting and being heated up by a burning fire. That is like my life—good days, bad days, days when I can do things and days when I can’t. And that is it—the puzzle and unpredictability of my life. My heart is a survivor. It defines me, it shaped my life, challenging me and those looking after me but rewarding me and them too, at times, as obstacles are overcome and battles are won, living life to the fullest. It is different, it stands out from crowd, bearing its scars and its medals, and makes me stand out of the crowd too. I also sometimes imagine that it is inscribed with words from comics that I read when I was little and books that I have read as I was growing up. So many things are important in my life—things that I keep close to my heart and that make me who I am. Comics, books, the theatre—all there as supports, comforters at times of stress, sources of strength, but also purely to be enjoyed. A Midsummer Night’s Dream, Great Expectations, the wooden sword I used to play with as a child, all mixed up.

nt in my life—things that I keep close to my heart and that make me who I am. Comics, books, the theatre—all there as supports, comforters at times of stress, sources of strength, but also purely to be enjoyed. A Midsummer Night’s Dream, Great Expectations, the wooden sword I used to play with as a child, all mixed up. My heart is unique, and strong, and fragile. And when I think of it I am reminded of a Latin saying that I learnt at school: ‘audentis fortuna iuvat’, fortune favours the brave. How true that is! There are times—I’m not going to lie—when I wish I didn’t have anything wrong with my heart, that I could be like a ladybug with the power to fly and be free of all problems, but my heart makes me who I am and leaves me with an overwhelming sense of pride and achievement. And yes, a feeling of good fortune.

ue that is! There are times—I’m not going to lie—when I wish I didn’t have anything wrong with my heart, that I could be like a ladybug with the power to fly and be free of all problems, but my heart makes me who I am and leaves me with an overwhelming sense of pride and achievement. And yes, a feeling of good fortune. Discussion CHD requires lifelong treatment and/or support, with a growing population of adults with CHD.10 11 Adolescence, when patients ultimately transition from the paediatric to the adult centre, is a particularly significant time in their care, which entails them ideally gaining independence, understanding of their condition, appreciation of complications and lifestyle adjustments and taking on increasing responsibility and ownership for their condition from their parents.12–14 Here, we suggest that an exploration of the narrative of young people with CHD could offer unique insight into the way they see their heart. Furthermore, as the way in which narratives are solicited from patients is important,9 we propose that a creative workshop led by an artist with participatory experience and supported by a multidisciplinary team can be a valuable way to begin exploring such narratives. Having adopted a composite first-person narrative approach, it was possible to identify themes that are central to the patients that were involved in the creative process. These include:The use of medical references: these comprise mainly scars, but also the idea of the ‘hole in the heart’, bandages, plasters and patches.

Having adopted a composite first-person narrative approach, it was possible to identify themes that are central to the patients that were involved in the creative process. These include:The use of medical references: these comprise mainly scars, but also the idea of the ‘hole in the heart’, bandages, plasters and patches. The resilience of these patients: images of strength and battle have clearly emerged, predominantly with one ‘soldier heart’ dressed in camouflage and decorated with medals, but also with the use of the Latin quote ‘Audentis fortuna iuvat’ (Fortune favours the brave). The use of writing in the design: participants incorporated elements of writing, whether keywords or entire sentences. Additionally, the use of colours was very unique and participants were given absolute freedom in choosing how to develop their design. Some used pastels and the results were denser and richer, some used watercolours resulting in more delicate and softer designs, while one participant focused on the design as an outline and eloquently included the piece of a jigsaw, suggestive of his personal view on surgical replacements and repairs that had taken place on his heart.

s and the results were denser and richer, some used watercolours resulting in more delicate and softer designs, while one participant focused on the design as an outline and eloquently included the piece of a jigsaw, suggestive of his personal view on surgical replacements and repairs that had taken place on his heart. Variables including ethnicity, social status, level of education or type of professional occupation (including parental education) are known to affect neuropsychological outcomes in patients with CHD.15 In our study, the sample size was too small to evaluate differences in some of the key variables at play, but we note that participants were all at an equivalent stage in their education and engaged well during the workshop process. Furthermore, we would advocate that the artistic process contributed to creating a bond between participants which, qualitatively, was demonstrated by their high level of engagement, willingness to share their stories and returning for a second workshop. It is important also to consider character traits typical of a young generation that tends to be techno-savvy and collaborative16 17 in support of adopting a creative and visual approach in a workshop setting to explore patients’ narratives.

ement, willingness to share their stories and returning for a second workshop. It is important also to consider character traits typical of a young generation that tends to be techno-savvy and collaborative16 17 in support of adopting a creative and visual approach in a workshop setting to explore patients’ narratives. The workshop was framed as an artistic activity and not as art therapy. This is an important distinction, as the artist leading the workshop was focusing on using the creative tools to stimulate and hold the narratives. Participants’ benefit, nevertheless, could be inferred from the feedback received via email after the activities, mostly referring to the possibility of sharing their accounts and to the opportunity of doing it with peers who also have a form of CHD. One participant eloquently reflected on the approach being “very useful when going through the transition clinic” as “[i]t made me feel like I still mattered as opposed to feeling like I was being forgotten and passed on without much thought”, and that “[t]he work with the artist allowed me to actually reflect on what my condition meant to me and how it impacted me growing up; this was a good way to mark the transition into being an adult patient”.

eel like I still mattered as opposed to feeling like I was being forgotten and passed on without much thought”, and that “[t]he work with the artist allowed me to actually reflect on what my condition meant to me and how it impacted me growing up; this was a good way to mark the transition into being an adult patient”. The use of a composite approach does not diminish the individual voice and contribution. Each individual account contains unique elements and should in itself be respected and hailed as significant.18 A composite approach does not intend to dilute this uniqueness or suggest that singular images or expressions should be removed in an amalgamation of common traits. Rather, the composite approach was chosen as a way to protect individual stories and identities, by combining all of them into one. Indeed, it is suggested that this method could lead to a ‘more embodied’ understanding of the phenomenon being represented, conveying its wholeness.8 The exploration and assimilation of stories of illness has been advocated to lead to better understanding and, as a result, improvement of healthcare.19 Taking into account the patient’s narrative can provide not only biographic or social references that might not otherwise emerge, but also allow for insights into the patient’s development to surface.20 In our case, it was important to identify a theme of resilience and the contrasted feelings (eg, a scarred heart vs a heart that is warming up) in a population at a crucial stage of their care.

or social references that might not otherwise emerge, but also allow for insights into the patient’s development to surface.20 In our case, it was important to identify a theme of resilience and the contrasted feelings (eg, a scarred heart vs a heart that is warming up) in a population at a crucial stage of their care. Taking into account the professionals’ perspective, it has been argued that the use of an artistic approach can lead to enriching the discourse of the practitioners.5 This has been beautifully elaborated by eminent narrative medicine scholar Rita Charon and described as a “Cézanne-like shift to the right or the left” that “gives […] sight of questions we are not wise enough to ask” to patients.21 In referring to the famous anecdote of Paul Cézanne painting over and over his subject of the Montagne Sainte-Victoire overlooking Aix-en-Provence having realised that just moving his sight a few inches to the left or to the right he was able to view his subject entirely afresh, Charon describes the subtle yet powerful insight that can be achieved by virtue of a narrative approach.

r and over his subject of the Montagne Sainte-Victoire overlooking Aix-en-Provence having realised that just moving his sight a few inches to the left or to the right he was able to view his subject entirely afresh, Charon describes the subtle yet powerful insight that can be achieved by virtue of a narrative approach. Such an approach can be complementary to the insight into CHD that can be nowadays gathered with advanced technologies such as exquisite medical imaging, refined computational modelling or 3D printing.22–24 It is complementary in that it allows us to go beyond crucial themes such as anatomy, function and complications, enabling us to start exploring the nature of health and the idea of pain and its sources, such concepts which have been referred to as central ontological and existential questions on individual uniqueness and human worth.25 Conclusion A composite first-person narrative from the perspective of adolescents with CHD was created following an artistic workshop during which patients explored imagery relating to their individuality and their heart. The authors would like to thank the participants and their families. Contributors: All authors were involved in running the workshops and interpreting workshops’ outcomes. GB drafted the manuscript and all authors contributed to and approved the final version.

Conclusion A composite first-person narrative from the perspective of adolescents with CHD was created following an artistic workshop during which patients explored imagery relating to their individuality and their heart. The authors would like to thank the participants and their families. Contributors: All authors were involved in running the workshops and interpreting workshops’ outcomes. GB drafted the manuscript and all authors contributed to and approved the final version. Funding: This work was supported by a Wellcome Trust Small Arts Award (grant ref. 107175/Z/15/Z) and a Royal Academy of Engineering Ingenious Grant for public engagement (grant ref. ING1415\9\154). The authors also acknowledge the generous support of the Blavatnik Family Foundation and GOSH Arts. Disclaimer: The views expressed in this publication are those of the authors and not necessarily those of the NHS, the National Institute for Health Research or the Department of Health. Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Evidence of an emerging/increasing female ‘excess’ in physical morbidity over the transition to adolescence is less well established than that in respect of psychological morbidity. Most studies of sex differences in child/adolescent physical morbidity focus on single/few conditions, so cannot investigate the degree to which results reflect a generalised pattern. What this study adds? We observed a predominant pattern of an emerging/increasing female ‘excess’ in a broad range of physical morbidity measures between ages 4 and 13 years (rather than no sex differences or an emerging/increasing male ‘excess’). Further studies are required to corroborate and explain these findings and understand long-term implications for sex differences in adult health. Introduction An emerging or increasing female ‘excess’ in psychological morbidity1 2 over the transition to adolescence is well recognised. Evidence of an emerging/increasing female ‘excess’ in several measures of common physical morbidity is less well established.3–5 Furthermore, since most epidemiological studies focus on single, or small groups of conditions, the degree to which this might reflect a generalised pattern in sex differences in physical morbidity has rarely been investigated.

emale ‘excess’ in several measures of common physical morbidity is less well established.3–5 Furthermore, since most epidemiological studies focus on single, or small groups of conditions, the degree to which this might reflect a generalised pattern in sex differences in physical morbidity has rarely been investigated. A previous systematic review and meta-analysis, based on a range of self-reported/routinely collected physical morbidity measures in Western children and adolescents, examined whether higher prevalence among males in childhood is replaced by higher prevalence among females in adolescence.5 It found an emerging/increasing female ‘excess’ with increasing age for self-reported general health and several specific symptoms. This pattern was strongest for headache, abdominal pain and tiredness and weaker for back pain and dizziness. It was also evident for self-reported migraine, but not two conditions based on routinely collected data, epilepsy (no sex differences) and type 1 diabetes (weak emerging/increasing male ‘excess’). The age when a female ‘excess’ was first evident varied by morbidity measure; around 6–8 years for self-assessed health, abdominal pain, dizziness and headache (earlier than expected given previous literature associating female puberty with these physical symptoms6–8), and around 9–11 years for back pain, sleeping problems/tiredness and migraine.

s’ was first evident varied by morbidity measure; around 6–8 years for self-assessed health, abdominal pain, dizziness and headache (earlier than expected given previous literature associating female puberty with these physical symptoms6–8), and around 9–11 years for back pain, sleeping problems/tiredness and migraine. More robust evidence for an emerging/increasing female ‘excess’ in self-report than routinely collected data5 suggests the need for careful consideration of the impact of data source in studies/reviews focusing on morbidity measured over the transition to adolescence. Changing sex differences with age may reflect underlying biological changes, or societal expectations, of which conditions, symptoms or infections are more ‘appropriate’ for either males or females at different life stages. Apparent sex-by-age differences in morbidity may therefore partly reflect whether data are self-reported (impossible at younger ages), proxy-reported (requiring awareness of symptoms by another, with or without them having been specifically informed by the sufferer) or routinely collected (requiring awareness of symptoms by the sufferer or another, followed by presentation to, and diagnosis by, health professionals).

rted (impossible at younger ages), proxy-reported (requiring awareness of symptoms by another, with or without them having been specifically informed by the sufferer) or routinely collected (requiring awareness of symptoms by the sufferer or another, followed by presentation to, and diagnosis by, health professionals). Further evidence from other reviews broadly suggests an emerging female ‘excess’ occurring around puberty in several measures of common physical morbidity, including asthma,9–12 eczema12 13 and respiratory infections,14 a consistent female excess in both urinary tract infections15 16 and musculoskeletal pain,17 and no clear sex-by-age pattern in food allergy.12 However, many such reviews do not consider the issue of data source,9 10 14 15 17 and while others refer to different methods/condition presentations, potential impacts on results are not considered systematically11 16 or at all.12 13 Similarly, some studies of sex-by-age differences in common childhood/adolescent conditions are unclear about data source,18 or based on parental report at younger ages and self-report thereafter, but without acknowledging this might have impacted on results.19 20 This paper presents analysis of sex differences in a wide range of parent/carer-reported (almost all mother-reported) physical morbidity measures (general health; conditions and symptoms; infections) between ages of 4 and 13 in a large UK birth cohort. This allows exploration of:Whether there is evidence of an emerging/increasing female ‘excess’ across these measures, reflecting a generalised pattern;

rted (almost all mother-reported) physical morbidity measures (general health; conditions and symptoms; infections) between ages of 4 and 13 in a large UK birth cohort. This allows exploration of:Whether there is evidence of an emerging/increasing female ‘excess’ across these measures, reflecting a generalised pattern; When any emerging female ‘excess’ occurs; Whether the sex-by-age patterns seen in the more restricted set of child/adolescent self-reported physical morbidity measures and described in a previous systematic review5 are replicated in parent-reported measures. Method Participants Data are from the Avon Longitudinal Study of Parents and Children (ALSPAC),21 22 a population-based cohort study in South-West England. Pregnant women with estimated delivery date 1 April 1991 to 31 December 1992, were invited to participate, resulting in a cohort of 14 541 pregnancies, with 13 988 singletons and first-born twins alive at 1 year. ALSPAC’s study website includes details of available data via a fully searchable data dictionary (www.bristol.ac.uk/alspac/researchers/access/). All data for this analysis were from ‘child-based’ questionnaires, completed by their main carer at roughly 1 year intervals, from approximately age 4 (57 months; n=4967 male and 5252 female questionnaires; 90% (n=9234) completed by mothers) to 13 years (166 months; n=3682 male and 3678 female questionnaires; 93% (n=6809) completed by mothers).

were from ‘child-based’ questionnaires, completed by their main carer at roughly 1 year intervals, from approximately age 4 (57 months; n=4967 male and 5252 female questionnaires; 90% (n=9234) completed by mothers) to 13 years (166 months; n=3682 male and 3678 female questionnaires; 93% (n=6809) completed by mothers). Measures Table 1 details the 32 morbidity measures analysed here and the child ages when they were included. Included measures comprised: three general health (health in past month and year, and health-related school absences); 19 conditions and symptoms (diarrhoea, vomiting, cough, high temperature, earache, ear discharge, stomach-ache, rash, wheezing, breathlessness, headache, constipation, lice/scabies, eczema, asthma, hay fever, pains in arms/legs, food/drink allergies, other allergies) and 10 infections (chicken pox, cold sores, eye infection, ear infection, chest infection, tonsillitis/laryngitis, influenza, cold, urinary infection, worm infections). Table 1 Morbidity measures and ages collected

Measures Table 1 details the 32 morbidity measures analysed here and the child ages when they were included. Included measures comprised: three general health (health in past month and year, and health-related school absences); 19 conditions and symptoms (diarrhoea, vomiting, cough, high temperature, earache, ear discharge, stomach-ache, rash, wheezing, breathlessness, headache, constipation, lice/scabies, eczema, asthma, hay fever, pains in arms/legs, food/drink allergies, other allergies) and 10 infections (chicken pox, cold sores, eye infection, ear infection, chest infection, tonsillitis/laryngitis, influenza, cold, urinary infection, worm infections). Table 1 Morbidity measures and ages collected Measure How coded for analysis/additional notes Age in months 57 65 69 81 91 103 128 140 157 166 General health How would you assess the health of your child in the past month? In the past year? ‘Sometimes quite ill’ or ‘almost always unwell’ (categorised here together as ‘poor’) vs ‘very healthy, no problems’ or ‘healthy, but a few minor problems’ X X X X X X X X X How many days the child had taken off school in the past year. For a range of health reasons (infections, asthma/eczema/hay fever, hospital or other investigations/admissions, other reasons). Any days off for any health reason vs none X X X X X Conditions and symptoms Has child had any of the following? Since age 3 (at 57 months); in the past 15 months (at 69 months); in the past year (all other ages) Diarrhoea X X X X X X X X Vomiting X X X X X X X X Cough X X X X X X X X High temperature X X X X X X X X Earache X X X X X X X X Ear discharge (pus) X X X X X X X X Stomach-ache(s) X X X X X X X X Rash X X X X X X X X Wheezing X X X X X X X X Breathlessness X X X X X X X X Headache(s) X X X X X X X X Constipation X X X X X X X X Lice or scabies Asked separately from 81 months; combined for consistency in analyses X X X X X X X X Eczema X X X X X X Asthma X X X X X X Hay fever X X X X X X Child often has pains in arms or legs X X X X X X Are there any foods or drinks that your child is allergic to? X X X X Apart from food and drink are there any other things to which child is allergic? X X X X Snuffles/cold* X X X X X Urinary infection† X X X X X Blood in stools‡ X X X X X X X X Convulsions/fits‡ X X X X X X X X Episodes of stopping breathing‡ X X X X X X X X Convulsion, fit or seizure due to epilepsy‡ X X X X X Accident§ X X X X X X X X Infections Has child had any of the following infections?

rgic? X X X X Snuffles/cold* X X X X X Urinary infection† X X X X X Blood in stools‡ X X X X X X X X Convulsions/fits‡ X X X X X X X X Episodes of stopping breathing‡ X X X X X X X X Convulsion, fit or seizure due to epilepsy‡ X X X X X Accident§ X X X X X X X X Infections Has child had any of the following infections? Since age 3 (at 57 months); in the past 15 months (at 69 months); in the past year (all other ages) Chicken pox X X X X X X X Cold sores X X X X X X X Eye infection X X X X X X X Ear infection X X X X X X X Chest infection X X X X X X X Tonsilitis/laryngitis X X X X Influenza X X X X Cold* X X X X Urinary infection† X X X X X X X Worm infections¶ X X X X X X X X Measles‡ X X X X X X X Mumps‡ X X X X X X X Meningitis‡ X X X X X X X Whooping cough‡ X X X X X X X German measles‡ X X X X Scarlet fever‡ X X X X Glandular fever** X *Reported by parent as ‘cold/snuffles’ (within conditions list) at 57, 69, 81, 91 and 103 months and ‘cold’ (within infections list) at 128, 157 and 166 months were combined and included here as ‘cold’ (categorised under infections). †Reported by parent within both conditions (57–103 months) and infections (57–166 months) lists, so only results in respect of infections data included here. ‡Excluded from analyses as reported in respect of very small numbers (fewer than 20 males and/or females). §Excluded from analyses as out with the scope of the paper. ¶Reported by parent within conditions list, but included here within infections. **Excluded from analysis as only one time point.

†Reported by parent within both conditions (57–103 months) and infections (57–166 months) lists, so only results in respect of infections data included here. ‡Excluded from analyses as reported in respect of very small numbers (fewer than 20 males and/or females). §Excluded from analyses as out with the scope of the paper. ¶Reported by parent within conditions list, but included here within infections. **Excluded from analysis as only one time point. Analyses For each morbidity measure, age-specific logistic regression analyses determined the odds of it being reported in respect of females (vs males). Sex-by-age interactions (testing for a changing sex difference with age) were then included in further logistic regression analyses, based on reports at all ages, with robust SEs to allow for non-independence of observations from the same child.

determined the odds of it being reported in respect of females (vs males). Sex-by-age interactions (testing for a changing sex difference with age) were then included in further logistic regression analyses, based on reports at all ages, with robust SEs to allow for non-independence of observations from the same child. Results are presented as graphs for each measure, showing female versus male ORs at each age. The graphs have logarithmic scaling (eg, ORs of 2 (females twice as likely) and 0.5 (females half as likely) are the same distance from 1 (no sex difference)). The graphs are presented in three sections (general health; conditions and symptoms; infections) and, within these, according to potential patterns of sex-by-age differences, which a previous systematic review conceptualised in terms of four ‘types’,5 and are defined here as:‘Type 1’: an emerging/increasing female ‘excess’, or disappearing male excess with age, occurring because female rates increase more than those of males or decrease less than those of males, resulting in a marked sex-by-age interaction. Type 1 patterns therefore include: (a) a male ‘excess’ reversing to a female ‘excess’, or no sex difference at younger ages, but a female ‘excess’ at older ages (emerging female ‘excess’); (b) a female ‘excess’ at younger ages, increasing with age (increasing female ‘excess’) or (c) a male ‘excess’ at younger ages, but no sex difference at older ages (disappearing male ‘excess’). For ‘type 1’ patterns, the odds of morbidity among females compared with males start below, at, or above unity and increase with age.

emale ‘excess’ at younger ages, increasing with age (increasing female ‘excess’) or (c) a male ‘excess’ at younger ages, but no sex difference at older ages (disappearing male ‘excess’). For ‘type 1’ patterns, the odds of morbidity among females compared with males start below, at, or above unity and increase with age. ‘Type 2’: (a) stable female ‘excess’; (b) stable lack of a sex difference or (c) stable male ‘excess’. For ‘type 2’ patterns, the odds of female versus male morbidity are consistently either above, below or at unity. ‘Type 3’: variations on an emerging/increasing male ‘excess’, or disappearing female ‘excess’ (the reverse of ‘type 1’). For ‘type 3’ patterns, the odds of female versus male morbidity start above, at or below unity and decrease with age. ‘Type 4’: mixed/unclassifiable patterns. (Note precise definitions are included as footnotes to the results graphs.)

‘Type 3’: variations on an emerging/increasing male ‘excess’, or disappearing female ‘excess’ (the reverse of ‘type 1’). For ‘type 3’ patterns, the odds of female versus male morbidity start above, at or below unity and decrease with age. ‘Type 4’: mixed/unclassifiable patterns. (Note precise definitions are included as footnotes to the results graphs.) Results are based on cross-sectional samples at each age. Two sets of sensitivity analyses were completed. One was restricted to a longitudinal subsample, selecting only those for whom data were available at all relevant ages, n=4454. The other was conducted for the ‘mother-only’ subsample, selecting only mother-completed data, to investigate whether completion by different carers at different ages might impact on the results. Online supplementary table 1 shows cross-sectional, longitudinal and mother-only sample sizes. Online supplementary table 2 shows the characteristics of those included/not included in the samples and demonstrates that due to sample attrition, those included were more likely to be first-born children of mothers in a first marriage, of higher socioeconomic status, and who had never smoked. Results from both the longitudinal and mother-only samples were almost identical to the cross-sectional results. Online supplementary tables 3–5 show results based on each sample. Finally, given that some have suggested Poisson as an alternative to logistic regression for analysis of cross-sectional studies with binary outcomes,23 we also conducted Poisson regression analyses on the cross-sectional samples. Results (expressed as risk ratios, rather than ORs) were very similar to those obtained via logistic regression (see online supplementary tables 6–8).

tive to logistic regression for analysis of cross-sectional studies with binary outcomes,23 we also conducted Poisson regression analyses on the cross-sectional samples. Results (expressed as risk ratios, rather than ORs) were very similar to those obtained via logistic regression (see online supplementary tables 6–8). 10.1136/bmjpo-2017-000191.supp1Supplementary file 1 Results General health Figure 1 shows a ‘type 1a’ female ‘excess’ emerging by 128 months for parent-reported poor general health in their child over the last month (OR=1.02, 95% CI 0.76 to 1.36 at 57 months; OR=1.97, 95% CI 1.26 to 3.10 at 166 months; sex-by-age interaction P=0.01) and the last year (OR=0.85, 95% CI 0.66 to 1.08 at 57 months; OR=1.50, 95% CI 1.07 to 2.11 at 166 months; interaction P<0.001). Days off school in the last year was unavailable at the youngest ages, but showed a small consistent female ‘excess’ (‘type 2a’ pattern) from 91 to 166 months (OR=1.15, 95% CI 1.04 to 1.28 at 91 months; OR=1.18, 95% CI 1.05 to 1.33 at 166 months; interaction P=0.68). Figure 1 General health measures (with P values for significance of sex-by-age interactions).

Results General health Figure 1 shows a ‘type 1a’ female ‘excess’ emerging by 128 months for parent-reported poor general health in their child over the last month (OR=1.02, 95% CI 0.76 to 1.36 at 57 months; OR=1.97, 95% CI 1.26 to 3.10 at 166 months; sex-by-age interaction P=0.01) and the last year (OR=0.85, 95% CI 0.66 to 1.08 at 57 months; OR=1.50, 95% CI 1.07 to 2.11 at 166 months; interaction P<0.001). Days off school in the last year was unavailable at the youngest ages, but showed a small consistent female ‘excess’ (‘type 2a’ pattern) from 91 to 166 months (OR=1.15, 95% CI 1.04 to 1.28 at 91 months; OR=1.18, 95% CI 1.05 to 1.33 at 166 months; interaction P=0.68). Figure 1 General health measures (with P values for significance of sex-by-age interactions). Conditions and symptoms Figure 2 shows that for 6 of the 20 conditions/symptoms measures, there was an emerging/increasing female ‘excess’, while one showed a disappearing male ‘excess’ (all ‘type 1’ patterns). Thus, although there were no sex differences at younger ages, a female ‘excess’ emerged in respect of rates of parent-reported high temperature by 128 months and, more markedly, rash by 81 months (interaction P=0.01 and <0.001, respectively). Earache, stomach-ache, headache and head lice/scabies were more likely to be reported in respect of females at younger ages, but this female ‘excess’ increased with age (interaction P<0.001). Reported breathlessness also showed a type 1 sex-by-age interaction, from a male ‘excess’ at younger ages which disappeared, resulting in no sex difference at older ages (OR=0.63, 95% CI 0.54 to 0.73 at 57 months; OR=1.05, 95% CI 0.87 to 1.26 at 166 months; interaction P<0.001).

ncreased with age (interaction P<0.001). Reported breathlessness also showed a type 1 sex-by-age interaction, from a male ‘excess’ at younger ages which disappeared, resulting in no sex difference at older ages (OR=0.63, 95% CI 0.54 to 0.73 at 57 months; OR=1.05, 95% CI 0.87 to 1.26 at 166 months; interaction P<0.001). Figure 2 Conditions/symptoms measures (with P values for significance of sex-by-age interactions). Ten conditions/symptoms showed stable (‘type 2’) sex differences/lack of sex differences with age. Thus, two (constipation and eczema) showed a female ‘excess’ at almost all ages and three a consistent male ‘excess’ (wheezing, asthma and hay fever). A further five were largely consistent in showing no marked sex difference at any age (cough, vomiting, ear discharge, food and other allergy). Finally, two conditions/symptoms (pain in arms/legs and diarrhoea) showed an emerging/increasing male ‘excess’ (‘type 3’ patterns). There was no sex difference in respect of pain in arms/legs at younger ages (OR=0.95, 95% CI 0.85 to 1.07 at 57 months), but a small male ‘excess’ emerged by 140 months and was maintained at 157 months (OR=0.81, 95% CI 0.74 to 0.90 at 157 months; interaction P=0.001). Diarrhoea showed an increasing male ‘excess’, being more likely to be reported in males at all ages, particularly older ages (OR=0.89, 95% CI 0.82 to 0.96 at 57 months; OR=0.77, 95% CI 0.69 to 0.85 at 166 months; interaction P=0.004).

ined at 157 months (OR=0.81, 95% CI 0.74 to 0.90 at 157 months; interaction P=0.001). Diarrhoea showed an increasing male ‘excess’, being more likely to be reported in males at all ages, particularly older ages (OR=0.89, 95% CI 0.82 to 0.96 at 57 months; OR=0.77, 95% CI 0.69 to 0.85 at 166 months; interaction P=0.004). Infections Figure 3 shows an emerging/increasing female ‘excess’ in 4 of the 11 parent-reported infections (‘type 1’ patterns), a consistent female ‘excess’ in 3, a consistent male ‘excess’ in 1 and no sex difference in 2 (‘type 2’ patterns). Figure 3 Infections measures (with P values for significance of sex-by-age interactions). Among infections showing an emerging/increasing female ‘excess’, there were no sex differences in parent-reported child eye and ear infections at younger ages, but a female ‘excess’ emerged for eye infections by 103 months and ear infections by 81 months (interaction P<0.001 for both). Parent-reported cold sores and, more markedly, urinary infections (note different scale on graph) were higher in females0 at all ages, but this sex difference increased with age (eg, for urinary infections, OR=2.35, 95% CI 1.97 to 2.81 at 57 months; OR=5.13, 95% CI 3.30 to 7.98 at 166 months; interaction P<0.001). Finally, among the infections showing ‘type 2’ patterns, there was a consistent female ‘excess’ in respect of reported tonsillitis, cold/snuffles and worm infections, no sex difference in chicken pox and influenza and a consistent male ‘excess’ in respect of chest infections.

o 7.98 at 166 months; interaction P<0.001). Finally, among the infections showing ‘type 2’ patterns, there was a consistent female ‘excess’ in respect of reported tonsillitis, cold/snuffles and worm infections, no sex difference in chicken pox and influenza and a consistent male ‘excess’ in respect of chest infections. Discussion This is the first study to examine sex differences in a range of parent-reported physical morbidity measures during childhood and adolescence and explore age-based changes.(Table 2) Table 2 Summary of patterns of sex-by-age differences in each morbidity measure

o 7.98 at 166 months; interaction P<0.001). Finally, among the infections showing ‘type 2’ patterns, there was a consistent female ‘excess’ in respect of reported tonsillitis, cold/snuffles and worm infections, no sex difference in chicken pox and influenza and a consistent male ‘excess’ in respect of chest infections. Discussion This is the first study to examine sex differences in a range of parent-reported physical morbidity measures during childhood and adolescence and explore age-based changes.(Table 2) Table 2 Summary of patterns of sex-by-age differences in each morbidity measure General health Conditions and symptoms Infections ‘Type 1’ 1a: Emerging female excess General health – Past year General health – Past month High temperature Rash Eye infection Ear infection 1b: Increasing female excess Earache Stomach-ache Headache Lice/scabies Cold sores Urinary infection 1c: Disappearing male excess Breathlessness ‘Type 2’ 2a: Consistent female excess Health-related days off school Constipation Eczema Tonsillitis/laryngitis Cold Worm infections 2b: Consistent no sex difference Vomiting Cough Ear discharge Food/drink allergies Non-food/drink allergies Chicken pox Influenza 2c: Consistent male excess Wheezing Asthma Hay fever Chest infection ‘Type 3’ 3a: Emerging male excess Pain in arms/legs 3b: Increasing male excess Diarrhoea Given evidence of a generalised pattern for psychological morbidity measures, we were interested in whether there was also a generalised pattern of an emerging/increasing female ‘excess’ across these physical morbidity measures and, if so, when this occurred. As summarised in table 2, of the 32 measures examined, only 7 showed no sex differences throughout the included age ranges. Six were categorised as showing an emerging female ‘excess’, six an increasing female ‘excess’ one a disappearing male ‘excess’ (‘type 1’ patterns) and six a consistent female ‘excess’. In contrast, only one showed an emerging male ‘excess’, one an increasing male ‘excess’ (‘type 3’ patterns) and four a consistent male ‘excess’. Thus, far more measures showed an emerging/increasing female ‘excess’ than an emerging/increasing male ‘excess’ or no sex difference.

ns) and six a consistent female ‘excess’. In contrast, only one showed an emerging male ‘excess’, one an increasing male ‘excess’ (‘type 3’ patterns) and four a consistent male ‘excess’. Thus, far more measures showed an emerging/increasing female ‘excess’ than an emerging/increasing male ‘excess’ or no sex difference. We also wished to know whether sex-by-age patterns in a more restricted set of child/adolescent self-reported physical morbidity measures described in a previous review,5 were replicated when these measures were parent-reported, as here. Three measures (general health, headache, abdominal pain) are included in both studies. The review found a marked female ‘excess’ in each, based on self-reports, from around 6–8 years. The current analysis, based on parent-reported measures, had broadly similar findings, but suggested an even earlier small female ‘excess’ in headache and stomach-ache.

eadache, abdominal pain) are included in both studies. The review found a marked female ‘excess’ in each, based on self-reports, from around 6–8 years. The current analysis, based on parent-reported measures, had broadly similar findings, but suggested an even earlier small female ‘excess’ in headache and stomach-ache. How can these complex patterns of sex differences in parent-reported morbidity be explained? Female puberty is associated with physical symptoms, including menstrual cramps and headaches.6–8 Although previous literature suggests higher male rates of asthma, eczema, respiratory infections and perhaps hay fever at younger ages, reversing around puberty,9–14 18 19 we observed a consistent male ‘excess’ in asthma, wheeze, hay fever and chest infections throughout the age range considered here. It is possible that a ‘reversal’ occurred later in puberty in the ALSPAC cohort. However, the consistently higher female eczema rates cannot be explained in this way. While potential explanations might be constructed for some findings (eg, the increasing female ‘excess’ in lice/scabies throughout childhood might result from girls’ often longer hair and/or greater time spent in physically close social interactions24), others, such as an emerging/increasing female ‘excesses’ in temperature, rash, earache/infection, eye infection or cold sores are harder to explain.

asing female ‘excess’ in lice/scabies throughout childhood might result from girls’ often longer hair and/or greater time spent in physically close social interactions24), others, such as an emerging/increasing female ‘excesses’ in temperature, rash, earache/infection, eye infection or cold sores are harder to explain. Another potential explanation is that some of these sex differences in parent-reported morbidity measures result from different illness-related attitudes/expectations (by both children and parents) in respect of males compared with females. Parental expectations about, and reinforcement of, their child’s emotional expressivity differ according to child sex,25 as do ratings of, and responses to, paediatric pain.26 27 Perhaps these translate into differences in acknowledging, recalling and reporting illness in respect of boys and girls. There are few studies in this area and stereotyped attitudes and expectations relating to sex differences may differ according to what aspect of child morbidity is being considered.

iatric pain.26 27 Perhaps these translate into differences in acknowledging, recalling and reporting illness in respect of boys and girls. There are few studies in this area and stereotyped attitudes and expectations relating to sex differences may differ according to what aspect of child morbidity is being considered. The strengths of this study include its large sample size and wide range of morbidity measures, enabling us to address previously unexplored questions. Analyses based on cross-sectional samples produced almost identical results to those limited to the longitudinal sample (which was subject to differential attrition) and mother-only samples (which eliminated different carers reporting on a child’s health at different ages). Findings such as the much higher rates of urinary infections among females at all ages and markedly increasing female ‘excess’ in stomach and headaches in early adolescence, are consistent with other research.6–8 15 16 However, we also draw attention to potential limitations. Our categorisation of patterns of sex-by-age differences in physical morbidity, based on graphs and interactions, could be regarded as simple, and the fact that we conducted analyses on 32 morbidity measures introduces the possibility of spurious (chance) significance for some sex-by-age interactions due to multiple testing. However, our focus was on consistent patterns of ORs with age and these are unlikely to have arisen purely by chance. In addition, there is no reason to think that spurious interactions would occur more often for measures showing a ‘type 1’ pattern. Another limitation is that this secondary analysis of existing data from a well-established cohort was inevitably limited by the specific measures chosen by the ALSPAC team at each age. In particular, lack of comparable data at older ages prevents extension of the analysis to mid-later adolescence. Furthermore, although there was little evidence of differential attrition according to child sex, the sample was more advantaged than the general population, thus potentially limiting generalisability of the findings.

ar, lack of comparable data at older ages prevents extension of the analysis to mid-later adolescence. Furthermore, although there was little evidence of differential attrition according to child sex, the sample was more advantaged than the general population, thus potentially limiting generalisability of the findings. Based on rigorous analysis questioning, we believe for the first time, whether age-based changes in sex differences in child and adolescent physical morbidity follow specific patterns (as proposed in a prior review5), this analysis suggests both substantive and methodological conclusions. Substantively, it is intriguing that sex differences are evident in respect of a wide range of parent-reported physical morbidity measures in childhood and early adolescence, generally indicating poorer health in girls. While some measures show consistent female or male ‘excesses’, many show an emerging/increasing female ‘excess’ in childhood, consistent with findings for psychological morbidity,1 2 which are evident prepuberty; almost none shows an emerging/increasing male ‘excess’. This pattern of ‘excess’ female morbidity by/before puberty highlights important inequities, with public health implications, not least in future health service usage. Many of these (changing) sex differences are hard to explain on the basis of existing literature, suggesting, as in adults,28 the need for further quantitative studies to corroborate these findings and to examine whether patterns differ across sociodemographic or cultural groups. There is also a need for further qualitative or experimental studies to examine the social and/or biological mechanisms underlying these findings.29 30 Methodologically, possible differences according to data source (routine data, child/proxy-reported) highlight the need for reviews of sex-by-age differences in child/adolescent morbidity to pay close attention to this issue and for studies which systematically compare results based on multiple sources.

findings.29 30 Methodologically, possible differences according to data source (routine data, child/proxy-reported) highlight the need for reviews of sex-by-age differences in child/adolescent morbidity to pay close attention to this issue and for studies which systematically compare results based on multiple sources. The authors are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. Contributors: HS conceptualised and designed the study with KH, lead the data request and drafted all versions of the manuscript. EW advised on the data request, carried out the analyses, produced the graphical results and commented on all versions of the manuscript. AT advised on the data request, conducted the data extraction and commented on later versions of the manuscript. KH conceptualised and designed the study with HS and commented on all versions of the manuscript. All authors approved the final manuscript as submitted.

ults and commented on all versions of the manuscript. AT advised on the data request, conducted the data extraction and commented on later versions of the manuscript. KH conceptualised and designed the study with HS and commented on all versions of the manuscript. All authors approved the final manuscript as submitted. Funding: This work was supported by the UK Medical Research Council, Wellcome Trust and University of Bristol. The UK Medical Research Council and Wellcome (Grant ref: 102215/2/13/2) and the University of Bristol provide core support for ALSPAC. HS, EW and KH are funded by the UK Medical Research Council MC_UU_12017/12 and MC_UU_12017/13. AT is supported by PEARL (Project to Enhance ALSPAC through Record Linkage), a programme of research funded by the Wellcome Trust (WT086118/Z/08/Z). Competing interests: None declared. Ethics approval: Ethical approval was obtained from the ALSPAC Law and Ethics Committee and local research ethics committee. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: ALSPAC’s study website includes details of available data via a fully searchable data dictionary (www.bristol.ac.uk/alspac/researchers/access/). Our analysis was based on variables which we specifically requested.

What is already known on this topic? Orlistat and metformin are both used as antiobesity drugs (AOD) in children and young people, yet only orlistat is approved in this age group. UK primary care prescribing data show increasing use and high levels of drug discontinuation, with half of orlistat prescriptions not being continued beyond 1 month. One qualitative study eshowed frequent cessation by families independent of their doctors, usually because the perceived advantages did not outweigh the medication side effects. What this study hopes to add? Orlistat was largely prescribed independently by general practitioners to patients aged 16 years and over without physical comorbidities. Metformin was largely initiated by specialists for subjects with comorbidities, including polycystic ovarian syndrome, insulin resistance and impaired glucose tolerance. Adherence to the National Institute for Health and Care Excellence guidelines for orlistat prescribing to children and young people was low.

What this study hopes to add? Orlistat was largely prescribed independently by general practitioners to patients aged 16 years and over without physical comorbidities. Metformin was largely initiated by specialists for subjects with comorbidities, including polycystic ovarian syndrome, insulin resistance and impaired glucose tolerance. Adherence to the National Institute for Health and Care Excellence guidelines for orlistat prescribing to children and young people was low. Introduction Little is known about use of medication for obesity in children and adolescents in the UK, particularly use in primary care. Orlistat is currently the only licensed antiobesity drug (AOD) in the UK since sibutramine was withdrawn due to concerns about cardiovascular safety.1 2 However, the most commonly used drug for obesity in children and young people (CYP) is metformin, an antidiabetes drug used off-licence to treat the metabolic sequelae of obesity in CYP, although not formally classed as an AOD.3 4 Both orlistat and metformin appear to offer small benefits for body mass index (BMI) loss in CYP; systematic reviews show small reductions in BMI compared with placebo, orlistat by 0.83 kg/m25 and metformin by 1.4 kg/m2 (at 6–12 months and 6 months, respectively).6

ty in CYP, although not formally classed as an AOD.3 4 Both orlistat and metformin appear to offer small benefits for body mass index (BMI) loss in CYP; systematic reviews show small reductions in BMI compared with placebo, orlistat by 0.83 kg/m25 and metformin by 1.4 kg/m2 (at 6–12 months and 6 months, respectively).6 In the UK, the National Institute for Health and Care Excellence (NICE) guidance recommends community-based lifestyle modification programmes as the first tier of weight management for childhood obesity, with pharmacotherapy as a second-line treatment.2 Their guidance, summarised in box 1, only covers use of orlistat, which they state should be prescribed only in exceptional circumstances for those with obesity-related comorbidities (life-threatening in those under 12 years of age) and only prescribed by teams with expertise in these conditions.Box 1 Summary of 2014National Institute for Health and Care Excellence guidance for prescribing of orlistat to children and young people 1.8.4—Drug treatment is not generally recommended for children younger than 12 years. 1.8.5—In children younger than 12 years, drug treatment may be used only in exceptional circumstances, if severe comorbidities are present. Prescribing should be started and monitored only in specialist paediatric settings.

In the UK, the National Institute for Health and Care Excellence (NICE) guidance recommends community-based lifestyle modification programmes as the first tier of weight management for childhood obesity, with pharmacotherapy as a second-line treatment.2 Their guidance, summarised in box 1, only covers use of orlistat, which they state should be prescribed only in exceptional circumstances for those with obesity-related comorbidities (life-threatening in those under 12 years of age) and only prescribed by teams with expertise in these conditions.Box 1 Summary of 2014National Institute for Health and Care Excellence guidance for prescribing of orlistat to children and young people 1.8.4—Drug treatment is not generally recommended for children younger than 12 years. 1.8.5—In children younger than 12 years, drug treatment may be used only in exceptional circumstances, if severe comorbidities are present. Prescribing should be started and monitored only in specialist paediatric settings. 1.8.6—In children aged 12 years and older, treatment with orlistat is recommended only if physical comorbidities (such as orthopaedic problems or sleep apnoea) or severe psychological comorbidities are present. Treatment should be started in a specialist paediatric setting, by multidisciplinary teams with experience of prescribing in this age group. 1.8.7—Do not give orlistat to children for obesity unless prescribed by a multidisciplinary team with expertise in drug monitoring, psychological support, behavioural interventions, interventions to increase physical activity and interventions to improve diet.

1.8.6—In children aged 12 years and older, treatment with orlistat is recommended only if physical comorbidities (such as orthopaedic problems or sleep apnoea) or severe psychological comorbidities are present. Treatment should be started in a specialist paediatric setting, by multidisciplinary teams with experience of prescribing in this age group. 1.8.7—Do not give orlistat to children for obesity unless prescribed by a multidisciplinary team with expertise in drug monitoring, psychological support, behavioural interventions, interventions to increase physical activity and interventions to improve diet. 1.8.8—Drug treatment may be continued in primary care, for example, with a shared care protocol if local circumstances and/or licensing allow. 1.9.2—Adults and children: If there is concern about micronutrient intake adequacy, a supplement providing the reference nutrient intake for all vitamins and minerals should be considered, particularly for vulnerable groups such as older people (who may be at risk of malnutrition) and young people (who need vitamins and minerals for growth and development). 1.9.11—If orlistat is prescribed for children, a 6-month to 12-month trial is recommended, with regular review to assess effectiveness, adverse effects and adherence.

1.9.2—Adults and children: If there is concern about micronutrient intake adequacy, a supplement providing the reference nutrient intake for all vitamins and minerals should be considered, particularly for vulnerable groups such as older people (who may be at risk of malnutrition) and young people (who need vitamins and minerals for growth and development). 1.9.11—If orlistat is prescribed for children, a 6-month to 12-month trial is recommended, with regular review to assess effectiveness, adverse effects and adherence. Randomised trial data on orlistat and metformin come from specialist clinical settings and largely from outside the UK. Very little is known about how these AODs are prescribed and used in actual practice. Pharmacoepidemiology studies of AOD prescribing in primary care in the UK show increasing use of AODs, but also high levels of drug discontinuation, with approximately half the prescriptions of orlistat not being continued beyond 1 month.5 The one qualitative study examining adolescent use of AOD showed frequent cessation by families independent of their doctors, usually because the perceived advantages did not outweigh the medication side effects that they endured with often minimal professional support.7 These data suggest that the effectiveness of AOD in ‘real life’ settings may be considerably less than shown in trials, and suggest a need to identify strategies to improve the effectiveness of AODs for CYP.

antages did not outweigh the medication side effects that they endured with often minimal professional support.7 These data suggest that the effectiveness of AOD in ‘real life’ settings may be considerably less than shown in trials, and suggest a need to identify strategies to improve the effectiveness of AODs for CYP. We undertook a questionnaire survey of general practitioners (GPs) prescribing AODs to CYP to better understand their use in primary care in the UK. We sought to characterise patient demographics, quantify adherence to NICE guidance and identify primary care perceptions of AOD with the long-term aim of optimising AOD prescribing and efficacy. Materials and methods We used routinely collected primary care data from The Health Improvement Network (THIN) database to identify CYP aged up to and including 18 years prescribed orlistat or metformin between 31 May 2010 and 31 May 2012. We excluded patients prescribed metformin for type 2 diabetes. THIN covers approximately 6% of the UK population, with 3.6 million active patients from 587 general practices using the Vision General Practice System.8 These practices are broadly representative of practices in the UK in respect of patients’ demographics and characteristics.9 Questionnaire administration was undertaken by THIN Additional Information Services (THIN AIS), an independent research organisation affiliated with THIN, with data protection firewalls.

System.8 These practices are broadly representative of practices in the UK in respect of patients’ demographics and characteristics.9 Questionnaire administration was undertaken by THIN Additional Information Services (THIN AIS), an independent research organisation affiliated with THIN, with data protection firewalls. A paper questionnaire was sent to the GP practices of all identified CYP to collect patient-level data (see online supplementary appendix file 1 for full questionnaire). The questionnaire was designed by two paediatricians, an academic GP, a psychologist, a pharmacist and a GP representative following recommendations for good practice in survey research.10 GPs were contacted up to three times over 3 months until the questionnaire was returned. GPs received a £35 payment for each completed and returned questionnaire. THIN AIS anonymised questionnaires prior to analysis by the study team. 10.1136/bmjpo-2017-000104.supp1Supplementary file 1 Year of birth, practice ID and region were provided by THIN AIS. All other data were provided by GPs using existing medical records, including ethnicity. We assumed AOD termination if no prescription had been issued within 3 months of the survey. Age at first prescription was calculated from the midpoint of birth year, as month and day of birth were not provided due to data protection restrictions.

ere provided by GPs using existing medical records, including ethnicity. We assumed AOD termination if no prescription had been issued within 3 months of the survey. Age at first prescription was calculated from the midpoint of birth year, as month and day of birth were not provided due to data protection restrictions. BMI and zBMI were calculated from GP-derived height and weight measurements using the LMS method and UK reference data.11 We audited orlistat use against NICE 2006 recommendations, which remain unchanged in the 2014 update, bar some text clarifications.2 12 For audits against NICE criteria only, we assumed birth date of 1 January to ensure that no subjects were misclassified as children if they were 18 years of age. Questionnaire responses were read by two researchers and any differences agreed. Analyses Analyses were conducted using STATA V.11.0. Simple descriptive statistics were used for the majority of data. Duration of drug use was compared using Wilcoxon-Mann-Whitney test (highly skewed data) and paired Likert scores using Wilcoxon signed-rank test. Handwritten free-text comments were read and coded using a general thematic coding methodology.13 Models were developed through an iterative process, in which the initial model was reviewed using constant comparison techniques (in which successive items of data are appraised and compared to ensure the code is reflective of all) and the models revised accordingly.

coded using a general thematic coding methodology.13 Models were developed through an iterative process, in which the initial model was reviewed using constant comparison techniques (in which successive items of data are appraised and compared to ensure the code is reflective of all) and the models revised accordingly. Results Patient demographics Figure 1 summarises patient sampling; 151 patients were identified on THIN database from 108 unique GP practices, with 79% GP response rate (119 of 151 identified patients) from 84 unique practices. A total of 94 subjects were eligible (86% female, 45% British, 31% white/Caucasian, 4% Asian, 4% other, 16% unknown ethnicity). The majority came from England (79%), with the remaining from Wales (12%), Scotland (7%) and Northern Ireland (2%). A total of 99 AOD initiations occurred in 94 subjects (five subjects were prescribed both orlistat and metformin), consisting of 44 metformin (47% of sample) and 55 orlistat (59%) prescriptions. Drugs were initiated in 68 practices, with 46 practices prescribing one drug each, 15 practices two drugs each, 6 practices three drugs each and 1 practice prescribing five drugs. Table 1 summarises baseline demographic and comorbidities by drug. Comorbidities appeared higher in those taking metformin. BMI and zBMI data were available for 91% (40/44) prescribed metformin and 89% (49/55) prescribed orlistat. All had BMI above the 98th centile (>2 SD). Prescriptions for metformin and orlistat increased with age, with orlistat largely prescribed to those aged 16 years or above.

appeared higher in those taking metformin. BMI and zBMI data were available for 91% (40/44) prescribed metformin and 89% (49/55) prescribed orlistat. All had BMI above the 98th centile (>2 SD). Prescriptions for metformin and orlistat increased with age, with orlistat largely prescribed to those aged 16 years or above. Figure 1 Patient sampling. THIN = The Health Improvement Network, GP = general practitioner, AOD = anti-obesity drug. Table 1 Demographics and comorbidities by drug Metformin Orlistat n (total, % GP-initiated) 44 (27%) 55 (89%) Female (n, %) 40 (91%) 46 (84%) BMI (mean kg/m2, SD) 35.9 (6.1) 37.6 (6.5) zBMI (mean, SD) 3.2 (0.7) 3.2 (0.6) Median age (range) 15.7 (6.5–19.2) 17.3 (13.8–18.8) <12 years (n) 5 0 12–15.9 years (n) 19 9 16–17.9 years (n) 16 29 ≥18 years (n) 4 17 Comorbidities (n,%) Hypertension 1 (2%) 0 Hyperinsulinism/insulin resistance 13 (30%) 0 Type 2 diabetes 0 3 (5%) Dyslipidaemia 1 (2%) 3 (5%) Emotional distress 12 (27%) 17 (31%) Sleep apnoea 0 1 (2%) Polycystic ovarian syndrome 32 (73%) 6 (11%) Orthopaedic issues 3 (7%) 3 (5%) Pervasive developmental disorder 3 (7%) 0 Hypothyroidism 1 (2%) 3 (5%) BMI, body mass index; Drug initiation Figure 2 summarises the frequency of drug prescription by age and drug initiator; 89% (49/55) of orlistat and 27% (12/44) metformin prescriptions were initiated in primary care independent of specialist advice. Orlistat was recommended by paediatricians (n=3), an adult physician, lipid clinic and dietitian, and metformin by paediatricians (n=19), gynaecologists (7), adult physicians (4) and endocrinologists (1).

f orlistat and 27% (12/44) metformin prescriptions were initiated in primary care independent of specialist advice. Orlistat was recommended by paediatricians (n=3), an adult physician, lipid clinic and dietitian, and metformin by paediatricians (n=19), gynaecologists (7), adult physicians (4) and endocrinologists (1). Indications for metformin initiation were obesity together with (1) polycystic ovarian syndrome (70%, 31/44), (2) insulin resistance (25%, 11/44), (3) impaired glucose tolerance/impaired fasting glucose (9%, 4/44) and (4) obesity without known comorbidity (7%, 3/44). Drug monitoring Medication monitoring in primary care was undertaken in 67% where initiated independently and 26% on specialist recommendation. GPs were made aware of adverse drug effects by two patients, both prescribed metformin; one had diarrhoea and the other nausea. Drug duration and termination Duration of drug prescription is summarised in figure 3. The median supply of metformin was 10.5 months (IQR 4–18.5 months) compared with 2.0 months (1.0–4.0) for orlistat (p≤0.001). Over half of all metformin prescriptions (25/44) but only 5% of orlistat prescriptions (3/55) were active at the time of survey, defined as a new prescription issued within the preceding 3 months. There was a disparity between reported length of drug prescription and the amount of drug prescribed, suggesting non-continuous use at dose prescribed. Figure 2 Bar graph summarising age at initiation by drug and initiator. GP, general practitioner.

Drug duration and termination Duration of drug prescription is summarised in figure 3. The median supply of metformin was 10.5 months (IQR 4–18.5 months) compared with 2.0 months (1.0–4.0) for orlistat (p≤0.001). Over half of all metformin prescriptions (25/44) but only 5% of orlistat prescriptions (3/55) were active at the time of survey, defined as a new prescription issued within the preceding 3 months. There was a disparity between reported length of drug prescription and the amount of drug prescribed, suggesting non-continuous use at dose prescribed. Figure 2 Bar graph summarising age at initiation by drug and initiator. GP, general practitioner. Figure 3 Kaplan-Meier survival curve demonstrating treatment duration of metformin and orlistat. Figure shows proportion still actively prescribed antiobesity drug for orlistat and metformin by time since initiation, beginning from time 0 (100% active prescriptions) out to 85 months (longer active prescription). Twenty-seven patients had only a single prescription issued from primary care, being 45% (25/55) of all orlistat and 5% (2/44) metformin treatments. None of these single prescriptions were issued in the 3 months prior to the survey, and all were given a maximum of 1-month supply, making ongoing use highly unlikely.

Figure 3 Kaplan-Meier survival curve demonstrating treatment duration of metformin and orlistat. Figure shows proportion still actively prescribed antiobesity drug for orlistat and metformin by time since initiation, beginning from time 0 (100% active prescriptions) out to 85 months (longer active prescription). Twenty-seven patients had only a single prescription issued from primary care, being 45% (25/55) of all orlistat and 5% (2/44) metformin treatments. None of these single prescriptions were issued in the 3 months prior to the survey, and all were given a maximum of 1-month supply, making ongoing use highly unlikely. The majority of all drug terminations were due to families not requesting repeat prescriptions (96% of orlistat and 89% metformin) rather than medically led terminations. GPs reported possible orlistat cessation in three cases due to lack of drug supply in pharmacies. Of four prescriptions actively terminated by a doctor (metformin=2, orlistat=2), two were due to lack of efficacy, one for lack of drug adherence, and the other two for reasons unknown. Adherence to NICE guidance We restricted NICE compliance analysis to recommendations for children; 23 subjects were identified as definitely aged less than 18 years at drug initiation. All subjects were aged over 12 years (recommendation 1.8.4), and no participants were prescribed orlistat for over 12 months (1.9.11).

The majority of all drug terminations were due to families not requesting repeat prescriptions (96% of orlistat and 89% metformin) rather than medically led terminations. GPs reported possible orlistat cessation in three cases due to lack of drug supply in pharmacies. Of four prescriptions actively terminated by a doctor (metformin=2, orlistat=2), two were due to lack of efficacy, one for lack of drug adherence, and the other two for reasons unknown. Adherence to NICE guidance We restricted NICE compliance analysis to recommendations for children; 23 subjects were identified as definitely aged less than 18 years at drug initiation. All subjects were aged over 12 years (recommendation 1.8.4), and no participants were prescribed orlistat for over 12 months (1.9.11). The following criteria were partially met: First, four (17%) were prescribed orlistat following specialist advice (1.8.5). Recommending specialists were paediatricians (n=3) and an adult physician, with one known to be part of a specialist multidisciplinary team (1.8.7). All prescriptions recommended by specialists were continued in primary care (1.8.8).

y met: First, four (17%) were prescribed orlistat following specialist advice (1.8.5). Recommending specialists were paediatricians (n=3) and an adult physician, with one known to be part of a specialist multidisciplinary team (1.8.7). All prescriptions recommended by specialists were continued in primary care (1.8.8). Second, comorbidities were reported in 57% of the sample (13/23) despite NICE requiring comorbidities to be present (1.8.6). These were emotional distress (7/23), hypothyroidism (3/23), type 2 diabetes (1/23), medulloblastoma (1/23), polycystic ovarian syndrome (1/23) or worsening of another chronic disease secondary to obesity (1/23). No patients had sleep apnoea. Low levels of comorbidity screening in primary care were reported, suggesting that higher number of comorbidities may have existed (6 of 23 were screened for psychosocial distress, 5 for hypertension, 2 each for type 2 diabetes and dyslipidaemia, and none for sleep apnoea). No patients were prescribed a multivitamin (1.9.2). We did not assess screening for micronutrient intake or risk of vitamin deficiencies. Improving prescribing in primary care GP confidence in prescribing AOD to CYP and adults is summarised in figure 4 and shows a skewed inverse relationship. Confidence was higher for prescribing to adults (median=8, IQR 8–9) than children (3, 1–5) (p<0.001). Figure 4 Histogram summarising general practitioner confidence in prescribing antiobesity drug to children and young people (left) and adults (right).

Improving prescribing in primary care GP confidence in prescribing AOD to CYP and adults is summarised in figure 4 and shows a skewed inverse relationship. Confidence was higher for prescribing to adults (median=8, IQR 8–9) than children (3, 1–5) (p<0.001). Figure 4 Histogram summarising general practitioner confidence in prescribing antiobesity drug to children and young people (left) and adults (right). GPs used NICE guidance (orlistat n=20, metformin n=7), British National Formulary (orlistat 17, metformin 8), local prescribing guidelines (orlistat 8, metformin 11) and specialist guidance (orlistat 1, metformin 7) to support prescribing. GPs perceived that 27% (n=12) of patients prescribed metformin and 13% (7) prescribed orlistat benefited from the drug, with half (50% metformin, 53% orlistat) reporting not knowing if there had been any benefits for the patient. Thirty-five GPs provided brief free-text reflections of their experiences prescribing orlistat (n=20) and metformin(n=14). Three main themes arose. First, the use of metformin was mostly ascribed to polycystic ovarian syndrome rather than as a weight loss drug. One GP stated that (s)he “wouldn’t normally prescribe this just for weight loss.” Second, there was controversy about whether AODs should be prescribed in primary care in this age group, with one saying (s)he “usually not prescribe for children” and another saying (s)he avoided orlistat “where possible.” Metformin was used either “on advice of specialist only,” or had specialist follow-up after initiation.

there was controversy about whether AODs should be prescribed in primary care in this age group, with one saying (s)he “usually not prescribe for children” and another saying (s)he avoided orlistat “where possible.” Metformin was used either “on advice of specialist only,” or had specialist follow-up after initiation. Third, GPs noted concern about the efficacy of these drugs. “Inadequate counseling,” lack of drug availability and patient compliance (“clearly patient was not able to comply”) were hypothesised reasons for ineffectiveness. Sixty-two GPs wanted improved support, primarily split into two main themes. First, they requested improved age-related guidance for prescribing AOD that is ‘realistic’, with ‘clear [and] concise’ advice including ‘flow diagrams’ and ‘stepwise advice’. This would include instructions on assessment prior to initiation, indication indications, contraindications, monitoring, safety advice, duration, targets and indications for stopping treatment. Second, they wanted improved guidance for managing patients with obesity, namely advice about lifestyle management and details of available interventions. GPs requested details of “non-drug treatments,” including “community support for adolescents” and “special clinics for monitoring and support of patients.” One GP highlighted that “non-drug treatments need to be key alongside drug treatment.”

ely advice about lifestyle management and details of available interventions. GPs requested details of “non-drug treatments,” including “community support for adolescents” and “special clinics for monitoring and support of patients.” One GP highlighted that “non-drug treatments need to be key alongside drug treatment.” Discussion This is a detailed study of primary care prescribing of AODs in CYP at the individual patient level. Small numbers of prescriptions were issued in this age group, with most practices surveyed prescribing just a single AOD to a child or young person. However, clear patterns were detected that can help guide prescribing of current and future generations of AODs. Recipients of an AOD were largely female, with two-thirds (65%) of the sample aged 16 or over. Two major prescribing patterns were seen: orlistat was largely initiated independently to those over 16 years, and metformin was largely recommended by specialists to girls with either polycystic ovarian syndrome or disturbances in glucose homeostasis. Comparison with NICE guidelines for orlistat showed low compliance with national prescribing recommendations, namely low prevalence of comorbidities and drug initiation without specialist advice. Given that most orlistat prescriptions were for those above 16 years, it could be hypothesised that those aged 16 years and over were treated as adults, with drugs prescribed in line with adult guidelines that do not necessitate presence of comorbidities.

of comorbidities and drug initiation without specialist advice. Given that most orlistat prescriptions were for those above 16 years, it could be hypothesised that those aged 16 years and over were treated as adults, with drugs prescribed in line with adult guidelines that do not necessitate presence of comorbidities. Our findings augment the very limited existing data relating to AOD rates of initiation and cessation14 and experiences of CYP prescribed an AOD.7 A paired study by our research team investigating patient experiences of AODs found high levels of side effects, low levels of professional support managing these side effects, and ultimately families deciding to stop the AOD due to the disadvantages of the side effects outweighing the perceived benefits of the drugs.7 This contrasts with findings from this study where no patients actively discussed side effect profiles with their GPs, and GPs being aware of side effects in only two patients. This study does not explain the disconnect between the experiences of patients and clinicians, and further work should examine ways to support families so they are able to effectively manage the side effects of these drugs.

d side effect profiles with their GPs, and GPs being aware of side effects in only two patients. This study does not explain the disconnect between the experiences of patients and clinicians, and further work should examine ways to support families so they are able to effectively manage the side effects of these drugs. GPs reported low confidence in prescribing AOD to CYP, despite high levels of confidence when prescribing to adults. This fits with findings from our paired study, which showed that families notice this unease in primary care, and which can result in heightened familial concerns about AOD usage.7 GPs reported a desire for improved guidance on drug initiation and monitoring, and on lifestyle interventions, implying low overall confidence in managing childhood obesity. These findings suggest that current national guidelines are inadequate for the needs of primary care, and further work is needed to understand how GPs can better support those with obesity.

on drug initiation and monitoring, and on lifestyle interventions, implying low overall confidence in managing childhood obesity. These findings suggest that current national guidelines are inadequate for the needs of primary care, and further work is needed to understand how GPs can better support those with obesity. Strengths and limitations We used data from a nationally representative data set to identify patients prescribed an AOD with high rate of completion of questionnaires by GPs. Data collection relied on retrospective notes-based recall by GPs, increasing the likelihood of missing data. Individual item completion rates were variable, with some having only a few questions answered. We assumed that unanswered questions implied lack of evidence to support the questions. We were unable to ascertain the exact age of subjects, resulting in risk of misclassification bias, and we are likely to have underestimated the number of subjects who were prescribed orlistat under 18 years of age. We limited the scope of project, and as such we are unable to comment on variation in drug dosing. We were unable to evaluate non-participant bias due to lack of information about individual GP practices.

and we are likely to have underestimated the number of subjects who were prescribed orlistat under 18 years of age. We limited the scope of project, and as such we are unable to comment on variation in drug dosing. We were unable to evaluate non-participant bias due to lack of information about individual GP practices. Conclusions Use of AOD including metformin in primary care is rare, particularly in men and those below 16 years. High rates of discontinuation were seen, primarily in those prescribed orlistat. Rates of compliance with NICE guidance for orlistat were low and GPs report low confidence in the use of AOD in this age group. Improved training and support for GPs is needed to guide AOD use in primary care, both for current and future generations of drugs. The authors would like to thank the team at THIN AIS for collecting the data. Contributors: BW devised the questionnaire with support from all authors and coordinated data collection. YH collated the data set from GP responses. BW analysed and interpreted the data set. BW and RV wrote the manuscript. All authors reviewed and contributed towards the final manuscript. All authors were involved in writing the paper and had final approval of the submitted and published versions. Funding: This work was supported by funding from the National Institute for Health Research (NIHR) in England under its Programme Grants for Applied Research (RP-PG-0608–10035). Competing interests: None declared. Ethics approval: This study was reviewed and approved by the NRES Committee London – Surrey Borders REC reference number11/LO/1020.

Funding: This work was supported by funding from the National Institute for Health Research (NIHR) in England under its Programme Grants for Applied Research (RP-PG-0608–10035). Competing interests: None declared. Ethics approval: This study was reviewed and approved by the NRES Committee London – Surrey Borders REC reference number11/LO/1020. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Original data are available from the authors.

What is already known on this topic? Frequent complaints of pain are common in children and the causes, but the social patterning is unclear. Clinical presentations are heterogeneous and are variously categorised, including as unexplained chronic pain. Frequent complaints of pain in children may have serious lifelong consequences for a child’s health, education and well-being. What this study hopes to add? About a third of children in the UK frequently complain of pain, with significantly higher prevalence in children living in less advantaged families. Poor maternal mental health and early indicators of poor childhood emotional and behavioural development are associated with frequent pain complaints. These differences in maternal and childhood mental health partially explain the social patterning.

What this study hopes to add? About a third of children in the UK frequently complain of pain, with significantly higher prevalence in children living in less advantaged families. Poor maternal mental health and early indicators of poor childhood emotional and behavioural development are associated with frequent pain complaints. These differences in maternal and childhood mental health partially explain the social patterning. Introduction Frequent complaints of pain (FCP) in children are reported in 25%–33% of children in industrialised countries.1 2 Terms including functional pain, frequent pain, unexplained chronic pain (UCP) and somatic symptoms are commonly used. Case definitions vary widely. Pain complaints often relate to regular, unexplained headaches, stomachaches or musculoskeletal complaints that are serious enough to prompt a general practitioner consultation for a third of these children.3 Children with UCP are time consuming within general practice, difficult to investigate and hard to treat,3 with a proportion of patients being referred to tertiary paediatric pain clinics. Some children with UCP suffer serious consequences, including disability and prolonged school absence.4 UCP may have important social consequences. Some children with UCP report problems in eating, sleeping, pursuit of hobbies, attending school and meeting their friends.5

g referred to tertiary paediatric pain clinics. Some children with UCP suffer serious consequences, including disability and prolonged school absence.4 UCP may have important social consequences. Some children with UCP report problems in eating, sleeping, pursuit of hobbies, attending school and meeting their friends.5 The aetiology of FCP is varied. Pain may occur after childhood cancer, traumatic or nerve injury, but most cases remain unexplained.4 International consensus diagnostic criteria allow categorisation for functional gastrointestinal (GI) disorders, but do not imply causation.6 Associations with parental pain experience and diet have been found for many of these disorders. A link between UCP and mental health conditions has been demonstrated, prompting theories invoking psychosomatic causation of these symptoms.7 It is expected that socioeconomic factors have a substantial impact on a child’s risk of FCP. This study therefore aimed to assess the social patterning of, and risk factors for, FCP in UK children. A secondary aim was to assess whether any risk factors identified attenuate inequalities in pain complaints.

e symptoms.7 It is expected that socioeconomic factors have a substantial impact on a child’s risk of FCP. This study therefore aimed to assess the social patterning of, and risk factors for, FCP in UK children. A secondary aim was to assess whether any risk factors identified attenuate inequalities in pain complaints. Method Design, setting and data source This study uses data from the UK Millennium Cohort Study (MCS), a nationally representative stratified multistage random sample of approximately 18 000 children born in the UK between September 2000 and January 2002, coordinated by the Centre for Longitudinal Studies, University of London.8–12 Oversampling ensured adequate power to assess outcomes in children living in deprived areas and across ethnic groups. This analysis makes use of data on singleton children across all waves of the study, from 9 months to 11 years. Outcome measure: parent-reported FCP Childhood pain was assessed by a Strengths and Difficulties Questionnaire (SDQ) item in survey 5 of the MCS at age 11 years, representing an age at which persistent pain that causes school absence would be disruptive to both social development and education. Main respondents (99.2% mothers) were asked whether their child ‘often complains of headaches, stomach-aches or sickness.’ Possible answers were: not true, somewhat true, certainly true or don’t know. This was recoded to compare those children for whom the description of frequent pain was certainly or somewhat true at age 11 (yes=1) with those for whom the statement was not true (no=0).

n complains of headaches, stomach-aches or sickness.’ Possible answers were: not true, somewhat true, certainly true or don’t know. This was recoded to compare those children for whom the description of frequent pain was certainly or somewhat true at age 11 (yes=1) with those for whom the statement was not true (no=0). Exposure measure: socioeconomic circumstances The primary exposure of interest was maternal education level (higher degree/fist degree/diploma/A, AS, S level/O level GCSE (General Certificate of Secondary Education) A–C/GCSE D–G/none with international and other qualifications excluded), used as a fixed measure of socioeconomic conditions (SECs) at birth. Maternal education level captures the advantages of SECs that accrue to a child,13 and was selected to represent SEC based on similar studies evaluating recurrent abdominal pain14 and other childhood diseases.15

nal and other qualifications excluded), used as a fixed measure of socioeconomic conditions (SECs) at birth. Maternal education level captures the advantages of SECs that accrue to a child,13 and was selected to represent SEC based on similar studies evaluating recurrent abdominal pain14 and other childhood diseases.15 Covariates Other exposure variables potentially associated with chronic pain were identified through literature review and matched to available data within the MCS. Maternal variables were measured within a year of childbirth and child variables at age 5 or 7 years (details in table 1). Demographic factors included child sex16 and ethnicity16 (coded as white/Asian/black/mixed or other). Potentially mediating covariates included: preterm birth <37 weeks (yes/no), infant behavioural problems17 (regular eating, sleeping yes/no), problematic crying (yes/no), maternal general (excellent/good/fair/poor), maternal complaint of bodily pain18–20 (none/very mild/mild/moderate/severe or very severe), doctor or nurse diagnosed maternal GI disease14 (yes/no), maternal mental health17 measured using the Kessler scale (normal/distressed), child mental health and behaviour19 21 22 using the SDQ (normal/borderline/abnormal), body mass index (BMI) group23 (underweight or normal range/overweight/obese) and daily fruit consumption23 (none or one/two/three or more).

ase14 (yes/no), maternal mental health17 measured using the Kessler scale (normal/distressed), child mental health and behaviour19 21 22 using the SDQ (normal/borderline/abnormal), body mass index (BMI) group23 (underweight or normal range/overweight/obese) and daily fruit consumption23 (none or one/two/three or more). Analysis strategy Following the Baron and Kenny steps to mediation,24 we explored the unadjusted association between maternal qualifications (primary exposure) and FCP at 11 years (outcome measure). The risk of FCP was then estimated based on exposures and mediators of interest, presenting unadjusted risk ratios (RRs) using Poisson regression. Following a life course approach, covariates that were significant at p<0.1 in univariable analysis were added to a multivariable regression model. We used a sequential approach to construct the adjusted models,25 first adding demographic variables, then perinatal factors and postnatal exposures in the order in which they were experienced by the child. Mediation was taken to be a reduction in, or elimination of, statistically significant RRs in a final complete case sample. Attenuation was calculated as a percentage reduction from baseline to fully adjusted RRs. Data were analysed using Stata V.1326 with survey commands used to weight data for sampling design and attrition.

on was taken to be a reduction in, or elimination of, statistically significant RRs in a final complete case sample. Attenuation was calculated as a percentage reduction from baseline to fully adjusted RRs. Data were analysed using Stata V.1326 with survey commands used to weight data for sampling design and attrition. Sensitivity analysis Sensitivity analyses were conducted with a stricter specification of the outcome variable, or teacher-reported pain complaints and considering household income and parental occupation instead of education as alternative measures of SECs. Further analyses were undertaken to examine the possibility that the result was influenced by collinearity between a mediator (SDQ) and the outcome (an item from the SDQ) and to explore the impact of missing data.27 Ethics Ethical approval for the MCS was received from a National Health Service research ethics committee prior to each survey.28 Our secondary data analysis did not require additional ethics approval. Results Up to 8463 children had complete data on all covariates of interest, 82.1% of children present at all sweeps up to age 11 years. Overall, 32.3% of children frequently complained of pain at age 11, and this differed significantly by childhood SECs (table 1). Up to 43.2% of children in the lowest maternal education group (no qualifications) reported experiencing frequent pain compared with 21.0% in the highest maternal education group (higher degree). A clear socioeconomic gradient is present, with higher reporting of frequent pain in more disadvantaged children (figure 1).

Up to 43.2% of children in the lowest maternal education group (no qualifications) reported experiencing frequent pain compared with 21.0% in the highest maternal education group (higher degree). A clear socioeconomic gradient is present, with higher reporting of frequent pain in more disadvantaged children (figure 1). Lower maternal qualifications are associated with deteriorations in maternal and child health covariates, with the exception of measures of maternal GI disease and infant behaviour. Fruit consumption, a possible protective factor, increases with increased maternal qualifications. Figure 1 Bar chart showing the percentage of children who frequently complained of pain at age 11 years by maternal education level. GCSE, General Certificate of Secondary Education. Table 1 Characteristics of the total study population, by level of maternal qualification at the birth of child

Lower maternal qualifications are associated with deteriorations in maternal and child health covariates, with the exception of measures of maternal GI disease and infant behaviour. Fruit consumption, a possible protective factor, increases with increased maternal qualifications. Figure 1 Bar chart showing the percentage of children who frequently complained of pain at age 11 years by maternal education level. GCSE, General Certificate of Secondary Education. Table 1 Characteristics of the total study population, by level of maternal qualification at the birth of child Maternal education level (Weighted number of observations) Higher degree (n=341), 3.6% Degree (n=1515), 15.1% Diploma (n=990), 9.6% A levels (n=1020), 10.2% GCSE-C (n=3744), 38.0% GCSE-G (n=1099), 11.5% None (n=1193), 12.0% Total (n=9902) χ2 p Value FCP outcome measure (11 years) Yes 21.0% 24.3% 29.2% 28.3% 33.2% 38.4% 43.2% 32.3% <0.001 Sex Female 52.0% 48.8% 47.7% 50.7% 49.6% 50.3% 48.5% 49.4% 0.876 Ethnicity White 84.8% 91.8% 91.6% 92.5% 93.3% 93.6% 89.8% 92.1% 0.004 Mixed or other 6.3% 3.7% 3.5% 2.9% 3.0% 2.3% 4.7% 3.4% Asian 5.4% 2.5% 2.1% 3.6% 2.2% 2.4% 4.0% 2.7% Black 3.5% 2.0% 2.7% 1.0% 1.4% 1.6% 1.5% 1.7% Gestation Premature (<37 weeks) 5.7% 6.0% 6.1% 4.8% 8.0% 6.8% 10.5% 7.3% 0.003 Baby eats and sleeps regularly (9 months) Disagree 1.3% 1.8% 2.2% 2.4% 2.2% 2.3% 3.5% 2.3% 0.257 Whether baby crying is a problem (9 months) Yes 5.5% 6.0% 4.6% 6.6% 5.3% 7.4% 10.3% 6.3% <0.001 Maternal GI conditions (9 months) Yes 11.9% 9.9% 11.3% 10.9% 9.7% 7.8% 8.2% 9.7% 0.176 Maternal mental health (Kessler scale at 3 years) Distressed 14.2% 9.5% 11.8% 15.6% 20.6% 24.4% 26.9% 18.5% <0.001 Maternal general health (5 years) Excellent 42.6% 43.6% 36.9% 36.0% 28.8% 25.3% 21.1% 31.8% <0.001 Good 45.7% 48.4% 50.9% 51.2% 53.9% 54.0% 55.0% 52.3% Fair 10.0% 7.3% 10.4% 11.4% 14.7% 16.9% 20.8% 13.6% Poor 1.7% 0.7% 1.8% 1.4% 2.5% 3.8% 3.2% 2.3% Self-reported maternal bodily pain (5 years) None 47.7% 50.1% 49.2% 48.5% 47.4% 50.0% 45.3% 48.1% <0.001 Very mild 29.1% 24.9% 18.3% 22.0% 20.1% 16.8% 15.6% 20.3% Mild 14.4% 13.0% 13.8% 12.1% 11.6% 13.2% 12.0% 12.4% Moderate 6.4% 10.0% 13.8% 12.6% 13.8% 12.1% 16.3% 12.9% Severe or very severe 2.4% 2.0% 4.9% 4.8% 7.1% 7.9% 10.7% 6.2% Child SDQ (5 years) Normal 95.8% 97.2% 94.1% 91.4% 90.9% 83.3% 78.0% 90.0% <0.001 Borderline 3.6% 1.6% 3.3% 5.3% 5.0% 8.2% 10.3% 5.3% Abnormal 0.6% 1.2% 2.6% 3.3% 4.0% 8.5% 11.8% 4.7% Child's BMI group (7 years) Normal or underweight 86.3% 85.8% 81.1% 82.2% 79.2% 76.8% 76.7% 80.4% <0.001 Overweight 11.7% 11.5% 14.7% 13.6% 14.8% 16.3% 17.8% 14.6% Obese 2.0% 2.7% 4.2% 4.2% 6.0% 6.9% 5.4% 5.0% Child's portions of fruit consumed per day (5 years) None or one 9.3% 7.9% 12.8% 13.1% 20.2% 26.8% 35.2% 19.0% <0.001 Two 22.5

p (7 years) Normal or underweight 86.3% 85.8% 81.1% 82.2% 79.2% 76.8% 76.7% 80.4% <0.001 Overweight 11.7% 11.5% 14.7% 13.6% 14.8% 16.3% 17.8% 14.6% Obese 2.0% 2.7% 4.2% 4.2% 6.0% 6.9% 5.4% 5.0% Child's portions of fruit consumed per day (5 years) None or one 9.3% 7.9% 12.8% 13.1% 20.2% 26.8% 35.2% 19.0% <0.001 Two 22.5 % 21.6% 21.9% 23.8% 27.6% 32.1% 26.0% 25.9% Three or more 68.2% 70.5% 65.2% 63.1% 52.2% 41.1% 38.9% 55.1% All figures are percentages adjusted for sampling design. Missing data reasons are detailed in section 4 of the online supplementary material. BMI, body mass index; FCP, frequent complaints of pain; GI, gastrointestinal; SDQ, Strengths and Difficulties Questionnaire. Univariable analysis In univariable Poisson regression (table 2), lower maternal qualifications were associated with increased risk of FCP. Children in families in which the mother had no qualifications were more likely than children whose mother had a higher degree to complain of FCP (RR 2.06, 95% CI 1.64 to 2.59). Lower maternal education qualifications, female sex, black ethnicity, problematic crying, maternal GI disease, poor maternal mental health, poor maternal general health, maternal bodily pain, childhood emotional difficulties and high BMI were all associated with an increased risk of childhood pain at age 11 years. Eating two or more portions of fruit daily was associated with lower risk of childhood pain. Dose–response relationships were shown for maternal education level, maternal general health, maternal bodily pain, child SDQ, high BMI and fruit consumption.

associated with an increased risk of childhood pain at age 11 years. Eating two or more portions of fruit daily was associated with lower risk of childhood pain. Dose–response relationships were shown for maternal education level, maternal general health, maternal bodily pain, child SDQ, high BMI and fruit consumption. Table 2 Univariable and multivariable regression showing RRs with 95% CIs

associated with an increased risk of childhood pain at age 11 years. Eating two or more portions of fruit daily was associated with lower risk of childhood pain. Dose–response relationships were shown for maternal education level, maternal general health, maternal bodily pain, child SDQ, high BMI and fruit consumption. Table 2 Univariable and multivariable regression showing RRs with 95% CIs Univariable RR (95% CI) Multivariable RR (95% CI) % (N) with outcome Maternal education (at birth, ref Higher degree) 21% (71) Degree 1.16 (0.93 to 1.44) 1.21 (0.98 to 1.50) 24% (368) Diploma 1.40 ** (1.10 to 1.78) 1.38 ** (1.09 to 1.74) 29% (290) A levels 1.35 * (1.07 to 1.70) 1.30 * (1.04 to 1.64) 28% (289) GCSE A–C 1.59 *** (1.29 to 1.95) 1.46 *** (1.19 to 1.79) 33% (1244) GCSE D–G 1.83 *** (1.46 to 2.30) 1.57 *** (1.25 to 1.98) 38% (421) None 2.06 *** (1.64 to 2.59) 1.70 *** (1.36 to 2.13) 43% (516) Sex (ref Male) 28% (1403) Female 1.31 *** (1.21 to 1.42) 1.34 *** (1.24 to 1.44) 37% (1796) Ethnicity (ref White) 32% (2915) Mixed or other 1.06 (0.86 to 1.31) 1 (0.81 to 1.23) 34% (115) Asian 1.1 (0.92 to 1.31) 0.97 (0.83 to 1.13) 35% (96) Black 1.39 ** (1.11 to 1.73) 1.27 * (1.03 to 1.57) 44% (75) Gestation (ref Term) 32% (2960) Premature 1.03 (0.88 to 1.21) 33% (239) Eats, sleeps regularly (9 months, ref Agree) 32% (3113) Disagree or strongly disagree 1.2 (0.97 to 1.48) 39% (87) Problematic crying (9 months, ref No) 32% (2942) Yes 1.30 *** (1.14 to 1.47) 1.11 (0.98 to 1.25) 41% (257) Maternal GI disease (9 months, ref No) 31% (2793) Yes 1.36 *** (1.23 to 1.49) 1.23 *** (1.12 to 1.36) 42% (407) Maternal Kessler scale (3 years, ref Normal) 29% (2346) Distressed 1.61 *** (1.50 to 1.73) 1.33 *** (1.23 to 1.44) 47% (853) Maternal health (5 years, ref Excellent) 26% (819) Good 1.25 *** (1.14 to 1.38) 1.12 * (1.02 to 1.23) 33% (1689) Fair 1.66 *** (1.49 to 1.85) 1.22 *** (1.09 to 1.36) 43% (582) Poor 1.90 *** (1.55 to 2.34) 1.21 (0.98 to 1.49) 49% (110) Maternal amount of bodily pain (5 years, ref None) 27% (1306) Very mild 1.17 ** (1.06 to 1.30) 1.16 ** (1.05 to 1.28) 32% (646) Mild 1.40 *** (1.27 to 1.55) 1.30 *** (1.17 to 1.44) 38% (474) Moderate 1.49 *** (1.34 to 1.67) 1.28 *** (1.14 to 1.43) 41% (524) Severe or very severe 1.48 *** (1.29 to 1.69) 1.21 ** (1.06 to 1.39) 41% (250) Child SDQ (5 years, ref Normal) 30% (2698) Borderline 1.59 *** (1.40 to 1.80) 1.35 *** (1.19 to 1.54) 48% (252) Abnormal 1.77 *** (1.54 to 2.04) 1.41 *** (1.21 to 1.65) 54% (249) BMI (7 years, ref Normal or underweight) 31% (2483) Overweight 1.14 ** (1.04 to 1.25) 1.06 (0.97 to 1.16) 36%

.21 ** (1.06 to 1.39) 41% (250) Child SDQ (5 years, ref Normal) 30% (2698) Borderline 1.59 *** (1.40 to 1.80) 1.35 *** (1.19 to 1.54) 48% (252) Abnormal 1.77 *** (1.54 to 2.04) 1.41 *** (1.21 to 1.65) 54% (249) BMI (7 years, ref Normal or underweight) 31% (2483) Overweight 1.14 ** (1.04 to 1.25) 1.06 (0.97 to 1.16) 36% (514) Obese 1.31 *** (1.15 to 1.50) 1.17 * (1.02 to 1.33) 41% (203) Daily fruit portions (5 years, ref None or one) 39% (742) Two 0.86 ** (0.77 to 0.96) 0.93 (0.83 to 1.03) 34% (866) Three or more 0.74 *** (0.68 to 0.81) 0.85 *** (0.78 to 0.93) 29% (1592) Unweighted n=8463. ***p<0.001; **p<0.01; *p<0.05. BMI, body mass index; GCSE, General Certificate of Secondary Education; GI, gastrointestinal; RR, risk ratio; SDQ, Strengths and Difficulties Questionnaire. Mediation analysis The RR for FCP comparing low versus high maternal education is attenuated by 17% after adjustment for covariates comparing the final fully adjusted model relative to a baseline model that included maternal education level, sex and ethnicity. The addition to the model of variables measuring problematic crying, maternal mental health, maternal general health, childhood SDQ and fruit consumption partially attenuated the impact of maternal education on FCP (figure 2).

model relative to a baseline model that included maternal education level, sex and ethnicity. The addition to the model of variables measuring problematic crying, maternal mental health, maternal general health, childhood SDQ and fruit consumption partially attenuated the impact of maternal education on FCP (figure 2). Figure 2 Attenuation of the effect size for low versus high maternal education as risk factors are added sequentially to the model as covariates. RR shown is for no qualifications relative to a higher degree. Error bars show 95% CIs around the survey estimates. BMI, body mass index; GI, gastrointestinal; RR, risk ratio; SDQ, Strengths and Difficulties Questionnaire. The greatest attenuations of the baseline RR occurred on addition of the child’s SDQ score and maternal Kessler scale, with smaller changes for maternal self-reported general health and fruit consumption. Attenuations are small, and lie within the 95% CIs surrounding the baseline RR. Fully adjusted model In the fully adjusted regression model, children of families in which the main respondent had no qualifications have more than twice the risk of FCP symptoms experienced by children in families where the main respondent had a higher degree (RR 1.70, 95% CI 1.36 to 2.13, p<0.001). Female sex and all measures of poor maternal health remained significant predictors of increased FCP (table 2 and figure 3). Black children have increased risk of FCP (RR 1.27, 95% CI 1.03 to 1.57).

Fully adjusted model In the fully adjusted regression model, children of families in which the main respondent had no qualifications have more than twice the risk of FCP symptoms experienced by children in families where the main respondent had a higher degree (RR 1.70, 95% CI 1.36 to 2.13, p<0.001). Female sex and all measures of poor maternal health remained significant predictors of increased FCP (table 2 and figure 3). Black children have increased risk of FCP (RR 1.27, 95% CI 1.03 to 1.57). Figure 3 Unadjusted and fully adjusted RRs of frequent complaints of pain prevalence with 95% CIs. BMI, body mass index; GCSE, General Certificate of Secondary Education; GI, gastrointestinal; RR, risk ratio; SDQ, Strengths and Difficulties Questionnaire. The strongest associations between child health measures and FCP are for child mental health, represented by the SDQ, in which abnormal SDQ scores, compared with normal scores, were associated with increased risk FCP (RR 1.77, 95% CI 1.54 to 2.04). Sensitivity analyses Repeating the analysis using occupational class or household income rather than maternal education as the measure of SECs produced similar results (online supplementary material 1). 10.1136/bmjpo-2017-000093.supp1Supplementary file 1

The strongest associations between child health measures and FCP are for child mental health, represented by the SDQ, in which abnormal SDQ scores, compared with normal scores, were associated with increased risk FCP (RR 1.77, 95% CI 1.54 to 2.04). Sensitivity analyses Repeating the analysis using occupational class or household income rather than maternal education as the measure of SECs produced similar results (online supplementary material 1). 10.1136/bmjpo-2017-000093.supp1Supplementary file 1 Repeating the analysis with a stricter definition of FCP resulted in similar results but wider CIs. The analysis using teacher (instead of parent) reported complaints of FCP also produced similar results, but with considerably larger RRs. Repeating the analysis without variables that contributed large amounts of missing data also produced similar results. Repeating the analysis using a recalculated SDQ measure without the frequent pain item produced near identical results to using the full SDQ score. Results of all sensitivity analyses are presented in online supplementary material.

variables that contributed large amounts of missing data also produced similar results. Repeating the analysis using a recalculated SDQ measure without the frequent pain item produced near identical results to using the full SDQ score. Results of all sensitivity analyses are presented in online supplementary material. Discussion Children from households in which the mother has no qualifications are twice as likely to frequently complain of pain at age 11 years than children from households in which the main respondent has a higher degree, and there is a clear socioeconomic gradient. This finding may be partially explained by the social patterning of risk factors for FCP, including poor parent and child mental health and indicators of poor diet. However, after adjusting for known risk factors, those exposed to the lowest SECs experience twice as many cases of FCP as those with the most advantageous SECs, suggesting a large effect of SECs on risk of FCP that is mediated by unknown risk factors. Both maternal and child mental health significantly predicted FCP. The effect of SECs on FCP may be partially mediated through the influence of SECs on maternal mental health and subsequently child mental health. Adults with fewer qualifications and lower incomes suffer a greater burden of depression.29 Existing cross-sectional studies also associate parental17 18 30 and childhood19 21 22 31 mental health problems with the development of FCP.

the influence of SECs on maternal mental health and subsequently child mental health. Adults with fewer qualifications and lower incomes suffer a greater burden of depression.29 Existing cross-sectional studies also associate parental17 18 30 and childhood19 21 22 31 mental health problems with the development of FCP. The effect of SECs on FCP may also be mediated through the influence of SECs on maternal GI disease and indicators of childhood diet. This partly replicates Malaty et al’s23 findings in their school-based US sample using a much larger and more representative UK study. This study provides further evidence that girls complain more frequently of pain than boys, replicating the findings of previous studies that considered recurrent pain or functional somatic symptoms in general,17 32 but not studies focused on abdominal23 or back pain.33 Strengths and limitations This study was carried out using secondary data from a representative and contemporary UK cohort. To our knowledge, this is the largest study to date of FCP in children. The study had ample statistical power to consider inequalities in pain as a health outcome, which was strengthened through the MCS’s sampling strategy and use of weights to compensate for attrition.

from a representative and contemporary UK cohort. To our knowledge, this is the largest study to date of FCP in children. The study had ample statistical power to consider inequalities in pain as a health outcome, which was strengthened through the MCS’s sampling strategy and use of weights to compensate for attrition. Through its longitudinal analysis, this study provides strong evidence that socioeconomic risk factors measured 10 years before the measurement of an outcome are associated with that outcome. It has demonstrated a large effect of SECs on FCP that was robust in an extensive sensitivity analysis. The MCS’s rich data set has enabled control for possible confounders, and comprehensive data on social conditions have allowed the effect of social inequalities on pain to be confirmed across all the key individual measures of SECs.

ted a large effect of SECs on FCP that was robust in an extensive sensitivity analysis. The MCS’s rich data set has enabled control for possible confounders, and comprehensive data on social conditions have allowed the effect of social inequalities on pain to be confirmed across all the key individual measures of SECs. The most important limitation of this study is the uncertain clinical validity of the outcome measure, which was unvalidated and did not assess whether the pain about which children complained was categorically chronic, or exclude children who may have an organic cause of pain. While chronic pain is common in children1 and most chronic pain remains unexplained,34–37 cases of severe UCP are much less common.2 Our reported FCP prevalence based on parental report (32.2%) was towards the upper end of that reported in UCP literature, and an existing study38 has suggested that parental report is inaccurate, based on disagreement between parent-reported and self-reported pain in children. These concerns are addressed through a sensitivity analysis in which teacher-reported pain was used as the outcome measure, which supported the main analysis despite finding less than half the prevalence (15.2%) of parent-reported pain.

, based on disagreement between parent-reported and self-reported pain in children. These concerns are addressed through a sensitivity analysis in which teacher-reported pain was used as the outcome measure, which supported the main analysis despite finding less than half the prevalence (15.2%) of parent-reported pain. Given the breadth of the MCS question on FCP, children who frequently complain of pain are reporting a heterogeneous phenomenon. While the term sickness, within the MCS question, may be interpreted as a general state of malaise, particularly in conjunction with complaints, some parents may instead interpret this as nausea. For some children, this pain seems likely to be psychogenic or psychosomatic in origin,30 39 yet for others it might be caused by an underlying predisposition to GI disease, predisease states14 or poor diet. The finding that both psychological and dietary factors independently predict FCP and may mediate the effects of SECs on FCP supports this conclusion. The mediating variables tested within this study were constrained by the questions asked within the MCS. It is possible that a future study using more sensitive measures of childhood and maternal diet and emotional functioning would show that these factors mediate a greater proportion of the association between SECs and FCP. It is possible that our results may be biased by including a mediator (SDQ) in a model for an outcome measure that is an SDQ item, but our sensitivity analysis using a recalculated SDQ score (without the pain item) produced near identical results.

rs mediate a greater proportion of the association between SECs and FCP. It is possible that our results may be biased by including a mediator (SDQ) in a model for an outcome measure that is an SDQ item, but our sensitivity analysis using a recalculated SDQ score (without the pain item) produced near identical results. A further limitation of this study is attrition within the MCS, common to any cohort study, but partially mitigated by weighting. A sensitivity analysis that removed the three variables that had the greatest amount of missing data showed similar results to the main analysis, suggesting that the exclusion of children due to missing data did not bias this study. Policy and practice This study suggests that a range of existing interventions might be able to reduce the overall prevalence of FCP, but might also be able to reduce the inequalities in FCP across the social gradient. Interventions that reduce differential exposure to risk factors that promote the development of paediatric FCP symptoms could include improved parental healthcare to address GI illness, improved parental mental healthcare and particularly perinatal mental health services. Population level interventions (typically regulatory changes) that address dietary change could also have an effect, particularly where these seek to improve choice of diet for the least advantaged or the development of services to promote early years’ psychological well-being and emotional resilience.

health services. Population level interventions (typically regulatory changes) that address dietary change could also have an effect, particularly where these seek to improve choice of diet for the least advantaged or the development of services to promote early years’ psychological well-being and emotional resilience. While targeting some of the mediating factors linking SECs to FCP is likely to reduce inequalities in FCP, given the large direct effect of SECs in this study, action is needed to address the underlying inequalities in the social determinants of child health. Contributors: BF designed the study, analysed the data and drafted the manuscript. GC, SLW and DCTR provided input on the study design. SLW and DCTR contributed to the analytical strategy. GC, SLW, BRB and DCTR commented on drafts of the manuscript. All authors read and approved the final manuscript. Funding: SLW is supported by a Wellcome Trust research fellowship. This piece of work was also supported by The Farr Institute for Health Informatics Research (MRC grant: MR/M0501633/1). The Millennium Cohort Study is funded by grants to former and current directors of the study from the Economic and Social Research Council (Professor HeatherJoshi, Professor Lucinda Platt and Professor Emla Fitzsimons) and a consortium of government funders. All researchers were independent of the funders, and the funders played no part in the study design, analysis or interpretation of the data, writing of the report or the decision to submit for publication. Competing interests: None declared. Patient consent: Obtained.

Funding: SLW is supported by a Wellcome Trust research fellowship. This piece of work was also supported by The Farr Institute for Health Informatics Research (MRC grant: MR/M0501633/1). The Millennium Cohort Study is funded by grants to former and current directors of the study from the Economic and Social Research Council (Professor HeatherJoshi, Professor Lucinda Platt and Professor Emla Fitzsimons) and a consortium of government funders. All researchers were independent of the funders, and the funders played no part in the study design, analysis or interpretation of the data, writing of the report or the decision to submit for publication. Competing interests: None declared. Patient consent: Obtained. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: The data used in this study are available from the UK data service.

What is already known on this topic? Research nurses are increasingly involved in all stages of the development of complex clinical trials conducted across paediatric specialties. It is important that research nurses are appropriately trained for the various regulatory, methodological, ethical and administrative aspects of clinical trial design. Few studies have been published focusing on the level of training and roles played by research nurses working in paediatric specialties across Europe. What this study hopes to add? Approximately two-thirds of research nurses felt that they would benefit from additional training in specific areas, with a clear relationship observed between length of time in post and level of training satisfaction. An increased level of nurse prescribing may be beneficial in paediatric specialties, with only 3% of participants prescribed investigational medicinal products in a clinical trial setting. Sharing of the information generated through national research nurse networks should be used to encourage the development of paediatric research nurse training programmes. Introduction There is a clear need to accelerate the development of drugs across a wide range of childhood disease specialties.1 In order for this to be achieved, high-quality ethical research on the safety and efficacy of medicines in children is needed. In this respect, the research nurse plays an increasingly pivotal role in the successful conduct of paediatric clinical trials.2

of drugs across a wide range of childhood disease specialties.1 In order for this to be achieved, high-quality ethical research on the safety and efficacy of medicines in children is needed. In this respect, the research nurse plays an increasingly pivotal role in the successful conduct of paediatric clinical trials.2 Clinical trials in children have an inherent default level of complexity relating to regulatory, methodological, ethical and administrative issues, with additional burdens commonly introduced for multinational studies.3 The role of the research nurse has developed in line with increasing numbers of often complicated research studies commonly built into trial design, to generate as much information as possible relating to the new treatment. Outside of the collection of blood samples for routine clinical analysis, samples are frequently requested for a range of substudies including clinical pharmacology and biomarker studies, pharmacogenomics, biobanking and cytogenetics.4 Clinical samples will commonly be requested at multiple time points and require the collection and recording of data and completion of clinical trial visit-associated case report forms (CRFs), often via trial-specific electronic data capture (EDC) systems. These studies are carried out alongside more routine research nurse responsibilities, including day-to-day study management, patient screening, provision of patient information sheets, appropriate collection of consent/assent and collaboration with other members of the multidisciplinary team required to ensure a positive experience for study patients.5

gside more routine research nurse responsibilities, including day-to-day study management, patient screening, provision of patient information sheets, appropriate collection of consent/assent and collaboration with other members of the multidisciplinary team required to ensure a positive experience for study patients.5 The European Network of Paediatric Research at the European Medicines Agency (Enpr-EMA) consists of a consortium of research networks, investigators and centres with expertise in performing clinical trials in children and adolescents.6 A working group was established by Enpr-EMA to investigate potential needs and gaps in research nurse training across specialties and countries. A questionnaire-based survey was carefully planned to compare experiences and seek the views of research nurses working in paediatric settings across Europe, in addition to generating information on the extent of involvement in clinical trial activities and the specific roles carried out by research nurses in different countries and specialties. Such an approach may allow the identification of potentially desirable training models for recommendation by Enpr-EMA at a European level.

ion to generating information on the extent of involvement in clinical trial activities and the specific roles carried out by research nurses in different countries and specialties. Such an approach may allow the identification of potentially desirable training models for recommendation by Enpr-EMA at a European level. Materials and methods Questionnaire design and preparation A structured, cross-sectional questionnaire survey, informed by previous research on nurse training and questionnaire development,7–9 was designed to cover four main areas of interest. The first section covered basic demographic information, including disease specialties, age of children being cared for and length of time as a research nurse. The second section focused on the level, frequency and method of training received. The third area focused on clinical trials experience with regards to involvement in different types of trials and various aspects of developing and running studies. Finally, paediatric research nurse roles and activities were investigated, from patient consent, prescribing and administration of investigational medicinal products (IMPs), through to sample collection, processing and transport.

vement in different types of trials and various aspects of developing and running studies. Finally, paediatric research nurse roles and activities were investigated, from patient consent, prescribing and administration of investigational medicinal products (IMPs), through to sample collection, processing and transport. The design incorporated fixed choice and open-ended questions and was piloted on an initial cohort of 20 paediatric research nurses to check for usability. Following minor alterations and additions to the questionnaire at this point, a final version was approved for dissemination. box 1 summarises the questions incorporated in the final questionnaire and the full questionnaire is provided as an online supplementary figure 1.Box 1 Summary of questions in final questionnaire (response required and follow-up questions) Across what specialties do you work in paediatrics? (select all that apply; if ‘other’, please provide details) What age of children do you work with? (select all that apply) How long have you worked as a research nurse? How many phase I, II, III or IV clinical trials have you participated in? Do you feel that you have received appropriate training for the role(s) you carry out in your position? Do you feel that you would benefit from additional training in some aspects of your job? (if ‘yes’, please comment) How would you best describe describe the training that you received for your role? (select one) If you have received Good Clinical Practice (GCP) training, has this been generic GCP training or paediatric-specific GCP training?

Do you feel that you would benefit from additional training in some aspects of your job? (if ‘yes’, please comment) How would you best describe describe the training that you received for your role? (select one) If you have received Good Clinical Practice (GCP) training, has this been generic GCP training or paediatric-specific GCP training? If you have received additional training, please provide further information with regards to the type of training. How frequently do you receive training in your current job? How frequently do you receive GCP certified training within your role? How would you best describe the training that you received when you first started in post? (select all that apply) How would you best describe the training updates that you receive? (select all that apply) Which of the following activities do you have experience of actively participating in? Within your role, do you participate in CTIMP (Clinical Trial of an Investigational Medicinal Product) studies? If ‘yes’, which of the following roles do you perform? If ‘yes’ to any of the above, have you received specific training for this? Are you involved in the following types of paediatric clinical trials? (tick all that apply) 10.1136/bmjpo-2017-000170.supp1Supplementary file 1 sf1

Within your role, do you participate in CTIMP (Clinical Trial of an Investigational Medicinal Product) studies? If ‘yes’, which of the following roles do you perform? If ‘yes’ to any of the above, have you received specific training for this? Are you involved in the following types of paediatric clinical trials? (tick all that apply) 10.1136/bmjpo-2017-000170.supp1Supplementary file 1 sf1 Participants and data collection The final questionnaire was made available via an electronic link through the internet-based survey tool provider Google Forms. National and disease specialty networks of paediatric research nurses were identified through Enpr-EMA networks and the identification of appropriate European groups through internet searches. A link to the Google Forms questionnaire was sent to lead network contacts alongside a letter from Enpr-EMA, explaining the purpose and aims of the survey, for dissemination to research nurses within individual networks or groups. The questionnaire was translated into French and Spanish as requested by specific networks and was made available for a 9 month period between April and December, 2016. Data analysis and statistical analysis Data entry and initial analysis were carried out using Microsoft Excel 2013 and Qlik Sense V.3.1 (Qlik International AB). Statistical analysis was carried out using the χ2 test as appropriate using SPSS statistical software.

Participants and data collection The final questionnaire was made available via an electronic link through the internet-based survey tool provider Google Forms. National and disease specialty networks of paediatric research nurses were identified through Enpr-EMA networks and the identification of appropriate European groups through internet searches. A link to the Google Forms questionnaire was sent to lead network contacts alongside a letter from Enpr-EMA, explaining the purpose and aims of the survey, for dissemination to research nurses within individual networks or groups. The questionnaire was translated into French and Spanish as requested by specific networks and was made available for a 9 month period between April and December, 2016. Data analysis and statistical analysis Data entry and initial analysis were carried out using Microsoft Excel 2013 and Qlik Sense V.3.1 (Qlik International AB). Statistical analysis was carried out using the χ2 test as appropriate using SPSS statistical software. Results Demographics of respondents The questionnaire was completed by 341 research nurses from 20 European countries. The respondents worked across 45 different disease specialties, with 34% working with children in a single specialty, 21% in two specialties, 14% across three or four specialties and the remaining respondents (31%) working across at least five specialties. The most common specialty areas were respiratory diseases, oncology and diabetes. Respondents spanned a wide range of experience levels, with approximately 1/3 of participants (31%) having worked as research nurses for <3 years and 38% having >6 years of experience. Table 1 provides a summary of the demographics of the research nurses who participated in the study.

y diseases, oncology and diabetes. Respondents spanned a wide range of experience levels, with approximately 1/3 of participants (31%) having worked as research nurses for <3 years and 38% having >6 years of experience. Table 1 provides a summary of the demographics of the research nurses who participated in the study. Table 1 Demographics of research nurse respondents Characteristic No. of respondents % Evaluable responses 341 Age of patients cared for (years) <1 284 83 1–3 278 82 4–10 295 87 11–16 287 84 >16 234 69 Length of time working as a research nurse (years) <3 106 31 3–5 90 26 6–10 81 24 >10 48 14 Unknown 16 4.7 Country UK 189 55 France 22 6.5 Germany 16 4.7 Norway 16 4.7 Spain 15 4.4 Ireland 14 4.1 Switzerland 13 3.8 Netherlands 11 3.2 Denmark 10 2.9 Austria 8 2.3 Finland 7 2.0 Belgium 6 1.7 Sweden 4 1.2 Portugal 3 0.9 Other* 7 2.3 Number of specialties 1 116 34 2 73 21 3–4 46 13 5–9 84 25 10–14 18 5.3 15–20 4 1.2 *Country grouping ‘other’ includes Italy (2), Greece (2), Hungary (1), Luxembourg (1) and Turkey (1). Training received and satisfaction with level of training Data collected on frequency of training suggested that research nurses received regular training, with 53% receiving formal training at 6 monthly, annual or 2 yearly intervals, and 38% being trained ‘as needed’. Less than 10% of respondents received training at intervals of >2 years.

satisfaction with level of training Data collected on frequency of training suggested that research nurses received regular training, with 53% receiving formal training at 6 monthly, annual or 2 yearly intervals, and 38% being trained ‘as needed’. Less than 10% of respondents received training at intervals of >2 years. A total of 147 research nurses (43%) were fully satisfied with the level of training received, with 32 (9%) respondents not satisfied and a further 145 (43%) satisfied that they were appropriately trained for the majority of tasks that they carried out; the remaining 17 (5%) participants failed to respond to this question. Further analysis suggested that a significantly higher percentage of research nurses within the first 3 years of taking up their post were dissatisfied with their level of training (16%) as compared with those with 3–6 years (8%) and >6 years (6%) of experience (p<0.001). Overall, there was a clear trend towards a relationship between length of time in post and level of training satisfaction (see figure 1). Looking at the results obtained geographically, for those countries with at least 10 respondents, higher percentages of nurses dissatisfied with the level of training received were observed in Norway (25%) and Denmark (20%). In contrast, 100% of respondents from Spain and the Netherlands were either fully satisfied or satisfied with the level of training received for the majority of tasks carried out. In response to the direct question as to whether they would benefit from extra training in some aspects of their job, 67% of all respondents indicated that this would be beneficial.

in and the Netherlands were either fully satisfied or satisfied with the level of training received for the majority of tasks carried out. In response to the direct question as to whether they would benefit from extra training in some aspects of their job, 67% of all respondents indicated that this would be beneficial. Figure 1 Level of satisfaction with training received by level of experience in terms of length of time in research nurse post. With regards to the type of training received, in terms of whether this was carried out online or in person, organised internally or run by an external organisation, institution or self-funded, there were no clear trends observed in terms of the level of training satisfaction (p>0.05). Similarly, there was no relationship between the frequency of training received and the level of satisfaction reported (p>0.05). Interestingly, 68% of research nurses received training updates online and this value was also high (54%) in terms of the initial method of training received when first in post. While there were no particularly striking findings observed when these data were analysed by country, there appeared to be a higher percentage of respondents who received self-funded training in mainland Europe, with reported values of 15%–20% in Germany, Norway, Switzerland, Denmark and the Netherlands, as compared with <1% in the UK and <4% in Ireland.

iking findings observed when these data were analysed by country, there appeared to be a higher percentage of respondents who received self-funded training in mainland Europe, with reported values of 15%–20% in Germany, Norway, Switzerland, Denmark and the Netherlands, as compared with <1% in the UK and <4% in Ireland. Specific questions were included in the questionnaire relating to the provision of Good Clinical Practice (GCP) training, with overall 96.5% (329/341) of research nurses having received GCP training, with 40% of these respondents (132/329) having received paediatric-specific GCP training and 60% (197/329) generic GCP training. These figures were comparable across specialties and countries.

ovision of Good Clinical Practice (GCP) training, with overall 96.5% (329/341) of research nurses having received GCP training, with 40% of these respondents (132/329) having received paediatric-specific GCP training and 60% (197/329) generic GCP training. These figures were comparable across specialties and countries. Additional training needs Information relating to areas of additional training that research nurses would benefit from could be categorised into the following general areas of training: regulatory issues (21% of respondents who indicated that additional training would be beneficial), clinical trial coordination/GCP-related (20%), nursing procedures (19%), information technology (IT) based (10%), data analysis (8%) and communications skills training (6%). In the area of clinical trials in particular, a wide range of training needs were identified including areas such as research governance, trial set-up, design and coordination, costing and finance. Similarly, requests for training in a wide range of nursing procedures and laboratory skills highlight the increasing requirement for research nurses to possess a wide range of skill sets consummate with complex clinical trial designs.

as research governance, trial set-up, design and coordination, costing and finance. Similarly, requests for training in a wide range of nursing procedures and laboratory skills highlight the increasing requirement for research nurses to possess a wide range of skill sets consummate with complex clinical trial designs. Clinical trial experience The questionnaire explored the role of the research nurse in various aspects of developing and running paediatric clinical trials. Approximately 1/3 of respondents (36%) were involved in the development of consent forms, 27% had experience of trial submission and 32% in the development of trial CRFs. Approximately half of participants (46%) had experience of developing patient information sheets. These data were consistent across countries and specialties. Research nurse roles and activities In terms of the activities in which participants are commonly involved, over 70% of research nurses actively participated in the collection and processing of blood samples (252 respondents) and the shipment/transport of clinical samples (242 respondents), with approximately 60% of respondents involved in the training and education of patients in terms of the administration of new medicines or procedures (207 respondents), and the administration of IMPs (205 respondents). Approximately 1/3 of research nurses who responded to the survey were involved in taking consent and/or assent from patients (126 respondents) and only 3% (11 respondents) prescribed IMPs in a clinical trial setting (see figure 2).

cines or procedures (207 respondents), and the administration of IMPs (205 respondents). Approximately 1/3 of research nurses who responded to the survey were involved in taking consent and/or assent from patients (126 respondents) and only 3% (11 respondents) prescribed IMPs in a clinical trial setting (see figure 2). Figure 2 Summary of activities carried out by research nurses working in a paediatric setting. IMPs, investigational medicinal products. Further analysis of these data by country identified wide ranges in percentages of research nurses actively involved in the defined roles and activities described above. While percentages were high across all countries for routine roles such as the collection, processing and transport of clinical samples, marked differences were seen in the percentage of respondents taking patient consent for clinical trial participation. For those countries with at least 10 respondents, less than 10% of research nurses in Germany and Spain took consent, whereas approximately half of UK research nurses (49%) and 80% of respondents from the Netherlands carried out this role. Percentages of research nurses who administered IMPs ranged from 15% in Switzerland to 87% in Spain, with small numbers of respondents (≤10%) responsible for prescribing IMPs in all countries except for Norway (13%) and Switzerland (23%). In terms of analysis of research nurse roles and activities by specialty area, there were no clear trends or differences observed.

MPs ranged from 15% in Switzerland to 87% in Spain, with small numbers of respondents (≤10%) responsible for prescribing IMPs in all countries except for Norway (13%) and Switzerland (23%). In terms of analysis of research nurse roles and activities by specialty area, there were no clear trends or differences observed. Discussion The current study was carried out to gather information relating to the training and roles of research nurses who conduct paediatric clinical trials across Europe. A questionnaire-based survey was proposed and executed by Enpr-EMA, with responses gathered from 341 respondents, encompassing 45 different disease specialties and 20 European countries. As there was no explicit sampling frame, in terms of a defined list of numbers of research nurses in the networks and centres who received the questionnaire to complete, it was not possible to address the extent of non-response bias and this represents an accepted limitation of the study. Similarly, while national and disease specialty networks of paediatric research nurses were identified through Enpr-EMA networks and the identification of appropriate European groups through internet searches, many lead network contacts were not research nurses and wider circulation of the study information and link to the survey may not always have been prioritised.

etworks of paediatric research nurses were identified through Enpr-EMA networks and the identification of appropriate European groups through internet searches, many lead network contacts were not research nurses and wider circulation of the study information and link to the survey may not always have been prioritised. Results generated from the questionnaire were generally encouraging, with 86% of respondents either fully satisfied with the level of training received, or satisfied that they were appropriately trained for the majority of tasks that they carried out. Indeed, a healthy percentage of respondents had either completed formal postbachelor education or training programmes or obtained more focused training in key areas, commonly provided by sponsors or pharmaceutical companies. However, 67% of respondents also felt that they would benefit from additional training, with a wide range of areas highlighted where training would be most beneficial. The most common areas for additional training reflected the increased complexities of modern day clinical trials and increasing requirement for research nurses to possess a wide range of skill sets.2 3 These included clinical trial set-up and management, IT skills, pharmacovigilance, CRF data entry and laboratory skills training. A number of respondents highlighted the challenges of keeping up to date with frequent changes to clinical trial practices relating to ever increasing GCP regulations.

ange of skill sets.2 3 These included clinical trial set-up and management, IT skills, pharmacovigilance, CRF data entry and laboratory skills training. A number of respondents highlighted the challenges of keeping up to date with frequent changes to clinical trial practices relating to ever increasing GCP regulations. In terms of how research nurses obtain training, 54% of respondents received their initial training online, when they first took up post, and 68% received training updates online. These figures highlight marked increases in online nurse training observed over the past 20 years, largely due to its convenience and flexibility. Online training can help to avoid problems relating to intensive workloads and working shifts, which could provide barriers to research nurses attending scheduled training sessions. In this respect, many studies have reported positive outcomes from online nurse training, with learning outcomes comparable or even improved as compared with face-to-face training events.8

g to intensive workloads and working shifts, which could provide barriers to research nurses attending scheduled training sessions. In this respect, many studies have reported positive outcomes from online nurse training, with learning outcomes comparable or even improved as compared with face-to-face training events.8 One interesting point raised by several participants, related to the expectation that they would gain relevant experience through ‘on the job’ training. This theory appears to be supported by the findings of the current study, with a clear relationship observed between length of time in post and level of training satisfaction reported. Several respondents highlighted the fact that they would have benefited from more training when they first started in post, at which time they were unaware of the training opportunities available. This may be particularly relevant to countries where higher levels of dissatisfaction were reported in terms of training received. For example, in some Nordic countries, more accessible research nurse training programmes have only relatively recently been developed, following studies highlighting a need for more relevant training to be made available.10 11 This represents an area that could be improved, through advertising and promotion of research nurse training events. The availability of induction packs for new research nurses, containing relevant information and useful links to networks where training is available, is commonplace in some countries and should be encouraged more widely. It is accepted that the role of the research nurse may differ significantly between countries and that the current survey, while relatively expansive in terms of the number of respondents and countries involved, did not include respondents from many other European countries.

ntries and should be encouraged more widely. It is accepted that the role of the research nurse may differ significantly between countries and that the current survey, while relatively expansive in terms of the number of respondents and countries involved, did not include respondents from many other European countries. Data generated from the current study highlight the integral role that research nurses play in the running of clinical trials in paediatric specialties,2 3 with significant numbers of participants being involved in various wide-ranging tasks. In terms of the practicalities of working in a clinical trial setting, high percentages of respondents were involved in taking patient consent, collection and processing of samples, patient training and the administration of IMPs. Despite relatively small numbers of respondents from individual countries, apparent differences were observed in the level of involvement in activities including administration of IMPs and taking consent. Such differences may reflect guidelines and philosophies within countries, with the role of taking consent being actively encouraged in the UK, but not being seen as an appropriate research nurse role in some European countries. Indeed, it is entirely feasible that while research nurses may feel comfortable in explaining clinical trials to patients and families, they may be less amenable to being responsible for the signing of consent forms. In this respect, it is important to understand that the appropriateness of research nurses taking on this responsibility may be related to the complexity of the particular clinical trial and the IMP involved. For research nurses working in environments where roles such as the taking of consent are encouraged, competency tools are commonly available to promote patient safety and may be included in research nurse induction packs referred to above.12

ated to the complexity of the particular clinical trial and the IMP involved. For research nurses working in environments where roles such as the taking of consent are encouraged, competency tools are commonly available to promote patient safety and may be included in research nurse induction packs referred to above.12 A key role absent from the activities of the vast majority of respondents related to the prescribing of IMPs, with only 3% of research nurses performing this role. For research nurses to prescribe medicines, including unlicensed and clinical trial drugs, they are required to have the appropriate qualifications.13 In the UK, registered nurses with a minimum of 3 years of clinical experience must undertake a recognised ‘Nursing and Midwife Council’ accredited prescribing course through a UK university. Other countries have their own research nurse qualifications, which may contrast in terms of the level of training involved and the prescribing responsibilities of the research nurse.14 15

linical experience must undertake a recognised ‘Nursing and Midwife Council’ accredited prescribing course through a UK university. Other countries have their own research nurse qualifications, which may contrast in terms of the level of training involved and the prescribing responsibilities of the research nurse.14 15 While nurse prescribing in the UK and Europe has developed significantly over the past decade, with an estimated 19 000 nurse prescribers registered in the UK in 2014,16 this remains a remarkably underdeveloped area. Indeed, experience in the UK indicates that differences exist between individual centres and health authorities, in terms of whether or not research nurses are able to prescribe IMPs, even if the appropriate level of training is in place. Therefore, the reported low numbers of research nurses responsible for prescribing IMPs may reflect local legislation, as opposed to a lack of desire for nurses to be involved in this area. Benefits of research nurse prescribing in the UK have been widely reported,17 18 and may represent an area where improvements could be made to the efficiency of running paediatric clinical trials. An increase in level of nurse prescribing would seem to represent a sensible way to optimise the skills and expertise of all health professionals working in increasingly stretched healthcare systems.19 It is unclear whether or not the observed increased research nurse prescribing rates in countries such as Norway and Switzerland reflects a real difference, possibly related to more accessible training and accreditation in these countries, as the numbers of respondents from the majority of countries was small. Similarly, differences highlighted between countries in the percentages of research nurses actively involved in taking patient consent and the administration of IMPs should be interpreted with caution.

training and accreditation in these countries, as the numbers of respondents from the majority of countries was small. Similarly, differences highlighted between countries in the percentages of research nurses actively involved in taking patient consent and the administration of IMPs should be interpreted with caution. Conclusion In summary, the study provides a useful overview of the current training status of research nurses working in paediatric medicine, highlighting potential training needs and summarising the roles and activities of research nurses across Europe. As higher percentages of respondents received self-funded training and/or were not satisfied with the level of training in some European countries, this would suggest that different training opportunities and historical working cultures may currently exist. While the level of training and general satisfaction levels expressed by research nurses is encouraging, approximately two-thirds of respondents felt that they would benefit from additional training, with commonly requested areas for further training highlighted. Increased availability and provision of research nurse training in these areas may facilitate an increased efficiency in the running of clinical trials in a paediatric setting. Sharing of the information generated in the current study through Enpr-EMA and national research nurse networks will be strongly encouraged, with a view to supporting, facilitating and developing new research training programme for paediatric research nurses across Europe. In this respect, Enpr-EMA should look to enhance the design of the European paediatric research nurse core curriculum, together with relevant European Nursing Associations, which could be then adopted across EU countries. Increased collaboration and discussion between key stakeholders will help to harmonise approaches to training and standardise the way that paediatric clinical trials are conducted across Europe, promoting improved ethical and clinical standards and the generation of robust results from clinical trials.

ss EU countries. Increased collaboration and discussion between key stakeholders will help to harmonise approaches to training and standardise the way that paediatric clinical trials are conducted across Europe, promoting improved ethical and clinical standards and the generation of robust results from clinical trials. We thank the research nurses from centres across Europe for participating in the survey and the paediatric networks for their enthusiastic dissemination of the survey. We would also like to thank Professor Elaine McColl for her insightful comments on the manuscript and for input into the statistical analysis of study data generated. Contributors: GJV, SM, ML, SMF, PL, MC and AC conceived the study and were responsible for the design of the study and appropriate dissemination of the final questionnaire across European centres. GV, CB, CL, AO, MR and PB were responsible for the collation and analysis of data obtained from the study. GV was responsible for writing of the manuscript. All authors provided input into manuscript review and approved the final version for submission. Funding: This research was supported by Cancer Research UK and the Experimental Cancer Medicine Centre Network. Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Traditionally, appendicectomy has been the gold standard treatment for acute appendicitis in children, but there has been increased interest in non-operative treatment (with antibiotics). Core outcome sets are developed and adopted to avoid inconsistencies in outcome selection, measurement and reporting that may otherwise exist. There is currently no core outcome set for the measurement of effectiveness of treatment interventions in children with acute uncomplicated appendicitis. What this study hopes to add? This project will involve defining a core outcome set for the measurement of effectiveness of treatment interventions in children with acute uncomplicated appendicitis. Considering outcomes of importance to patients, parents of patients and health professionals is crucial for paediatric appendicitis research to be meaningful and relevant.

What this study hopes to add? This project will involve defining a core outcome set for the measurement of effectiveness of treatment interventions in children with acute uncomplicated appendicitis. Considering outcomes of importance to patients, parents of patients and health professionals is crucial for paediatric appendicitis research to be meaningful and relevant. Background A lack of knowledge and understanding regarding which outcomes are important to patients and clinicians may result in important outcomes being omitted from clinical trials. Differences in outcome selection and reporting between studies and how outcomes are defined and measured also make it difficult, sometimes impossible, to synthesise results of studies (eg, meta-analysis) and apply them in a meaningful way. To address these problems, core outcome sets (COS) have been proposed as a means of standardising outcome selection, measurement and reporting in healthcare research and in clinical trials in particular.1 2 The development of a COS and its adoption by researchers is intended to help avoid inconsistencies in outcome selection, measurement and reporting that may otherwise exist. If trials do not adopt an established COS, they risk selecting suboptimal outcomes and are unlikely to contribute usable information.3

rticular.1 2 The development of a COS and its adoption by researchers is intended to help avoid inconsistencies in outcome selection, measurement and reporting that may otherwise exist. If trials do not adopt an established COS, they risk selecting suboptimal outcomes and are unlikely to contribute usable information.3 For children with acute uncomplicated appendicitis, appendicectomy has traditionally been the gold standard treatment, but there has been growing interest in alternatives to appendicectomy. In particular, non-operative treatment of appendicitis, with antibiotics, has been proposed as a potential treatment. A small number of randomised controlled trials (RCTs) in adults4–6 and, more recently, RCTs and prospective preference studies in children7 8 suggest that antibiotic treatment may be a valid alternative to appendicectomy.9 However, there is currently insufficient data to justify its widespread use. Prior to performing further efficacy studies of the treatment of appendicitis in children, it is imperative to identify the most relevant outcomes for inclusion in the design of comparative studies. This is of particular importance when evaluating a novel treatment approach since the outcomes of importance may differ from those commonly reported with traditional therapies. A review of the relevant literature and electronic resources failed to identify a COS for children with appendicitis.10 Furthermore, a wide range of outcomes were reported, and a range of different primary outcomes were used across studies.

importance may differ from those commonly reported with traditional therapies. A review of the relevant literature and electronic resources failed to identify a COS for children with appendicitis.10 Furthermore, a wide range of outcomes were reported, and a range of different primary outcomes were used across studies. In order to advance our understanding of which outcomes are important and to fulfil an unmet need in our future research programme, the aim of this study is to develop a COS for the measurement of effectiveness of treatment interventions in children (<18 years) with acute uncomplicated appendicitis. Methods The COS development is a component of a wider project, CONservative TReatment of Appendicitis in Children - a randomised controlled Trial (Feasibility) (CONTRACT) study (http://www.nets.nihr.ac.uk/projects/hta/1419290). The COS development study was registered with the COMET Initiative in May 2017 (http://www.comet-initiative.org/studies/details/987).

t of a wider project, CONservative TReatment of Appendicitis in Children - a randomised controlled Trial (Feasibility) (CONTRACT) study (http://www.nets.nihr.ac.uk/projects/hta/1419290). The COS development study was registered with the COMET Initiative in May 2017 (http://www.comet-initiative.org/studies/details/987). Scope of the COS The COS is intended to be used to evaluate the overall success of operative and non-operative treatment among children who are assigned a clinical and/or radiological diagnosis of acute uncomplicated appendicitis. The COS will include outcome measures identified as important within 12 months of treatment initiation and longer term outcomes if applicable. The COS focuses specifically on treatment of acute uncomplicated appendicitis; the treatment of perforated appendicitis (with or without abscess) and appendix mass is outside the scope of this COS. The key objectives of the study are:to determine which outcomes have previously been reported in studies comparing treatments for acute uncomplicated appendicitis in children to prioritise treatment outcomes of children with acute uncomplicated appendicitis from key stakeholder groups’ perspectives (including paediatric surgeons, general surgeons, patients (12–18 years old) and parents of children who have had acute uncomplicated appendicitis) to compare and contrast paediatric acute uncomplicated appendicitis treatment outcomes prioritised by key stakeholder groups (as detailed above)

to prioritise treatment outcomes of children with acute uncomplicated appendicitis from key stakeholder groups’ perspectives (including paediatric surgeons, general surgeons, patients (12–18 years old) and parents of children who have had acute uncomplicated appendicitis) to compare and contrast paediatric acute uncomplicated appendicitis treatment outcomes prioritised by key stakeholder groups (as detailed above) to achieve consensus between key stakeholder groups on a COS to evaluate overall success of treatment for acute uncomplicated appendicitis in children. Design COS development will entail four key stages:a systematic review to identify previously reported acute uncomplicated appendicitis treatment outcomes assembly of stakeholder panels a three-phase online Delphi process consensus meeting. The CONTRACT study, as part of the feasibility study, will involve in-depth qualitative research. Interviews will be conducted with patients, parents and caregivers to explore experiences of treatment for appendicitis. We will explore families’ perceptions of meaningful and important outcomes, which will help to optimise COS participants’ project understanding and engagement throughout the COS development.

esearch. Interviews will be conducted with patients, parents and caregivers to explore experiences of treatment for appendicitis. We will explore families’ perceptions of meaningful and important outcomes, which will help to optimise COS participants’ project understanding and engagement throughout the COS development. Systematic review The COMET Initiative recommend the use of systematic reviews in informing the first phase of the Delphi process.11 Two recent systematic reviews will be used to inform the initial list of potential outcomes to be considered for the COS. The first review identified outcomes used in studies of paediatric appendicitis,10 and the second review aimed to determine safety and efficacy of non-operative treatment for acute appendicitis.9 Identified outcomes were combined with closely similar outcomes from the operative treatment systematic review10 to develop an initial list of outcomes (online supplementary appendix 1). Online supplementary appendix 1 describes the eligibility criteria for inclusion of papers. All relevant articles will therefore be included that reported any non-operative treatment regimen for acute uncomplicated appendicitis in children with or without a comparative group of children undergoing surgical treatment. All outcomes identified through these reviews and an updated literature search will inform the initial list of outcomes and will be considered by the stakeholder groups as part of the Delphi process. 10.1136/bmjpo-2017-000151.supp1Supplementary file 1

Systematic review The COMET Initiative recommend the use of systematic reviews in informing the first phase of the Delphi process.11 Two recent systematic reviews will be used to inform the initial list of potential outcomes to be considered for the COS. The first review identified outcomes used in studies of paediatric appendicitis,10 and the second review aimed to determine safety and efficacy of non-operative treatment for acute appendicitis.9 Identified outcomes were combined with closely similar outcomes from the operative treatment systematic review10 to develop an initial list of outcomes (online supplementary appendix 1). Online supplementary appendix 1 describes the eligibility criteria for inclusion of papers. All relevant articles will therefore be included that reported any non-operative treatment regimen for acute uncomplicated appendicitis in children with or without a comparative group of children undergoing surgical treatment. All outcomes identified through these reviews and an updated literature search will inform the initial list of outcomes and will be considered by the stakeholder groups as part of the Delphi process. 10.1136/bmjpo-2017-000151.supp1Supplementary file 1 Finalising and appropriate wording of initial outcomes To inform and support the CONTRACT study, a Study Specific Advisory Group (SSAG) has been assembled, comprising 15–20 young people and parents. Young people recruited are children who have had appendicitis or children from the existing Clinical Research Network (Children) Young Persons Advisory Groups. Parents are parents of children who have had appendicitis. An SSAG meeting will be held to present the initial list of treatment outcomes, to inform the addition and wording of outcomes and to ensure outcomes are appropriately presented. Three versions of the stakeholder-facing materials will be developed for all rounds of the Delphi process, for each stakeholder panel (clinicians, young people and parents), using appropriate language identified and agreed by the SSAG. The CONTRACT qualitative study will also inform the initial outcomes.

ely presented. Three versions of the stakeholder-facing materials will be developed for all rounds of the Delphi process, for each stakeholder panel (clinicians, young people and parents), using appropriate language identified and agreed by the SSAG. The CONTRACT qualitative study will also inform the initial outcomes. Stakeholder panel assembly: identification and recruitment For the COS to be meaningful and relevant to those involved in the treatment of acute appendicitis, the COS needs to reflect the views of patients who have been treated for acute appendicitis, their parents and relevant clinicians. As these groups may have different priorities that could obstruct reaching consensus on a COS, the stakeholders will be separated into three panels, which we intend to be equally weighted: (1) patients; (2) parents; and (3) paediatric surgeons and general surgeons. Initially, potential members of each stakeholder panel known to the research team will be invited to participate, and we will develop strategies to identify further experts (see table 1). Table 1 Core outcome set stakeholder groups and methods of approaching potential participants Stakeholder group Selection criteria Method of approach Patients and parents Patients aged 12–18 years who have been treated for acute uncomplicated appendicitis in the preceding 12–24 months. Parents of children (aged 5–18 years) who have been treated for acute uncomplicated appendicitis in the preceding 12–24 months. Families may or may not have participated in CONTRACT.

Stakeholder group Selection criteria Method of approach Patients and parents Patients aged 12–18 years who have been treated for acute uncomplicated appendicitis in the preceding 12–24 months. Parents of children (aged 5–18 years) who have been treated for acute uncomplicated appendicitis in the preceding 12–24 months. Families may or may not have participated in CONTRACT. Patient and parent panels will specifically include children and parents treated initially by non-operative management as well as those treated operatively. Invited to participate via clinical teams from the three sites that are participating in the CONTRACT study. Identified to participate via further participant identification sites. Paediatric surgeons All practising consultant paediatric surgeons in the UK who treat children with acute uncomplicated appendicitis will be considered potential participants. Invited to participate via the mailing list of the British Association of Paediatric Surgeons and through personal contacts of the investigators. General surgeons Adult general surgeons in the UK who regularly treat children with acute uncomplicated appendicitis will be considered potential participants. This will include those identified as having an interest in the treatment of children. Invited to participate via existing personal contacts and through regional paediatric surgical networks within the UK. CONTRACT, CONservative TReatment of Appendicitis in Children—a randomised controlled Trial (Feasibility).

General surgeons Adult general surgeons in the UK who regularly treat children with acute uncomplicated appendicitis will be considered potential participants. This will include those identified as having an interest in the treatment of children. Invited to participate via existing personal contacts and through regional paediatric surgical networks within the UK. CONTRACT, CONservative TReatment of Appendicitis in Children—a randomised controlled Trial (Feasibility). Initial contact with potential participants will explain the study and why they have been identified as a potential participants. It will contain a plain language summary of the study aims and procedures, emphasising the importance of commitment to the panel. The wording of this initial contact will be tailored to meet the panel category. It will also contain a link to an online form to enable participants to express their interest in participation in the study and to provide further information on their experience of the treatment of acute appendicitis. We will ask participants to commit to completing three rounds of questionnaires anticipated to take approximately 10 min each to complete.

ne form to enable participants to express their interest in participation in the study and to provide further information on their experience of the treatment of acute appendicitis. We will ask participants to commit to completing three rounds of questionnaires anticipated to take approximately 10 min each to complete. The process of invitation and enrolment will continue until the optimal number of stakeholders have expressed an interest to participate (with at least 10 in each panel). There is no consensus on the optimal sample size for a Delphi study; recruitment will therefore be based on previous Delphi studies.12 We will aim to achieve 75–100 participants in the first round of the Delphi with at least as many parents/children as clinicians. Efforts will be made to invite a diverse range of participants to each stakeholder group. We will aim to send the first questionnaire to all participants on the same day they each confirm their desire to participate.13 Participants will be sent a link to a customised online database hosted on a secure server, from which they can access and complete phase one questionnaire of the Delphi process. To limit attrition, appropriate procedures will be completed,12 14 including reminder emails.

hey each confirm their desire to participate.13 Participants will be sent a link to a customised online database hosted on a secure server, from which they can access and complete phase one questionnaire of the Delphi process. To limit attrition, appropriate procedures will be completed,12 14 including reminder emails. Participation in this COS development process will be limited to surgeons, patients and parents from the UK. We will not recruit paediatric or general surgeons from outside the UK because the treatment pathway for appendicitis in the UK differs to that in other countries. Furthermore, if we were to recruit surgeons from outside the UK, we would also need to recruit patients and parents from outside the UK to avoid bias; this would become increasingly challenging logistically. Definition of consensus During the process, participants will be asked to score each outcome using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) scale, which is recommended by the COMET Initiative.15 The scale will be presented in the format 1–9, with 1–3 labelled ‘not important’, 4–6 labelled ‘important but not critical’ and 7–9 labelled ‘critical’.16 ‘Consensus in’ will be defined as ≥70% of participants rating the outcome 7–9 and <15% rating it as 1–3. Outcomes will defined as ‘consensus out’ if >70% participants rate it 1–3 and <15% rate it 7–9. Outcomes not meeting these definitions will be classified as ‘no consensus’.

t critical’ and 7–9 labelled ‘critical’.16 ‘Consensus in’ will be defined as ≥70% of participants rating the outcome 7–9 and <15% rating it as 1–3. Outcomes will defined as ‘consensus out’ if >70% participants rate it 1–3 and <15% rate it 7–9. Outcomes not meeting these definitions will be classified as ‘no consensus’. Delphi process: phase one data collection A customised online data system (developed by the National Perinatal Epidemiology Unit (University of Oxford, UK)) will be used to conduct a three-phase Delphi process run in parallel across the stakeholder panels. Participants will be presented with the initial list of outcomes, grouped by domains. As described, the initial list will comprise outcomes identified in our recent systematic review,10 updated review of the literature (online supplementary appendix 1) and any additional outcomes identified during qualitative interviews with key stakeholders. There will also be an option for experts to add further outcomes, but these outcomes will not be scored in phase one. Surgeon participants will be asked the key question ‘How important do you consider the following outcomes to be when considering which treatment to offer children with uncomplicated acute appendicitis?’ A similar question will be posed to patient and parent stakeholder panels, with the wording altered as necessary based on our SSAG input.

will be asked the key question ‘How important do you consider the following outcomes to be when considering which treatment to offer children with uncomplicated acute appendicitis?’ A similar question will be posed to patient and parent stakeholder panels, with the wording altered as necessary based on our SSAG input. Participants will be asked to complete each round of the Delphi exercise within 3 weeks, and two subsequent reminder emails will be sent. Participants who have not completed the questionnaire within 4 weeks of being requested to complete the questionnaire will be deemed not to have completed that phase. Delphi process: phase one data analysis The number of participants who were invited to participate in phase one and the response rate from each stakeholder group will be recorded. Outcomes will be analysed separately for each panel, and descriptive statistics will be calculated. All outcomes will be carried forward to phase two. Additional outcomes provided by participants will be reviewed by two members of the COS team to ensure they represent new outcomes and will be included in phase two if they were proposed by at least two participants. Delphi process: phase two data collection Participants who completed phase one will be invited to participate in phase two. Participants will be individually presented with their own scores, the distribution of scores for each outcome from their stakeholder group in phase one, and will be asked to rescore each outcome. Participants will be asked to score any new outcomes identified in phase one.

will be invited to participate in phase two. Participants will be individually presented with their own scores, the distribution of scores for each outcome from their stakeholder group in phase one, and will be asked to rescore each outcome. Participants will be asked to score any new outcomes identified in phase one. Delphi process: phase two data analysis The data analysis process described for phase one will be repeated. Bias from loss of experts between phases will be assessed. Any outcomes that meet the criteria of ‘consensus out’ will be removed from the outcomes list prior to phase three. All other outcomes from phase two will be carried forward to phase three. Delphi process: phase three data collection Participants who completed phases one and two will be invited to participate in phase three. The data collection process described for phase two will be repeated; however, participants will also be shown scores for their own stakeholder panel and separately for each other panel. This will allow participants to consider other stakeholder groups’ views before rescoring the outcomes.17 Participants will be asked to identify the one outcome which they believe is the most important for informing their treatment choice, and if they cannot identify a single outcome, a combination of essential outcomes. Delphi process: phase three data analysis The data analysis process described for phase two will be repeated. All outcomes from phase three will be carried forward to the consensus meeting.

Delphi process: phase three data collection Participants who completed phases one and two will be invited to participate in phase three. The data collection process described for phase two will be repeated; however, participants will also be shown scores for their own stakeholder panel and separately for each other panel. This will allow participants to consider other stakeholder groups’ views before rescoring the outcomes.17 Participants will be asked to identify the one outcome which they believe is the most important for informing their treatment choice, and if they cannot identify a single outcome, a combination of essential outcomes. Delphi process: phase three data analysis The data analysis process described for phase two will be repeated. All outcomes from phase three will be carried forward to the consensus meeting. Final consensus meeting The aim of the consensus meeting is to ratify outcomes where consensus (‘in’ or ‘out’) has been achieved, to discuss outcomes where consensus could not be achieved and to finalise the COS. All participants who have completed all three rounds of the Delphi exercise will be invited to attend the consensus meeting. We will aim to have a minimum of 40 stakeholders confirm their attendance with equally weighted panels and disciplines. Representatives from each stakeholder panel will be required in order for the consensus meeting to be quorate.

all three rounds of the Delphi exercise will be invited to attend the consensus meeting. We will aim to have a minimum of 40 stakeholders confirm their attendance with equally weighted panels and disciplines. Representatives from each stakeholder panel will be required in order for the consensus meeting to be quorate. During the meeting, stakeholders will be provided with an overview of the results of phase three including presentation of each outcome scored, how it was scored by each stakeholder panel and its consensus status. Following moderated discussion, each outcome will be anonymously rescored using the same scoring system as the Delphi process. For outcomes for which ‘no consensus’ was achieved across all stakeholder panels at the end of the Delphi exercise, and for which consensus was achieved in at least one but not all stakeholder groups, further discussion will take place, following which attendees will be asked to score each outcome anonymously. Following rescoring at the consensus meeting, outcomes reaching ‘consensus in’ will be included in the finalised COS. All others will be excluded.

s was achieved in at least one but not all stakeholder groups, further discussion will take place, following which attendees will be asked to score each outcome anonymously. Following rescoring at the consensus meeting, outcomes reaching ‘consensus in’ will be included in the finalised COS. All others will be excluded. Finalising the COS The Outcome Measures in Rheumatology initiative provides a COS development framework that is a useful across various healthcare domains.18 In developing the current COS, we will draw on this framework, which comprises three core domains (death, life impact and pathophysiological manifestations) and one strongly recommended domain (resource use). The framework recommends inclusion of at least one applicable measurement instrument for each core domain. It also recommends inclusion of ‘adverse events’. Overall, we aim to achieve a manageable COS with a maximum of approximately 10 outcomes. Beyond the current project, further work may be undertaken to establish optimal methods of measuring each core outcome18 and subsequent work may be undertaken to determine whether the UK-based COS is valid internationally. Data management Experts will enter data directly into the customised database when they complete each questionnaire at each phase of the Delphi process. Anonymised data will be stored securely and will be managed as per standard operating procedures. Only selected members of the research team will have access to the data.

Data management Experts will enter data directly into the customised database when they complete each questionnaire at each phase of the Delphi process. Anonymised data will be stored securely and will be managed as per standard operating procedures. Only selected members of the research team will have access to the data. Ethics and dissemination Full ethical approval for CONTRACT was granted by South Central - Hampshire A Research Ethics Committee (REC ref: 16/SC/0596) in November 2016. The finalised COS will be disseminated via publication, events and the COMET Initiative website (http://www.comet-initiative.org/). Supplementary Material Reviewer comments Author's manuscript All authors gratefully acknowledge funding from the National Institute for Health Research Health Technology Assessment Programme—the CONTRACT trial, including the COS Development. SE gratefully acknowledges funding from Great Ormond Street Children’s Charity. Contributors: FCS, SE and NJH developed a first draft of the protocol. All authors advised on the development of the protocol and contributed towards the revision of the protocol for final submission. Funding: The CONTRACT study is supported by the United Kingdom National Institute for Health Research Health Technology Assessment Programme (Grant number: 14/192/90 http://www.nets.nihr.ac.uk/projects/hta/1419290). Competing interests: None declared. Ethics approval: South Central—Hampshire A Research Ethics Committee (REC ref: 16/SC/0596). Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? Paediatric admissions are rising year by year in the UK. Evidence for interventions to better manage paediatric acute care and therefore reduce avoidable admissions is lacking. What this study hopes to add? In a single hospital, an advice and guidance phone line was associated with fewer less than 1 day admissions, but an increase in overall bed-days. Short stay paediatric assessment unit (SSPAU) was associated with a reduction in ward admissions, less than 1 day admissions and overall bed-days. There are indications that advice and guidance and SSPAU, as examples of interventions reducing time taken to senior clinician review, are effective in better managing paediatric acute care. Introduction Background Avoiding excess unplanned admissions is a UK National Health Service priority, with acute paediatric admissions rising year by year since 2003.1 2 While rates vary by area (and indeed in the site included in this study admission rates are flat),3 such increases are unsustainable and remain a research priority.4 5 Admission to hospital is an undesirable outcome for children and their parents for many reasons, including disruption to family life, increased emotional distress and exposure to infections. There are also significant cost implications of a hospital admission. The six most common conditions resulting in the presentation for paediatric acute care are the ‘big 6’ conditions: bronchiolitis/croup, fever, gastroenteritis, head injury, wheezy child/asthma and abdominal pain.6 7

stress and exposure to infections. There are also significant cost implications of a hospital admission. The six most common conditions resulting in the presentation for paediatric acute care are the ‘big 6’ conditions: bronchiolitis/croup, fever, gastroenteritis, head injury, wheezy child/asthma and abdominal pain.6 7 While not well understood, the reasons for increased admissions are likely to be linked to changes in primary care provision, risk aversion among junior clinicians, a ‘defensive model’ of admission, advances in care reducing length of stay, funding arrangements and reduced parental experience in dealing with childhood illness.8 What constitutes a hospital ‘admission’ has also changed.9–13 Coon et al 9 examined the evidence for interventions intended to reduce acute paediatric admissions, including trials examining the effectiveness of five common initiatives: (1) consultant versus trainee decision on admission, (2) consultant telephone triage, (3) short stay/observation/assessment units, (4) algorithm-based care at admission and (5) next-day paediatric clinics. The evidence identified was weak and results equivocal; no firm conclusions could be drawn on effective initiatives for reducing admissions while avoiding negative impacts on those discharged. However, many hospitals are trying to change the organisation of care based on existing evidence and clinical experience.

nics. The evidence identified was weak and results equivocal; no firm conclusions could be drawn on effective initiatives for reducing admissions while avoiding negative impacts on those discharged. However, many hospitals are trying to change the organisation of care based on existing evidence and clinical experience. The Royal Devon and Exeter Hospital worked with the NIHR CLAHRC South West Peninsula (PenCLAHRC) and the South West Strategic Clinical Network (SWSCN) to implement an evidence-driven ‘best guess’ change in paediatric service delivery. This comprised the establishment of a short stay paediatric assessment unit (SSPAU), the design of which is largely derived from adult clinical decision units.14 15 This change was associated with an 18% fall in the number of overnight admissions in 2013 compared with the preceding 4-year period 2009–2012 (Martin et al submitted). As a result, the SWSCN partnered with PenCLAHRC to build a broader evidence base by mapping and assessing the impact of interventions in the region. This paper reports the first phase, a pilot study conducted in South Devon Healthcare NHS Foundation Trust’s Torbay Hospital, focusing on two interventions considered impactful, delivered at critical clinical and service decision points in patient care: (1) an advice and guidance (A&G) phone line on which a paediatrician is available for general practitioners (GPs) and the emergency department (ED) at all times and (2) SSPAU.

Torbay Hospital, focusing on two interventions considered impactful, delivered at critical clinical and service decision points in patient care: (1) an advice and guidance (A&G) phone line on which a paediatrician is available for general practitioners (GPs) and the emergency department (ED) at all times and (2) SSPAU. Objective The aim of this study was to describe and assess the impact of an A&G phone line and a SSPAU in reducing GP-referred attendances, admissions (including short-stay admissions) and length of stay of unplanned cases in Torbay Hospital. Methods Design We used a 7-year series of routine observational data to assess the impact of the two interventions implemented in sequence in Torbay. Intervention specifications were collected through telephone interview with the clinical lead (RT) and operations manager (GS). Routinely collected daily data relating to attendance, admission and length of stay (outcomes) were collected for the time period April 2009–December 2015. We prespecified data definitions and coding through collaboration with the clinical network and the local Academic Health Sciences Network (see online Supplementary material). 10.1136/bmjpo-2017-000235.supp1Supplementary file 1

Methods Design We used a 7-year series of routine observational data to assess the impact of the two interventions implemented in sequence in Torbay. Intervention specifications were collected through telephone interview with the clinical lead (RT) and operations manager (GS). Routinely collected daily data relating to attendance, admission and length of stay (outcomes) were collected for the time period April 2009–December 2015. We prespecified data definitions and coding through collaboration with the clinical network and the local Academic Health Sciences Network (see online Supplementary material). 10.1136/bmjpo-2017-000235.supp1Supplementary file 1 Setting Torbay Hospital is a foundation trust, medium-sized district general hospital, with paediatric services comprising a 19 bed/cot inpatient ward, including a two bed high dependency unit and six bed adolescent unit. Staffing consists of 13 acute consultants, eight level 1 training grades plus six middle tier trainees. The study population was all children (<18 years old) in the catchment area, estimated at 27 740 (figure from Local Authority Joint Strategic Needs Assessment, 2015). Importantly, this population increases in the summer months, but no robust estimate exists for this increase or health service use. Admission rates at Torbay, in the face of a national increase, are relatively flat; despite this, the clinical team are still implementing strategies to reduce unplanned acute admissions.

Importantly, this population increases in the summer months, but no robust estimate exists for this increase or health service use. Admission rates at Torbay, in the face of a national increase, are relatively flat; despite this, the clinical team are still implementing strategies to reduce unplanned acute admissions. Intervention A number of local interventions were mapped and two initiatives selected as the focus, based on anticipated impact: an A&G phone line, established in April 2014, on which a paediatrician is available for GPs and the ED at all times. At the commencement of the A&G phase, there was an increase from one to two consultants available for acute service provision. This increase was partly to enable more consultant input and partly to compensate for reduction in numbers of middle-grade paediatric staff owing to rota gaps. The purpose of A&G was to enable timely, robust communication with hospital-based paediatricians to agree most appropriate direction for unwell children. The phone line was a single phone held by a consultant (09:00–21:00 weekdays, 09:00–15:00 weekends) and middle-grade doctors outside of these hours. From November 2014, calls could result in: referrals to the newly established SSPAU to be seen that day (Monday–Friday) or where it was felt that immediate assessment was not required: A&G to GPs and parents enabling them to manage at home, sometimes with further review; booked review on SSPAU early the next day or booked into urgent (1–2 week) slots in consultant or registrar clinics. Calls were logged with an A&G clinic code and a summary placed with the patient notes.

assessment was not required: A&G to GPs and parents enabling them to manage at home, sometimes with further review; booked review on SSPAU early the next day or booked into urgent (1–2 week) slots in consultant or registrar clinics. Calls were logged with an A&G clinic code and a summary placed with the patient notes. The second intervention—a SSPAU—was established in November 2014, operating at full capacity immediately (five beds, one cubicle (SSPAU reduced the number of ward beds by two)). SSPAU was intended to be a place between primary care, ED and the paediatric ward to reduce admission to the ward of those not requiring lengthy care/review. The unit opened 09:00–20:00 weekdays, with last admission at 19:00. Those present at 21:00 either stayed late to complete care, were admitted to the ward or, if no beds, were kept in the SSPAU and counted as ‘overflow’. No patients were returned to ED but accepted referrals later than 19:00 were redirected to ED.

view. The unit opened 09:00–20:00 weekdays, with last admission at 19:00. Those present at 21:00 either stayed late to complete care, were admitted to the ward or, if no beds, were kept in the SSPAU and counted as ‘overflow’. No patients were returned to ED but accepted referrals later than 19:00 were redirected to ED. The unit was staffed by senior nursing staff (Band 6), healthcare assistants and additional consultant (taking total to 13) and targeted acutely unwell children (referred by GPs and/or ED) plus routine and review cases. SSPAU has not changed outpatient management of chronic conditions but acute deterioration would go to SSPAU. Children needing resuscitation on arrival and those being sent by ambulance all went to ED first. GPs could refer acute concerns directly via the A&G phone or less urgent concerns by letter or fax. The consultant responsible for the SSPAU was also the individual holding the A&G phone line. During implementation of these intervention, there were a number of changes to the GP landscape, which we are unable to account for due to a lack of robust data but remain important contextual factors. Primary outcomes We analysed routine hospital data for four service parameters for children under 18 years:GP-referred attendances; Paediatric ward admissions; Less than 1 day admissions; Length of stay on paediatric ward/s.

The unit was staffed by senior nursing staff (Band 6), healthcare assistants and additional consultant (taking total to 13) and targeted acutely unwell children (referred by GPs and/or ED) plus routine and review cases. SSPAU has not changed outpatient management of chronic conditions but acute deterioration would go to SSPAU. Children needing resuscitation on arrival and those being sent by ambulance all went to ED first. GPs could refer acute concerns directly via the A&G phone or less urgent concerns by letter or fax. The consultant responsible for the SSPAU was also the individual holding the A&G phone line. During implementation of these intervention, there were a number of changes to the GP landscape, which we are unable to account for due to a lack of robust data but remain important contextual factors. Primary outcomes We analysed routine hospital data for four service parameters for children under 18 years:GP-referred attendances; Paediatric ward admissions; Less than 1 day admissions; Length of stay on paediatric ward/s. The A&G phone line was evaluated on all four parameters. The SSPAU was evaluated for paediatric ward admissions, less than 1 day admissions and length of stay. We had no rationale for believing SSPAU had an impact on GP-referred attendances so we did not test this. We originally planned to analyse 48 hours readmission, but were unable to as our reclassification of SSPAU admissions as attendances meant that robust comparison data could not be collected given local system constraints.

The A&G phone line was evaluated on all four parameters. The SSPAU was evaluated for paediatric ward admissions, less than 1 day admissions and length of stay. We had no rationale for believing SSPAU had an impact on GP-referred attendances so we did not test this. We originally planned to analyse 48 hours readmission, but were unable to as our reclassification of SSPAU admissions as attendances meant that robust comparison data could not be collected given local system constraints. Data source Daily data were retrieved from local systems by a business intelligence specialist at the hospital (RR) and aggregated into monthly totals for analysis. We define an admission as presence in the hospital at midnight. Importantly, these are paediatric ward overnight admissions not simply hospital admissions (ie, SSPAU admissions are reclassified as attendances) and so are consistent preintervention and postintervention. Owing to collection method, admissions include elective and non-elective cases; however, electives were similar across all included years, both as a raw value and as a proportion of total admissions (range 17.2%–23.5%). Length of stay on the ward was measured in whole days and we distinguished between short stays (1 day or less) and other lengths of stay (2+ days), although we recognise others have defined this differently (eg, <2 days).16

included years, both as a raw value and as a proportion of total admissions (range 17.2%–23.5%). Length of stay on the ward was measured in whole days and we distinguished between short stays (1 day or less) and other lengths of stay (2+ days), although we recognise others have defined this differently (eg, <2 days).16 Data analysis Outcomes were measured with two distinct time periods: preintervention and postintervention, for each intervention. To assess the impact of the A&G phone line (introduced April 2014), the period April 2014–October 2014 was compared with the same period April–October in the preceding years combined for which data were available (2009–2013). To assess the impact of the SSPAU (introduced November 2014), the period November 2014–October 2015 was compared with November–October from 2009 to 2012, excluding the period November 2013–October 2014, which was confounded by the opening of the A&G phone line. Thus, the respective impact of the A&G phone line and the effect of the bundled A&G and SSPAU was assessed. For each outcome, preintervention and postintervention monthly totals were summarised using means and SD. Two sample t-tests were used to compare the outcomes between the preintervention and postintervention phases. Results of these analyses are reported as estimated differences in means (postintervention−preintervention), with 95% CIs and p values.

on and postintervention monthly totals were summarised using means and SD. Two sample t-tests were used to compare the outcomes between the preintervention and postintervention phases. Results of these analyses are reported as estimated differences in means (postintervention−preintervention), with 95% CIs and p values. Hypotheses We specified hypotheses following discussion with the clinical leads:A&G phone line: assessed in isolation. We anticipated a decrease in GP-referred attendances. We also anticipated a decrease in all admissions and a reduction in short-stay admissions. SSPAU: we anticipated a reduction in all admissions and short stay admissions. Combined, therefore, we anticipated a decrease in attendances, admissions and short-stay admissions. Ethics As a service evaluation, R&D management approval was sought and obtained from the Hospital R&D department. Results Figure 1 shows the total number of attendances and admissions for the hospital between January 2010 and December 2014 (data not shown for incomplete years: 2009 and 2015). In the face of national increases,1 both the total number of attendances and admissions remain relatively constant in Torbay. Figure 1 Total number of attendances and admissions at Torbay hospital each year, between January 2010 and December 2014.

Figure 1 shows the total number of attendances and admissions for the hospital between January 2010 and December 2014 (data not shown for incomplete years: 2009 and 2015). In the face of national increases,1 both the total number of attendances and admissions remain relatively constant in Torbay. Figure 1 Total number of attendances and admissions at Torbay hospital each year, between January 2010 and December 2014. Advice and guidance (A&G) phone line We assessed the impact of the A&G phone line on GP-referred attendances, ward admissions, less than 1 day admissions and overall bed-days using the time periods specified. Figure 2 shows total GP-referred attendances, ward admissions and short stays for each April–October period. Figure 2 Total number of general practitioner (GP)-referred attendances, ward admissions and short stays at Torbay hospital, for each April–October part-year period between April 2009 and October 2014. Red vertical line indicates introduction of intervention. There was little evidence of a change in monthly total GP-referred attendances postintervention (difference in means (post−pre) −17.1 (95% CI 5.6 to −39.8); p=0.1) or in monthly total ward admissions (difference in means (post−pre) −3.7 (95% CI 14.5 to −21.8); p=0.7) (table 1). Table 1 Summary of main results

Figure 2 Total number of general practitioner (GP)-referred attendances, ward admissions and short stays at Torbay hospital, for each April–October part-year period between April 2009 and October 2014. Red vertical line indicates introduction of intervention. There was little evidence of a change in monthly total GP-referred attendances postintervention (difference in means (post−pre) −17.1 (95% CI 5.6 to −39.8); p=0.1) or in monthly total ward admissions (difference in means (post−pre) −3.7 (95% CI 14.5 to −21.8); p=0.7) (table 1). Table 1 Summary of main results Intervention Outcome Preintervention mean (SD) monthly total Postintervention mean (SD) monthly total % Change in mean Mean change post−pre (95% CI) P values Apr–Oct 2009–2013 Apr–Oct 2014 A&G GP-referred attendances 206.8 (27.8) 189.7 (22.5) −8.3% −17.1 (5.6 to −39.8) 0.1 Ward admissions 252.9 (22.6) 249.3 (15.8) −1.5% −3.7 (14.5 to −21.8) 0.7 <1 day admissions 194.1 (20.3) 177.6 (14.2) −8.5% −16.6 (−0.2 to −32.9) 0.05 Overall bed-days 341.9 (62.5) 414.4 (55.6) 21.2% 72.5 (124.0 to 21.0) 0.01 SSPAU Nov–Oct 2009–2012 Nov–Oct 2014–2015 Ward admissions 248.0 (21.5) 213.4 (16.6) −14.0% −34.6 (-21.3 to −48.0) 0.0001 <1 day admissions 186.6 (21.8) 164.9 (14.7) −11.6% −21.7 (-8.4 to −35.1) 0.002 Overall bed-days 345.3 (60.4) 295.1 (52.5) −14.5% −50.2 (-12.1 to −88.3) 0.01 On average, monthly short-stay (less than 1 day) admissions reduced by 8.5% postintervention, from a mean monthly total of 194.1 (SD 20.3) to 177.6 (SD 14.2); difference in means −16.6 (95% CI −0.2 to −32.9); p=0.04 (table 1).

to −35.1) 0.002 Overall bed-days 345.3 (60.4) 295.1 (52.5) −14.5% −50.2 (-12.1 to −88.3) 0.01 On average, monthly short-stay (less than 1 day) admissions reduced by 8.5% postintervention, from a mean monthly total of 194.1 (SD 20.3) to 177.6 (SD 14.2); difference in means −16.6 (95% CI −0.2 to −32.9); p=0.04 (table 1). Monthly overall bed-days increased by 21.2% post-intervention, from a mean monthly total of 341.9 (SD 62.6) to 414.4 (SD 55.6); difference in means 72.5 (95% CI 21.0 to 124.0); p=0.01 (table 1). Short stay paediatric assessment unit We assessed the introduction of the SSPAU in November 2014, therefore what is assessed is the bundling of A&G and SSPAU. Figure 3 presents total ward admissions and short stays for each November–October period. Figure 3 Total number of ward admissions and short stays at Torbay hospital, for each November–October period between November 2009 and October 2015. Red vertical line indicates introduction of intervention. There was strong evidence of a reduction in monthly admissions following the introduction of the SSPAU, from a mean monthly total of 248.0 (SD 21.5) to 213.4 (SD 16.6); difference in means (post–pre) −34.6 (95% CI −21.3 to −48.0); p=0.0001 (table 1). On average, monthly short-stay (less than 1 day) admissions also reduced, by 11.6% postintervention, from a mean monthly total of 186.6 (SD 21.8) to 164.9 (SD 14.7); difference in means (post–pre) −21.7 (95% CI −8.4 to −35.1); p=0.002 (table 1).

There was strong evidence of a reduction in monthly admissions following the introduction of the SSPAU, from a mean monthly total of 248.0 (SD 21.5) to 213.4 (SD 16.6); difference in means (post–pre) −34.6 (95% CI −21.3 to −48.0); p=0.0001 (table 1). On average, monthly short-stay (less than 1 day) admissions also reduced, by 11.6% postintervention, from a mean monthly total of 186.6 (SD 21.8) to 164.9 (SD 14.7); difference in means (post–pre) −21.7 (95% CI −8.4 to −35.1); p=0.002 (table 1). Monthly overall bed-days reduced by 14.5% postintervention, from a mean monthly total of 345.3 (SD 60.4) to 295.1 (SD 52.5); difference in means (post−pre) −50.2 (95% CI −12.1 to −88.3); p=0.01 (table 1). Discussion We anticipated decreased GP-referred attendances following the introduction of the A&G phone line. There was little evidence of a real change although the size of the reduction is consistent with the clinical view (RT) that around 10% of calls avoid admission through discussion. Importantly, the A&G line increased partnership working between paediatrics and primary care, enabling more responsive and flexible care, with GPs valuing consultant contact and the ability to manage acute illness through discussion.

th the clinical view (RT) that around 10% of calls avoid admission through discussion. Importantly, the A&G line increased partnership working between paediatrics and primary care, enabling more responsive and flexible care, with GPs valuing consultant contact and the ability to manage acute illness through discussion. We anticipated reduced admissions following introduction of the A&G line, of which there was some suggestion, but again little statistical evidence. This fits clinical description as, prior to the SSPAU, there was nowhere to manage cases other than ED or the ward. We anticipated short-stay admissions would reduce, with results indicating that this was significantly lowered postintervention. With the introduction of SSPAU, we anticipated a reduction in admissions, less than 1 day admissions and bed-days. There was evidence that all of these outcomes reduced postintervention, and there are likely to be linked financial benefits, however, hospital-specific funding arrangements make robust assessments difficult; these interventions improve quality rather than simply reducing costs, with savings offset by the greater expense of providing additional consultant presence. There was an increase in overall bed-days after the introduction of A&G, probably due to fluctuating numbers of long-stay cases, likely a direct impact of including mental health cases (something which is also likely to have had an impact on the decrease in ward admissions following the introduction of SSPAU).

With the introduction of SSPAU, we anticipated a reduction in admissions, less than 1 day admissions and bed-days. There was evidence that all of these outcomes reduced postintervention, and there are likely to be linked financial benefits, however, hospital-specific funding arrangements make robust assessments difficult; these interventions improve quality rather than simply reducing costs, with savings offset by the greater expense of providing additional consultant presence. There was an increase in overall bed-days after the introduction of A&G, probably due to fluctuating numbers of long-stay cases, likely a direct impact of including mental health cases (something which is also likely to have had an impact on the decrease in ward admissions following the introduction of SSPAU). These reductions in assessments in hospital care represent, we believe, not only an improvement for those individuals but also greater consultant involvement in assessment and management has reduced investigations and interventions. It is possible that some parents whose children were not admitted experienced increased anxiety managing them at home, but we believe that consultant review before discharge and safety netting allays most fears.

so greater consultant involvement in assessment and management has reduced investigations and interventions. It is possible that some parents whose children were not admitted experienced increased anxiety managing them at home, but we believe that consultant review before discharge and safety netting allays most fears. Limitations/further research Results presented here would usefully be broken down by injury/illness to assess the impact of true admissions against summer/visitor accidents. Additionally, amending the age profile to less than 10 years16 could reduce the impact of mental health cases on length of stay. It may be beneficial to examine these data by Big 66 conditions; SSPAU may prevent asthma overnight stays, while A&G review may benefit fever cases. Conclusion The introduction of an A&G phone line for GPs to contact paediatric consultants at Torbay hospital was associated with a decrease in less than 1 day admissions and an increase in overall bed-days. The later addition of a SSPAU alongside the A&G phone line was associated with a reduction in ward admissions, less than 1 day admissions and overall bed-days. Further work should explore these results by age, condition and injury/illness status. We would like to thank the South West Strategic Clinical Network’s Maternity and Children’s Urgent Care Working Group, whose members helped conceive the project as well as facilitate the process throughout.

Conclusion The introduction of an A&G phone line for GPs to contact paediatric consultants at Torbay hospital was associated with a decrease in less than 1 day admissions and an increase in overall bed-days. The later addition of a SSPAU alongside the A&G phone line was associated with a reduction in ward admissions, less than 1 day admissions and overall bed-days. Further work should explore these results by age, condition and injury/illness status. We would like to thank the South West Strategic Clinical Network’s Maternity and Children’s Urgent Care Working Group, whose members helped conceive the project as well as facilitate the process throughout. Contributors: SL conceived the project. VB managed and led the project and KH was the main researcher. RT is a consultant paediatrician at the hospital, where GS is the operations manager and both assisted with data interpretation. RR identified, collected and processed local data. SB and OCU provided statistical advice. Funding: Intervention costs and RT, GS and RR’s time were funded by the Torbay and South Devon Foundation Hospital. KH, VB, SB, OCU and SL’s time was funded by the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care South West Peninsula at the Royal Devon and Exeter NHS Foundation Hospital. Disclaimer: The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. Competing interests: RT, GS and RR are employed by the hospital delivering these interventions. Patient consent: Not required.

Funding: Intervention costs and RT, GS and RR’s time were funded by the Torbay and South Devon Foundation Hospital. KH, VB, SB, OCU and SL’s time was funded by the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care South West Peninsula at the Royal Devon and Exeter NHS Foundation Hospital. Disclaimer: The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. Competing interests: RT, GS and RR are employed by the hospital delivering these interventions. Patient consent: Not required. Provenance and peer review: Not commissioned; externally peer reviewed.

What is already known on this topic? The consequences of permanent childhood hearing loss (PCHL) can include impairment in language skills and academic achievement which become more marked with more severe PCHL. Birth during periods with universal newborn hearing screening (UNHS) is associated with benefits to language and reading abilities. The societal costs of prelingual PCHL at 7–9 years increase with its severity and are inversely related to language abilities. What this study hopes to add? Total annual costs in adolescents with bilateral PCHL > 40 dB are, on average, 2.7 fold higher than in those with normal hearing. PCHL plus specified additional medical conditions is associated with a doubling of annual cost in adolescence compared to that of PCHL alone. In adolescents with PCHL, superior language skills are associated with significantly lower societal costs in adolescence but birth in periods with UNHS is not. Introduction Permanent childhood hearing loss (PCHL) is the most common sensory impairment. It is affecting more than 112 per 100 000 children at birth1 and incurs substantial economic costs to society,2 including those related to special education,3 4 employment,3 4 vocational rehabilitation,3 5 hearing aids, cochlear implants and other medical interventions.3 5 These costs to society are particularly high in severe and profound PCHL of prelingual onset and early intervention might confer substantial lifetime financial gains.5 Universal newborn hearing screening (UNHS) has been shown to increase the proportion of cases of PCHL that are detected early.1 6–8

al interventions.3 5 These costs to society are particularly high in severe and profound PCHL of prelingual onset and early intervention might confer substantial lifetime financial gains.5 Universal newborn hearing screening (UNHS) has been shown to increase the proportion of cases of PCHL that are detected early.1 6–8 UNHS adds to financial costs in the first year of life, both because of the cost of administering a UNHS programme, estimated in 1998 as £13 881 per annum for a district with 1000 births,9 and because of the additional costs of management of identified cases during the months that would otherwise precede identification of the permanent childhood hearing impairment (PCHI). On the other hand, earlier identification of children born with PCHL can facilitate earlier access to linguistic input and better language and literacy skills10–16 and may thus reduce cost subsequent to infancy. More research into the long-term cost-effectiveness of UNHS is needed17 and rigorous data on long-term economic consequences of PCHL are required to conduct cost-effectiveness evaluation of UNHS programmes that take into account the long-term consequences of hearing loss.18 There is, however, very little direct evidence regarding the long-term economic implications of PCHL18 or the effect on them of UNHS for PCHL.

-term economic consequences of PCHL are required to conduct cost-effectiveness evaluation of UNHS programmes that take into account the long-term consequences of hearing loss.18 There is, however, very little direct evidence regarding the long-term economic implications of PCHL18 or the effect on them of UNHS for PCHL. Prior to 2000, attempts to model the long-term costs and outcomes of PCHL were limited by lack of data3 and uncertainty regarding the effectiveness of UNHS.5 19 From 2003 onwards, UNHS has been implemented in the UK, USA and numerous other countries in the light of high-grade evidence of the benefits of UNHS14 20 and there has been significant progress in the provision of paediatric audiological services.21 22 In 2009, an estimated 5073 cases of PCHL were detected by UNHS23 and accounted for over 43% of the confirmed cases of all 29 medical conditions for which universal newborn screening is mandated in the USA.23 The hypothesis that early detection of PCHL reduces the costs of education in the long term3 12 21 24 and thus offsets the initial costs of UNHS incurred in infancy warrants examination.

nted for over 43% of the confirmed cases of all 29 medical conditions for which universal newborn screening is mandated in the USA.23 The hypothesis that early detection of PCHL reduces the costs of education in the long term3 12 21 24 and thus offsets the initial costs of UNHS incurred in infancy warrants examination. We have previously reported that the economic costs of bilateral PCHL in the preceding year of life among participants in the present study when they were aged 5–10 years were £14 092 for children with PCHL compared with £4207 for normally hearing children. Furthermore, each unit increase in the z score for receptive language among children with PCHL was associated with a statistically significant £2553 reduction in cost in the preceding year. Compared with birth during periods without UNHS, birth during periods with UNHS was associated with a smaller cost reduction of £2213, which fell short of statistical significance.25 The participants in that study were subsequently further evaluated at ages 13–20 years in the Hearing Outcomes at Teen Age (HOT) project. We report here on the effects of the severity of their PCHL, of birth during periods with UNHS, of early confirmation of PCHL and of their language ability and reading skills on the societal costs of PCHL.

t study were subsequently further evaluated at ages 13–20 years in the Hearing Outcomes at Teen Age (HOT) project. We report here on the effects of the severity of their PCHL, of birth during periods with UNHS, of early confirmation of PCHL and of their language ability and reading skills on the societal costs of PCHL. Participants and methods Study sample The study sample was drawn from 157 000 children born in two birth cohorts in eight districts of Southern England between 1992 and 1997. The Wessex cohort was born over a 36-month period in four districts that formed the population for the Wessex Trial; a quasiexperimental trial in which UNHS was or was not undertaken in alternating 4–6 month periods in two pairs of hospitals, with UNHS equipment and personnel moving back and forth between the paired hospitals. UNHS increased the rate of early identification for infants with PCHL.1 6 14 The Greater London birth cohort was born in two pairs of health districts in Greater London over a 60-month period. Each pair included one of the only two districts in the UK offering UNHS at that time and an immediately neighbouring district.

UNHS increased the rate of early identification for infants with PCHL.1 6 14 The Greater London birth cohort was born in two pairs of health districts in Greater London over a 60-month period. Each pair included one of the only two districts in the UK offering UNHS at that time and an immediately neighbouring district. The language, reading, behaviour and resource use in children with PCHL in these two birth cohorts, and in a normally hearing comparison group (HCG) was assessed in 183 children (120 with PCHL and 63 in the HCG) at a mean age of 7.9 years.13 15 25 26 Further assessment of 114 (73 PCHL and 37 HCG) of the sample was undertaken in the Hearing Outcomes at Teen Age (HOT) project at a mean age of 16.9 years.27–29 A flow diagram of participants through completion of the HOT study was published in our report of the effect of UNHS on reading comprehension, the primary outcome.28 The design included one participant in a normal HCG for every two participants with PCHL in the expectation of providing three equally sized groups: participants with PCHL exposed and not exposed to a UNHS programme and participants in the HCG. Written informed consent was obtained from principal caregivers and the teenage participants.

The language, reading, behaviour and resource use in children with PCHL in these two birth cohorts, and in a normally hearing comparison group (HCG) was assessed in 183 children (120 with PCHL and 63 in the HCG) at a mean age of 7.9 years.13 15 25 26 Further assessment of 114 (73 PCHL and 37 HCG) of the sample was undertaken in the Hearing Outcomes at Teen Age (HOT) project at a mean age of 16.9 years.27–29 A flow diagram of participants through completion of the HOT study was published in our report of the effect of UNHS on reading comprehension, the primary outcome.28 The design included one participant in a normal HCG for every two participants with PCHL in the expectation of providing three equally sized groups: participants with PCHL exposed and not exposed to a UNHS programme and participants in the HCG. Written informed consent was obtained from principal caregivers and the teenage participants. Measures Severity of PCHL was classified according to average pure tone thresholds across four frequencies of sound in the better ear as moderate (> 40–70 decibels (dB) hearing level), severe (71–95 dB) or profound (>95 dB). Intellectual disability (defined by non-verbal ability scores), genetic syndrome, visual loss and cerebral palsy were recorded as additional medical conditions (AMC). Methods of assessing participants’ reading comprehension, receptive language ability, non-verbal ability and other outcomes have been reported previously.27–29 Occupation of the head of the household and maternal educational level were, as in our previous 2001–2004 assessment,25 described using UK 2001 national census categories.30

sessing participants’ reading comprehension, receptive language ability, non-verbal ability and other outcomes have been reported previously.27–29 Occupation of the head of the household and maternal educational level were, as in our previous 2001–2004 assessment,25 described using UK 2001 national census categories.30 Resource use and costs Resource use was considered from the healthcare provider, National Health Service (NHS), Personal Social Services (PSS) and societal perspective, including costs borne by the family. It was estimated by retrospective examination of each child’s audiology records coupled with data on resource use in seven domains (vide infra) obtained by the four study research assistants at interviews of parents in their homes using instruments previously developed for our 2001–2004 study of the same families.25 These data covered use of a range of services during the preceding 6 months, a period short enough for recall to be reliable, and were extrapolated to provide an estimate of annual cost. The research staff involved in the follow-up study were unaware of the age of initial referral and management and, in the case of the Wessex subgroup, blind to whether or not the child was born in a period with UNHS.

od short enough for recall to be reliable, and were extrapolated to provide an estimate of annual cost. The research staff involved in the follow-up study were unaware of the age of initial referral and management and, in the case of the Wessex subgroup, blind to whether or not the child was born in a period with UNHS. All unit costs adopted in the analysis were based on 2012/2013 price indexes. Health and Community Health Services pay and price indices were used to inflate costs, where appropriate.31 Published sources of unit costs included NHS Reference Costs32 and the PSS Resource Unit estimates.31 Unit costs for schooling were accessed individually for each school (from the UK Department for Education for state schools and from individual schools for the private sector) and the mean unit cost estimates for each type of school were included in the analysis.33 Other unit cost estimates were obtained from local authorities and local suppliers. Costs, estimated at the individual person level, are presented in the form of group means and SDs in seven domains: hospital outpatient and inpatient services, including cochlear implantation; community health and social care services; respite and foster care; local authority loaned/provided equipment and home adaptations; educational services including special educational needs provision; parents’ lost productivity; and other household-borne costs, including household purchased equipment and home adaptations.

mmunity health and social care services; respite and foster care; local authority loaned/provided equipment and home adaptations; educational services including special educational needs provision; parents’ lost productivity; and other household-borne costs, including household purchased equipment and home adaptations. Statistical analysis The primary outcome variable for the economic study reported here was total costs which were compared between the teenagers with PCHL and the HCG. As the time frame for the cost analysis was 1 year, discounting applied to economic evaluations in excess of a 1-year time frame was not necessary. The target sample size of 96 children with PCHL for the HOT project, that is, 80% of the participants with PCHL that had been assessed at a mean age of 7.9 years, was estimated to provide 90% power at 5% significance level (two tailed) to detect a 0.67 SD effect size of UNHS on reading comprehension, the prespecified primary outcome measure in the HOT study, in participants with PCHL. Sample size was determined by the above power calculation rather than any separate power calculation relating to power to detect group differences in costs, the secondary outcome reported here.

t size of UNHS on reading comprehension, the prespecified primary outcome measure in the HOT study, in participants with PCHL. Sample size was determined by the above power calculation rather than any separate power calculation relating to power to detect group differences in costs, the secondary outcome reported here. Among participants with PCHL, the effect on costs was assessed in four regression models, each with one independent variable of interest: birth during periods with UNHS; ‘early’ confirmation of PCHL; receptive language ability z-score; and reading comprehension z-score. These effects are presented unadjusted and adjusted in two regression models. The first model adjusted for cochlear implantation and the presence of AMCs and the second model added severity of PCHI into that regression model. As violation of normality was confirmed (Shapiro-Wilk P<0.05), mean differences between groups are presented with 95% CIs estimated by bootstrapping (1000 replications). In addition to conventional ordinary least squares (OLS) regression analysis, generalised linear models (GLM) using non-normal distributions, and the alternative model specifications were examined for robustness to deviations from normality and equality of variance in costs.34 OLS and GLM analyses gave very similar results so we used OLS findings with robust SEs. All analyses were carried out using STATA V.12 and R V.3.1.1.35–37

ing non-normal distributions, and the alternative model specifications were examined for robustness to deviations from normality and equality of variance in costs.34 OLS and GLM analyses gave very similar results so we used OLS findings with robust SEs. All analyses were carried out using STATA V.12 and R V.3.1.1.35–37 Results Four of 114 participants in the HOT study did not return the completed economic questionnaire and resource use is therefore reported in 110 participants in this economic study. The mean (SD) age of the participants was 16.9 (1.4) years. Of the participants with PCHL, 32 (44%), 18 (25%) and 23 (32%) had moderate, severe and profound PCHL, respectively (table 1). For the PCHL group there were no significant differences of gender, severity of PCHL, mother’s educational qualifications, or English as the main language at home between participants and those lost to follow-up in the larger sample of 120 children with PCHL, who had been assessed at 7.9 years.27 Additional demographic characteristics by UNHS status and by timing of confirmation of PCHL are presented in online supplementary appendix table 1. Online supplementary appendix table 1 and our previous reports indicate that the (approximately) half of our study population with PCHL that was born in periods with UNHS was similar to the other half born in periods without UNHS with respect to the severity of their PCHL. That is to say severity of PCHL was not a confounder of UNHS status when considering the effect of UNHS on costs. Resource use (table 2) was combined with unit costs (table 3) to derive total costs in all participants (table 4).

to the other half born in periods without UNHS with respect to the severity of their PCHL. That is to say severity of PCHL was not a confounder of UNHS status when considering the effect of UNHS on costs. Resource use (table 2) was combined with unit costs (table 3) to derive total costs in all participants (table 4). 10.1136/bmjpo-2017-000228.supp1Supplementary file 1 Table 1 Sociodemographic and clinical characteristics of study participants

to the other half born in periods without UNHS with respect to the severity of their PCHL. That is to say severity of PCHL was not a confounder of UNHS status when considering the effect of UNHS on costs. Resource use (table 2) was combined with unit costs (table 3) to derive total costs in all participants (table 4). 10.1136/bmjpo-2017-000228.supp1Supplementary file 1 Table 1 Sociodemographic and clinical characteristics of study participants Variable Bilateral permanent childhood hearing loss >40 dB HCG n=37 Moderate n=32 Severe n=18 Profound n=23 Total n=73 Age mean (SD) in years 16.9 (1.4) 17.5 (1.4) 16.8 (1.5) 17.0 (1.4) 16.3 (1.2) Female, n (%) 16 (50) 9 (50) 10 (44) 35 (48) 13 (35) Mode of communication, n (%) Oral 22 (69) 10 (56)* 11 (48)* 43 (59)   – Sign   0   0 1 (4) 1 (1)   – More than one mode 10 (31) 8 (44) 11 (48) 29 (40)   – UNHS status, n (%) Born in periods without UNHS 14 (44) 10 (56) 14 (61) 38 (52)   – Age PCHL confirmed, n (%) >9 completed months 16 (50) 12 (67) 11 (48) 39 (53)   – English main language at home, n (%) 31 (97) 15 (83) 18 (78) 64 (88)   – Mother’s educational qualifications†, n (%) No qualifications 3 (9) 2 (11) 2 (9) 7 (10) 2 (5) <5 O-level examinations 4 (12) 1 (6) 4 (17) 9 (12) 3 (8) ≥5 O-level examinations 9 (28) 7 (39) 8 (35) 24 (33) 13 (35) Some A-level examinations 8 (25) 4 (22) 2 (9) 14 (19) 1 (3) ≥University degree 8 (25) 4 (22) 7 (30) 19 (26) 18 (49) Social class‡, n (%) Higher occupations 15 (47) 10 (56) 11 (48) 36 (49) 26 (70) Intermediate occupations 10 (31) 3 (17) 5 (22) 18 (25) 8 (22) Lower occupations 4 (12)    0 5 (22) 9 (12) 3 (8) Never worked and long term unemployed 3 (9) 5 (28) 2 (9) 10 (14)    0 Family income, n (%) <10 000 4 (13) 2 (12)    0 6 (9)    0 10 000–20 000 6 (20) 2 (12) 7 (33) 15 (22) 4 (11) 21 000–30 000 2 (7) 4 (24) 2 (10) 8 (12) 7 (19) 31 000–40 000 7 (23) 2 (12) 4 (19) 13 (19) 4 (11) 41 000–50 000 3 (10) 2 (12) 1 (5) 6 (9) 5 (14) >50 000 8 (27) 5 (29) 7 (33) 20 (29) 17 (46) Additional medical conditions§, n (%) 9 (28) 3 (11) 4 (17) 16 (22) 1 (3) Hearing aids, n (%) No aid 3 (9) 2 (17) 11 (48) 16 (22)   – One aid 4 (12) 1 (6) 3 (13) 8 (11)   – Two aids 25 (78) 15 (83) 9 (39) 49 (67)   – Number of cochlear implant(s), n (%) None 32 (100) 17 (94) 11 (48) 59 (81)   – 1 – 1 (6) 7 (30) 9 (12)   – 2 – – 5 (22) 5 (7)   – *Language other than English in one participant.

earing aids, n (%) No aid 3 (9) 2 (17) 11 (48) 16 (22)   – One aid 4 (12) 1 (6) 3 (13) 8 (11)   – Two aids 25 (78) 15 (83) 9 (39) 49 (67)   – Number of cochlear implant(s), n (%) None 32 (100) 17 (94) 11 (48) 59 (81)   – 1 – 1 (6) 7 (30) 9 (12)   – 2 – – 5 (22) 5 (7)   – *Language other than English in one participant. †O level refers to ‘ordinary levels’ UK qualification achieved at 16 years. A level refers to ‘advanced levels’ UK qualification achieved at 18 years. ‡Classified according to UK National Census 2002. §These were severe visual impairment, cerebral palsy, mental retardation and genetic syndrome. HCG, hearing comparison group; PCHL, permanent childhood hearing loss >40 dB; UNHS, universal newborn hearing screening. Table 2 Estimated group mean resource use in preceding 12 months

‡Classified according to UK National Census 2002. §These were severe visual impairment, cerebral palsy, mental retardation and genetic syndrome. HCG, hearing comparison group; PCHL, permanent childhood hearing loss >40 dB; UNHS, universal newborn hearing screening. Table 2 Estimated group mean resource use in preceding 12 months Resource items* Bilateral permanent childhood hearing loss >40 dB HCG (n=37) Moderate (n=32) Severe (n=18) Profound (n=23) Total (n=73) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Community and social care services contacts General practitioner 2.25 (3.04) 2.89 (3.51) 2.70 (2.80) 2.55 (3.06) 1.46 (2.14) Practice nurse 0.56 (1.37) 0.44 (1.10) 16.52 (74.90) 5.56 (42.09) 0.16 (0.55) Community nurse – 0.44 (1.89) 0.26 (1.25) 0.19 (1.16) 0.05 (0.33) Community paediatrician 0.13 (0.49) 0.11 (0.47) 0.09 (0.42) 0.11 (0.46) 0.11 (0.66) Dentists 1.44 (1.27) 1.67 (1.41) 1.39 (1.12) 1.48 (1.25) 1.35 (1.16) Orthodontist 0.88 (2.15) 0.67 (1.94) 1.57 (2.95) 1.04 (2.38) 1.51 (2.77) Optician 0.63 (0.94) 0.78 (1.00) 1.22 (1.00) 0.85 (1.00) 0.65 (0.95) Chiropodist 0.44 (1.50) – – 0.19 (1.01) – Physiotherapist 0.63 (1.86) 0.22 (0.65) 2.35 (10.03) 1.07 (5.76) 0.38 (1.62) Speech and language 1.94 (9.25) 17.56 (56.85) 18.35 (38.43) 10.96 (36.27) 0.05 (0.33) Health visitor 1.44 (1.70) 1.11 (1.23) 1.30 (1.66) 1.32 (1.57) – Home visitor 0.31 (1.15) 0.44 (1.46) 0.96 (2.75) 0.55 (1.86) – Social worker – 5.33 (15.52) – 1.32(7.89) – Counsellor 0.81 (4.25) – 0.61 (2.04) 0.55 (3.02) 0.05 (0.33) Community psychologist – – – – 0.11 (0.46) Community psychiatrist – 0.11 (0.47) 0.09 (0.42) 0.05 (0.33) 0.05 (0.33) Osteopath – – – – – Audiologist 0.44 (0.84) 0.44 (1.10) 0.17 (0.58) 0.36 (0.84) – Other 0.06 (0.25) 0.11 (0.47) – 0.05 (0.28) – Other care service Respite care (days) 0.38 (2.12) 4.00 (12.35) 4.13 (13.20) 2.45 (9.70) – Foster care (days) 6.66 (37.70) – – 2.92 (24.90) – Hospital outpatient, attendances Category 1 (ENT) 0.19 (0.78) 0.44 (1.46) 0.09 (0.42) 0.22 (0.92) 0.22 (1.03) Category 2 (A&E) 0.81 (2.02) 0.56 (1.50) 0.87 (2.40) 0.77 (2.02) 0.86 (1.86) Category 3 (other) 0.19 (0.78) 0.11 (0.47) 0.26 (1.25) 0.19 (0.89) 0.22 (0.79) Hospital inpatient admissions (days) Cochlear implant – – 0.17 (0.39) 0.05 (0.23) – Total days 0.22 (0.94) – 0.17 (0.49) 0.15 (0.68) – Education, number of children attending: n (%) Mainstream school 19 (65.5) 5 (35.7) 4 (17.4) 28 (42.4) 33 (100.0) Mainstream school with unit for deaf 3 (10.3) 5 (35.7) 5 (21.7) 13 (19.7) – Special school for deaf 1 (3.5) 2 (

hlear implant – – 0.17 (0.39) 0.05 (0.23) – Total days 0.22 (0.94) – 0.17 (0.49) 0.15 (0.68) – Education, number of children attending: n (%) Mainstream school 19 (65.5) 5 (35.7) 4 (17.4) 28 (42.4) 33 (100.0) Mainstream school with unit for deaf 3 (10.3) 5 (35.7) 5 (21.7) 13 (19.7) – Special school for deaf 1 (3.5) 2 ( 14.3) 11 (47.8) 14 (21.2) – Other special school 5 (17.2) 2 (14.3) 2 (8.7) 9 (13.6) – Other school 1 (3.5) – 1 (4.4) 2 (3.0) – Residential school 1 (3.4) 2 (14.3) 10 (43.5) 13 (19.7) – *Medication costs are not included. Thirty-four of 110 reported having used medication which was unnamed in 16. Dose and frequency information was seldom available. A&E, Accident and Emergency; ENT, Ear, Nose and Throat; HCG, hearing comparison group. Table 3 Unit costs of resource items Resource items Unit cost or range* Source of unit cost Community and social care services, per contact hour Practice nurse 41.0 (35.0–53.0) Curtis31 Community nurse 39.0 (33.0–43.0) Curtis31 Community paediatrician 223.0 NHS Reference Costs32 Dentists 115.0 NHS Reference Costs32 Orthodontist 45.0 NHS Reference Costs32 Optician 138.0 NHS Reference Costs32 Chiropodist 41.0 (33.0–45.0) Curtis31 Physiotherapist 47.0 (37.0–53.0) Curtis31 Speech and language 74.0 (52.0–87.0) Curtis31 Health visitor/research therapist 44.0 (33.0–54.0) Curtis31 Social worker 54.0 (34.0–150.0) Curtis31 Counsellor 35.6–90.1 Inflated PSSRU, 200734 Community psychologist 60.0–136.0 Curtis31 Community psychiatrist 60.0 Curtis31 Osteopath 35.0–50.0 NHS Reference Costs32 Audiologist 150.0 NHS Reference Costs32

Speech and language 74.0 (52.0–87.0) Curtis31 Health visitor/research therapist 44.0 (33.0–54.0) Curtis31 Social worker 54.0 (34.0–150.0) Curtis31 Counsellor 35.6–90.1 Inflated PSSRU, 200734 Community psychologist 60.0–136.0 Curtis31 Community psychiatrist 60.0 Curtis31 Osteopath 35.0–50.0 NHS Reference Costs32 Audiologist 150.0 NHS Reference Costs32 General practitioner, per consultation 53.0 (43.0–63.0) Curtis31 Other care service, per week Residential respite care 268.0 (71.0–413.0) Inflated PSSRU, 201135 Foster care 637.0 Inflated PSSRU, 201134 Hospital outpatient, per attendance† Category 1 (ENT) 71.7 (45.0–98.0) NHS Reference Costs32 Category 2 (A&E) 137.6 (106.0–197.0) NHS Reference Costs32 Category 3 (other) 268.6 (205.0–351.0) NHS Reference Costs32 Hospital inpatient admissions, per admission Cochlear implant‡ 20 333.0–30 709.0 NHS Reference Costs32 Paediatric ward 757.0–12 281.0 NHS Reference Costs32 Other 545.0–1846.0 NHS Reference Costs32 Education, per year Mainstream school 4581.0 Department of Education, 201236 Mainstream school with special unit 4819.0 Department of Education, 201236 Special school for the physically disabled 17 795.0–27 000.0 Local authority (Southampton) Residential school 61 859.0–167 268.0 NASS33 and individual schools Special school for learning difficulties/deaf 15 580.0–25 833.0 Local authority (Southampton) Equipment loaned, per year Digital hearing aid 126 NHS Reference Costs32

al school for the physically disabled 17 795.0–27 000.0 Local authority (Southampton) Residential school 61 859.0–167 268.0 NASS33 and individual schools Special school for learning difficulties/deaf 15 580.0–25 833.0 Local authority (Southampton) Equipment loaned, per year Digital hearing aid 126 NHS Reference Costs32 Wheelchair 172.0 Inflated PSSRU, 2011 Loop system 137.0–1200.0 Local provider Vibrating alarm clock 15.0–85.0 Local provider Doorbell/light 8.5–59.9 Local provider Fire alarm and flashing lights 7.7–138.0 Local provider Light-up phone 34.8–70.8 Local provider Local authority provided home adaptations, unit cost Bathing equipment 4539 Local authority (Southampton) Adapted shower 5000 Local authority (Southampton) Accessible kitchen built 483 Curtis31 Values are £2013. *Ranges of unit costs are specified where unit costs varied according to location or intensity of care provided. †Hospital outpatient attendances are categorised as low, medium and high cost services. ‡Includes cost of cochlear implant equipment and surgical procedure and other inpatient costs. A&E, Accident and Emergency; ENT, Ear, Nose and Throat; NASS, National Association of Independent Schools and Non-Maintained Special Schools; NHS, National Health Service; PSSRU, Personal Social Services Resource Unit. Table 4 Annual mean costs by cost category for health and social service use and UNHS status

‡Includes cost of cochlear implant equipment and surgical procedure and other inpatient costs. A&E, Accident and Emergency; ENT, Ear, Nose and Throat; NASS, National Association of Independent Schools and Non-Maintained Special Schools; NHS, National Health Service; PSSRU, Personal Social Services Resource Unit. Table 4 Annual mean costs by cost category for health and social service use and UNHS status Cost domain Children with PCHL HCG (n=37) Mean (SD) PCHL versus HCG Mean difference (Bootstrap SE) 95% CI Moderate (n=32) Severe (n=18) Profound (n=23) All PCHL (n=73) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Hospital outpatient 168.94 (330.48) 125.11 (274.74) 202.96 (693.85) 168.85 (461.30) 141.19 (307.21) 27.66 (76.15) −121.60 to 176.92 No UNHS 238.57 (383.15) 85.00 (191.79) 60.00 (224.50) 132.37 (291.70) UNHS 114.78 (282.50) 175.25 (361.72) 425.33 (1072.67) 208.46 (595.76) Hospital inpatient 165.59 (712.58) – 65.83 (218.09) 93.33 (487.75) – 93.33 (60.86) −25.96 to 212.61 No UNHS 108.14 (404.63) 54.07 (202.32) 59.76 (271.61) UNHS 210.28 (892.13) 84.11 (252.33) 129.77 (648.75) Cochlear implant – – 2652.13 (7001.67) 835.60 (4064.28) – 835.6 (415.43) 21.37 to 1649.83 No UNHS 2904.71 (7383.66) 1070.16 (4601.24) UNHS 2259.22 (6777.67) 580.94 (3436.90) Hospital total 400.16 (911.06) 191.78 (353.71) 2947.00 (7170.03) 1151.21 (4195.99) 141.19 (307.21) 1010.02 (578.76) −124.33 to 144.37 No UNHS 432.43 (529.46) 205.00 (366.22) 3040.21 (7508.91) 1333.34 (4657.81) UNHS 375.06 (1139.16) 175.25 (361.72) 2802.00 (7050.75) 953.46 (3687.43) Community and social care 776.09 (916.37) 2433.32 (4906.58) 2590.92 (3608.70) 1756.51 (3284.68) 503.79 (437.67) 1252.72 (402.05) 464.71 to 2040.73 No UNHS 930.67 (1185.75) 2799.16 (6461.98) 3061.47 (4450.85) 2207.41 (4312.64) UNHS 655.86 (648.94) 1976.01 (2085.33) 1858.94 (1621.98) 1266.97 (1460.70) Respite care 100.13 (566.39) – 174.13 (835.10) 98.75 (596.21) – 98.75 (51.45) −2.08 to 199.59 No UNHS 228.86 (856.31) 286.07 (1070.38) 189.71 (820.98) UNHS – – – Foster care 604.77 (3421.08) – – 265.10 (2265.05) – 265.10 (226.56) −178.93 to 709.14 No UNHS – – UNHS 1075.14 (4561.45) 552.93 (3271.18) Equipment and home adaptations 129.51 (220.89) 200.33 (238.68) 63.48 (122.87) 126.17 (204.22) – 126.17 (22.49) 82.10 to 170.24 No UNHS 168.64 (286.39) 223.00 (282.84) 36.43 (72.39) 134.24 (237.31) UNHS 99.07 (154.97) 172.00 (183.95) 105.56 (172.43) 117.41 (163.97) Educational services 9250.30 (7817.55) 9743.14 (9356.48) 17 752.05 (10 077.91) 12 082.50 (9655.09) 5330.24 (1690.63) 6752.26 (1117.46) 4562.08 to 8942.44 No UNHS 12 020.40 (9894.91) 9940.71 (

68.64 (286.39) 223.00 (282.84) 36.43 (72.39) 134.24 (237.31) UNHS 99.07 (154.97) 172.00 (183.95) 105.56 (172.43) 117.41 (163.97) Educational services 9250.30 (7817.55) 9743.14 (9356.48) 17 752.05 (10 077.91) 12 082.50 (9655.09) 5330.24 (1690.63) 6752.26 (1117.46) 4562.08 to 8942.44 No UNHS 12 020.40 (9894.91) 9940.71 ( 9096.11) 17 080.49 (10 828.66) 13 429.16 (10 251.11) UNHS 7095.77 (5033.60) 9520.87 (10 268.11) 18 796.71 (9313.30) 10 658.89 (8907.32) Lost productivity 16.88 (59.48) – 82.61 (315.73) 33.42 (182.09) 12.97 (78.91) 20.45 (24.39) −27.35 to 68.25 No UNHS 21.43 (80.18) 107.14 (400.89) 47.37 (246.86) UNHS 13.33 (38.81) 44.44 (101.38) 18.29 (58.69) Other household 521.25 (2040.87) 461.22 (912.33) 710.09 (1289.46) 565.95 (1583.72) 38.92 (236.73) 527.03 (153.73) 225.72 to 828.34 No UNHS 927.43 (3071.52) 208.60 (390.24) 686.57 (1372.79) 649.53 (2023.91) UNHS 205.33 (373.21) 777.00 (1272.98) 746.67 (1227.84) 475.20 (915.15) Total costs excluding education 2548.77 (4914.31) 3286.65 (5160.29) 6568.22 (8890.53) 3997.12 (6633.79) 696.88 (674.54) 3300.24 (644.21) 2037.62 to 4562.86 No UNHS 2709.45 (4331.12) 3435.76 (6541.88) 7217.90 (10 191.51) 4561.59 (7602.96) UNHS 2423.80 (5445.76) 3100.26 (3094.22) 5557.61 (6833.21) 3384.26 (5435.40) Total costs including education 11 799.07 (10 186.17) 12 488.50 (13 179.43) 24 320.28 (16 381.95) 15 914.10 (14 167.56) 5883.05 (2075.50) 10 031.05 (1822.27) 6459.47 to 13 602.62 No UNHS 14 729.85 (12 598.90) 12 382.40 (13 892.31) 24 298.39 (19 877.75) 17 637.35 (16 401.04) UNHS 9519.57 (7432.83) 12 621.12 (13 178.09) 24 354.32 (9794.52) 14 043.15 (11 198.32) HCG, hearing comparison group; PCHL, permanent childhood hearing loss > 40 dB; UNHS, universal newborn hearing screening.

47 to 13 602.62 No UNHS 14 729.85 (12 598.90) 12 382.40 (13 892.31) 24 298.39 (19 877.75) 17 637.35 (16 401.04) UNHS 9519.57 (7432.83) 12 621.12 (13 178.09) 24 354.32 (9794.52) 14 043.15 (11 198.32) HCG, hearing comparison group; PCHL, permanent childhood hearing loss > 40 dB; UNHS, universal newborn hearing screening. Comparison between those with PCHL and the HCG The mean (SD) cost estimates for the teenagers with PCHL and the HCG were £15 914 (14 168) and £5883 (2076), respectively (mean difference (95% CI) £10 031 (£6459 to £13 603), P<0.001) (table 4). Both educational costs and the sum of all other costs differed significantly between these groups. The educational cost difference of £6752 was the main cost driver (table 4).

PCHL and the HCG were £15 914 (14 168) and £5883 (2076), respectively (mean difference (95% CI) £10 031 (£6459 to £13 603), P<0.001) (table 4). Both educational costs and the sum of all other costs differed significantly between these groups. The educational cost difference of £6752 was the main cost driver (table 4). Comparisons within the PCHL group Effect of severity of PCHL Moderate, severe and profound PCHL were associated with mean costs of £11 799, £12 489 and £24 320, respectively (table 4). The mean cost differences from the HCG for moderate, severe and profound PCHL were £5916, £6605 and £18 437, respectively. The mean difference (95% CI) in the cost of profound compared with other severities of PCHL was £12 273 (£4808 to £19 738) (P=0.002). The higher cost of attendance at boarding and independent special schools was the main cost driver: the percentage (95% CI) of teenagers with moderate, severe and profound PCHL attending residential schools was 8 (1% to 42%), 15 (4% to 47%) and 77 (46% to 93%), respectively, and the percentage (95% CI) attending mainstream schools was 68 (48% to 83%), 18 (7% to 37%) and 14 (5% to 33%), respectively. Community health, social care and hospital-based service costs all increased significantly with severity (table 4). For those with profound PCHL, the cost of cochlear implantation was only incurred during the assessed period of resource use in a small proportion but the group mean cost of cochlear implant (£2652) was nevertheless a key cost driver (table 4).

al-based service costs all increased significantly with severity (table 4). For those with profound PCHL, the cost of cochlear implantation was only incurred during the assessed period of resource use in a small proportion but the group mean cost of cochlear implant (£2652) was nevertheless a key cost driver (table 4). Effect of AMCs Of those participants with AMC, 56%, 19% and 25% had moderate, severe, and profound PCHL, respectively. Overall, the presence of a medical condition additional to PCHL was associated with higher mean costs (95% CI) by £15 385 (£8533 to £22 238). This cost difference was £21 876 (£13 024 to £30 728) for loss of vision, £14 200 (£5620 to £22 780) for genetic syndromes, £11 728 (£5456 to £18 000) for intellectual disability and £1642 (−£8013 to £11 296) for cerebral palsy. For the moderate, severe and profound groups, the mean costs by severity for the PCHL group with and without AMC (n=1657) were £22 436, £36 318, £33 990 and £7637, £7723, £22 285, respectively. Effect of birth during periods with UNHS In children with PCHL, the total mean annual costs associated with birth in periods with and without UNHS were £14 043 and £17 637, respectively (mean difference £3594, 95% CI −£2918 to £10 106, P=0.28) (table 5). The cost difference was mainly associated with placement of a higher percentage (95% CI) of those born in periods with UNHS in local mainstream schools, 61 (41% to 78%) compared with 39 (22% to 59%). Table 5 Total costs in preceding year in relation to presence of UNHS, confirmation by age 9 months, receptive language ability and reading

Effect of birth during periods with UNHS In children with PCHL, the total mean annual costs associated with birth in periods with and without UNHS were £14 043 and £17 637, respectively (mean difference £3594, 95% CI −£2918 to £10 106, P=0.28) (table 5). The cost difference was mainly associated with placement of a higher percentage (95% CI) of those born in periods with UNHS in local mainstream schools, 61 (41% to 78%) compared with 39 (22% to 59%). Table 5 Total costs in preceding year in relation to presence of UNHS, confirmation by age 9 months, receptive language ability and reading Variable  Unadjusted costs  Costs adjusted for cochlear implantation and presence of an additional medical condition  Costs adjusted for cochlear implantation, presence of an additional medical condition and severity of PCHL β coefficient (95% CI)  P  β coefficient (95% CI)  P  β coefficient (95% CI)  P Born during periods with UNHS −3594.21 (−10 105.9 to 2917.6) 0.28 −970.46 (−6099.42 to 4158.50) 0.71 −228.35 (−5303.44 to 4846.74) 0.93 Confirmation of PCHL at >9 months −2824.11 (−9381.5 to 3733.3) 0.39 −408.01 (−5662.35 to 4846.32) 0.88 47.99 (−5098.08 to 5194.06) 0.98 Receptive language z-score −1615.84 (−2389.3 to −842.4) <0.001 −1011.33 (−2086.17 to 63.50) 0.065 −649.90 (−1745.77 to 445.97) 0.24 Reading comprehension z-score −1886.94 (−3797.4 to 23.5) 0.05 −669.91 (−2516.30 to 1176.48) 0.47 −328.15 (−2478.69 to 1822.39) 0.76 PCHL, permanent childhood hearing loss > 40 dB hearing level (HL); UNHS, universal newborn hearing screening.

4) <0.001 −1011.33 (−2086.17 to 63.50) 0.065 −649.90 (−1745.77 to 445.97) 0.24 Reading comprehension z-score −1886.94 (−3797.4 to 23.5) 0.05 −669.91 (−2516.30 to 1176.48) 0.47 −328.15 (−2478.69 to 1822.39) 0.76 PCHL, permanent childhood hearing loss > 40 dB hearing level (HL); UNHS, universal newborn hearing screening. Effect of early confirmation Early confirmation of PCHL, like birth in periods with UNHS, was not associated with a significant difference in cost (cost difference £2824, 95% CI £3733 to £9382, P=0.39) (table 5). The cost difference remained non-significant when only participants without an AMC (n=57) were included in the analysis (difference £1487, 95% CI −£5164 to £8138). As there is an association between early confirmation of PCHL and greater severity of PCHL, it is necessary to adjust for severity in multivariate analysis. In that analysis (right hand model in table 5) no effect of early confirmation of PCHL on costs is apparent.

e analysis (difference £1487, 95% CI −£5164 to £8138). As there is an association between early confirmation of PCHL and greater severity of PCHL, it is necessary to adjust for severity in multivariate analysis. In that analysis (right hand model in table 5) no effect of early confirmation of PCHL on costs is apparent. Effects of language and reading z-score Each unit increase in receptive language ability z-score was associated with significantly lower annual costs by £1616 (95% CI £842 to £2389) (P<0.001). Similarly, each unit increase in reading ability z-score was associated with marginally significantly lower annual costs by £1887 (95% CI −£1234 to £3516, P=0.053) (table 5). Both of these effects fell short of statistical significance in multivariate analysis, although the effect of receptive language score on costs remained marginally significant (0.05<P<0.1) after adjusting for the effects of cochlear implantation and AMC on costs but not when additional adjustment for severity of PCHL was added into the model (table 5). This is considered further in the Discussion section. Discussion Compared with participants of similar age with normal hearing, the presence of bilateral PCHL > 40 dB at ages 13–20 years was associated with 2.7-fold higher costs in the preceding 12-month period and the presence of prespecified medical conditions in addition to PCHL increased costs almost twofold. No statistically significant association was found between either birth in periods with UNHS or early confirmation of PCHL and costs in adolescence.

ociated with 2.7-fold higher costs in the preceding 12-month period and the presence of prespecified medical conditions in addition to PCHL increased costs almost twofold. No statistically significant association was found between either birth in periods with UNHS or early confirmation of PCHL and costs in adolescence. In participants with PCHL, superior language skills were associated with lower costs and this remained marginally significant (0.05<P<0.1) after adjustment for the presence of a cochlear implant or an AMC but not after adjustment for severity of PCHL. Superior receptive language scores were associated with significantly lower costs in the same birth cohort when assessed 9 years earlier at ages 5–10 years.25 The costs associated with superior language at ages 5–10 years remained significantly lower after adjustment for severity of PCHL (ref 25, table 5) but not in the present study at ages 13–20 years. However, the latter regression model may represent overadjustment if the effect of severity of PCHL on cost is mediated by the well-recognised inverse relationship between severity of PCHL and receptive language,2–5 as seems particularly likely in the case of educational costs. In that earlier report, the reduction in cost associated with a unit increase in the z-score for receptive language was equivalent to 28.6% of total excess group mean annual costs associated with PCHL, whereas in the present study it was shown that the equivalent figure had fallen to 16.1% of those costs in adolescence. The absence of significant difference in the current evaluation may be due to this lower contribution of receptive language to total excess annual cost in adolescence compared with children 5–10 years old, but the total overall costs (from 0 to 20 years) may still be affected by receptive language score and further economic evaluations are required to assess this. Taken together, the previous and present evaluations in our study cohort do provide some support for the hypothesis that superior language scores are associated with lower costs in PCHL during childhood and adolescence. A full economic evaluation integrating all costs from birth to adolescence and the costs of screening would be required to better assess the cost-effectiveness of universal hearing screening.

pport for the hypothesis that superior language scores are associated with lower costs in PCHL during childhood and adolescence. A full economic evaluation integrating all costs from birth to adolescence and the costs of screening would be required to better assess the cost-effectiveness of universal hearing screening. The 7-year range of age at time of assessment is consequent on the study design, that is, an evaluation of a 5-year birth cohort conducted over a 3-year study period. This is both a limitation, in that it makes the estimate for any 1 year of age less precise, and a strength, in that it makes the findings more generalisable to teenagers in general. The two principal limitations of the study were the modest study size with 120 participants at the outset and further reduction in its power to look at subgroups (eg, severities of PCHL) by slow but steady loss of participants over the 17 years of follow-up, although without apparent attrition bias.27 28 This long period of follow-up provides data that are rare because of the extreme difficulty of obtaining them and also limits the generalisability of these findings to babies currently being born because newborns will now be offered paediatric audiology services that have adapted to UNHS in the 25 years since recruitment of newborns into our study began.

es data that are rare because of the extreme difficulty of obtaining them and also limits the generalisability of these findings to babies currently being born because newborns will now be offered paediatric audiology services that have adapted to UNHS in the 25 years since recruitment of newborns into our study began. The receptive language skills, which we found to be superior in children with PCHL born in periods with UNHS in our study population,13 should receive greater benefit in current and future birth cohorts than those observed in our study cohort because of the much clearer care pathways that now lead from UNHS to early intervention for PCHL. We therefore predict that the lower costs associated with superior language skills that we observed in this study population in childhood and adolescence will be more strongly associated with UNHS in a current or future birth cohort. In other words, management strategies made possible by UNHS could have the potential to lead to significantly reduced future costs as a result of superior language skills. Reports of societal costs in more recent and larger birth cohorts exposed to UNHS, such as those reported in the Netherlands8 and Australia,38 are therefore awaited to confirm and extend our observations. Future research should, in addition, consider extracting resource utilisation from large national databases as a cost-effective approach to economic evaluation of UNHS.

irth cohorts exposed to UNHS, such as those reported in the Netherlands8 and Australia,38 are therefore awaited to confirm and extend our observations. Future research should, in addition, consider extracting resource utilisation from large national databases as a cost-effective approach to economic evaluation of UNHS. The messages for policymakers of the associations observed, both at 5–10 years and in adolescence, in the birth cohort reported here include confirmation of the association between PCHL and significantly increased cost to society and the suggestion that interventions that improve language skills may bring benefit to the individual with PCHL and may be seen as a financial investment that should bring longer term cost savings through reductions in educational spending. These findings need confirmation in other larger birth cohorts.

ost to society and the suggestion that interventions that improve language skills may bring benefit to the individual with PCHL and may be seen as a financial investment that should bring longer term cost savings through reductions in educational spending. These findings need confirmation in other larger birth cohorts. We thank the research assistants Eleanore Coulthard, Joanne Pickersgill, Lisa Shipway and Zahra Taghizadeh; the other members of the HOT study steering group who had sight of the first draft of the manuscript (Jim Stevenson, Jana Kreppner, Emmanouela Terletski and HM Yuen); the audiologists Margaret Baldwin, Alyson Bumby, Adrian Dighe, Harpreet Nijar, David Reed, Joy Roberts, Sue Robinson, Salim Suleman, Rosbin Syed and Huw Thomas; and the other medical and educational professionals who supported this study. We thank particularly the participating teenagers and their families. We thank Jo Lord at the Southampton Health Technology Assessment Centre for commenting on an earlier draft of the manuscript. No remuneration was offered for these acknowledgements.

ther medical and educational professionals who supported this study. We thank particularly the participating teenagers and their families. We thank Jo Lord at the Southampton Health Technology Assessment Centre for commenting on an earlier draft of the manuscript. No remuneration was offered for these acknowledgements. Contributors: MC drafted the manuscript, conducted the economic analysis and approved the final manuscript. HP oversaw the conduct and analysis of the study. MM and SW assisted in the design and supervision of the study, assisted with manuscript preparation and approved the final manuscript. CRK designed and supervised the study, assisted in manuscript preparation and approved the final manuscript. MC and CRK had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Funding: This work was funded by The Wellcome Trust (grant number: 089251/Z/09/Z). Competing interests: None declared. Patient consent: Parental/guardian consent obtained. Ethics approval: Southampton and South West Hampshire Research Ethics Committee. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Unpublished data from the study are available upon reasonable request to the corresponding author.

What is already known on this topic? There is growing awareness that psychosocial risk and resilience factors in early life play a key role in influencing later health. Most work has been done in high-income settings, rather than low-income and middle-income countries (LMICs), where the majority of the global childhood population resides. The few studies with well-defined cohorts in LMICs have employed various methods and measures, making comparisons across studies challenging. What this study hopes to add? The Drakenstein Child Health Study aims to provide an understanding of the effects of multiple risk and mitigating factors on child health and development in a LMIC. Introduction Risk and resilience factors encountered during the early years of life have an enduring influence on later physiological and psychological outcomes.1–3 A number of risk factors are already apparent in utero; for example, antepartum maternal psychological distress and depression can adversely affect infant physical, neurocognitive and socioemotional developmental outcomes.4–6 During early childhood, exposure to stressors such as familial violence and abuse has been associated with increased risk of behaviour problems, autoimmune disorders, cardiovascular disease and premature mortality.7–10 In low-income and middle-income countries (LMICs), key risk factors such as HIV infection and prenatal maternal malnutrition are responsible for millions of children failing to reach their full developmental potential.11–13 Poor child outcomes may have intergenerational effects, so exacerbating their impact.12

ity.7–10 In low-income and middle-income countries (LMICs), key risk factors such as HIV infection and prenatal maternal malnutrition are responsible for millions of children failing to reach their full developmental potential.11–13 Poor child outcomes may have intergenerational effects, so exacerbating their impact.12 At the same time, protective factors may be associated with increased resilience, and so with positive mental health and developmental outcomes in the face of stressors.14 Resilience is thought to arise from the interplay between factors at the individual, family and community levels.15 Protective factors can be highly context specific and can exert different effects at different time points.15 16 Thus, longitudinal studies provide the best opportunities to identify protective factors at various stages of development, as well as sensitive periods for intervention. However, most studies on resilience have been done in high-income countries, where contextual factors may be different.

at different time points.15 16 Thus, longitudinal studies provide the best opportunities to identify protective factors at various stages of development, as well as sensitive periods for intervention. However, most studies on resilience have been done in high-income countries, where contextual factors may be different. Indeed, the vast majority of previously reported studies have focused on psychobiological and psychosocial risk profiles in well-resourced countries. These profiles differ considerably in LMICs. For instance, there is, in general, considerably higher prevalence of low birth weight, childhood malnutrition and infectious diseases in LMICs.17 18 In addition, critical psychosocial factors that are known to have impact on child development such as maternal depression and exposure to violence frequently have a greater prevalence in these high-risk communities.12 13 19 There is considerable work in LMICs, including work that is longitudinal and that is culturally appropriate. However, various methods and measures have been used, so that cross-cohort comparison is not always possible.

and exposure to violence frequently have a greater prevalence in these high-risk communities.12 13 19 There is considerable work in LMICs, including work that is longitudinal and that is culturally appropriate. However, various methods and measures have been used, so that cross-cohort comparison is not always possible. The Drakenstein Child Health Study (DCHS) is a multidisciplinary longitudinal study investigating the early-life determinants of child health in two periurban communities in the Western Cape Province, South Africa.20 The early-life component focuses on a broad spectrum of child health outcomes, including physical health and growth as well as neurodevelopmental, cognitive and psychological health. The study investigates the role and interaction of environmental, infectious, psychosocial, nutritional, genetic, maternal and immunological risk and protective factors for development. The DCHS follows an extensively phenotyped cohort over critical early years, aiming to provide an understanding into the effects of multiple risk and mitigating factors, and their interactions, on child health and development in a LMIC. This paper describes the methodology for the infant and child measures of the psychosocial component of the DCHS. By documenting the measures used and our reasons for choosing these, we hope to improve future harmonisation between cohort studies.

ctors, and their interactions, on child health and development in a LMIC. This paper describes the methodology for the infant and child measures of the psychosocial component of the DCHS. By documenting the measures used and our reasons for choosing these, we hope to improve future harmonisation between cohort studies. Methods and analysis Design and setting The DCHS is located in the periurban Drakenstein district, Western Cape, South Africa. The communities surrounding the Mbekweni and TC Newman clinics are ethnically, culturally and linguistically heterogeneous. Mbekweni consists of a mostly isiXhosa-speaking black African population, whereas TC Newman consists of a mostly Afrikaans-speaking mixed race population.20 However, both communities are characterised by low socioeconomic status and feature a high prevalence of multiple psychosocial risk factors, including single-parent households, high rates of psychological distress and exposure to violence, HIV and substance abuse.21 In particular, risk factors that may be highly prevalent in these communities and similar communities in the region include high rates of HIV exposure,20 22 23 high prevalence of maternal psychological distress and depression,5 21 24–26 high rates of drug and alcohol usage,21 24 27 28 high levels of violence and intimate partner violence29 and low levels of employment and educational attainment.21 The population is stable, with little immigration or emigration. More than 90% of people in the district use the public health system. In this regard, the communities are representative of many other communities in South Africa and other LMICs.

rtner violence29 and low levels of employment and educational attainment.21 The population is stable, with little immigration or emigration. More than 90% of people in the district use the public health system. In this regard, the communities are representative of many other communities in South Africa and other LMICs. Participants Pregnant women were recruited while attending routine antenatal care at Mbekweni or TC Newman clinic between March 2012 and March 2015. Women were enrolled in the DCHS at 20–28 weeks’ gestation and were followed through birth and postnatally. Pregnant mothers were eligible for the study if they were 18 years or older, planned to attend antenatal care at one of the two clinics and intended to remain in the area for at least a year. Expecting mothers provided informed written consent at enrolment and were reconsented annually following childbirth.

ostnatally. Pregnant mothers were eligible for the study if they were 18 years or older, planned to attend antenatal care at one of the two clinics and intended to remain in the area for at least a year. Expecting mothers provided informed written consent at enrolment and were reconsented annually following childbirth. Mothers were provided informed consent in their preferred language, English, Afrikaans or isiXhosa, by trained study staff from the community. Informed consent forms described the scope and aims of the study, including potential harm or benefits. In total, 1137 mother–child dyads were enrolled in the study, of which four mothers had twins and one had triplets. Thus, 1143 children were enrolled in the study. A conservative cumulative attrition of 20% over 5 years was estimated, and the sample size was calculated accordingly. Enrolment criteria was broad to ensure generalisability and that the cohort would be representative of the general population. Majority of Drakenstein subdistrict births occur at Paarl Hospital, with an average of 4800 births per year; the study enrolled approximately 10% of catchment area births. Of pregnant women in the catchment area, who were provided information relating to study enrolment, 1471 mothers were determined to be ineligible (based on age, gestation or non-attendance at study clinics). A further 674 mothers were eligible but were not interested in study enrolment. Where data were available for the Cape Winelands district, the study population has comparable levels of education, partnership status and household size. Based on these sociodemographic variables and the broad enrolment criteria used, we believe that the study population is representative of the source population.

nt. Where data were available for the Cape Winelands district, the study population has comparable levels of education, partnership status and household size. Based on these sociodemographic variables and the broad enrolment criteria used, we believe that the study population is representative of the source population. Measures Mothers were followed during pregnancy and childbirth. Following birth, infants and mothers returned to the clinics and were asked to complete self-report and clinician-administered measures at time points ranging from 2 weeks to 4 weeks to 5.5 years (see figure 1). At the time of submission, the oldest children in the cohort are 5 years old and the youngest children are currently 2 years old. Figure 1 Time line for child assessment. The infant and child developmental and psychosocial measures are described here. The overview methods of the larger DCHS are described elsewhere,20 as is the maternal and paternal psychosocial evaluation component of the study.21 Broadly, the child measures assessed (1) social and biological risk and protective factors, (2) general neurobiologicalneurobiological (3) cognitive development and (4) socioemotional risk and protective factors for child health outcomes. Measures were translated from English to isiXhosa and using the standard forward and backwards-translation method.30 Assessments were conducted at community centres located near the two clinics.

ogicalneurobiological (3) cognitive development and (4) socioemotional risk and protective factors for child health outcomes. Measures were translated from English to isiXhosa and using the standard forward and backwards-translation method.30 Assessments were conducted at community centres located near the two clinics. The tests used are detailed in table 1 with detailed description in online supplementary appendix 1. The measures of social and biological risk and protective factors (eg, exposure toviolence, alcohol and tobacco use, parenting practices, and attachment) were completed by the mothers during several antenatal and postnatal study visits. Additionally, dyadic interactions between the mother and child were recorded. Basic demographic data and health information for the infants were obtained from participants’ hospital records. The child’s cognitive and general development was assessed directly at several different time points. Early child development was assessed by trained physiotherapists and occupational therapists at the clinics supervised by a paediatric neurodevelopmental specialist. Cognitive assessments, which include language, fine motor, executive functioning, memory and general and social cognitive ability, were administered by trained research assistants, with the aid of trained interpreters when necessary. Child socio-emotional development was captured through a combination of observational, self-report and parent-report measures (see online supplementary appendix 1). The socio-emotional assessment included measures of emotion regulation, affective arousal and social attention-allocation, empathy, morality, prosocial behaviour, temperament and callous–unemotional traits. Additionally, a subgroup of children in the DCHS cohort underwent multimodal neuroimaging assessment at the Cape Universities Body Imaging Centre (CUBIC). The imaging modalities performed included: (1) structural MRI with T1-weighting and T2-weighting to examine cortical and subcortical volumes; (2) diffusion tensor imaging for white matter microstructure; magneticresonance spectroscopy; and (4) resting state functional MRI for regional connectivity.

ging Centre (CUBIC). The imaging modalities performed included: (1) structural MRI with T1-weighting and T2-weighting to examine cortical and subcortical volumes; (2) diffusion tensor imaging for white matter microstructure; magneticresonance spectroscopy; and (4) resting state functional MRI for regional connectivity. 10.1136/bmjpo-2018-000282.supp1Supplementary file 1 Table 1 Child psychosocial and neurodevelopmental measures

ging Centre (CUBIC). The imaging modalities performed included: (1) structural MRI with T1-weighting and T2-weighting to examine cortical and subcortical volumes; (2) diffusion tensor imaging for white matter microstructure; magneticresonance spectroscopy; and (4) resting state functional MRI for regional connectivity. 10.1136/bmjpo-2018-000282.supp1Supplementary file 1 Table 1 Child psychosocial and neurodevelopmental measures Domain Measure Demographic data Household income and assets, maternal education and employment. Infant health information Weight, height and diagnoses. Infant neurodevelopment MRI, DTI, MRS and fMRI. AnfMRItal risk factors Birth planning; partner support; maternal depression; alcohol, nicotine and illicit substance use; maternal childhood trauma; intimate partner violence; exposure to stressful events; psychological distress; and symptoms of peritraumatic and post-traumatic stress. Trauma and exposure Emotional distress Paediatric emotional distress scale. Exposure to violence Child exposure to community violence. Survey for exposure to community violence. Parenting Dyadic interaction* Global rating scale. Emotional availability scale. Parenting practices Parenting and family adjustment scale. Attachment Brockington postpartum bonding questionnaire. Resilience Child and youth resilience measure. General development Western Cape screen. Bayley Scales of Infant and Toddler Development, Third Edition (Bayley-III), Third Edition (Bayley-III). General cognitive function Kaufman Assessment Battery for Children (KABC-II). Problem solving KABC-II Conceptual thinking. Visual-spatial memory/processing KABC-II Face recognition. Visual-spatial processing and problem solving KABC-II Triangles. Working memory and motor sequencing KABC-II Hand movements. Language Peabody Picture Vocabulary Test, Fourth Edition. KABC-II Expressive vocabulary. Memory Learning KABC-II Atlantis. Delayed recall KABC-II Atlantis delayed. Executive function Working memory Wechsler preschool and primary scale of intelligence, Fourth Edition: picture memory Inhibition Day–night task. Cognitive flexibility Dimension change card sort. Attention Test of Everyday Attention: sky search. Motor control Bayley-III: fine motor. Grooved peg board. Social cognition Theory of mind Diverse desires. Diverse beliefs. Understanding intentions. Perception-leads-to-knowledge. Location-change false belief. Unexpected contents false belief. Belief emotion. Hidden emotions. Emotion recognition A Developmental NEuroPSYchological Assessment (NEPSY-II): affect recognition. Effortful control/emotion regulation Snack delay.

. Diverse beliefs. Understanding intentions. Perception-leads-to-knowledge. Location-change false belief. Unexpected contents false belief. Belief emotion. Hidden emotions. Emotion recognition A Developmental NEuroPSYchological Assessment (NEPSY-II): affect recognition. Effortful control/emotion regulation Snack delay. Gift-in-bag. Whisper. Rydell’s emotion questionnaire. Social competence Affective arousal Pupil dilation. Attention allocation Gaze fixation. Empathy Chicago empathy for pain task. Questionnaire of cognitive and affective empathy. Prosocial behaviour Dictator game. Internalising and externalising behaviour Child behaviour checklist. Temperament/personality Temperament Rothbart Infant and Early Child Behavior Questionnaire.* Callous-unemotional traits Callous unemotional screening device. References for the measures are given in the appendix. *Very short forms. DTI, diffusion tensor imaging; fMRI, functional MRI; MRS, magnetic resonance spectroscopy. Descriptive statistics We use descriptive statistics (medians, IQRs, counts and percentages) to present the sociodemographicstatistics on sites TC Newman and Mbekweni. χ2 and Mann-Whitney tests were used to test for differences between the sites. As the study is still ongoing, the data presented here was collected antenatally and at birth.

s We use descriptive statistics (medians, IQRs, counts and percentages) to present the sociodemographicstatistics on sites TC Newman and Mbekweni. χ2 and Mann-Whitney tests were used to test for differences between the sites. As the study is still ongoing, the data presented here was collected antenatally and at birth. Child sociodemographics The cohort followed 1137 mothers over the course of 3 years during the initial recruitment period. During this time, there was a total of 1143 live births including four sets of twins and one set of triplets (see table 2). Within this group, 17.2% of children were born preterm (definepretermas less than 37 weeks’ gestation). The birth weight and lengths of the children have been converted to z-scores according to WHO standardisation and adjusted for gestation. In line with our previous reports,31 32 the children at TC Newman clinic were born significantly smaller in weight compared with Mbekweni children; however, the IQR falls within two z-scores of 0 in both clinics. Table 2 Child sociodemographisociodemographics Variable Mbekweni TC Newman Total P values Number (% of each clinic) Mothers 628 509 1137 Live births 634 509 1143 Twin sets 4 0 4 Triplet sets 1 0 1 Preterm births 107 (16.88) 83 (16.31) 190 (16.32) 0.797 Child’s race Black 627 (99.05) 6 (0.95) 633 (55.38) <0.001 Mixed 7 (1.11) 503 (98.82) 510 (44.62) Median (IQR) Birth weight z-score* −0.4 (−1.2–0.2) −0.7 (−1.4 to −0.1) −0.5 (−1.3–0.1) <0.001 Birth length z-score* 0.1 (−0.9–1.0) −0.03 (−0.9–0.8) 0.003 (−0.9–0.9) 0.151 *Adjusted for gestation.

(16.31) 190 (16.32) 0.797 Child’s race Black 627 (99.05) 6 (0.95) 633 (55.38) <0.001 Mixed 7 (1.11) 503 (98.82) 510 (44.62) Median (IQR) Birth weight z-score* −0.4 (−1.2–0.2) −0.7 (−1.4 to −0.1) −0.5 (−1.3–0.1) <0.001 Birth length z-score* 0.1 (−0.9–1.0) −0.03 (−0.9–0.8) 0.003 (−0.9–0.9) 0.151 *Adjusted for gestation. Maternal and family sociodemographics The maternal and family characteristics are presented in table 3. Socioeconomic status was low across both clinics as shown by the low household income levels, maternal employment rates of 27% and maternal educational attainment. The median maternal age at enrolment was 26 years (IQR 22–31), and the majority of mothers had completed some secondary schooling by this point. The mothers reported that over 65% of pregnancies were unplanned. Approximately 40% were currently married or cohabiting with their partner, and a high proportion reported that partners were supportive, although this differed across clinics. Table 3 Maternal and family sociodemographisociodemographics

Maternal and family sociodemographics The maternal and family characteristics are presented in table 3. Socioeconomic status was low across both clinics as shown by the low household income levels, maternal employment rates of 27% and maternal educational attainment. The median maternal age at enrolment was 26 years (IQR 22–31), and the majority of mothers had completed some secondary schooling by this point. The mothers reported that over 65% of pregnancies were unplanned. Approximately 40% were currently married or cohabiting with their partner, and a high proportion reported that partners were supportive, although this differed across clinics. Table 3 Maternal and family sociodemographisociodemographics Variable Cape Winelands* Mbekweni TC Newman Total P values n=787 490 Number (% of each clinic) Maternal educational attainment 0.038 Primary 49 (7.8) 37 (7.3) 86 (7.6) Some secondary 340 (54.1) 266 (52.3) 606 (53.3) Completed secondary 193 723 (24.6) 189 (30.1) 183 (36.0) 372 (32.7) Any tertiary 82 686 (10.5) 50 (8.0) 23 (4.5) 73 (6.4) Currently employed 675 666 (85.8) 157 (25.0) 149 (29.3) 306 (27.0) 0.106 Married/cohabiting 282 709 (35.9) 237 (37.8) 221 (43.4) 458 (40.3) 0.054 Unplanned pregnancy 366 (68.0) 286 (62.7) 652 (65.6) 0.079 Partner support <0.001 Not at all/slightly supportive 42 (7.8) 46 (10.13) 88 (8.9) Moderately supportive 86 (16.0) 19 (4.2) 105 (10.6) Considerably/extremely supportive 409 (76.1) 389 (85.7) 798 (80.5) Reliance on partner for help <0.001 Not at all/not very often 53 (9.9) 46 (10.1) 99 (10.0) Some of the time 170 (31.7) 28 (6.2) 198 (20.0) Most/all of the time 314 (58.5) 380 (83.7) 694 (70.0) Monthly income 119 536/year (average) <0.001 <R1000 (<$75) 263 (41.9) 167 (32.8) 430 (37.8) R1000–R5000 ($75–374) 299 (47.6) 254 (49.9) 553 (48.6) >R5000 (>$374) 66 (10.5) 88 (17.3) 152 (13.5) Median (IQR) Household size 3.7 persons 4 (3–6) 5 (4–7) 4 (3–6) <0.001 Mother’s age at enrolment 27 (22–32) 25 (21–29) 26 (22–31) <0.001 Census 2011 Municipal Report Western Cape. Statistics South Africa. Report no. 03-01-49. http://www.statssa.gov.za.

254 (49.9) 553 (48.6) >R5000 (>$374) 66 (10.5) 88 (17.3) 152 (13.5) Median (IQR) Household size 3.7 persons 4 (3–6) 5 (4–7) 4 (3–6) <0.001 Mother’s age at enrolment 27 (22–32) 25 (21–29) 26 (22–31) <0.001 Census 2011 Municipal Report Western Cape. Statistics South Africa. Report no. 03-01-49. http://www.statssa.gov.za. Mothers and children in both communities were frequently exposed to community violence (see table 4). Levels of violence exposure were higher in TC Newman than in Mbekweni, but both communities from this cohort reported greater exposure to violence than reported in a previous study in South Africa.33 Exposure to violence is thus highly prevalent in these communities both within the home and external community environment. Both communities reported less consistent parenting (ie, higher consietency scores) but otherwise similar parenting styles and family adjustment, compared with European samples.34 Mothers in Mbekweni reported more coercive parenting, less consistency and less encouragement, but also better family relationships and parental teamwork than in TC Newman. Table 4 Family risk and protective factors

Mothers and children in both communities were frequently exposed to community violence (see table 4). Levels of violence exposure were higher in TC Newman than in Mbekweni, but both communities from this cohort reported greater exposure to violence than reported in a previous study in South Africa.33 Exposure to violence is thus highly prevalent in these communities both within the home and external community environment. Both communities reported less consistent parenting (ie, higher consietency scores) but otherwise similar parenting styles and family adjustment, compared with European samples.34 Mothers in Mbekweni reported more coercive parenting, less consistency and less encouragement, but also better family relationships and parental teamwork than in TC Newman. Table 4 Family risk and protective factors Variable Mbekweni TC Newman Total P values Mean (SD) PAFAS Parenting Consistency 7.95 (2.01) 6.24 (2.92) 7.17 (2.61) <0.001 Coercive parenting 6.24 (3.38) 5.19 (3.98) 5.76 (3.70) <0.001 Encouragement 1.23 (1.66) 1.00 (1.50) 1.13 (1.59) 0.113 Parent–child relationship 0.63 (1.48) 1.37 (1.76) 0.97 (1.65) <0.001 PAFAS Family Adjustment Parental adjustment 2.72 (2.67) 2.36 (2.86) 2.56 (2.76) 0.025 Family relationships 1.46 (2.01) 2.45 (2.85) 1.91 (2.48) <0.001 Parental teamwork 1.32 (1.67) 1.93 (2.08) 1.59 (1.89) <0.001 SECV Total 21.27 (6.57) 26.92 (7.46) 23.84 (7.53) <0.001 CECV Total 38.65 (3.88) 40.13 (4.21) 38.65 (3.88) <0.001 Note: the questionnaires were administered at 2.5 years. Higher scores on the PAFAS indicate higher levels of dysfunction, that is, higher consistency scores indicate less consistent parenting and higher coercive parenting scores indicate more coercive parenting.

CECV Total 38.65 (3.88) 40.13 (4.21) 38.65 (3.88) <0.001 Note: the questionnaires were administered at 2.5 years. Higher scores on the PAFAS indicate higher levels of dysfunction, that is, higher consistency scores indicate less consistent parenting and higher coercive parenting scores indicate more coercive parenting. CECV, Child Exposure to Community Violence Checklist; PAFAS, Parenting and Family Adjustment Scale; SECV, Survey for Exposure to Community Violence. Discussion This paper highlights the rationale and approach to assessing both psychosocial risk and protective factors impacting the development of children in a high-risk South African communities. The study follows a multilevel approach that targets developmental, cognitive, socioemotional socioemotionalcal outcomes. As can be seen from some of our baseline data (n = 1143), there are a broad number of sociodemographic risk and resilience factors for children in this region. Demographic and sociodemographic data show that although these communities are in close proximity, they differ substantively in social and financial resources.

As can be seen from some of our baseline data (n = 1143), there are a broad number of sociodemographic risk and resilience factors for children in this region. Demographic and sociodemographic data show that although these communities are in close proximity, they differ substantively in social and financial resources. Given these sorts of risk and resilience factors, it is important to assess child outcomes using a multidimensional approach.35 This includes three important components that are built into this cohort. First, the DCHS collects biomedical and psychosocial risk factors across a wide range of factors in the prenatal period and first years of life. These include both factors that are known to put children at risk for poor outcome such as maternal mental health, substance use disorders, poor nutrition and factors known to be protective or hypothesised to potentially support development including early infant feeding choices, pregnancy support and maternal bonding and attachment styles. Second, the outcome measures described in this manuscript are also multidimensional and allow examination of outcomes in terms of the dyadic relationship and the family system into which these children are born. Third, the cohort is following these mother infant pairs over time. Longitudinal data (with repeat measures) allows the investigation of developmental, cognitive and socioemotional trajectories as well as the interactions between exposures within the context of this cohort. The investigation of the timing of maximum windows of vulnerability becomes possible with this approach.

er time. Longitudinal data (with repeat measures) allows the investigation of developmental, cognitive and socioemotional trajectories as well as the interactions between exposures within the context of this cohort. The investigation of the timing of maximum windows of vulnerability becomes possible with this approach. Attention to the ethical issues requires consideration in a study of this type. The DCHS maintains an active programme focused on community engagement, including regular engagement with study participants for feedback on study involvement, active dissemination of research results to key local stakeholders and distribution of health promotion information to study participants. Given the context, a key ethical obligation includes screening within the study population for physical and mental health issues, in both mothers and children. Screening is done in the DCHS in conjunction with an active referral system and is bolstered by close relationships between study staff and provincial health staff. All women involved in the study, independent of identified mental or physical health problems, receive information regarding service providers in the area. The network of investigators in the DCHS with strong and relevant clinical background in the South African public health environment is a strength of the research and has allowed realistic and integrated referral systems to be developed and implemented as part of the study.

tion regarding service providers in the area. The network of investigators in the DCHS with strong and relevant clinical background in the South African public health environment is a strength of the research and has allowed realistic and integrated referral systems to be developed and implemented as part of the study. Very few cohorts are reported that take into account the composite effects of multiple factors on health and development in the early years. This is especially true of cohorts from LMICs where young children are exposed to overlapping epidemics of infectious and non-communicable diseases. The P-MaMie birth cohort in Ethiopia36 collected information on maternal mental health as well as growth and developmental outcomes in very young infants but represents an almost entirely rural community in Africa with attendant risk factors likely to vary from the periurban community described in the DCHS. With urbanisation representing a critical current epidemiological phenomenon, documenting communities in this context becomes increasingly important. The Pelotas cohorts in Brazil37 represent one of the longest standing set of population-based cohorts in the global south. These are large cohorts capturing whole communities have been documenting maternal and child outcomes for over 20 years. The most recent of these cohorts starting in 2015 is the first of these to collect prospective antenatal data on the mothers. Though smaller, the DCHS, at present, has been able to collect the most comprehensive set of biological, psychosocial, environmental, maternal and child data and so carefully measures outcomes through use of sensitive modalities including brain imaging in a high-risk setting. Sensitivity to exposures, individually and together, in both external and internal environments, are different at specific ages and periods in development. The developmental window spanning the 40 weeks of pregnancy and the first years of life appear to be a critical period where environmental exposure may cause embedded effects that may have impact across the lifespan. There is a critical need for research in this field to elucidate the underlying factors that inform risk for and resilience from poor developmental outcomes in infants born into high risk communities that may ultimately inform effective preventative and ameliorating intervention strategies appropriate to this context.

There is a critical need for research in this field to elucidate the underlying factors that inform risk for and resilience from poor developmental outcomes in infants born into high risk communities that may ultimately inform effective preventative and ameliorating intervention strategies appropriate to this context. From a public health perspective, a better understanding of the relevant mechanisms is critical, as this may ultimately drive preventative and targeted therapeutic approaches. The United Nation’s Sustainable Development Goals (SDG) were officially adopted in 2015. These cross-cutting SDGs consistently forefront the importance of programmes targeting maternal and child health (in particular, the theme of early child growth and development), as being key in the global strategy to optimise human health and well-being across the lifespan. In the South African setting, where children make up around a third of the population, expectant mothers and their young infants are a particularly important focus. Much more attention is needed to address maternal and infant health, in order to decrease early mortality and later morbidity in this vulnerable population.

n the South African setting, where children make up around a third of the population, expectant mothers and their young infants are a particularly important focus. Much more attention is needed to address maternal and infant health, in order to decrease early mortality and later morbidity in this vulnerable population. Limitations Limitations of the study include the fact that though extremely well characterised, the cohort is a modest size. Though measures investigating aspects of child health, development and cognition have been administered to as much of the cohort as possible, certain measures have been administered only on a subset of the group (eg, neuroimagineg dneuroimagingipant burden and the cost of assessment. Although care was taken to translate measures into Afrikaans and isiXhosa, there willalways be some difficulty in interpreting the results of measures designed in a different language and cultural context. Every effort was made to use tools that minimised problems in this area. Supplementary Material Reviewer comments Author's manuscript We would like to thank the study staff in Paarl, the study data team and lab teams, the clinical and administrative staff of the Western Cape Government Health Department at Paarl Hospital and at the clinics for support of the study. We acknowledge the advice from members of the study International Advisory Board and our collaborators. We would like to thank the families and children who participated in this study.

nd administrative staff of the Western Cape Government Health Department at Paarl Hospital and at the clinics for support of the study. We acknowledge the advice from members of the study International Advisory Board and our collaborators. We would like to thank the families and children who participated in this study. Contributors: This study requires multidisciplinary expertise in the execution of measures of this type. DJS is PI of the psychosocial aspects of the Drakenstein Child Health Study (DCHS) cohort and contributed to the design and decision making involving psychosocial tools and measures used as well as general study design. HJZ is PI of the umbrella DCHS cohort and played a central role in the operational aspects and design of the study. KAD, PI of the child psychosocial aspects of the study, was involved in the design of the study and operational aspects of the study and played a key role in the child psychosocial measures used. SM-S, MH and CPdP contributed to decision making involving tools used, training and operational aspects of the child assessments. NH, WB and CJW contributed to the operational aspects of the study, QC of data described and data management. RTN contributed to data management. Authors contributed to sections relating to their area of expertise in the manuscript. All authors reviewed and approved the final version of this manuscript.

ld assessments. NH, WB and CJW contributed to the operational aspects of the study, QC of data described and data management. RTN contributed to data management. Authors contributed to sections relating to their area of expertise in the manuscript. All authors reviewed and approved the final version of this manuscript. Funding: This study was funded by National Institute on Alcohol Abuse and Alcoholism (R21AA023887), US Brain and Behaviour Foundation (24467) and Bill and Melinda Gates Foundation (OPP 1017641), Collaborative Initiative on Fetal Alcohol Spectrum Disorders (U24 AA014811), Medical Research Council of South Africa, South African Medical Research Council, Wellcome Trust (203525/Z/16/Z), National Research Foundation, and Newton Fund (NAF002\1001). Competing interests: None declared. Patient consent: Parental/guardian consent obtained. Ethics approval: The study was approved by the faculty of Health Sciences, Human Research Ethics Committee, University of Cape Town (401/2009), Stellenbosch University (N12/02/0002) and the Western Cape Provincial Health Research committee (2011RP45). Provenance and peer review: Not commissioned; externally peer reviewed.

Patient consent: Parental/guardian consent obtained. Ethics approval: The study was approved by the faculty of Health Sciences, Human Research Ethics Committee, University of Cape Town (401/2009), Stellenbosch University (N12/02/0002) and the Western Cape Provincial Health Research committee (2011RP45). Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Collaborations for the analysis of data are welcome. The DCHS has a large and active group of investigators and postgraduate students, and many have successfully partnered with researchers from other institutions. In particular, we encourage collaborations that lead to skills transfer and capacity building for postgraduate students. Researchers who are interested in datasets or collaborations can contact the PI, Heather Zar (heather.zar@uct.ac.za) with a concept note outlining the request. More information can be found on our website (http://www.paediatrics.uct.ac.za/scah/dclhs).

What is already known on this topic? Reduced vision in young children is commonly related to refractive error, strabismus or amblyopia. Following identification of reduced vision at vision screening attendance for eye care appointments is poor with 30% of children failing to attend. Treatment consists of spectacle wear and may be combined with wearing an eye patch but adherence to treatment is inconsistent. What this study hopes to add? Reasons for non-attendance and non-adherence are complex with both pragmatic factors and health beliefs interlinked in parental decision making. The decision to adhere to spectacle wear was based primarily on the perceived severity of the visual reduction with the perceived benefit outweighing any negative consequences. The role of schools as facilitators should be considered when developing interventions to promote spectacle wear in young children.

What this study hopes to add? Reasons for non-attendance and non-adherence are complex with both pragmatic factors and health beliefs interlinked in parental decision making. The decision to adhere to spectacle wear was based primarily on the perceived severity of the visual reduction with the perceived benefit outweighing any negative consequences. The role of schools as facilitators should be considered when developing interventions to promote spectacle wear in young children. Introduction Decreased vision in young children is commonly due to the presence of refractive error, strabismus and/or amblyopia.1 The presence of one or more of these conditions can result in a visual deficit in the developing infant. Most young children with refractive error or amblyopia do not however demonstrate obvious signs or symptoms, making it difficult for parents to identify, and vision screening is recommended in the United Kingdom at age 4–5 years.2 Early detection programmes can only be effective if those identified with poor vision are appropriately treated. In deprived communities both locally3 and internationally4 attendance rates following vision screening are reported between 30% to 40%.3 In addition, studies have shown poor adherence to spectacle wear5–7; therefore, a high proportion of children with identified needs do not access ophthalmic services and subsequent treatment.

ved communities both locally3 and internationally4 attendance rates following vision screening are reported between 30% to 40%.3 In addition, studies have shown poor adherence to spectacle wear5–7; therefore, a high proportion of children with identified needs do not access ophthalmic services and subsequent treatment. Socioeconomic factors have been reported to influence access to children’s eye care8 and adherence with spectacle wear.9 In USA, difficulty navigating the healthcare system, cost and loss and breakage of spectacles has been found to prevent attendance and adherence leading to inequalities in visual health.10 Factors such as cultural sensitivity and health beliefs have been associated with differences in access, uptake and adherence decisions in adults.11 People’s perception of the risk of health problems also influences healthcare utilisation.12 Current literature therefore provides evidence of a number of factors impacting on access and uptake of children’s eye services. However, unlike adult ophthalmic services13 where patients’ experiences have been documented, there is little reported qualitative evidence that would provide an understanding of parents’ experiences of their child’s eye care. This study explores in-depth parents’ experiences and understanding of their children’s eye care in order to better comprehend why there is relatively low uptake of services and variable adherence to treatment.

d qualitative evidence that would provide an understanding of parents’ experiences of their child’s eye care. This study explores in-depth parents’ experiences and understanding of their children’s eye care in order to better comprehend why there is relatively low uptake of services and variable adherence to treatment. Methods Participant recruitment The population-based vision screening programme in Bradford, a large multiethnic city in England, is offered to children in their year of school entry, aged 4–5 years. Children failing to achieve the pass criterion set by the UK National Screening Committee2 (≤0.20 logarithm of the minimum angle of resolution (logMAR) in both eyes) are referred for examination, either to a community optometrist or the hospital eye service (HES) and subsequently followed up in order to measure the visual acuity (VA)  with prescribed spectacles. As part of a separate longitudinal research programme,14 children were followed up between 2013 and 2015 to explore the impact of adherence to spectacle wear on VA and early literacy. Parents of children participating in the study were invited for interview between November 2015 and May 2016. Parents were purposively selected if their child had failed vision screening in school years 2013–2014 or 2014–2015 and their child was either a non-attender (never attended follow-up appointments at their local optometrist or the HES) or their child was non-adherent (not wearing prescribed spectacles at unannounced visits in school and/or spectacle wear was not corroborated by documentation in the medical notes). For comparison purposes, parents of four children identified as adherent (wearing prescribed spectacles at unannounced visits in school) were also invited to participate.

(not wearing prescribed spectacles at unannounced visits in school and/or spectacle wear was not corroborated by documentation in the medical notes). For comparison purposes, parents of four children identified as adherent (wearing prescribed spectacles at unannounced visits in school) were also invited to participate. Initially, 40 letters with prepaid reply slips were sent inviting parents (mother or father) of children participating in the longitudinal study to participate. No responses were obtained with this method. The letters were then distributed to the parents via their child’s school. All interviews were undertaken in the school that their child was attending. Bradford is ranked the fifth highest deprived city in the UK; the five schools located across Bradford all had a Townsend score between 1 and 4 indicating lower socioeconomic status of the school communities. Four of the parents required interpreters (3 Urdu and 1 Parsi); the interpreters were teaching assistants working in the schools.

the fifth highest deprived city in the UK; the five schools located across Bradford all had a Townsend score between 1 and 4 indicating lower socioeconomic status of the school communities. Four of the parents required interpreters (3 Urdu and 1 Parsi); the interpreters were teaching assistants working in the schools. The study was informed by the Health Belief Model (HBM), a theoretical concept developed in the arena of public health (figure 1).15 It consists of six key constructs that together provide an explanatory framework for the adoption of preventative health behaviour.16 HBM was developed from the study of the factors influencing adults attendance at health screening programmes,15 so it was chosen as a starting point to inform the topic guide for this study (table 1). The topic guide was used to ensure details of the parents’ experiences and understanding of their child’s eye care was captured, in particular their views on vision and spectacle wear in children and the influence of their own and the experiences of family members (table 1). Figure 1 Illustration of Health Belief Model. Table 1 Topic guide (based on the Health Belief Model14)

The study was informed by the Health Belief Model (HBM), a theoretical concept developed in the arena of public health (figure 1).15 It consists of six key constructs that together provide an explanatory framework for the adoption of preventative health behaviour.16 HBM was developed from the study of the factors influencing adults attendance at health screening programmes,15 so it was chosen as a starting point to inform the topic guide for this study (table 1). The topic guide was used to ensure details of the parents’ experiences and understanding of their child’s eye care was captured, in particular their views on vision and spectacle wear in children and the influence of their own and the experiences of family members (table 1). Figure 1 Illustration of Health Belief Model. Table 1 Topic guide (based on the Health Belief Model14) Concept Definition Areas of questioning Perceived susceptibility An individual’s assessment of their chances of developing a condition. Experience of eye conditions encountered in children. Experience of eye tests and or treatment as a child/adult. Eye care advice received from parents/friends. Advice they would give their child. Perceived severity An individual’s opinion as to the seriousness of the condition. Importance of vision: worries, fears and attitudes. Attitudes to children wearing glasses. Attitudes to children wearing eye patches. Knowledge of impact of poor vision on employment, driving and other everyday activities. Perceived benefits An individual’s opinion as to whether a new behaviour is better than current behaviour. Importance of good vision. Impact of poor vision in children. Impact of vision on a child’s ability to learn. Perceived barriers An individual’s opinion as to what will prevent them from adopting a new behaviour. Costs and relative costs. Satisfaction with health services. Number of visits to optometrist or hospital. Obstacles to attending. Eye clinic opportunities. Cosmetic impact. Self-efficacy Belief in one’s own ability to perform an action. Capability of arranging appointments. Knowledge of children’s eye tests. Capability of performing the treatment (ensuring your child wears glasses or eye patch every day to improve their vision). Cues to action Factors that will prompt a person into changing behaviour. Family attitude to eye care. Attitude of school to eye care. Any associated cultural or religious practices in relation to children’s or adult eye care. Semistructured interviews were conducted by one member of the study team (AB), a female orthoptist and postdoctoral research fellow, who was unknown to the parents, in the child’s school at a time suitable for the parent, two mothers in one school arrived at the same time and a combined interview was undertaken. Both the mother and father of one child were interviewed together; the father was the main contributor; therefore, only his discourse is included in the analysis.

parents, in the child’s school at a time suitable for the parent, two mothers in one school arrived at the same time and a combined interview was undertaken. Both the mother and father of one child were interviewed together; the father was the main contributor; therefore, only his discourse is included in the analysis. Interviews were audio recorded and transcribed verbatim; in addition, the non-English transcripts were listened to by a bilingual research assistant to confirm accuracy of the interpretation. Two members of the study team (AB and TS) reviewed the transcripts independently and then together developed a thematic analysis, using constant comparison between the transcripts, identifying key themes and concepts.17 18 Our approach to analysis was based on the ‘constant comparative’ methodology derived from grounded theory. We analysed the interviews in search of similarities and differences, which led to the development of themes in the data. Thematic analysis was chosen to allow for the emergence of opinions that were not anticipated in advance. No member checking was performed with this hard-to-recruit group of parents; however, a focus group with a small number of mothers confirmed the interpretation of the findings. Written consent was obtained prior to each interview.

s was chosen to allow for the emergence of opinions that were not anticipated in advance. No member checking was performed with this hard-to-recruit group of parents; however, a focus group with a small number of mothers confirmed the interpretation of the findings. Written consent was obtained prior to each interview. Results In-depth interviews were completed with 20 parents (17 mothers and 3 fathers) of 19 children from five schools; 14 parents were South Asian, 5 were white British and 1 African heritage (table 2). No further interviews were undertaken after saturation of the emerging themes from the 19 interviews. Our analysis identified themes that highlighted pragmatic reasons for non-attendance and non-adherence and also those related to health beliefs about illness, disease and consultation behaviour; illustrative quotes are presented in the text. The themes identified present the complexity of parental decision making. Table 2 Participant characteristics

Results In-depth interviews were completed with 20 parents (17 mothers and 3 fathers) of 19 children from five schools; 14 parents were South Asian, 5 were white British and 1 African heritage (table 2). No further interviews were undertaken after saturation of the emerging themes from the 19 interviews. Our analysis identified themes that highlighted pragmatic reasons for non-attendance and non-adherence and also those related to health beliefs about illness, disease and consultation behaviour; illustrative quotes are presented in the text. The themes identified present the complexity of parental decision making. Table 2 Participant characteristics Parent Ethnicity Attended and adherent Attended and non-adherent Non-attendance and non-adherent Mother 1 (i) South Asian Yes Mother 2 (i) South Asian Yes Mother 3 South Asian Yes Mother 4 African Yes Mother 5 South Asian Yes Father 1 White British Yes Mother 6 White British Yes Mother 7 South Asian Yes Father 2 South Asian Yes Mother 8 South Asian Yes Mother 9 White British Yes Mother 10 White British Yes Mother 11 White British Yes Mother 12 South Asian Yes Mother 13* South Asian Father 3 South Asian Yes Mother 14 (i) South Asian Yes Mother 15 South Asian Yes Mother 16 South Asian Yes Mother 17 South Asian Yes *Mother 13 and father 3 are parents of the same child and were interviewed together; only the fathers discourse is reported. i, interpreter.

Parent Ethnicity Attended and adherent Attended and non-adherent Non-attendance and non-adherent Mother 1 (i) South Asian Yes Mother 2 (i) South Asian Yes Mother 3 South Asian Yes Mother 4 African Yes Mother 5 South Asian Yes Father 1 White British Yes Mother 6 White British Yes Mother 7 South Asian Yes Father 2 South Asian Yes Mother 8 South Asian Yes Mother 9 White British Yes Mother 10 White British Yes Mother 11 White British Yes Mother 12 South Asian Yes Mother 13* South Asian Father 3 South Asian Yes Mother 14 (i) South Asian Yes Mother 15 South Asian Yes Mother 16 South Asian Yes Mother 17 South Asian Yes *Mother 13 and father 3 are parents of the same child and were interviewed together; only the fathers discourse is reported. i, interpreter. Health beliefs Deep-rooted health beliefs played an important part in explaining attitudes towards vision care in children starting primary school. Doubt along with other cultural health beliefs gave rise to uncertainty surrounding the potential benefits of spectacle wear, viewed by some as unnecessary. Confidence in the test Reduced vision in young children is not generally associated with overt signs and symptoms, and parents found it difficult to accept their child had a vision problem. Adherence to spectacle wear was supported by parents when they perceived the vision test to be reliable. It was really hard when she was young and it was the first time she had her eyes tested and they were putting some things in front of her eyes and I think she found that difficult. (Mother 3, adherent)

What is already known on this topic? Young people’s well-being may be affected by multiple individual and contextual factors. Key determinants of adolescent well-being remain unclear. Few studies have examined a wide range of potential determinants while adjusting for area-level deprivation. What this study hopes to add? Findings support current policy foci on bullying, physical activity and screen-time as correlates of well-being among young people. Sleep and eating behaviours may also be important policy targets for promoting adolescent well-being. A coherent policy framework to promote adolescent well-being needs to be multifaceted and consider a range of health factors in young people’s lives. Introduction There are growing concerns about the well-being of young people in modern societies, particularly in the UK where there is evidence that young people’s well-being is lower than in many comparable developed countries.1 2 Well-being is defined as ‘the state of being comfortable, healthy or happy’.3 From a holistic perspective, well-being incorporates different dimensions of adolescent lives including social relationships and individual functioning.4 However, the determinants of adolescent well-being is a relatively understudied area in comparison with the large literature on factors associated with mental health problems, as well-being concept has only been on greater focus over the past two decades.5 6

Confidence in the test Reduced vision in young children is not generally associated with overt signs and symptoms, and parents found it difficult to accept their child had a vision problem. Adherence to spectacle wear was supported by parents when they perceived the vision test to be reliable. It was really hard when she was young and it was the first time she had her eyes tested and they were putting some things in front of her eyes and I think she found that difficult. (Mother 3, adherent) You do find the children do get bored very, very quickly but they dealt with them in a pleasant manner but sometimes it makes me wonder whether they actually could tell what she was seeing properly because, with them being so young. (Mother 6, adherent) He’s never had any problems with his work to say that he’s got bad eyes. (Mother 10, non-attender) As he’s got a bit older and he’s obviously done more eye tests ….I’m a lot happier that (you know) he is properly diagnosed. (Mother 8, non-adherent) The view that children ‘should not’ require spectacles and the lack of confidence in the accuracy of testing young children led to parents testing their own children at home and canvasing significant others to verify that the child’s vision was good without spectacles. Mums tried her at home to check & test to see if she needs them but there is no difference so it’s not like she needs them at home at all. (Mother 3, adherent) She does gymnastics as well and she doesn’t wear her glasses and she can see on the beams and the floor and everything fine. (Mother 6, adherent)

The view that children ‘should not’ require spectacles and the lack of confidence in the accuracy of testing young children led to parents testing their own children at home and canvasing significant others to verify that the child’s vision was good without spectacles. Mums tried her at home to check & test to see if she needs them but there is no difference so it’s not like she needs them at home at all. (Mother 3, adherent) She does gymnastics as well and she doesn’t wear her glasses and she can see on the beams and the floor and everything fine. (Mother 6, adherent) …. she’s fine at home she doesn’t need the glasses at home. She goes to the mosque as well and she will be doing reading there as well and she doesn’t need them at mosque. She’s asked the teacher there as well and she says she’s fine and she doesn’t need them there as well. (Mother 2, non-adherent) I says to her can you see without them and she says yeh I can see without them, yeh… she’s had no problems at all, she can see alright, fine. She’s had no pain in her eyes or anything like that, she says she doesn’t really need them she can see everything without them. (Father 2, non-adherent) Community and social influences The advice of family or community members was influential, particularly when it matched parents’ personal beliefs. Where there was a positive family history of wearing spectacles parents reported attending and adherence with spectacle wear.

I says to her can you see without them and she says yeh I can see without them, yeh… she’s had no problems at all, she can see alright, fine. She’s had no pain in her eyes or anything like that, she says she doesn’t really need them she can see everything without them. (Father 2, non-adherent) Community and social influences The advice of family or community members was influential, particularly when it matched parents’ personal beliefs. Where there was a positive family history of wearing spectacles parents reported attending and adherence with spectacle wear. Well you do have more pressure from older people like my parents, ‘why is he wearing glasses, his eye sight is going to get worse, he’ll get dependant on them’. (Mother 8, non-adherent) My mum has said it, like they’re too young… I do have this from all my family, …. he is obviously too young and once he wears them they’re going to be like you’re going to have to wear them all the time. (Mother 17, non-attender) He (father) doesn’t feel that there is a problem so that’s the reason why he hasn’t taken her (child). (Mother 1, non-attender) No one has discouraged the children from wearing glasses, everybody has asked … does she need glasses is that why she’s wearing them…. so she gets a lot of encouragement. (Mother 3, adherent) She was quite happy actually, because she’s wearing them (glasses) like her older sister and brother and daddy…. (Mother 12, non-adherent)

No one has discouraged the children from wearing glasses, everybody has asked … does she need glasses is that why she’s wearing them…. so she gets a lot of encouragement. (Mother 3, adherent) She was quite happy actually, because she’s wearing them (glasses) like her older sister and brother and daddy…. (Mother 12, non-adherent) Understanding Parents of those children who were adhering to spectacle wear believed that failure to wear the spectacles would lead to a deterioration of both the child’s vision and their ability to learn, and many parents believed that by wearing the spectacles, the vision would improve and then spectacles would no longer be required. …. if he takes them off then his vision is going to get really weak. Glasses will help him so he has to keep them on all the time and just take them off when he really has to. (Mother 3, adherent) … if a kid cannot see properly and keeps getting headaches because you know he’s trying to concentrate, it’s going to affect him in his education, it’s going to affect him when he’s playing and during life as well, so it’s very important I think. (Father 3, adherent) … if she keeps wearing them now that when she gets to a teenager she might not need them. (Mother 6, adherent) Stigma The stigma of spectacle wear was reported by parents of adherent and non-adherent children, and this contributed significantly in explaining parents’ resistance to the child’s spectacle wear.

… if a kid cannot see properly and keeps getting headaches because you know he’s trying to concentrate, it’s going to affect him in his education, it’s going to affect him when he’s playing and during life as well, so it’s very important I think. (Father 3, adherent) … if she keeps wearing them now that when she gets to a teenager she might not need them. (Mother 6, adherent) Stigma The stigma of spectacle wear was reported by parents of adherent and non-adherent children, and this contributed significantly in explaining parents’ resistance to the child’s spectacle wear. To be honest yesterday she took her glasses out of school and she thought oh I’m a very different girl, my friends are saying you’re a different […] to be honest she doesn’t need really, she’s OK without glasses. (Mother 5, non-adherent) … in Zimbabwe people who wear glasses they are seen as people who are really well educated. They say, ‘You’re wearing glasses, you think you’re sooo… educated’. (Mother 4, non-adherent) … because they’re teasing him he’s retaliated and we’ve had a few issues. (Father 3) Pragmatism Access to appointments was influenced by the parent’s ability to organise appointment times around busy working schedules, the children’s school attendance and the availability of transport. Awkward, really awkward because it’s generally in school time and she doesn’t like missing any time at school…. (Mother 6, adherent) When I was at work it was difficult obviously because you know when you’re at work you can’t really get the appointments during the day and all the kids are at school…(Mother 8, non-adherent)

Awkward, really awkward because it’s generally in school time and she doesn’t like missing any time at school…. (Mother 6, adherent) When I was at work it was difficult obviously because you know when you’re at work you can’t really get the appointments during the day and all the kids are at school…(Mother 8, non-adherent) No, it’s just main problem is parking, I just missed two appointments because I couldn’t get the parking on time. (Mother 7, non-adherent) Information The amount and detail of information provided to parents varied and parents suggested that more information was always helpful. Two mothers suggested that the eye care professional should provide parents with an indication of the improvement in VA at each visit. We’ve always had leaflets given to us every time we’ve been so we were given plenty of information. (Mother 6, adherent) I wasn’t fully clear of the letter and what information I did take from it. (Mother 1, non-attender) They (eye professionals) need to understand a little better and give a percentage, so we (parents) can compare vision to the last test. (Mother 16, non-adherent) It would be good to be updated on whether her eyesight was getting better or worse. (Mother 2, non-adherent) Parents whose first language was not English did not always fully comprehend the information provided or relied on family members to access care and interpret clinical advice. … my older daughters can read now so that they can read and interpret letters. (Mother 3, adherent)

It would be good to be updated on whether her eyesight was getting better or worse. (Mother 2, non-adherent) Parents whose first language was not English did not always fully comprehend the information provided or relied on family members to access care and interpret clinical advice. … my older daughters can read now so that they can read and interpret letters. (Mother 3, adherent) I used to take my sister in law, who can understand English, so she can explain it and ask questions. (Mother 14, non-adherent) Supporting strategies The children’s schools played a key role in supporting the parents in developing strategies to both attend  ophthalmic appointments and ensure the childs spectacle wear. Some parents were able to negotiate appropriate time for their child to be absent during school hours. They were fine, the times that they gave me have always been in the afternoon instead of the morning so it’s easy… and I can take them out of school. (<other 2, non-adherent) So (erm), many times many occasions I’ve forgot glasses at home and then I gave one pair to school as well to resolve this issue. (Mother 12, non-adherent) My son needs [glasses] all the time, so I have to say all the time to class teacher ‘please remind him to wear his glasses’. (Mother 16, non-adherent)

They were fine, the times that they gave me have always been in the afternoon instead of the morning so it’s easy… and I can take them out of school. (<other 2, non-adherent) So (erm), many times many occasions I’ve forgot glasses at home and then I gave one pair to school as well to resolve this issue. (Mother 12, non-adherent) My son needs [glasses] all the time, so I have to say all the time to class teacher ‘please remind him to wear his glasses’. (Mother 16, non-adherent) Discussion The decision to attend following vision screening and to adhere to spectacle wear was primarily based on the perceived severity of the visual reduction with the benefit of spectacle wear outweighing any negative consequences. Parental observations played an important role in validating the professional assessment, when parental perception did not match clinical opinion adherence was less likely.19 Parental knowledge tended to dominate decisions where there was discordance between professional and lay knowledge20 with parents having difficulty accepting clinical advice, particularly when their child did not appear to require spectacles for reading and writing. It has been reported that limited experience of a health condition affects judgement of its severity21; this may also have influenced the parents’ reasoning to support spectacle wear in their children. Parents whose children had failed vision screening and had not taken their child for further assessment were keen to justify their actions by normalising the eye condition19 22; in addition, they did not report any support from either family members or from their child’s school; this lack of support may have contributed to non-attendance. Stigma played a major part in explaining parents’ resistance to clinical advice.23 Appearing ‘normal’ seemed to outweigh potential benefits24 and was a strong driver for maintaining a socially acceptable image of the family within the community22; a child with spectacles seemed to threaten this image suggesting that the eye condition was not the sole factor impacting on adherence.25 Parents who did not consider their child’s vision to be impaired, who described barriers such as organising transport or social stigma, did not follow recommendations and deemed treatment unnecessary.

acles seemed to threaten this image suggesting that the eye condition was not the sole factor impacting on adherence.25 Parents who did not consider their child’s vision to be impaired, who described barriers such as organising transport or social stigma, did not follow recommendations and deemed treatment unnecessary. Parents with limited English language ability relied on family members to access care and interpret clinical advice; this use of the family network is reported to be unreliable, leading to misinformation and has been reported to impact on the child’s care.12 The practical considerations of attending regular appointments particularly in cases of uncertainty about the benefit of treatment therefore posed a dilemma, contributing to and perhaps being the tipping point for non-attendance and non-adherence. Where support was available either from family members or from the child’s school, this provided a cue to action prompting parents to take the first step in attending appointments.

e benefit of treatment therefore posed a dilemma, contributing to and perhaps being the tipping point for non-attendance and non-adherence. Where support was available either from family members or from the child’s school, this provided a cue to action prompting parents to take the first step in attending appointments. Strengths and limitations This qualitative study helps to explain the reasons for non-attendance and poor adherence to spectacle wear in young children following vision screening. The strengths of the study include the purposive sample from a multiethnic community, including parents whose children were known to fail to attend and those whose children failed to adhere to prescribed spectacle wear; these are hard-to-reach groups of parents. The additional inclusion of parents whose children had attended and adhered to spectacle wear allowed insights into both barriers and enablers to attendance and adherence.

dren were known to fail to attend and those whose children failed to adhere to prescribed spectacle wear; these are hard-to-reach groups of parents. The additional inclusion of parents whose children had attended and adhered to spectacle wear allowed insights into both barriers and enablers to attendance and adherence. The children were participants in the Born in Bradford (BiB) cohort, and this could have encouraged positive responses in recruitment and in the views provided; however, no parent responded solely from the initial invitation to participate generated via BiB. The schools were crucial to the successful recruitment of parents, and this may have positively influenced the parental responses regarding the support provided by schools; however, all the parents declined to be interviewed in their own home and would only be interviewed in school. The participating parents were all recruited from deprived communities, and this study may not be representative of the experiences of parents in more affluent areas; however non-attendance and non-adherence is known to occur predominantly in areas of lower socioeconomic status,26 and therefore these findings will represent the experiences of this under reported, hard-to-reach group of parents.

his study may not be representative of the experiences of parents in more affluent areas; however non-attendance and non-adherence is known to occur predominantly in areas of lower socioeconomic status,26 and therefore these findings will represent the experiences of this under reported, hard-to-reach group of parents. Existing literature The findings from this study provide a comprehensive picture of the experiences of parents whose young children have been referred following vision screening, further explaining the reasons for non-attendance and non-adherence. Previous population-based studies reporting children’s adherence to spectacle wear have focused on the factors found to be associated with non-attendance, reporting higher proportions of children from families of lower socioeconomic status26 failing to attend and older children demonstrating a greater rate of non-adherence to spectacle wear.27 In a study of young children (under 8 years) in Berkshire,28 compliance with spectacle wear was reported to be good; however, the place where the study was undertaken is in a relatively affluent area, and the parents were regular attenders to the HES.

n demonstrating a greater rate of non-adherence to spectacle wear.27 In a study of young children (under 8 years) in Berkshire,28 compliance with spectacle wear was reported to be good; however, the place where the study was undertaken is in a relatively affluent area, and the parents were regular attenders to the HES. Similar to our findings, studies of the experiences of adults in screening programmes describe that a lack of perceived severity of the condition influences attendance.13 29 This study, in the field of children’s eye care, adds to this, reporting that where the severity of the condition is in doubt, the parent will perform their own confirmatory vision test in their home environment. The practical barriers to attendance we report such as access to appointments and transport also confirm the findings of previous studies.13 29 In addition, importantly, this study highlights the influence that support from significant others such as family or school have in helping overcome barriers to attendance and adherence.

s including social relationships and individual functioning.4 However, the determinants of adolescent well-being is a relatively understudied area in comparison with the large literature on factors associated with mental health problems, as well-being concept has only been on greater focus over the past two decades.5 6 A wide range of factors have been shown to be related to adolescent well-being, including a range of cognitive and relational factors such as bullying,7 family structure and relationships,8 peer support9 and school connectedness.10 Other behaviours also influence well-being, including substance use (alcohol, drugs and smoking habits),11–13 fruit and vegetable consumption,14 breakfast consumption,15 physical activity,16 sleep duration,17 sedentary behaviour7 18 and leisure time activities.19 However, published studies use a wide range of well-being measures, resulting in conflicting findings.20 21 Furthermore, studies have largely focused on single risk factors and not explored how factors including behavioural factors interact to influence well-being. Additionally, given that many such behaviours are strongly socially patterned, studies have thus far paid little attention to confounding by socioeconomic position and issues relating to the clustering of behaviours and well-being within localities.

cal barriers to attendance we report such as access to appointments and transport also confirm the findings of previous studies.13 29 In addition, importantly, this study highlights the influence that support from significant others such as family or school have in helping overcome barriers to attendance and adherence. Implications for clinicians and policy makers Pragmatic strategies are required to improve attendance and adherence. The desire for condition-specific information is underestimated by clinicians,30 and provision of personalised information, following vision screening, highlighting the benefits of treatment in young children, could improve attendance. The current national development of information leaflets specifically aimed at parents providing information following vision screening may influence attendance.31 Concordance around treatment should involve discussion between clinician and parent through which an informed decision regarding treatment can take place.32 This has been reported in general practice with patients less likely to take medications if their own concerns are not initially addressed.20 Further study is required into how eye care professionals communicate with parents, and how they present information33 to ensure the parents can make informed decisions regarding adherence. Stigma was commonly cited by parents as a barrier to adherence.34 Strategies both at the individual and community level are required with the role of schools as facilitators in reducing stigma and promoting adherence, an area highlighted by parents and requiring future study.

ake informed decisions regarding adherence. Stigma was commonly cited by parents as a barrier to adherence.34 Strategies both at the individual and community level are required with the role of schools as facilitators in reducing stigma and promoting adherence, an area highlighted by parents and requiring future study. These findings provide insight into the reasons parents either rejected or resisted therapeutic advice, not solely reflecting levels of knowledge, but reflecting an active evaluation of the potential severity of the vision loss and the perceived loss of normality and social stigma compared with the benefit of the treatment. Our results illustrate the complexity around attendance and adherence patterns and provide a greater understanding that can inform the redesign of children’s eye care pathways. Supplementary Material Reviewer comments Author's manuscript We would like to thank all the families and schools who took part in this study. Contributors: AB initiated the project, designed the study, conducted the interviews, analysed the transcripts and drafted and revised the paper. She is guarantor. TS analysed the transcripts and revised the draft paper. TAS initiated the project and revised the draft paper.

Supplementary Material Reviewer comments Author's manuscript We would like to thank all the families and schools who took part in this study. Contributors: AB initiated the project, designed the study, conducted the interviews, analysed the transcripts and drafted and revised the paper. She is guarantor. TS analysed the transcripts and revised the draft paper. TAS initiated the project and revised the draft paper. Funding: AB is funded by a National Institute for Health Research Post-Doctoral Fellowship Award (PDF-2013-06-050). The research was funded bythe NIHR Collaboration for Leadership in Applied Health Research and CareYorkshire and Humber (NIHR CLAHRC YH). The views and opinions expressed are those of the author(s), and notnecessarily those of the NHS, the NIHR or the Department of Health. Disclaimer: The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. Competing interests: None declared. Patient consent: Not required. Ethics approval: Ethics approval was obtained from the National Research Ethics Committee Yorkshire and the Humber- South Yorkshire UK (Ref 13/YH/0379). Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional unpublished data are available.

Introduction Background Acute appendicitis is one of the most common acute surgical emergencies in children. According to the National Schedule of Reference Costs/Healthcare Resource Groups (HRG) data, almost 14 000 operations are performed every year in the UK (<18 years). Appendicectomy procedure for a patient under 18 years old costs on average from £3072 to £5992 and accounts for an annual total cost of approximately £50 million to the National Health Service (NHS).1

e Costs/Healthcare Resource Groups (HRG) data, almost 14 000 operations are performed every year in the UK (<18 years). Appendicectomy procedure for a patient under 18 years old costs on average from £3072 to £5992 and accounts for an annual total cost of approximately £50 million to the National Health Service (NHS).1 For many years, appendicectomy has been considered the standard treatment for appendicitis, both in adults and in children. However, there is great current interest in the role of non-operative treatment (with antibiotics alone) of adults and children with acute appendicitis. Alongside evaluations of the clinical effectiveness of this alternative treatment option, it is equally important to explore the economic implications. A small number of studies have reported (mainly clinical) outcomes of non-operative treatment of acute appendicitis in children, but very little is published on how this translates into economic outcomes.2–5 Therefore, the economic impact of different treatment options on the health system (eg, NHS) if adopted remains largely unknown. We plan to conduct a definitive randomised controlled trial (RCT) comparing these alternative treatment options. Since recruitment to such an RCT will be challenging, we are first conducting a feasibility RCT in the UK.6 Alongside this feasibility RCT, we will conduct a health economic feasibility substudy to inform a future cost-effectiveness and cost–utility analysis (CUA) within our definitive RCT. Herein, we describe the protocol for this health economic substudy. We aim to address the following research questions: what are the cost implications of treating childhood appendicitis non-operatively as compared with surgery; how the costs of both treatment options compare with widely used NHS Reference Costs; and what could be the implications of differing cost methods in assessing the cost-effectiveness of appendicitis. The study will also assess two preference-based quality of life (QoL) questionnaires widely used in paediatric research, using as reference case, clinical outcomes identified through the feasibility study.

and what could be the implications of differing cost methods in assessing the cost-effectiveness of appendicitis. The study will also assess two preference-based quality of life (QoL) questionnaires widely used in paediatric research, using as reference case, clinical outcomes identified through the feasibility study. Study design, participants, interventions and outcomes The CONservative TReatment of Appendicitis in Children – randomised controlled Trial (CONTRACT) study is a feasibility RCT that aims to explore whether it is feasible and acceptable to conduct a multicentre RCT testing the effectiveness and cost-effectiveness of a non-operative treatment pathway for the treatment of acute uncomplicated appendicitis in children. The study is being conducted in three specialist NHS Paediatric Units in England and participants are children (age 4–15 years) with a clinical diagnosis of acute uncomplicated appendicitis. It has been estimated that 52–65 participants in the feasibility RCT will be adequate to test treatment pathway procedures. The health technology under assessment involves treatment with antibiotics (intravenous followed by oral) and regular clinical review to determine disease resolution without appendicectomy or appendicectomy in those whom the disease worsens or fails to resolve. The broad inclusion criteria reflect current clinical practice and enable the generalisability of our results for routine inpatient care. The principal outcome of the feasibility study will be recruitment rate. A full protocol of the CONTRACT feasibility RCT is published elsewhere.6

disease worsens or fails to resolve. The broad inclusion criteria reflect current clinical practice and enable the generalisability of our results for routine inpatient care. The principal outcome of the feasibility study will be recruitment rate. A full protocol of the CONTRACT feasibility RCT is published elsewhere.6 Scope of the economic substudy protocol The economic substudy will provide evidence and guidance for determining data collection tools measuring cost and benefit outcomes for our future RCT assessing the cost-effectiveness of the non-operative treatment of appendicitis. Data management will be performed by the Southampton Clinical Trials Unit, and anonymised data will be delivered for the health economic analysis. This protocol describes methods for incorporating economic evidence into an early stage of the study and has been conceptually divided into two parts: (1) measuring resource utilisation and conducting microcosting, and (2) measuring QoL and assessment of health-related QoL (HRQoL) instruments. Methods and analysis Part I: resource utilisation and costs Costs are an important part of any economic evaluation but is a term that has different meaning across different disciplines. In Health Economics, costs are related to opportunity costs, and the question of interest is the choice between two alternatives; in other words, there are always forgone opportunities when choosing to invest in a new medical technology or health service rather than in a current treatment.7 8

oss different disciplines. In Health Economics, costs are related to opportunity costs, and the question of interest is the choice between two alternatives; in other words, there are always forgone opportunities when choosing to invest in a new medical technology or health service rather than in a current treatment.7 8 Principles of costing within CONTRACT The quality of the economic evaluation depends on the quality of the measurement of costs and outcomes.8–10 There are two main approaches used to measure healthcare costs: the ‘macrocosting/top-down’ and the ‘microcosting/bottom-up’ approach.8 11–14 The gross or macrocosting method is commonly used in cost analysis providing an overview of the effect of costs, but it has been argued that it is not appropriate in many cases for economic evaluation, because it provides limited accuracy and detail.8 14 In the UK, it is common practice to use the NHS Reference Costs/HRGs, which provides national tariffs as the unit costs for different services and procedures. These unit costs are calculated using the mean costs among patients and hospitals. However, these national tariffs might not represent good estimates of real costs especially for new interventions, and some reports show cost estimate discrepancies between macrocosting and microcosting, ranging between 9% and 66%.15 Advocates of the macrocosting approach highlight the advantages of this method, namely generalisability, easy to use and less time-consuming. Microcosting reflects the individual patient costs by identifying and collecting the actual individual resources used and estimating the economic costs of resources used. However, despite the micro-costing approach being regarded as more accurate and transparent, it is considered time-consuming and not easily applied in some settings.14 A general rule recommended by Beecham and Knapp16 is that the broader and most accurate approach is collecting individual data through a microcosting method or adopting a ‘reduced list costing’,17 which implies the selection of a limited number of services considered to be of most significance.

lied in some settings.14 A general rule recommended by Beecham and Knapp16 is that the broader and most accurate approach is collecting individual data through a microcosting method or adopting a ‘reduced list costing’,17 which implies the selection of a limited number of services considered to be of most significance. Detailed recommended methodology on costing is still lacking, but several guidelines have emerged13 18–21 such as a series of task force reports on methodological issues in costing methods by the International Society for Pharmacoeconomics and Outcomes Research.20 22 However, guidelines still vary in terms of recommended methods, and variations of costing methods affect the validity and comparability of economic evaluation results.23 24 Our approach in this economic study is to compare the two methods in an attempt to minimise costing bias and improve choice of instruments that will be used in our future RCT. Aims To develop and assess resource use data collection tools in support of the future economic evaluation within our definitive future RCT. To conduct microcosting of both treatment pathways and to explore what are the determinants of variation in costs across settings and methods (macrocosting vs microcosting). To provide an economic rationale for the use of the most appropriate resource use identification, valuation and data collection tools.

Aims To develop and assess resource use data collection tools in support of the future economic evaluation within our definitive future RCT. To conduct microcosting of both treatment pathways and to explore what are the determinants of variation in costs across settings and methods (macrocosting vs microcosting). To provide an economic rationale for the use of the most appropriate resource use identification, valuation and data collection tools. Identifying what costs are to be included A microcosting approach to data identification and collection will be adopted. This approach will allow identification of resource use, meaning it will be focused and able to provide rich data about the resources used in relation to managing paediatric acute appendicitis in secondary care. Each stage of the data collection refers to event pathways for activity costing so that context and information is not lost in the final outcome of each activity. The process will include identification of services, how the service works and which components of costs are incurred on delivery of each service. We will design and map processes involved in service delivery in order to identify all relevant resource use. Therefore, the details of the inpatient resource use will be collected through the design and implementation of case report forms that will be informed from hospital records (from which medical history and previous and concurrent medication will be summarised), clinical and office charts, laboratory and pharmacy records, diaries, microfiches, radiographs and correspondence. Detailed analysis of patients’ hospital records will be undertaken for the first 10 patients recruited into CONTRACT across all three participating sites. This will enable an initial inclusive list of resource use items to be created and updated based on actual patient data using microcosting principles. This work will inform the resource use data collection tool that will be used to collect data for the remainder of the recruited patients. Additionally, patient diary cards will be used to record resource use during the 14 days immediately following discharge from hospital. These will be used to collect data on use of antibiotics, pain medications and anti-inflammatory or other relevant medications, as well as productivity loss and absence from school information. Finally, a modified version of the Client Service Receipt Inventory (CSRI)16 25 questionnaire will be used to collect other resource use data.

ed to collect data on use of antibiotics, pain medications and anti-inflammatory or other relevant medications, as well as productivity loss and absence from school information. Finally, a modified version of the Client Service Receipt Inventory (CSRI)16 25 questionnaire will be used to collect other resource use data. The CSRI is a research instrument developed in the mid-1980s to collect information on service utilisation, income, accommodation and other cost-related variables. This will include healthcare appointments and additional family-borne costs, as reported by parents of participants at 6 weeks following discharge and at 6 months. Measuring and valuing resource use Following identification of the patient’s pathway and the services used, we will design a comprehensive list of resource use items that will be included in our resource inventory collection tool. This approach will form a comprehensive health profile of service utilisation that will form the outcome of this study and will lead to identification of the main cost drivers that need to be collected in our future definitive RCT. In this part of the study, we will use a mixed method approach for the valuation of the resources used. This valuation will use unit costs from both the Personal Social Services Research Unit and the NHS Reference Costs data. Additionally, as part of our microcosting approach, we will collect and compare unit cost data from participating hospitals.14

ll use a mixed method approach for the valuation of the resources used. This valuation will use unit costs from both the Personal Social Services Research Unit and the NHS Reference Costs data. Additionally, as part of our microcosting approach, we will collect and compare unit cost data from participating hospitals.14 Data collection and analysis After choosing the items of resource use to be included in this study, we will classify them in different components depending on the characteristics of the care pathway and service systems involved. Classifying resource use and costs implies focusing on variation at individual and aggregate level by trial arm. The economic substudy at this stage will allow us to verify the relevance of this variation.

fy them in different components depending on the characteristics of the care pathway and service systems involved. Classifying resource use and costs implies focusing on variation at individual and aggregate level by trial arm. The economic substudy at this stage will allow us to verify the relevance of this variation. Both datasets of resource use and costs, at individual and aggregate level, will allow us to identify the main factors that influence the cost of the intervention and will form the basis for considering the main cost drivers and methodology for inclusion in our definitive RCT. We will assess data quality and missing data identifying the most appropriate approach collecting economic data alongside randomised clinical studies. Descriptive statistics will be performed to summarise data and problems identified will be discussed and presented in a relevant publication. External validation will be achieved by comparing the outcomes from our bottom-up microcosting to the NHS Reference Costs, and the HRG tariff to identify the most appropriate costing method. We envisage that in case of significant variation in the costing methods, we will be able to adopt both methods in our future cost-effectiveness analysis (CEA) in the form of sensitivity analysis. Given the importance of costs in any CEA, this proposed work will allow defining uncertainty around the CEA results and will provide an evidence base for future research.

on in the costing methods, we will be able to adopt both methods in our future cost-effectiveness analysis (CEA) in the form of sensitivity analysis. Given the importance of costs in any CEA, this proposed work will allow defining uncertainty around the CEA results and will provide an evidence base for future research. Part II: preference-based HRQoL instruments and the quality-adjusted life year (QALY) framework The most commonly cited and used paediatric preference-based generic HRQoL instruments are the Health Utility Index (HUI),26 Euroqol 5 dimensions youth version (EQ-5D-Y)27 and CHU-9D.28 In the UK, there is a tendency towards the use of the EQ-5D-Y due to recommendation by NICE for adult population (EQ-5D-3L29 30). The EQ-5D-Y comprises the same 3 L as the adult version with improved wording for children despite not having a child specific value set. Euroqol states that ‘Recent research has indicated that regular EQ-5D-3L value sets cannot be used for children and adolescents. The main reason is that health states are valued differently when described for an adult or a child.’.

ersion with improved wording for children despite not having a child specific value set. Euroqol states that ‘Recent research has indicated that regular EQ-5D-3L value sets cannot be used for children and adolescents. The main reason is that health states are valued differently when described for an adult or a child.’. More recently, a relatively new paediatric instrument, CHU-9D, has become more widely used in the UK. This is the only preference-based HRQoL measure specifically designed and developed with children using UK general population value sets. The HUI, although a paediatric instrument, was initially developed for patients with cancer and is not used as much in the UK. We believe this is due to two reasons: first, it relies on preference values obtained from the Canadian general population and not a UK population, which might introduce some differences. Second, there is a cost attached to the use of HUI, and this could be an issue for consideration when research studies need to operate under reasonably limited budget.

ns: first, it relies on preference values obtained from the Canadian general population and not a UK population, which might introduce some differences. Second, there is a cost attached to the use of HUI, and this could be an issue for consideration when research studies need to operate under reasonably limited budget. Principles of HRQoL assessment within CONTRACT To enable detection of any effect of our intervention on HRQoL, we will collect data using two preference-based quality of life measures. The proposed measures are: (1) the EQ-5D-5L,31 which comprises the same five dimensions as the EQ-5D-3L and EQ-5D-Y but five levels of severity, which is considered to significantly increase reliability and sensitivity (discriminatory power)31 32 and (2) the CHU-9D, the paediatric generic quality of life measure specifically designed for use in studies with children, which comprises nine dimensions.28 33–36 Both measures will be obtained from parent/carer proxy responses and children if 7 years or older. Aims To compare two alternative preference-based generic HRQoL measures commonly used in paediatric studies. To identify the most appropriate HRQoL instrument for economic evaluations alongside clinical studies for children with acute uncomplicated appendicitis in tertiary care settings. To assess the variation and impact of time of data collection on utility values and the QALY framework when used in this population.

Aims To compare two alternative preference-based generic HRQoL measures commonly used in paediatric studies. To identify the most appropriate HRQoL instrument for economic evaluations alongside clinical studies for children with acute uncomplicated appendicitis in tertiary care settings. To assess the variation and impact of time of data collection on utility values and the QALY framework when used in this population. Data collection and analysis We will collect both HRQoL measures at baseline, discharge, 2 weeks (to determine any short-term difference in QoL that may not be apparent at later follow-up), 6 weeks, 3 months and 6 months to define the most appropriate timing of assessment in relation to other health outcomes. Evidence from this work will support the decision for the most appropriate HRQoL instrument to be used but also will provide valuable information adopting and reporting results in our future CUA in terms of cost per QALY gained. Any imbalances detected will inform sensitivity analyses and, therefore, will enrich the results from the future definitive trial. We will also assess the appropriateness of using the QALY framework in this population, in terms of identifying aspects that are excluded from the conventional QALY framework, and aspects that the QALY framework could be sensitive in regards to timing of data collection.

ch the results from the future definitive trial. We will also assess the appropriateness of using the QALY framework in this population, in terms of identifying aspects that are excluded from the conventional QALY framework, and aspects that the QALY framework could be sensitive in regards to timing of data collection. Concluding remarks Costs of different interventions are an important part of any economic evaluation to determine whether a particular intervention is better placed, in terms of the outcomes it generates, in comparison with standard care.7 8 The two most commonly used methods of collecting cost data are either ‘macro/top-down’ or ‘micro/bottom-up’ costing. The macrocosting uses the total budget to produce average costs per patient. This method is quicker but assumes that all patients have the same diagnosis, severity and treatment. Microcosting measures resource use by individual patient and therefore is considered more accurate detecting cost variability among patients. This method produces better quality costs but can be time-consuming and expensive.14 In this study, we will assess two HRQoL measures and the implications of adopting the QALY framework in our future economic evaluation. Incorporating the outcomes from this economic substudy into the feasibility stage of our RCT, and the microcosting method we adopt in doing so, we believe it will enhance our results and their applicability for healthcare decision making and for future economic evaluations.

framework in our future economic evaluation. Incorporating the outcomes from this economic substudy into the feasibility stage of our RCT, and the microcosting method we adopt in doing so, we believe it will enhance our results and their applicability for healthcare decision making and for future economic evaluations. Supplementary Material Reviewer comments Author's manuscript Contributors: The lead health economist (MC) designed the Health Economic Analysis Plan and prepared this manuscript; the lead statistician (IR) contributed to the design of the study; the principal investigator (NH) is involved in all aspects of the study; all coauthors contributed to the preparation and approval of this manuscript. Funding: The authors acknowledge funding received from the UK National Institute for Health Research (NIHR) Health Technology Assessment (HTA) Board (14/192/90). Competing interests: None declared. Patient consent: Patient/parent/gardian consent obtained. Ethics approval: The study, including the economic substudy, has been approved by the South Central – Hampshire A Research Ethics Committee (16/SC/0596). Provenance and peer review: Not commissioned; internally peer reviewed.

What is already known on this topic? Stunting is the most common cause of early childhood developmental delays, and nearly half of Guatemalan children are stunted, with higher prevalence among the Guatemalan indigenous Maya population. Stunting early in life has deleterious long-term impacts on educational attainment, intellectual outcomes and adult economic earning potential. Few investigations of these effects exist for the Guatemalan indigenous Maya population. Complementary feeding practices interventions are the cornerstone for stunting prevention and treatment; however, investigation of the effects of these interventions on developmental outcomes are limited. What this study hopes to add? Both standard and augmented behaviour change approaches to improved complementary feeding practices by caregivers have important positive impacts on the development of stunted children. Current public malnutrition reduction strategies being implemented in indigenous Maya communities in Guatemala can improve childhood developmental outcomes. Developmental assessments can be feasibly adapted and contextualised to a rural, indigenous population at risk.

What this study hopes to add? Both standard and augmented behaviour change approaches to improved complementary feeding practices by caregivers have important positive impacts on the development of stunted children. Current public malnutrition reduction strategies being implemented in indigenous Maya communities in Guatemala can improve childhood developmental outcomes. Developmental assessments can be feasibly adapted and contextualised to a rural, indigenous population at risk. Introduction In low-income and middle-income countries, 43% of children under 5 are at risk of not reaching their developmental potential, and 70% of the attributable risk for delays in early childhood development (ECD) is due to stunting, or low height-for-age.1 2 Guatemala, a Central American country of 17 million inhabitants, has the highest rate of stunting in the Western Hemisphere.3 The burden of stunting disproportionately affects the rural, indigenous Maya population, where stunting often exceeds 70% and complementary feeding and dietary diversity indicators are poor.4–6

Guatemala, a Central American country of 17 million inhabitants, has the highest rate of stunting in the Western Hemisphere.3 The burden of stunting disproportionately affects the rural, indigenous Maya population, where stunting often exceeds 70% and complementary feeding and dietary diversity indicators are poor.4–6 Early scientific research in Guatemala was critical to the international community’s understanding of stunting’s impact on human capital. The Institute of Nutrition of Central America and Panama cohort study—conducted from 1969 to 1977, with long-term follow-up—demonstrated the deleterious impacts of stunting on educational attainment, intellectual outcomes and adult economic earning potential.7 Subsequently, there have been extensive investigations of stunting’s nutritional and sociodemographic correlates in the indigenous Maya population. However, there have been few investigations of developmental outcomes or of the impact of contemporary nutritional interventions on ECD.8–11

lt economic earning potential.7 Subsequently, there have been extensive investigations of stunting’s nutritional and sociodemographic correlates in the indigenous Maya population. However, there have been few investigations of developmental outcomes or of the impact of contemporary nutritional interventions on ECD.8–11 Given the gaps in adequate complementary feeding practices and diet quality, interventions to promote caregiver knowledge around infant-child feeding are an important strategy in Guatemala.5 6 12 This strategy is supported by international evidence, with several studies showing that linear growth is strongly correlated with better development and that complementary feeding education interventions improve stunting.11 13–15 However, there has been limited direct investigation of the effect of complementary feeding on developmental outcomes.16–18 This is an important knowledge gap, because feeding interventions may at times have small direct impacts on linear growth velocity but conceivably larger impacts on developmental outcomes. Therefore, directly measuring these developmental outcomes may generate additional evidence supporting the interventions, which could have important policy implications in a country like Guatemala, where debate over the value of standard child nutrition programme offerings is ongoing.19 20

ts on developmental outcomes. Therefore, directly measuring these developmental outcomes may generate additional evidence supporting the interventions, which could have important policy implications in a country like Guatemala, where debate over the value of standard child nutrition programme offerings is ongoing.19 20 In this study, we contribute by assessing developmental outcomes in indigenous Maya infants in rural Guatemala. We hypothesised that an individualised, intensive approach to caregiver complementary feeding education would improve outcomes over usual care. We individually randomised child-caregiver dyads (aged 6–24 months, height-for-age z-score ≤−2.5) to 6 months of usual care, which included home-based growth monitoring and micronutrient and food supplementation, versus usual care augmented with individualised complementary feeding education, using monthly structured interviews and active goal-setting to promote incremental improvements. Methods Study context This study was conducted at Maya Health Alliance (MHA), a primary healthcare organisation working in rural Maya communities. At MHA, nutrition community health workers (CHW) provide home-based services to children aged 6–24 months with stunting.20 The study was conducted in Tecpán, Chimaltenango (population 95 000), with a settlement of agricultural Kaqchikel Maya families living 25 km from the town centre. The study was conducted according to the principles of Declaration of Helsinki.

HW) provide home-based services to children aged 6–24 months with stunting.20 The study was conducted in Tecpán, Chimaltenango (population 95 000), with a settlement of agricultural Kaqchikel Maya families living 25 km from the town centre. The study was conducted according to the principles of Declaration of Helsinki. Trial design This was a planned substudy on developmental outcomes within a larger, individually randomised, two-arm trial comparing individualised complementary feeding caregiver education with usual care (Clinicaltrials.gov Identifier: NCT02509936), described in detail elsewhere.20 Briefly, eligibility criteria included: children aged between 6 and 24 months and length-for-age z-score ≤−2.5 SD by WHO standards.21 Exclusion criteria were acute malnutrition (weight-for-length z-score of ≤−2 SD) or severe medical illness. Study interventions were delivered by two CHW teams, each consisting of two CHWs. One team provided usual care while the other provided the intervention. Child-caregiver dyads were recruited and written informed consent obtained by a study staff member not involved in the intervention. At enrolment, child’s anthropometric and dietary feeding practices as well as household demographic and socioeconomic characteristics were obtained. Household developmental stimulation data was gathered at enrolment using the Family Care Indicators (FCI) interview.22 Due to the considerable time and expense constraints related to administering psychometric testing, a subset of subjects, namely those consecutively recruited for the larger study during the first 5 months (n=210, figure 1), were invited to participate in the substudy. Study outcomes (subscale scores on the Bayley Scales of Infant Development, Third Edition (BSID-III))23 were obtained at 0 and 6 months by trained study psychologists from the Universidad del Valle de Guatemala, supported by MHA interpreters (Maya Kaqchikel/Spanish). Both psychologists and interpreters were blinded to study allocation.

ubscale scores on the Bayley Scales of Infant Development, Third Edition (BSID-III))23 were obtained at 0 and 6 months by trained study psychologists from the Universidad del Valle de Guatemala, supported by MHA interpreters (Maya Kaqchikel/Spanish). Both psychologists and interpreters were blinded to study allocation. Figure 1 Subject enrolment, randomisation and follow-up. BSID-III, Bayley Scales of Infant Development, Third Edition. Study interventions The study duration was 6 months. As described elsewhere,24 the usual care arm received a monthly home-visit regimen from a team of CHWs including growth monitoring, daily multiple micronutrient powder supplement (ferrous fumarate 12.5 mg, zinc gluconate 5 mg, retinol acetate 300 μg, folic acid 160 μg and ascorbic acid 30 mg; Prodipa, Guatemala City, Guatemala); and a biweekly food ration (beans 1000 g, eggs 20 units and Incaparina 900 g (a common corn and soy-based food supplement; Alimentos, Guatemala City, Guatemala)). For the intervention arm, subjects received usual care and monthly individualised caregiver counselling focused on improving meal frequency and dietary diversity for the child,25 from a separate CHW team.

20 units and Incaparina 900 g (a common corn and soy-based food supplement; Alimentos, Guatemala City, Guatemala)). For the intervention arm, subjects received usual care and monthly individualised caregiver counselling focused on improving meal frequency and dietary diversity for the child,25 from a separate CHW team. Study outcomes and measure The primary outcome was a change in z-score for subscales of the BSID-III, calculated as described below. In preparation for use of the BSID-III, our team conducted a review of prior studies of psychometric testing with Guatemalan children, with the majority using previous versions of the BSID.9–11 26 Despite its technical complexity, the BSID is widely used as a gold-standard instrument worldwide.27 28 Additionally, our lead psychologist (MPG) had extensive prior experience successfully using the tool in Guatemala. For these reasons, the BSID-III was chosen.

the majority using previous versions of the BSID.9–11 26 Despite its technical complexity, the BSID is widely used as a gold-standard instrument worldwide.27 28 Additionally, our lead psychologist (MPG) had extensive prior experience successfully using the tool in Guatemala. For these reasons, the BSID-III was chosen. In preparation for testing, our team (interpreters, linguists, psychologists, anthropologists, nurses and physicians) reviewed existing BSID-III English and Spanish materials.23 29 Modifications for idiomatic Guatemalan Spanish and translations into Kaqchikel Maya were produced through group discussion and audience testing with volunteer caregivers and health workers. Forward translation, back translation, harmonisation and cognitive debriefing were used to ensure accuracy and cultural adaptation. A visual Likert scale assisted with administration of socioemotional scales (see online supplementary figure 1). A small number of vocabulary and visual stimulus items were substituted to be culturally or context appropriate (eg, sedan car substituted for pick-up truck). 10.1136/bmjpo-2018-000314.supp1Supplementary file 1

In preparation for testing, our team (interpreters, linguists, psychologists, anthropologists, nurses and physicians) reviewed existing BSID-III English and Spanish materials.23 29 Modifications for idiomatic Guatemalan Spanish and translations into Kaqchikel Maya were produced through group discussion and audience testing with volunteer caregivers and health workers. Forward translation, back translation, harmonisation and cognitive debriefing were used to ensure accuracy and cultural adaptation. A visual Likert scale assisted with administration of socioemotional scales (see online supplementary figure 1). A small number of vocabulary and visual stimulus items were substituted to be culturally or context appropriate (eg, sedan car substituted for pick-up truck). 10.1136/bmjpo-2018-000314.supp1Supplementary file 1 Nine psychologists (bilingual Spanish/English; three postgraduate and six graduate or undergraduate) who had formal training in infant/child assessments as part of their education participated in a 1-week long training in rural data collection and multicultural competencies, including active BSID-III practice sessions. Five Maya interpreters (Kaqchikel Maya/Spanish speaking; the majority with a high school teaching degree) participated in a 1-week training with active BSID-III practice sessions. Senior study psychologist (MPG), with significant professional experience in infant/child assessments, supervised BSID-III training and data collection procedures. Testing was performed by a team of one psychologist and one interpreter, supervised by MPG and the principal study coordinator (BM).

-III practice sessions. Senior study psychologist (MPG), with significant professional experience in infant/child assessments, supervised BSID-III training and data collection procedures. Testing was performed by a team of one psychologist and one interpreter, supervised by MPG and the principal study coordinator (BM). Sample size and randomisation For the primary outcome, we calculated that enrolment of 72 subjects per arm would detect a change of 0.5 SD on the BSID-III composite score, with an α of 0.05, power of 80% and 15% lost to follow-up. All subjects recruited for the larger study were allocated by simple randomisation from a computer-generated list of random numbers, no further randomisation or allocation were performed for this substudy. Statistical analyses Descriptive statistics were calculated using Stata V.13. Family poverty scores (possible score range: 0–100) were calculated as described.30 Overall FCI scores were calculated as the sum of 18 item scores.22 Differences between arms were assessed using the Student’s t-test or Wilcoxon-Mann-Whitney U test for continuous variables and the Χ2 or Fisher’s exact tests for categorical variables. Analysis was by intention-to-treat, except where loss to follow-up prevented collecting outcome data or baseline age-band BSID-III data were unavailable to calculate exit BSID-III z-scores.

nt’s t-test or Wilcoxon-Mann-Whitney U test for continuous variables and the Χ2 or Fisher’s exact tests for categorical variables. Analysis was by intention-to-treat, except where loss to follow-up prevented collecting outcome data or baseline age-band BSID-III data were unavailable to calculate exit BSID-III z-scores. Raw scores for each BSID-III subscale (cognitive, receptive and expressive language, fine and gross motor, socioemotional) were calculated as the sum of assessment items performed. BSID-III norms do not exist for Guatemala, and there are concerns about the validity of applying BSID-III norms from a different population. This is the case because, for example, some test items are most appropriate for urban, literate populations (eg, use of books in the testing process) and because some language items require adaptation to the syntactic structure of Mayan languages.31 32 Furthermore, after data collection for the study was completed, interim ethics board review requested that we not use external norms in our analysis plan. Therefore, study-internal z-scores for each BSID-III subscale were calculated based on the raw scores distribution of each age-band during baseline measurements, an approach used in other studies.33 34 These were then used to calculate individual subject z-scores for both 0 and 6 month timepoints. Since all subjects recruited were younger than 24 months, baseline BSID-III z-score data distribution was unavailable for subjects in older age bands (>25.5 months) at study exit, hence subjects older than 25.5 months were excluded from the primary outcome analysis.

l subject z-scores for both 0 and 6 month timepoints. Since all subjects recruited were younger than 24 months, baseline BSID-III z-score data distribution was unavailable for subjects in older age bands (>25.5 months) at study exit, hence subjects older than 25.5 months were excluded from the primary outcome analysis. We conducted an adjusted exploratory analysis to investigate the effect of prespecified covariates on the primary outcomes, including maternal parity and education, gender, number of under-5 children in the home, household socioeconomic status and age and length-for-age at enrolment. We constructed a hierarchical linear model for BSID-III subscales (MIXED function, Stata V.13). Non-significant covariates were removed from the model using serial likelihood ratio tests.35 The best-fit model for all BSID-III subscales was chosen using the Breusch-Pagan Langrange multiplier test. Results Participants Eligible participants were recruited in rolling fashion from August to December 2015. Final participants exited in July 2016. A total of 210 children were eligible for the substudy (104 intervention arm, 106 usual care arm). Eighty-four children in the intervention and 91 in the control arm received BSID-III evaluation at study entry. Baseline characteristics of participants in the two study arms were well balanced, except for children in the intervention arm having greater minimum dietary diversity (table 1). Table 1 Baseline demographic, clinical and BSID-III characteristics of study participants

Results Participants Eligible participants were recruited in rolling fashion from August to December 2015. Final participants exited in July 2016. A total of 210 children were eligible for the substudy (104 intervention arm, 106 usual care arm). Eighty-four children in the intervention and 91 in the control arm received BSID-III evaluation at study entry. Baseline characteristics of participants in the two study arms were well balanced, except for children in the intervention arm having greater minimum dietary diversity (table 1). Table 1 Baseline demographic, clinical and BSID-III characteristics of study participants Characteristic* Individualised education (intervention) arm (n=84) Usual care arm (n=91) P values† Maternal characteristics Age (years) 26.08±6.84 27.15±6.59 0.29 Education (years) 2 (0–3.5) 2 (0–4) 0.93 Literacy, no. (%) 44 (51.65) 47 (52.38) 0.92 Parity 2.5 (1–5) 3 (2–5) 0.43 Child characteristics Male, no. (%) 49 (58.33) 57 (62.64) 0.56 Age at BSID-III evaluation (months) 16.32±4.93 15.84±5.21 0.53 Height-for-age z-score −3.40±0.69 −3.41±0.73 0.91 Weight-for-age z-score −2.01±0.74 −1.94±0.82 0.54 Weight-for-length z-score −0.26±0.86 −0.13±0.90 0.35 Feeding practices indicators Minimum dietary diversity, no. (%) 59 (70.24) 49 (53.85) 0.03 Minimum meal frequency, no. (%) 68 (80.95) 81 (89.01) 0.13 Minimum acceptable diet, no. (%) 51 (60.71) 43 (47.25) 0.07 Household characteristics No. of children under 5 years 2 (1–2) 2 (1–2) 0.18 Family poverty score 28.33±11.64 27.35±9.17 0.53 Family care indicators score 8.59±2.63 8.47±2.15 0.74 BSID-III subscales z-scores‡ Cognitive −0.05±1.00 0.04±0.97 0.53 Receptive language −0.02±0.96 0.02±1.00 0.81 Expressive language 0.01±0.94 −0.01±1.02 0.88 Fine motor −0.15±0.94 0.14±1.00 0.06 Gross motor −0.10±0.91 0.10±1.04 0.18 Socioemotional 0.05±1.15 −0.02±0.95 0.65 *Data presented as means±SD, median (IQR) or no. (%), as appropriate.

scores‡ Cognitive −0.05±1.00 0.04±0.97 0.53 Receptive language −0.02±0.96 0.02±1.00 0.81 Expressive language 0.01±0.94 −0.01±1.02 0.88 Fine motor −0.15±0.94 0.14±1.00 0.06 Gross motor −0.10±0.91 0.10±1.04 0.18 Socioemotional 0.05±1.15 −0.02±0.95 0.65 *Data presented as means±SD, median (IQR) or no. (%), as appropriate. †Student’s t-test or Wilcoxon-Mann-Whitney U test for continuous variables, Χ2 or Fisher’s exact tests for categorical variables, as appropriate. ‡ BSID-III, Bayley Scales of Infant Development, Third Edition.

scores‡ Cognitive −0.05±1.00 0.04±0.97 0.53 Receptive language −0.02±0.96 0.02±1.00 0.81 Expressive language 0.01±0.94 −0.01±1.02 0.88 Fine motor −0.15±0.94 0.14±1.00 0.06 Gross motor −0.10±0.91 0.10±1.04 0.18 Socioemotional 0.05±1.15 −0.02±0.95 0.65 *Data presented as means±SD, median (IQR) or no. (%), as appropriate. †Student’s t-test or Wilcoxon-Mann-Whitney U test for continuous variables, Χ2 or Fisher’s exact tests for categorical variables, as appropriate. ‡ BSID-III, Bayley Scales of Infant Development, Third Edition. The cumulative incidence of loss to follow-up was 18% (n=15) in the intervention and 14% (n=13) in the usual care arm (figure 1). One child in the control arm discontinued treatment. Furthermore, 22 children in the intervention arm and 25 in the control arm with BSID-III data at study exit were older than 25.5 months. Baseline z-score data distribution was not available for this older age band. Online supplementary table 1 compares the characteristics of subjects who completed the study and both BSID-III evaluations (n=147) with those lost to follow-up or who did not complete both assessments (n=63). There were no statistically significant differences for major characteristics. Online supplementary table 2 compares the baseline characteristics of subjects excluded from the outcomes analysis due to age at study exit with those included. Subjects excluded were similar to those included, except for having lower minimum meal frequency (72% for excluded vs 88% for included subjects) and being older (21.97±1.71 vs 13.60±3.95 months). Difference in age was an expected finding, since our analysis was restricted to subjects younger than 25.5 months at study end point, since baseline BSID-III data distribution for z-scores calculation was unavailable for the older age bands.

88% for included subjects) and being older (21.97±1.71 vs 13.60±3.95 months). Difference in age was an expected finding, since our analysis was restricted to subjects younger than 25.5 months at study end point, since baseline BSID-III data distribution for z-scores calculation was unavailable for the older age bands. 10.1136/bmjpo-2018-000314.supp2Supplementary file 2 10.1136/bmjpo-2018-000314.supp3Supplementary file 3 Outcomes The analysis of primary outcomes was by intention-to-treat for subjects younger than 25.5 months at study exit in this substudy, including one child who discontinued treatment. One hundred subjects were included in the primary analysis (47 intervention arm, 53 usual care arm; figure 1). For developmental outcomes (table 2), we observed positive changes in most BSID-III subscales z-scores over the 6-month period in both study arms (median duration between measurements 189 days (IQR 182–189)). Mean change for subscales was 0.45 (95% CI 0.23 to 0.67) z-scores in the intervention arm, and 0.43 (95% CI 0.25 to 0.61) z-scores in the usual care arm. No statistically significant differences were observed between the two study arms. Table 2 Key developmental outcomes

Outcomes The analysis of primary outcomes was by intention-to-treat for subjects younger than 25.5 months at study exit in this substudy, including one child who discontinued treatment. One hundred subjects were included in the primary analysis (47 intervention arm, 53 usual care arm; figure 1). For developmental outcomes (table 2), we observed positive changes in most BSID-III subscales z-scores over the 6-month period in both study arms (median duration between measurements 189 days (IQR 182–189)). Mean change for subscales was 0.45 (95% CI 0.23 to 0.67) z-scores in the intervention arm, and 0.43 (95% CI 0.25 to 0.61) z-scores in the usual care arm. No statistically significant differences were observed between the two study arms. Table 2 Key developmental outcomes z-Scores change for BSID-III subscales*‡ Individualised education (intervention) arm (n=47) Usual care arm (n=53) Difference Cognitive 0.28 (−0.14 to 0.71) 0.38 (0.05 to 0.72) −0.10 (−0.63 to 0.43) Receptive language 0.49 (0.11 to 0.86) 0.56 (0.21 to 0.92) −0.08 (−0.58 to 0.43) Expressive language 0.69 (0.38 to 0.99) 0.63 (0.34 to 0.93) 0.05 (−0.36 to 0.48) Fine motor 0.40 (0.04 to 0.76) 0.27 (−0.09 to 0.63) 0.13 (−0.38 to 0.63) Gross motor 0.70 (0.30 to 1.10) 0.51 (0.18 to 0.85) 0.19 (−0.32 to 0.70) Socioemotional 0.20 (−0.13 to 0.53) 0.44 (0.07 to 0.81) −0.24 (−0.73 to 0.25) Mean change 0.45 (0.23 to 0.67) 0.43 (0.25 to 0.61) 0.02 (−0.25 to 0.30) *Values are mean z-score change (95% CI). ‡BSID-III, Bayley Scales of Infant Development, Third Edition.

z-Scores change for BSID-III subscales*‡ Individualised education (intervention) arm (n=47) Usual care arm (n=53) Difference Cognitive 0.28 (−0.14 to 0.71) 0.38 (0.05 to 0.72) −0.10 (−0.63 to 0.43) Receptive language 0.49 (0.11 to 0.86) 0.56 (0.21 to 0.92) −0.08 (−0.58 to 0.43) Expressive language 0.69 (0.38 to 0.99) 0.63 (0.34 to 0.93) 0.05 (−0.36 to 0.48) Fine motor 0.40 (0.04 to 0.76) 0.27 (−0.09 to 0.63) 0.13 (−0.38 to 0.63) Gross motor 0.70 (0.30 to 1.10) 0.51 (0.18 to 0.85) 0.19 (−0.32 to 0.70) Socioemotional 0.20 (−0.13 to 0.53) 0.44 (0.07 to 0.81) −0.24 (−0.73 to 0.25) Mean change 0.45 (0.23 to 0.67) 0.43 (0.25 to 0.61) 0.02 (−0.25 to 0.30) *Values are mean z-score change (95% CI). ‡BSID-III, Bayley Scales of Infant Development, Third Edition. Exploratory analyses We used a hierarchical linear regression model to estimate changes in each BSID-III subscale z-score as a function of prespecified covariates, while controlling for subject-level variation over time. Our final model included gender, age at enrolment, maternal parity and school attendance, number of children under 5 in the household and study arm. The adjusted analysis was consistent with our unadjusted primary analysis, with statistically significant improvements in developmental outcomes over the study period in all subscales (table 3), despite no difference between the study arms. Maternal school attendance was associated with greater positive expressive language and gross motor developmental changes (change at 6 months of 0.34, 95% CI 0.06 to 0.62, and 0.31, 95% CI 0.04 to 0.59 z-scores, respectively). Improvements in the cognitive subscale z-scores were more pronounced for males (0.28, 95% CI 0.03 to 0.53).

l attendance was associated with greater positive expressive language and gross motor developmental changes (change at 6 months of 0.34, 95% CI 0.06 to 0.62, and 0.31, 95% CI 0.04 to 0.59 z-scores, respectively). Improvements in the cognitive subscale z-scores were more pronounced for males (0.28, 95% CI 0.03 to 0.53). Table 3 Estimates and 95% CI from hierarchical linear mixed models for z-scores change in BSID-III subscales‡

l attendance was associated with greater positive expressive language and gross motor developmental changes (change at 6 months of 0.34, 95% CI 0.06 to 0.62, and 0.31, 95% CI 0.04 to 0.59 z-scores, respectively). Improvements in the cognitive subscale z-scores were more pronounced for males (0.28, 95% CI 0.03 to 0.53). Table 3 Estimates and 95% CI from hierarchical linear mixed models for z-scores change in BSID-III subscales‡ Covariates Cognitive (n=176) Receptive language (n=176) Expressive language (n=174) Fine motor (n=174) Gross motor (n=175) Socioemotional (n=185) Adjusted z-score change from study entry to exit 0.41* (0.09 to 0.73) 0.50** (0.18 to 0.81) 0.69** (0.42 to 0.96) 0.32* (0.01 to 0.63) 0.58** (0.27 to 0.89) 0.52** (0.22 to 0.82) Gender (0=female, 1=male) 0.28* (0.03 to 0.53) −0.07 (−0.32 to 0.18) −0.08 (−0.33 to 0.18) 0.03 (−0.21 to 0.26) −0.05 (−0.30 to 0.21) −0.06 (−0.32 to 0.19) Age at enrolment (months) 6–12 – – – – – – 13–18 0.08 (−0.23 to 0.39) 0.06 (−0.24 to 0.37) −0.13 (−0.45 to 0.18) 0.13 (−0.16 to 0.42) 0.07 (−0.24 to 0.38) −0.09 (−0.41 to 0.24) 19–24 0.22 (−0.10 to 0.53) 0.07 (−0.24 to 0.38) 0.06 (−0.26 to 0.37) 0.20 (−0.10 to 0.50) 0.20 (−0.11 to 0.52) −0.15 (−0.48 to 0.17) Maternal parity (≤2, ≥3) −0.26 (−0.05 to 0.51) −0.02 (−0.32 to 0.28) −0.09 (−0.39 to 0.22) −0.14 (−0.43 to 0.15) −0.09 (−0.40 to 0.21) −0.25 (−0.54 to 0.05) Maternal school attendance (none, some) −0.02 (−0.30 to 0.26) 0.10 (−0.17 to 0.38) 0.34* (0.06 to 0.62) 0.14 (−0.12 to 0.40) 0.31* (0.04 to 0.59) −0.05 (−0.33 to 0.24) No. of children under 5 1 – – – – – – 2 0.23 (−0.05 to 0.51) 0.02 (−0.26 to 0.30) 0.00 (−0.28 to 0.28) 0.14 (−0.12 to 0.41) 0.02 (−0.26 to 0.30) −0.25 (−0.54 to 0.05) ≥3 0.36 (−0.10 to 0.82) −0.13 (−0.59 to 0.32) 0.19 (−0.27 to 0.66) 0.08 (−0.36 to 0.51) 0.28 (−0.18 to 0.74) −0.16 (−0.63 to 0.32) Study arm (0=usual care, 1=intervention) −0.08 (−0.38 to 0.21) −0.04 (−0.32 to 0.25) 0.01 (−0.27 to 0.29) −0.27 (−0.55 to 0.00) −0.20 (−0.49 to 0.09) −0.01 (−0.30 to 0.28) Likelihood Ratio test 0.11 0.13 0.0008 0.20 0.05 0.007 *p<0.05, **p<0.01.

7 to 0.66) 0.08 (−0.36 to 0.51) 0.28 (−0.18 to 0.74) −0.16 (−0.63 to 0.32) Study arm (0=usual care, 1=intervention) −0.08 (−0.38 to 0.21) −0.04 (−0.32 to 0.25) 0.01 (−0.27 to 0.29) −0.27 (−0.55 to 0.00) −0.20 (−0.49 to 0.09) −0.01 (−0.30 to 0.28) Likelihood Ratio test 0.11 0.13 0.0008 0.20 0.05 0.007 *p<0.05, **p<0.01. ‡BSID -III, Bayley Scales of Infant Development, Third Edition. Discussion In this paper, we describe developmental outcomes from a substudy of a larger individually randomised clinical trial of a complementary feeding intervention in a rural indigenous area of Guatemala with high stunting prevalence. We found significant improvements across multiple developmental subscales over the study period for children in both the usual care (mean change of 0.43 (95% CI 0.25 to 0.61) z-scores) and the intervention (mean change of 0.45 (95% CI 0.23 to 0.67) z-scores) arms. These improvements remained highly statistically significant after controlling for important covariates (table 3) and occurred despite the larger study showing only non-significant improvements in linear growth, as previously reported.24

s) and the intervention (mean change of 0.45 (95% CI 0.23 to 0.67) z-scores) arms. These improvements remained highly statistically significant after controlling for important covariates (table 3) and occurred despite the larger study showing only non-significant improvements in linear growth, as previously reported.24 No statistically significant difference in improvements was observed between the study arms, suggesting that both usual care and intensive approaches were equally effective in promoting development. Important limitations of our study lead to two alternative explanations for this finding. First, when designing the trial, usual care was to be delivered by an existing public sector rural programme. However, widespread closures of these public services happened during study preparation.19 Therefore, MHA’s CHWs agreed to institute the usual care arm. Since these CHWs conduct activities using home visits (rather than the public sector’s facility-based approach), the quality of usual care may have been greater than anticipated, leading to better outcomes. Additionally, the number of children included in the primary analysis was underpowered to detect a difference between the study arms. This loss of power occurred for two reasons. First, an ethics board interim review request after completing data collection led to use study-internal baseline z-scores for comparison and exclusion of some older children from the analysis. Second, the difficult rural geography and lengthy time-requirements for BSID-III assessments led to more caregivers than expected declining to complete follow-up assessments. Although there were few differences in baseline characteristics between subjects included versus excluded for these two reasons (see online supplementary tables 1–2), the proportion of children included in the analysis still remains only 43% of the originally randomised sample (figure 1). This potentially limits the generalisability of our findings, especially for the older children (aged 25–30 months at study exit), who represent the largest proportion of subjects excluded from the analysis.

tion of children included in the analysis still remains only 43% of the originally randomised sample (figure 1). This potentially limits the generalisability of our findings, especially for the older children (aged 25–30 months at study exit), who represent the largest proportion of subjects excluded from the analysis. Our study also has two important strengths. First, despite extensive work on stunting in rural indigenous Maya populations, there have been few contemporary efforts to investigate developmental outcomes in this population.8–11 Our study represents, to our knowledge, the first report of the impact of a nutrition intervention on development in Maya infants. Second, despite its impact on overall study power, the use of study-internal z-scores for comparisons overcomes some concerns about the validity and reliability of the BSID-III for this indigenous population, allowing for robust internal comparisons, although the results we report here cannot be directly compared with populations from other studies.

verall study power, the use of study-internal z-scores for comparisons overcomes some concerns about the validity and reliability of the BSID-III for this indigenous population, allowing for robust internal comparisons, although the results we report here cannot be directly compared with populations from other studies. Our study contributes to the literature on complementary feeding interventions, a cornerstone of stunting prevention and treatment efforts in low-income and middle-income countries.13–15 Although multiple studies and meta-analyses have demonstrated the importance of complementary feeding interventions for promoting linear growth,13–15 there are fewer studies examining their impact on development.17 18 36 Here, we show that both usual care (lower intensity) and a higher-intensity approach to complementary feeding have important developmental effects. Our finding of no difference between higher-intensity and lower-intensity approaches contrasts with a large cluster-randomised trial in India, where a more-intensive approach showed improved developmental outcomes.37 Given our study’s low power, and the fact that complementary feeding indicators significantly improved in our larger trial,24 we believe that the final role of individualised, intensive approaches to complementary feeding in our setting is not yet settled.

re-intensive approach showed improved developmental outcomes.37 Given our study’s low power, and the fact that complementary feeding indicators significantly improved in our larger trial,24 we believe that the final role of individualised, intensive approaches to complementary feeding in our setting is not yet settled. Our study has two important implications for child nutrition policy in Guatemala. Guatemala’s indigenous Maya population has some of the poorest nutritional outcomes in the world and, despite a recent resurgence public interest in this problem,38 39 chronic political and financial instability threatens sustained public and private commitments. Our study demonstrates, for the first time, the developmental impacts that such interventions can have for Maya children at risk and will help to advance this critical national conversation. Furthermore, our study shows that even low-intensity interventions modelled on existing public policy guidelines can be of great benefit.40 Second, we demonstrate here that, despite the cultural and linguistic challenges of developmental assessments in this population, such evaluations are feasible and can show important developmental impact even when growth outcomes are equivocal.24 We call for other nutrition researchers and programme implementers in Guatemala to more routinely incorporate developmental outcomes into their planned evaluations.

mental assessments in this population, such evaluations are feasible and can show important developmental impact even when growth outcomes are equivocal.24 We call for other nutrition researchers and programme implementers in Guatemala to more routinely incorporate developmental outcomes into their planned evaluations. Future research priorities for our group include large-scale well-powered investigations of complementary and responsive feeding interventions, as well as more comprehensive, integrated nutrition and ECD interventions by CHWs in rural Guatemala. This latter point is especially important, since it is increasingly recognised that comprehensive, multisectorial interventions are most likely to generate sustained positive impact. Design and evaluation of comprehensive wrap-around interventions is also most in-line with the Nurturing Care Framework for childhood recently put forth by WHO and Unicef.41 42 In addition, we are currently planning a re-enrolment study of this trial cohort to see if further growth or developmental benefits emerge or are sustained with longer follow-up, since an inherent weakness of this study was its short follow-up time. Finally, we plan to publish in greater detail our methodology for adapting the BSID-III and related instruments to this population.

f this trial cohort to see if further growth or developmental benefits emerge or are sustained with longer follow-up, since an inherent weakness of this study was its short follow-up time. Finally, we plan to publish in greater detail our methodology for adapting the BSID-III and related instruments to this population. Supplementary Material Reviewer comments Author's manuscript The authors would like to thank the subjects, their families and communities for their participation in this study. The authors would also like to thank the interpreters at MHA and the graduate and undergraduate psychology students at Universidad del Valle de Guatemala for their collaboration performing the BSID-III assessments. Contributors: BM supervised the study, analysed and interpreted data and produced the first manuscript draft. SC cured, analysed and interpreted data. PR, ML, AG collected data and supervised BSID-III data collection. MFW and MPG supervised the study and cultural adaptation for the BSID-III. MPG and PR designed and supervised the study. No individual was given any form of payment to produce the article. All authors critically revised the manuscript and made substantial contributions to the final draft. Funding: This work was supported by Grand Challenges Canada, grant number SB-1726251050. Disclaimer: The study sponsor had no role in study design; the collection, analysis and interpretation of data; the writing of the report and the decision to submit the paper for publication.

Contributors: BM supervised the study, analysed and interpreted data and produced the first manuscript draft. SC cured, analysed and interpreted data. PR, ML, AG collected data and supervised BSID-III data collection. MFW and MPG supervised the study and cultural adaptation for the BSID-III. MPG and PR designed and supervised the study. No individual was given any form of payment to produce the article. All authors critically revised the manuscript and made substantial contributions to the final draft. Funding: This work was supported by Grand Challenges Canada, grant number SB-1726251050. Disclaimer: The study sponsor had no role in study design; the collection, analysis and interpretation of data; the writing of the report and the decision to submit the paper for publication. Competing interests: BM, MFW and PR are employees or volunteers at Wuqu’ Kawoq, Maya Health Alliance, the non-governmental organisation that provided logistical support for this study in Guatemala. SC, PR, ML, AG and MPG are doctoral students or employees of Universidad del Valle de Guatemala. The authors declare no other relationships or activities that could have influenced the submitted work. Patient consent: Not required. Ethics approval: Institutional Review Boards of the Universidad del Valle de Guatemala and Maya Health Alliance. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Study protocol, replication dataset and statistical code are available on publication at: https://doi.org/10.7910/DVN/ATJ3NW.

What is already known on this topic? Multiple outcomes can be measured in infants that receive neonatal care. It is not known which outcomes are considered important by former neonatal patients, parents and healthcare professionals, or whether these differ between groups. What this study hopes to add? The predominant outcomes identified by parents, former patients and health professionals related to holistic concepts (such as ‘normality’). Significant differences were identified in outcomes discussed by parents, patients and health professionals. Differences in neonatal outcomes prioritised by parents, patients and health professionals should be recognised when planning research. Introduction In high-resource settings approximately 1 in 10 babies will require care in a neonatal unit.1 Conditions such as preterm birth affect patients’ long-term outcomes: consequences include cardiovascular disease in adulthood,2 neurosensory impairment,3 respiratory disease4 and lower rates of employment and marriage.5 Infants born more prematurely tend to have worse outcomes.6 As neonatal survival for babies of all gestational ages improves long-term outcomes become more important.

actors including behavioural factors interact to influence well-being. Additionally, given that many such behaviours are strongly socially patterned, studies have thus far paid little attention to confounding by socioeconomic position and issues relating to the clustering of behaviours and well-being within localities. Policy initiatives to improve well-being among young people have largely focused on cognitive and psychological factors related to resilience to adversity, and have paid little attention to the contribution of non-psychological modifiable factors, such as other lifestyle behaviours. Understanding the potential contribution of modifiable behavioural factors to adolescent well-being may inform different strategies to improve young people’s well-being. We used a very large recent nationally representative and population-based survey of English adolescents aged 15 years to examine the contribution of individual-level modifiable behaviours to well-being, including potentially protective (sleep, reading and physical activity) and risk behaviours (substance use, unhealthy eating habits and excessive screen time). Our objective was to identify modifiable behavioural factors for mental well-being in boys and girls using an adolescent-specific measure and accounting for deprivation, ethnicity and clustering within local authorities (LAs).

s: consequences include cardiovascular disease in adulthood,2 neurosensory impairment,3 respiratory disease4 and lower rates of employment and marriage.5 Infants born more prematurely tend to have worse outcomes.6 As neonatal survival for babies of all gestational ages improves long-term outcomes become more important. An outcome is the measured effect that illness or treatment has on an individual.7 Parents and patients are rarely involved in outcome selection in paediatric research.8 Poor outcome selection causes research waste9: research produced is not relevant to patients’ lives. Neonatal care, and the underpinning research, should focus on outcomes important to those it affects most: former neonatal patients, parents and healthcare professionals.9 10 Identifying these outcomes is crucial to ensure research is relevant and efficient.9 11 Qualitative research provides a rich description of complex phenomena such as neonatal care.12 One commonly used approach to identify outcomes of importance to stakeholders is primary qualitative research. Considerable qualitative research exploring how parents and health professionals perceive neonatal care has been conducted previously13 14; therefore, by systematically reviewing published qualitative research it is possible to map the outcomes discussed by different groups. This review does not include all research on how stakeholders perceive neonatal care: it is focused on how former patients, parents and health professionals perceive the outcomes of this care.

e, by systematically reviewing published qualitative research it is possible to map the outcomes discussed by different groups. This review does not include all research on how stakeholders perceive neonatal care: it is focused on how former patients, parents and health professionals perceive the outcomes of this care. In this study we aimed to map the range of outcomes identified in qualitative literature by different stakeholder groups: parents, ex-neonatal patients and healthcare professionals. We also wanted to test the hypotheses that stakeholder groups prioritise outcomes differently, and that outcomes identified differ by infant gestational age category. This work is a component of a wider programme to compile a core outcomes set for neonatology.15 A core outcomes set is an agreed collection of important outcomes identified through robust consensus methods by all key stakeholder groups.7 The results of this study will be combined with the results of a systematic review of outcomes reported in clinical trials.16 These will be used as the starting point for the consensus process to determine a core outcomes set.15

ortant outcomes identified through robust consensus methods by all key stakeholder groups.7 The results of this study will be combined with the results of a systematic review of outcomes reported in clinical trials.16 These will be used as the starting point for the consensus process to determine a core outcomes set.15 Methods We registered this systematic review prospectively on PROSPERO (prospective register of systematic reviews): CRD42016037874.17 We conducted the review according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.18 We searched the following databases: Medical Literature Analysis and Retrieval System Online (MEDLINE), Cumulative Index to Nursing and Allied Health Literature (CINAHL), Excerpta Medica Database (EMBASE), Psychological Information Database (PsycINFO) and Applied Social Sciences Index and Abstracts (ASSIA). Qualitative or mixed methods studies were included if they contained outcomes identified by stakeholders in the context of babies admitted to a neonatal unit. Full inclusion and exclusion criteria are listed in online supplementary eTable 1. We considered all studies published from 1 January 1997 to 1 January 2017 in a peer review journal in all languages (where necessary a translation was obtained). The databases were last searched on 14 February 2017. The search strategy used for MEDLINE is described in online supplementary eFigure 1. The terms derived from this search strategy were translated to other databases. 10.1136/bmjpo-2018-000343.supp1Supplementary data

Methods We registered this systematic review prospectively on PROSPERO (prospective register of systematic reviews): CRD42016037874.17 We conducted the review according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.18 We searched the following databases: Medical Literature Analysis and Retrieval System Online (MEDLINE), Cumulative Index to Nursing and Allied Health Literature (CINAHL), Excerpta Medica Database (EMBASE), Psychological Information Database (PsycINFO) and Applied Social Sciences Index and Abstracts (ASSIA). Qualitative or mixed methods studies were included if they contained outcomes identified by stakeholders in the context of babies admitted to a neonatal unit. Full inclusion and exclusion criteria are listed in online supplementary eTable 1. We considered all studies published from 1 January 1997 to 1 January 2017 in a peer review journal in all languages (where necessary a translation was obtained). The databases were last searched on 14 February 2017. The search strategy used for MEDLINE is described in online supplementary eFigure 1. The terms derived from this search strategy were translated to other databases. 10.1136/bmjpo-2018-000343.supp1Supplementary data All identified papers were screened by title and abstract and then by full text. After double-screening a sample of papers and agreeing criteria all screening was completed by one researcher (JW). For quality assurance, a second researcher screened a random 10% sample of abstracts and titles (CG). Agreement between reviewers was assessed by Cohen’s kappa coefficient.19

stract and then by full text. After double-screening a sample of papers and agreeing criteria all screening was completed by one researcher (JW). For quality assurance, a second researcher screened a random 10% sample of abstracts and titles (CG). Agreement between reviewers was assessed by Cohen’s kappa coefficient.19 After screening all papers were coded independently by two researchers (JW and CG or GB) using Eppi-Reviewer V.4 software.20 Any disagreement was resolved by a third researcher (CG or GB). Data on study design, stakeholder demographics, infant birth characteristics and verbatim text relating to neonatal care outcomes were extracted and stored. Quality assessment of individual studies was not undertaken as it is a controversial area of uncertain value in relation to qualitative research.21

her (CG or GB). Data on study design, stakeholder demographics, infant birth characteristics and verbatim text relating to neonatal care outcomes were extracted and stored. Quality assessment of individual studies was not undertaken as it is a controversial area of uncertain value in relation to qualitative research.21 All outcomes were grouped according to a previously defined framework of organ systems22 using the following domains: cardiovascular, respiratory, gastrointestinal, neurological, genitourinary, infection, skin and development. All three reviewers jointly refined this framework using methods incorporating thematic analysis.23–25 Where narrative data did not fit clearly into the domains, dialogue between all reviewers was used to develop new domains. Outcome domains were thematically analysed to develop higher order categories. A new hierarchy was developed to group outcomes because established hierarchies either did not relate well to neonatal care26–28 or missed key concepts.7 This outcome hierarchy is described in box 1.Box 1 An example of an outcome hierarchy Text extracts to identify or infer a result of clinical care, the ‘outcome’ such as ‘Bonding with parents’. Similar ‘outcomes’ were grouped into thematically linked ‘domains’ such as ‘Relationships with others’. ‘Outcome domains’ relating to similar concepts were grouped into ‘categories’ such as ‘Social’. We did not address the ways in which an outcome was measured. For example, the ‘outcome’ ‘Parental bonding’ could be measured using parent-reported scores or an external assessment.

Similar ‘outcomes’ were grouped into thematically linked ‘domains’ such as ‘Relationships with others’. ‘Outcome domains’ relating to similar concepts were grouped into ‘categories’ such as ‘Social’. We did not address the ways in which an outcome was measured. For example, the ‘outcome’ ‘Parental bonding’ could be measured using parent-reported scores or an external assessment. We analysed whether outcomes identified differed by stakeholder groups and by infant gestational age category (using WHO definitions of prematurity).29 We used permutation testing30 to test for an association between the frequency that outcomes in different domains were identified and the stakeholder group involved. We performed 5000 replications to generate the distribution of the test statistic under the null hypothesis of no association, and compared our results with this distribution. We performed a similar analysis to test for an association between infant gestational age and frequency of outcome reporting. If a significant association was found we explored this further in a post hoc analysis to identify where the observed results differed most from the frequencies expected under the hypothesis of no association established by the permutation analysis. Results Database searches produced 1130 results which were screened and assessed for eligibility (figure 1). After applying inclusion and exclusion criteria 62 studies containing the views of 4100 stakeholders were analysed. Agreement between reviewers was high (Cohen’s kappa coefficient=0.81).19 Figure 1 PRISMA flowchart of study selection.

Results Database searches produced 1130 results which were screened and assessed for eligibility (figure 1). After applying inclusion and exclusion criteria 62 studies containing the views of 4100 stakeholders were analysed. Agreement between reviewers was high (Cohen’s kappa coefficient=0.81).19 Figure 1 PRISMA flowchart of study selection. The 62 included studies reported data from 15 countries; 9 related to full-term infants, 31 to preterm infants (born <37 weeks’ gestational age) and 20 to extremely preterm infants (born <28 weeks’ gestational age). A range of methodologies was used including direct observation (13 studies) and individual (25 studies) or group interviews (13 studies). Questionnaires were used in 21 studies, two of which were Delphi processes. Included studies are described in online supplementary eTable 2. Included studies involved over 4100 participants. Parents were the most frequently involved stakeholder group (1969 parents in 40 studies; 65%); former neonatal patients were less commonly included (368 patients in 5 studies; 8%). Nurses and midwives were the professional group involved most often (1096 involved in 24 studies; 39%). Three hundred and sixteen doctors were involved in 18 studies (29%). We also identified 351 additional participants consisting of other family members, teachers, social workers and allied health professionals. In many studies, particularly those employing observation of clinical practice, the total number of research participants was not recorded.

tors were involved in 18 studies (29%). We also identified 351 additional participants consisting of other family members, teachers, social workers and allied health professionals. In many studies, particularly those employing observation of clinical practice, the total number of research participants was not recorded. One hundred and forty-six distinct outcomes were extracted from the included studies. Fifty-eight outcomes related to organ systems within the original framework; we were unable to categorise 88 outcomes within the original framework. The final framework is shown in table 1. An example of the thematic analysis leading to the expanded framework is shown in box 2. Table 1 Final outcome framework Outcome domain categories Outcome domains Organ system outcomes Cardiovascular Respiratory Gastrointestinal Neurological Genitourinary Infection Skin Developmental Holistic outcomes Survival Growth Pain Suffering Normality Other outcomes Parent-focused outcomes Parental support Other outcomes Healthcare delivery outcomes Healthcare workers—knowledge and competence Healthcare workers—Communication Other outcomes Economic outcomes Healthcare utilisation Other outcomes Social outcomes Psychiatric outcomes Relationships with others Other outcomes Outcome domain categories and outcome domains added to the original framework marked in italics. Box 2 Example of framework synthesis related to the outcome of ‘Normality’. Thematic analysis of verbatim extracts identified a recurring theme ‘The mother also worried that…Lisa would not have a normal life.’ 41

Relationships with others Other outcomes Outcome domain categories and outcome domains added to the original framework marked in italics. Box 2 Example of framework synthesis related to the outcome of ‘Normality’. Thematic analysis of verbatim extracts identified a recurring theme ‘The mother also worried that…Lisa would not have a normal life.’ 41 ‘Being reassured that he was on line for how old he was…Just reassurance he was doing well.’ 42 ‘Finally, a mother called it a developmental land-mark when an older sister dared show her irritation towards her little brother, ‘no longer treating him as if he were made of glass.’ 43 From this and similar text the outcome of ‘Normality’ was derived by thematic analysis. It did not fit within the existing outcome hierarchy but was reported extensively, so a new domain was added to the framework again called ‘Normality’. This outcome domain relating to the overall status of the infant was similar to outcome domains like ‘survival’, ‘vitality’ and ‘growth’, so these domains were grouped together as an outcome domain category called: ‘Holistic outcomes’. The full inventory of outcomes is listed in online supplementary eTable 3. A table of all outcomes in each study (with verbatim text extracts) is shown in online supplementary eTable 4.

From this and similar text the outcome of ‘Normality’ was derived by thematic analysis. It did not fit within the existing outcome hierarchy but was reported extensively, so a new domain was added to the framework again called ‘Normality’. This outcome domain relating to the overall status of the infant was similar to outcome domains like ‘survival’, ‘vitality’ and ‘growth’, so these domains were grouped together as an outcome domain category called: ‘Holistic outcomes’. The full inventory of outcomes is listed in online supplementary eTable 3. A table of all outcomes in each study (with verbatim text extracts) is shown in online supplementary eTable 4. Outcomes were identified relating to all of the organ systems included in the original framework and assigned to an organ system outcome domain category (table 2). The organ system outcome domains most frequently discussed at the study level were ‘development’ (32 studies; 52%) and ‘gastrointestinal’ (24 studies, 39%). The individual organ system outcomes most frequently discussed were ‘language disorders’ (8 studies, 13%), ‘visual impairment’ (7 studies, 11%) and ‘breast feeding’ (7 studies, 11%). Table 2 Organ system outcomes Organ system outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Developmental 32 Ability to walk 3 ‘He walked four, my mother never forgot, she says it was a miracle of God.’ 44

Outcomes were identified relating to all of the organ systems included in the original framework and assigned to an organ system outcome domain category (table 2). The organ system outcome domains most frequently discussed at the study level were ‘development’ (32 studies; 52%) and ‘gastrointestinal’ (24 studies, 39%). The individual organ system outcomes most frequently discussed were ‘language disorders’ (8 studies, 13%), ‘visual impairment’ (7 studies, 11%) and ‘breast feeding’ (7 studies, 11%). Table 2 Organ system outcomes Organ system outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Developmental 32 Ability to walk 3 ‘He walked four, my mother never forgot, she says it was a miracle of God.’ 44 Difficulties with activities of daily living 4 ‘A lot of times I have to put myself in certain positions to do things, like opening a jar—I have to squeeze it in between my arms instead of gripping it with my hand.’ 45 Hearing impairment 5 ‘I told the parents that he will never be able to see, to hear and I will get more data to show them how bad things are.’ 46 Issues of development and motor skills 5 ‘We both looked at my child, research, experience and what I could expect.’ 47 Language disorders 8 ‘I also had a hard time learning to talk.’ 48 Social difficulties 2 Social communication is difficult because of his hearing and speech problems, and he is described as having a few friends and no experience in dating. 49

Issues of development and motor skills 5 ‘We both looked at my child, research, experience and what I could expect.’ 47 Language disorders 8 ‘I also had a hard time learning to talk.’ 48 Social difficulties 2 Social communication is difficult because of his hearing and speech problems, and he is described as having a few friends and no experience in dating. 49 Visual impairment 7 At the time of the interviews, the only major sequel was in one child with seriously impaired vision. 43 Other outcomes only in 1 paper Ability to feed themselves: ability to undertake sport: need for physical therapy: normal hearing: retinopathy of prematurity Gastrointestinal 24 Breast feeding 7 ‘I fully breastfed for 4 months—100%—and I am so proud of it.’ 47 Choice of milk for feeding 2 ‘It’s like they [scn providers] didn’t inform us when they were trying to feed my daughter [formula].’ 50 Feeding difficulties 5 ‘We kept on saying to the staff on neonatal unit that it was only Gray’s feeding that was stopping him from going home, everything else wasfine.’51 Feeding practices 2 Mothers had difficulty understanding these instructions and seemed hesitant to liberalise their infant’s intake after discharge. 52 Initiating enteral feeds 2 ‘MEF [minimal enteral feeds] should be initiated in first 2–3 days of life as long as the baby is stable.’53 Oral feeding 3 ‘[The] very first time [feeding the baby] was just great, to tell you thetruth.’54

Feeding practices 2 Mothers had difficulty understanding these instructions and seemed hesitant to liberalise their infant’s intake after discharge. 52 Initiating enteral feeds 2 ‘MEF [minimal enteral feeds] should be initiated in first 2–3 days of life as long as the baby is stable.’53 Oral feeding 3 ‘[The] very first time [feeding the baby] was just great, to tell you thetruth.’54 Other outcomes only in 1 paper Choking during feeding: eating disorder: fistulas: frequency of defecation: liver failure: necrotising enterocolitis: nutritional intake: other gastrointestinal malformations: regurgitation: short gut syndrome Respiratory 12 Frequent respiratory illnesses 2 ‘There were lots of masks and nebulisers during thoseyears.’43 Mechanical ventilation 5 Over 30% of all infant descriptions were about babies who had tracheostomies and were unable to be weaned off a ventilator. 55 Oxygen dependence 5 ‘My babies did not fit into the criteria for going home early due to one of the twins still being dependent on oxygen.’ 51 Other outcomes only in 1 paper Asthma: breathlessness: chronic lung disease: excessive secretions: nasal congestion: pneumothorax Neurological 11 Brain damage (not further specified) 2 ‘Brain injury is laden with a lot more emotions and moral concerns for sure.’ 56 Neurological symptoms 2 ‘Can’t feel some—my left—this is numb right here.’ 57 Seizures 2 ‘I explained this to the doctor. And he was the one that said it could possibly be seizures. So we’re going to take him in and have him tested.’ 58

Other outcomes only in 1 paper Asthma: breathlessness: chronic lung disease: excessive secretions: nasal congestion: pneumothorax Neurological 11 Brain damage (not further specified) 2 ‘Brain injury is laden with a lot more emotions and moral concerns for sure.’ 56 Neurological symptoms 2 ‘Can’t feel some—my left—this is numb right here.’ 57 Seizures 2 ‘I explained this to the doctor. And he was the one that said it could possibly be seizures. So we’re going to take him in and have him tested.’ 58 Significant IVH 2 ‘Although she has a grade IV bleed, the resident says that she moves and looks around, and he thinks the odds are quite good.’ 46 Sleep disorders 4 Subsequent to an account of the son’s disturbed sleep at night for several months after discharge, which was an enormous challenge to the couple.43 Other outcomes only in 1 paper Neurological care Surgical 5 Appearance of scars 2 “I do not like the scar on my belly […]; I was at the beach and everyone kept staring at me like ‘That is a big scar’!” 48 Need for multiple operations 2 The mother also worried that there would be more surgeries.41 Other outcomes only in 1 paper Care for surgical babies: need for ileostomy Infection 5 Sepsis 3 Decrease bloodstream infections selected as key performance indicator 59 Other outcomes only in 1 paper Prevention of infection: susceptibility to infection Skin 4 Appearance of scars 2 In addition, hospitalisation and different interventions in their first days of life have left marks on their bodies. 48

Other outcomes only in 1 paper Care for surgical babies: need for ileostomy Infection 5 Sepsis 3 Decrease bloodstream infections selected as key performance indicator 59 Other outcomes only in 1 paper Prevention of infection: susceptibility to infection Skin 4 Appearance of scars 2 In addition, hospitalisation and different interventions in their first days of life have left marks on their bodies. 48 Other outcomes only in 1 paper Burns: extravasation injuries: pressure sores: skin care Cardiovascular 1 Other outcomes only in 1 paper Hypotension: presence of patent ductus arteriosus Genitourinary 1 Other outcomes only in 1 paper Urological disorders IVH, intraventricular haemorrhage. The majority of outcomes identified did not relate to individual organ systems. Some related to the overall status of the infant and were assigned to a holistic outcome domain category (table 3). Other domains related to the effects experiencing neonatal care has on parents; these were assigned to a ‘Parent focused’ outcome domain category (table 4). Another group of domains related to the neonatal care delivered; these were assigned to a ‘Healthcare delivery’ outcome domain category (table 5). A group of domains was identified relating to the cost of neonatal care; these were assigned to an ‘Economic’ outcome category (table 6). Finally, a group of outcome domains was identified relating to the relationships neonatal patients develop with others; these were assigned to a ‘Social’ outcome domain category (table 7). Table 3 Holistic outcomes

The majority of outcomes identified did not relate to individual organ systems. Some related to the overall status of the infant and were assigned to a holistic outcome domain category (table 3). Other domains related to the effects experiencing neonatal care has on parents; these were assigned to a ‘Parent focused’ outcome domain category (table 4). Another group of domains related to the neonatal care delivered; these were assigned to a ‘Healthcare delivery’ outcome domain category (table 5). A group of domains was identified relating to the cost of neonatal care; these were assigned to an ‘Economic’ outcome category (table 6). Finally, a group of outcome domains was identified relating to the relationships neonatal patients develop with others; these were assigned to a ‘Social’ outcome domain category (table 7). Table 3 Holistic outcomes Holistic outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Normality 22 Ability to lead a normal life 2 ‘The mother also worried that… Lisa would not have a normal life.’ 41 Normality 16 “A major focus for parents was seeking information that told them that what was happening was ‘normal’ and that everything was going to be ‘fine’.”60 Other outcomes only in 1 paper Being treated normally: inability to create a normal life: normal health: thriving Suffering 15 Comfort 4 ‘You can almost feel what it’s like in the incubator, lying on the lambskin, that it’s how I would want to have laid and… Well, it looks very comfortable.’61

Normality 16 “A major focus for parents was seeking information that told them that what was happening was ‘normal’ and that everything was going to be ‘fine’.”60 Other outcomes only in 1 paper Being treated normally: inability to create a normal life: normal health: thriving Suffering 15 Comfort 4 ‘You can almost feel what it’s like in the incubator, lying on the lambskin, that it’s how I would want to have laid and… Well, it looks very comfortable.’61 Suffering 9 ‘This infant’s short life was never comfortable…I frequently felt we were torturing the child just doing daily care.’ 55 Other outcomes only in 1 paper Ex-patients’ separation from their suffering: symptom control Survival 14 Survival 11 ‘It hurts. I didn’t know, I didn’t know if they were going to make it or not.’58 Survival with disability 3 ‘It isn’t up to us to say what is quality of life, because parents might think that even if the child was disabled, that it was better than not having a child at all.’62 Survival without disability 4 ‘And afterwards you are worried about how they are going to survive. If they would have impairments, and so on.’ 43 Growth 8 Growth 8 ‘She was born so early, it’s nice to see that she’s finally catching up with how she’s growing.’42 Pain 7 Pain 4 ‘It like hurts when you wake up in the morning.’57 Pain management 2 Research priorities identified: identifying effective interventions to prevent or reduce pain or stress 63

Survival without disability 4 ‘And afterwards you are worried about how they are going to survive. If they would have impairments, and so on.’ 43 Growth 8 Growth 8 ‘She was born so early, it’s nice to see that she’s finally catching up with how she’s growing.’42 Pain 7 Pain 4 ‘It like hurts when you wake up in the morning.’57 Pain management 2 Research priorities identified: identifying effective interventions to prevent or reduce pain or stress 63 Other outcomes only in 1 paper Chronic pain Other outcomes Overall health state 2 ‘We try to think of the whole life consequence. We talk about, you know, strength and cognitive capacity, but also life and communication and feeding yourself and getting around.’ 56 Vitality 2 ‘The doctor said that, whatever we do, however good we are, it is mostly up to the infant himself. No matter how small they are, they can have something within themselves.’64 Physical appearance 7 Both mothers and fathers found their infant’s appearance and behaviour to be the stressors with the most impact. 65 Other outcomes only in 1 paper Physiological stability Table 4 Parent-focused outcomes Parent-focused outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Parental support 30 Coping with maternal illness 5 One nurse spoke of her belief that mothers could be diagnosed with depressive conditions. 62 Culture differences 2 Three families felt strongly that their stress derived from differences in the medical management approaches between the USA and their homeland. 65

Parent-focused outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Parental support 30 Coping with maternal illness 5 One nurse spoke of her belief that mothers could be diagnosed with depressive conditions. 62 Culture differences 2 Three families felt strongly that their stress derived from differences in the medical management approaches between the USA and their homeland. 65 Parental ability to work 2 They liked being back at work because it helped occupy their minds, but they reported being exhausted. 66 Parental competence 4 ‘We learned everything we needed and knew what we had to do, I was quite comfortable when we went home.’ 47 Parental involvement 10 ‘During our baby’s stay in the neonatal unit both myself and Peter were fully involved in our son’s care and did most of the caring such as nappy changing and NGT feeds.’ 51 Support from family and friends 5 ‘My mother in law and my mother both would watch my older daughter that first year quite a bit while I would take my daughter to therapy.’45 Support from fathers 2 “Fathers ranged from being very supportive, ‘[we’re] in this together,’ to being deterrents or completely absent.” 66 Support from healthcare professionals 6 ‘The nursing staff, the doctors…they really know what they’re doing…not only medically, but dealing with us personally…that helped a lot.’ 67

Support from fathers 2 “Fathers ranged from being very supportive, ‘[we’re] in this together,’ to being deterrents or completely absent.” 66 Support from healthcare professionals 6 ‘The nursing staff, the doctors…they really know what they’re doing…not only medically, but dealing with us personally…that helped a lot.’ 67 Other outcomes only in 1 paper Balancing caring for themselves and their baby: barriers to parental involvement: care provided close to home: maintaining hope: online support: preparation for NICU admission: support from faith Other outcomes Long-term effects on parents 2 We should be looking at Postnatal Depression after the baby goes home… Once they actually get a baby home, that’s when the reality sets in. All the triggers are there. 62 Other outcomes only in 1 paper Support beyond NICU: parental perception of uncertainty NGT, nasogastric tube; NICU, neonatal intensive care unit. Table 5 Healthcare delivery outcomes Healthcare delivery outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Healthcare workers—communication 30 Communicating in challenging settings 10 When they arrived at hospital…some had difficult conversations with clinical staff…NICUs commonly set boundaries around the care that they offer. 68 Communicating information effectively 7 Other parents experienced stress from unknown medical terminology. 65

Healthcare delivery outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Healthcare workers—communication 30 Communicating in challenging settings 10 When they arrived at hospital…some had difficult conversations with clinical staff…NICUs commonly set boundaries around the care that they offer. 68 Communicating information effectively 7 Other parents experienced stress from unknown medical terminology. 65 Communication about discharge 3 Parents/caregivers may be inadequately prepared for home care and management of fragile neonates due to a lack of consistent and early communication. 69 Communication with parents 2 ‘When you’re talking to parents while you’re doing cares and everything, you’re not really talking to them, … you’re having a vague conversation across the room.’ 70 Developing a caring relationship 5 As the providers gave support to families, families also were described as supporting the staff. 55 Keeping parents informed 7 ‘I asked so many questions and read all the charts every day, and I probably angered them. Squeaky wheel gets the oil, as they say.’ 50 Treating parents with respect 3 ‘I got yelled at by a nurse at [the scn] for rubbing my son’s foot [even though that was okay at the nicu].’ 50

Developing a caring relationship 5 As the providers gave support to families, families also were described as supporting the staff. 55 Keeping parents informed 7 ‘I asked so many questions and read all the charts every day, and I probably angered them. Squeaky wheel gets the oil, as they say.’ 50 Treating parents with respect 3 ‘I got yelled at by a nurse at [the scn] for rubbing my son’s foot [even though that was okay at the nicu].’ 50 Other outcomes only in 1 paper Allowing time for conversation: awareness of parental views: candour: communication with ex-neonatal patients: language barrier: using aids to communication Healthcare workers—knowledge and competence 23 Consistency of decisions 6 ‘Everybody had a different point of view but they were opinions, not facts. So that was huge, don’t even get me started on that, that was just a nightmare.’ 60 Ethical decision-making 5 “…but when you actually mention this, say, ‘Well, in fact you know, we don’t really know what’s the best treatment,’ it is a delicate moment.” 71 Healthcare professionals’ behaviour 5 ‘It wasn’t a nurse related conversation it was just a casual conversation… Like I felt a bit [sic] she wasn’t their priority.’ 60 Healthcare professional competence 7 Most of the parents recalled specific incidents that they perceived as poor medical care; typically, these incidents involved technical procedures or medical knowledge. 72 Identifying who is responsible for care 3 ‘Sometimes we’re not real clear who to follow up with.’ 50

Healthcare professionals’ behaviour 5 ‘It wasn’t a nurse related conversation it was just a casual conversation… Like I felt a bit [sic] she wasn’t their priority.’ 60 Healthcare professional competence 7 Most of the parents recalled specific incidents that they perceived as poor medical care; typically, these incidents involved technical procedures or medical knowledge. 72 Identifying who is responsible for care 3 ‘Sometimes we’re not real clear who to follow up with.’ 50 Staffing levels 2 It was especially helpful for the parents to see so many nurses and physicians in the NICU. 73 Other outcomes only in 1 paper Expertise in palliative care: medical errors: staff insecurity Other outcomes Iatrogenic harm 3 ‘There are potential toxicities, very real toxicities associated with it.’ 71 Inclusion in research 2 Parents were often interested in the research, and some would have liked more contact and information than they actually received. 68 NICU, neonatal intensive care unit. Table 6 Economic outcomes Economic outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Healthcare utilisation 15 Frequent appointments 2 ‘I felt left out, I was always missing school because I had to go to the hospital for check-ups.’ 48 Frequent readmissions 4 The prolonged hospitalisations experienced by children with BPD and the frequent interactions of families with medical personnel may result in increased access and opportunities for services for parents of children with BPD. 74

Economic outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Healthcare utilisation 15 Frequent appointments 2 ‘I felt left out, I was always missing school because I had to go to the hospital for check-ups.’ 48 Frequent readmissions 4 The prolonged hospitalisations experienced by children with BPD and the frequent interactions of families with medical personnel may result in increased access and opportunities for services for parents of children with BPD. 74 Inappropriate treatments 2 Community providers…may lack the required knowledge and skills to manage complex infants, leading to suboptimal office-based care and perceived overutilisation of the emergency system. 69 Need for frequent treatments 3 ‘There were lots of masks and nebulisers during those years.’ 43 Need for lifelong care 3 ‘When the outcome is disastrous they just expect parents to take home severely handicapped babies and deal with life-long problems.’ 75 Recurrent sickness 1 ‘We’ve only put him with other children for the past month. The biggest worry right now is when he is going to get sick.’ 58 Other outcomes Duration of admission 2 Decrease length of stay selected as key performance indicator 59 Healthcare resources 3 Although respondents frequently discussed the emotional toll to all concerned, the monetary cost of long-term stays was very rarely (<1%) mentioned. 55 BPD, borderline personality disorder. Table 7 Social outcomes

Other outcomes Duration of admission 2 Decrease length of stay selected as key performance indicator 59 Healthcare resources 3 Although respondents frequently discussed the emotional toll to all concerned, the monetary cost of long-term stays was very rarely (<1%) mentioned. 55 BPD, borderline personality disorder. Table 7 Social outcomes Social outcome domains Number of studies discussing outcome domain (n=62) Outcome Number of studies discussing outcome (n=62) Verbatim text extract Relationships with others 19 Bonding with family and friends 3 ‘The only thing we might have done…some of our closest friends…it would have been nice to have them there as well.’ 67 Bonding with parents 8 ‘I find it a great joy when the mums do hold the baby against their chest.’ 76 Effects on family and friends 7 Almost all parents acknowledged the emotional adjustment of other family members in response to raising a child with physical impairment. 45 Family resources 2 Three families felt overwhelmed by a lack of resources (especially in the area of family support). 65 Peer acceptance 2 I’ve had 4 year-olds tell me the other kids don’t want to play with them cause they have a dumb arm. 57 Other outcomes only in 1 paper Childhood happiness: overprotective parent–child relationship: psychological coping Psychiatric 7 Need for educational support 7 The patient is at an age-appropriate grade level but attends resource classes in math and achieves only average grades in other areas. 49 Psychiatric disorder 3 The mother is very focused on the boys' physical and emotional symptoms. 49

Other outcomes only in 1 paper Childhood happiness: overprotective parent–child relationship: psychological coping Psychiatric 7 Need for educational support 7 The patient is at an age-appropriate grade level but attends resource classes in math and achieves only average grades in other areas. 49 Psychiatric disorder 3 The mother is very focused on the boys' physical and emotional symptoms. 49 Other outcomes only in 1 paper Autism: behavioural disturbances: dyslexia: mood disorders Other outcomes Other outcomes only in 1 paper Schooling: self-identifying as premature From these outcome domains the most frequently discussed at study level were ‘parental support’ (30 studies, 48%) and ‘healthcare workers—communication’ (30 studies, 48%). The individual outcomes most frequently discussed were ‘normality’ (16 studies, 26%) and ‘survival’ (11 studies, 18%). Permutation test analysis showed a statistically significant association (p=0.037) between different stakeholder groups and outcome categories identified (online supplementary eFigure 3). The frequency with which patients discussed the outcomes was most divergent from the other groups. In particular, patients discussed outcomes relating to the genitourinary, surgical, developmental and pain outcome domains more than would be expected by chance. We found no statistically significant association (p=0.114) between gestational age and outcome categories (online supplementary eFigure 2).

Permutation test analysis showed a statistically significant association (p=0.037) between different stakeholder groups and outcome categories identified (online supplementary eFigure 3). The frequency with which patients discussed the outcomes was most divergent from the other groups. In particular, patients discussed outcomes relating to the genitourinary, surgical, developmental and pain outcome domains more than would be expected by chance. We found no statistically significant association (p=0.114) between gestational age and outcome categories (online supplementary eFigure 2). Discussion We have systematically reviewed and synthesised the outcomes reported in qualitative research by those with lived experience of neonatal care: patients, parents and healthcare professionals. We show that the patterns of outcomes discussed by former neonatal patients, parents and healthcare professionals are different. This is in keeping with previous single-centre research31 and case reports.32 This indicates that healthcare professionals should consider whether the outcomes they discuss align with patients and parents’ concerns.33 Acceptance of the differences shown should form part of the process of shared decision-making in clinical practice.34 Poor outcome selection is also a known problem in paediatric research,8 35 involving patients and parents will help   reduce research waste.36 37

hey discuss align with patients and parents’ concerns.33 Acceptance of the differences shown should form part of the process of shared decision-making in clinical practice.34 Poor outcome selection is also a known problem in paediatric research,8 35 involving patients and parents will help   reduce research waste.36 37 The outcomes identified extend beyond the organ system-specific outcomes commonly reported in clinical trials and include global concepts such as ‘normality’ of the child in later life, the impact on an infant’s family and the healthcare team, financial and time costs and how patients interact with wider society. Our findings are in keeping with observational studies illustrating the wide-reaching consequences of neonatal illness.38–40 Another feature of the outcomes identified is that rather than relating to a specific diagnosis or disease many reflect the global status of the child. Diagnoses like necrotising enterocolitis or retinopathy of prematurity were mentioned less frequently than their consequences, such as feeding difficulties or visual impairment. In general, the outcomes identified indicate that pathological processes and diagnoses are less relevant to patients and parents than the effects they have on day-to-day life. Priority should be given to identifying efficient ways of measuring more global outcomes of neonatal conditions throughout childhood and later life, for example, through robust linkage of neonatal data with education databases.

ses are less relevant to patients and parents than the effects they have on day-to-day life. Priority should be given to identifying efficient ways of measuring more global outcomes of neonatal conditions throughout childhood and later life, for example, through robust linkage of neonatal data with education databases. This more holistic approach should extend to how babies are categorised. Our work included an undoubtedly heterogeneous population, but this was driven by discussions with former neonatal patients and parents at the planning stages of this project. They strongly stated that ‘a sick baby is a sick baby’ regardless of birth weight or gestational age: a statement that is supported by our finding that there was no significant difference in how frequently outcomes were discussed in relation to babies of differing gestational ages. Splitting research populations by arbitrary landmarks not recognised by parents or former patients32 may be a source of research heterogeneity.

atement that is supported by our finding that there was no significant difference in how frequently outcomes were discussed in relation to babies of differing gestational ages. Splitting research populations by arbitrary landmarks not recognised by parents or former patients32 may be a source of research heterogeneity. The strengths of our study included identification and synthesis of outcomes from an international and methodologically diverse range of studies, relating to babies of all gestational ages, and a wide range of stakeholders. We included outcomes that stakeholders spontaneously identified. As a result, we were able to include data from a wider range and diversity of stakeholders than a primary research study could. We followed a preregistered protocol with reporting in line with PRISMA guidelines.18 It has been argued that quality assessment is needed in ‘mapping’ reviews to aid in interpretation and uptake of findings,24 but the value of this approach is uncertain.21 The consultation phase of our core outcomes set development work will provide the opportunity to critically reflect on the contribution of these findings to our understanding of what constitutes an ‘important’ outcome in neonatal research.

etation and uptake of findings,24 but the value of this approach is uncertain.21 The consultation phase of our core outcomes set development work will provide the opportunity to critically reflect on the contribution of these findings to our understanding of what constitutes an ‘important’ outcome in neonatal research. A limitation of our study is that, in line with many systematic reviews, we are synthesising data from studies that did not explicitly address the research question we are asking. This meant that we combined data about which outcomes parents, patients or healthcare professionals mentioned during research. As a result, we described how frequently outcomes were discussed, rather than the importance assigned by groups to each outcome. Many outcomes were only discussed in a single study. We present them here to show the range and breadth of outcomes discussed, but cannot comment on whether they are more or less important than more frequently mentioned outcomes. Another limitation is that the researchers who undertook the primary qualitative research in the included studies will have influenced our review through their analysis; we reviewed data that was a step removed from the opinions of the stakeholders themselves. However, by following rigorous methodology and employing a comprehensive search strategy we have combined all available data to produce this mapping review.

included studies will have influenced our review through their analysis; we reviewed data that was a step removed from the opinions of the stakeholders themselves. However, by following rigorous methodology and employing a comprehensive search strategy we have combined all available data to produce this mapping review. Trying to measure all of the varied outcomes identified in this work in research would be impractical, if not impossible. This work supports the importance of identifying a core outcomes set, and highlights the importance of input from all stakeholder groups. In other fields, core outcomes sets have successfully aligned patient and healthcare professional research priorities.36 Conclusion Parents, patients and clinicians report a wide range of neonatal care outcomes. Parents and patients focus on different outcomes than health professionals. Outcomes reported do not map to organ systems commonly addressed in clinical trials, many are global outcomes. We suggest that the views of former patients and parents should be taken into consideration by researchers and funding bodies. Supplementary Material Reviewer comments Author's manuscript The authors are grateful to Louise Wann (West Middlesex University Hospital) for her contributions running the database searches.

Conclusion Parents, patients and clinicians report a wide range of neonatal care outcomes. Parents and patients focus on different outcomes than health professionals. Outcomes reported do not map to organ systems commonly addressed in clinical trials, many are global outcomes. We suggest that the views of former patients and parents should be taken into consideration by researchers and funding bodies. Supplementary Material Reviewer comments Author's manuscript The authors are grateful to Louise Wann (West Middlesex University Hospital) for her contributions running the database searches. Contributors: JW and CG conceived this systematic review. This protocol was created by JW, GB and CG. Searches were performed by LW. All search results were reviewed by JW and assessed by the eligibility criteria. Quality assurance was completed by CG. Coding and result synthesis was completed by JW, GB and CG. Statistical analysis was completed by NL. The first draft of the manuscript was written by JW, CG, GB and NL. NM edited and reviewed the manuscript. It was approved by JW, CG, GB, SA, LW, NL, NM and the COIN Steering Group. Funding: This research is sponsored by Imperial College London and supported by an MRC Clinician Scientist Fellowship award to CG (MR/N008405/1) and salary support for JW from the Portland Hospital. Disclaimer: The Imperial College London, the MRC and the Portland Hospital had no involvement in the research or this publication. Competing interests: None declared. Patient consent: Not required. Provenance and peer review: Not commissioned; externally peer reviewed.

Funding: This research is sponsored by Imperial College London and supported by an MRC Clinician Scientist Fellowship award to CG (MR/N008405/1) and salary support for JW from the Portland Hospital. Disclaimer: The Imperial College London, the MRC and the Portland Hospital had no involvement in the research or this publication. Competing interests: None declared. Patient consent: Not required. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Requests for access to data should be addressed to the corresponding author. Collaborators: COIN Project Steering Group: Elsa Afonso; Iyad Al-Muzaffar; Ginny Brunton; James Duffy; Chris Gale; Anne Greenough; Nigel Hall; Marian Knight; Jos Latour; Neil Marlow; Neena Modi; Laura Noakes; Julie Nycyk; Mehali Patel; Angela Richard-Londt; James Webbe; Ben Wills-Eve.

vity) and risk behaviours (substance use, unhealthy eating habits and excessive screen time). Our objective was to identify modifiable behavioural factors for mental well-being in boys and girls using an adolescent-specific measure and accounting for deprivation, ethnicity and clustering within local authorities (LAs). Methods Study design and sample The What About Youth study is a large-scale youth-oriented survey funded by the Department of Health in England and carried out by NHS Digital in 2014.22 The primary aim of the survey was to collect robust LA-level data on youth health behaviours and general health to improve their health outcomes. Study participants were those who turned 15 years old in the academic year 2013/2014. A random sampling methodology was employed to draw 298 080 participants from the National Pupil Database. The sample size was calculated to attain 1000 young people in each of 152 LAs in England; two LAs were merged with their nearest neighbours due to small size. The achieved sample was 120 115 individuals, of whom 16% responded online and 84% via postal means (2835 opted out). The response rates differed by gender, with adjusted response rates of 35% in boys and 49% in girls, and by deprivation, ethnicity and LA. Non-response weights using these factors were calculated to provide alignment between the achieved and target samples.22 We obtained a fully anonymised cohort data electronically from the UK Data Service website.22

nder, with adjusted response rates of 35% in boys and 49% in girls, and by deprivation, ethnicity and LA. Non-response weights using these factors were calculated to provide alignment between the achieved and target samples.22 We obtained a fully anonymised cohort data electronically from the UK Data Service website.22 Outcome variable Mental well-being was measured using the Warwick-Edinburgh Mental Wellbeing Scale,23 a population-level well-being measure. It is validated to use in adolescents aged 13 years or more and focus primarily on the positive aspects of mental well-being (internal consistency, α=0.90). Participants indicate how often they feel like each of the 14 items using a 5-point scale that ranges from 5 ‘all the time’ to 1 ‘none of the time’.24 Total scores ranged from 14 to 70 and were calculated by summing each participant’s responses. The potential explanatory behavioural variables were identified from the literature review of previous publication. A detailed description of each variable is given in supplementary appendix A. 10.1136/bmjpo-2018-000335.supp1Supplementary file 1

Outcome variable Mental well-being was measured using the Warwick-Edinburgh Mental Wellbeing Scale,23 a population-level well-being measure. It is validated to use in adolescents aged 13 years or more and focus primarily on the positive aspects of mental well-being (internal consistency, α=0.90). Participants indicate how often they feel like each of the 14 items using a 5-point scale that ranges from 5 ‘all the time’ to 1 ‘none of the time’.24 Total scores ranged from 14 to 70 and were calculated by summing each participant’s responses. The potential explanatory behavioural variables were identified from the literature review of previous publication. A detailed description of each variable is given in supplementary appendix A. 10.1136/bmjpo-2018-000335.supp1Supplementary file 1 A composite variable for risk behaviour index for substance use was constructed by the summation of three dichotomous risk behaviour variables: (a) smoking: if currently smokes, (b) drinking alcohol: if drinks once a month or more frequently, (c) cannabis use: if ever tried cannabis. Based on the number of risk behaviours, we categorised it from ‘none’ to ‘three’. Similarly, composite unhealthy eating habit index was derived from (a) skipping breakfast: if avoided breakfast in last 7 days, (b) poor diet: if consumed less than five portions of fruits and vegetables a day, (c) takeaway food: if consumed takeaway food in past 7 days. Based on a combination of unhealthy behaviours, the composite score was categorised into 0=none, 1=only one, 2=any two and 3=all three. Physical active for 60+ min for at least 5 days were classified as ‘physically active’ and the rest ‘physically inactive.’ This threshold was defined in line with government recommendations,25 26 except for the intensity of exercise which was not available in the dataset. The selected threshold was taken at 5 days a week, as only 13% reported being physically active for 7 days a week. A digital screen time variable was computed based on reported weekend and weekday usage of television, internet, smartphone and computer games. Subjects were categorised into ‘≥7 hours/day’, ‘about 5–6 hours/day’, ‘about 3–4 hours/day’, ‘about 2 hours/day’ and ‘≤1 hours/day’. Time spent reading on weekends and weekdays had response options ranging from none to 7 hours per day. Based on the distribution of data, we recoded the variable as ‘none’, ‘about half an hour/day’, ‘about 1 hour/day’ and ‘2 hours /day’. The frequency of 8 hours sleep in the last 7 days was coded as ‘every day’, ‘most days’, ‘some days’ and ‘not in the past 7 days’. Bullying was measured with Olweus Bully/Victim Questionnaire, a reliable 8-item scale used to assess the bullying victimisation.27 We combined responses to create one overall measure of bullying experience (yes/no). In line with a previous study,28 adolescents who were bullied more than ‘two or three times a month’ were categorised as bullying victims.

Bully/Victim Questionnaire, a reliable 8-item scale used to assess the bullying victimisation.27 We combined responses to create one overall measure of bullying experience (yes/no). In line with a previous study,28 adolescents who were bullied more than ‘two or three times a month’ were categorised as bullying victims. Ethnicity, deprivation and mode of questionnaire completion were selected as confounders in the relationship between well-being and potential explanatory variables as shown in previous studies.7 Ethnicity was self-identified by participants’ and was an adaptation of the 2001 UK census categories, supplemented by questions on the national group. English Index of Multiple Deprivation (IMD) was used as a measure of relative deprivation for small areas.29 IMD scores were divided into three deprivation categories as defined by quintiles of the national distribution: 1 and 2 (high deprivation), 3 (average), 4 and 5 (low). Participants were allowed to choose between online or postal modes of questionnaire completion.

as a measure of relative deprivation for small areas.29 IMD scores were divided into three deprivation categories as defined by quintiles of the national distribution: 1 and 2 (high deprivation), 3 (average), 4 and 5 (low). Participants were allowed to choose between online or postal modes of questionnaire completion. Analyses We conducted unadjusted and adjusted multilevel regression, in Stata V.14. For well-being scores, the interaction between gender and health behaviours was statistically significant (p<0.001) and therefore, analyses were stratified by gender. All variables were plotted to check the distribution and normality was checked with the Kolmogorov-Smirnov test. All estimates were weighted by representativeness of participants to compensate for the disproportionate selection of sample and non-response bias. Pearson χ2 tests were used to compare differences in the distribution of explanatory variables by gender. An unadjusted analysis was run to test the association between each independent variable (substance use, unhealthy eating habits, screen time, reading, bullying, physical activity and sleeping hours) and the outcome adolescent well-being (model 1). The analyses were repeated in a multivariable analysis where ethnicity, mode of questionnaire delivery and IMD were added as confounders between each risk factor and outcome (model 2). In model 3, all explanatory factors and covariates were included simultaneously to obtain associations between each variable and well-being scores after adjusting for confounders and other explanatory variables. LAs were treated as random effects in all models. Models were also tested for significant quadratic terms signifying curvilinear relationships; this was only significant for reading. The intraclass correlations coefficient, being the proportion of total variance attributable to differences at the LA level, was estimated using multilevel models for well-being with adjustment for IMD, ethnicity and mode of completion.

terms signifying curvilinear relationships; this was only significant for reading. The intraclass correlations coefficient, being the proportion of total variance attributable to differences at the LA level, was estimated using multilevel models for well-being with adjustment for IMD, ethnicity and mode of completion. Results In total, 57 153 boys (47.82%) and 62 962 girls (52.18%) participated in the study (table 1). Of these, boys had higher average well-being score in comparison with girls. The intracluster correlation coefficient for the well-being score was 0.032 for girls and 0.024 for boys in the adjusted model, suggesting that variance in adolescent well-being is small at LA level. Table 1 Descriptive statistics for well-being scores and explanatory variables under study, by gender Total Boys Girls N N Mean (SD)/% N Mean (SD)/% P values Full sample (N %) 120 115 57 153 47.58% 62 962 52.42%

Results In total, 57 153 boys (47.82%) and 62 962 girls (52.18%) participated in the study (table 1). Of these, boys had higher average well-being score in comparison with girls. The intracluster correlation coefficient for the well-being score was 0.032 for girls and 0.024 for boys in the adjusted model, suggesting that variance in adolescent well-being is small at LA level. Table 1 Descriptive statistics for well-being scores and explanatory variables under study, by gender Total Boys Girls N N Mean (SD)/% N Mean (SD)/% P values Full sample (N %) 120 115 57 153 47.58% 62 962 52.42% WEMWBS scores 117 842 56 352 47.82% 61 490 52.18% <0.0001 Mean (SD) 50 (8.60) 45 (9.66) Substance use* 71 133 32 516 45.71% 38 617 54.29% <0.0001 None 17 133 52.69 19 932 51.61 One 10 783 33.16 12 150 31.46 Two 3010 9.26 3785 9.8 Three 1590 4.89 2750 7.12 Unhealthy eating habits† 115 918 55 289 47.70% 60 629 52.30% <0.0001 None 12 081 21.85 11 951 19.71 One 20 287 36.69 20 389 33.63 Two 16 373 29.61 19 064 31.44 Three 6548 11.84 9225 15.22 Sleeping hours (>8 hours) 117 516 56, 207 47.83% 61 307 52.17% <0.0001 Not in the past 7 days 3676 6.54 6922 11.29 Some days 11 051 19.66 16 051 26.18 Most days 20, 121 35.8 20 928 34.14 Everyday 21, 361 38 17 406 28.39 Bullying 117 744 56 309 47.82% 61 435 52.18% <0.0001 No 45 959 81.62 45 094 73.40 yes 10 350 18.38 16 341 26.60 Physical activity 118 450 56 674 47.85 61 776 52.15 <0.0001 Physically active 48 172 85 44 348 71.79 Inactive 8502 15 17 428 28.21 Screen time 118 845 56 892 47.87 61 943 52.13% <0.0001 About 2 hours/day 3609 6.34 3469 5.6 ≤1 hours/day 912 1.6 929 1.5 About 3–4 hours/day 18 621 32.73 17 311 27.95 About 5–6 hours/day 17 198 30.23 17 850 28.82 ≥7 hours/day 16 559 29.1 22 387 36.14 Reading 118 140 56 513 47.84 61 627 52.16 <0.0001 None 14 278 25.26 9875 16.02 About Half an hour/day 17 572 31.09 15 363 24.93 About 1 hour/day 12 505 22.13 13 442 21.81 ≥ 2  hours/day 12 158 21.51 22947 37.24 Figures in bold refer to full sample distribution, the proportion of males and females in the study.

87 36.14 Reading 118 140 56 513 47.84 61 627 52.16 <0.0001 None 14 278 25.26 9875 16.02 About Half an hour/day 17 572 31.09 15 363 24.93 About 1 hour/day 12 505 22.13 13 442 21.81 ≥ 2  hours/day 12 158 21.51 22947 37.24 Figures in bold refer to full sample distribution, the proportion of males and females in the study. *Risk behaviours include smoking, drinking and cannabis use. †Unhealthy eating habits include skipping breakfast, not having five portions of fruits and vegetables and consumption of takeaway food. WEMWBS, Warwick-Edinburgh Mental Wellbeing score. Table 1 shows that there were significant differences in the distribution of potential explanatory variables between boys and girls. Girls had higher risk factors such as substance use, unhealthy eating habits, screen time and reported more bullying than boys. Protective variables such as sleeping hours (more than 8 hours) and reading were also significantly higher in girls than in boys. All explanatory variables and IMD were significantly associated with well-being in both sexes (table 2). Well-being in both sexes decreased with the use of substance use, unhealthy eating habits, bullying, physical activity and longer screen time in both sexes. Protective factors, such as, sleeping more than 8 hours and reading more than 2 hours were associated with higher well-being in both sexes. Table 2 Univariate analysis between well-being and explanatory variables, by gender

All explanatory variables and IMD were significantly associated with well-being in both sexes (table 2). Well-being in both sexes decreased with the use of substance use, unhealthy eating habits, bullying, physical activity and longer screen time in both sexes. Protective factors, such as, sleeping more than 8 hours and reading more than 2 hours were associated with higher well-being in both sexes. Table 2 Univariate analysis between well-being and explanatory variables, by gender Model† Boys Girls B 95 % CI B 95 % CI Substance use None Reference Reference One −0.34* (−0.56 to –0.13) −1.48** (−1.70 to –1.27) Two −2.19** (−2.60 to –1.78) −4.96** (−5.35 to –4.57) Three −3.56** (−4.16 to –2.95) −5.89** (−6.47 to –5.30) Unhealthy eating habits None Reference Reference One −1.65** (−1.88 to –1.43) −2.63** (−2.91 to –2.34) Two −3.36** (−3.60 to –3.12) −4.85** (−5.13 to –4.58) Three −5.00** (−5.39 to –4.61) −6.49** (−6.83 to –6.14) Sleeping hours (>8 hours) Not in the past 7 days Reference Reference Some days 2.99** (2.51 to 3.47) 4.71** (4.39 to 5.04) Most days 5.59** (5.15 to 6.03) 8.39** (8.06 to 8.72) Everyday 7.49** (7.07 to 7.87) 10.70** (10.35 to 11.06) Bullying No Reference Reference Yes −4.70** (−3.59 to –3.31) −5.84** (−5.25 to –4.90) Physical activity Physically active Reference Reference Inactive −3.85** (−4.11 to –3.61) −2.77** (−3.04 to –2.51) Screen time About 2 hours/day Reference Reference ≤1 hours/day 0.07 (−0.59 to 0.73) 0.08 (−0.69 to 0.85) About 3–4 hours/day −1.22** (−1.60 to –0.84) −1.27** (−1.71 to –0.83) About 5–6 hours/day −2.17** (−2.51 to –1.83) −3.06** (−3.52 to –2.60) ≥7 hours/day −3.72** (−4.12 to –3.32) −5.38** (−5.80 to –4.97) Reading None Reference Reference About Half an hour/day 1.61** (1.41 to 1.82) 2.17** (1.88 to 2.46) About 1 hour/day 2.25** (2.06 to 2.44) 2.54** (2.26 to 2.82) ≥2 hours/day 2.27*** (2.01 to 2.52) 2.35** (2.05 to 2.66) IMD scores High deprivation Reference Reference Average deprivation 0.57** (0.31 to 0.81) 0.35** (0.10 to 0.590 Least deprivation 1.05** (0.87 to 1.21) 1.16** (0.92 to 1.40) Ethnicity White Reference Reference Mixed 0.18 (−0.29 to 0.65) −0.44* (−0.82 to –0.06) Asian −0.54* (−0.85 to –0.23) 0.98** (0.59 to 1.38) Black 0.91** (0.42 to 1.39) 1.10** (0.67 to 1.52) Other −0.45* (−0.84 to –0.04) 0.30 (−0.09 to 0.70) *P<0.05; **P<0.001.

east deprivation 1.05** (0.87 to 1.21) 1.16** (0.92 to 1.40) Ethnicity White Reference Reference Mixed 0.18 (−0.29 to 0.65) −0.44* (−0.82 to –0.06) Asian −0.54* (−0.85 to –0.23) 0.98** (0.59 to 1.38) Black 0.91** (0.42 to 1.39) 1.10** (0.67 to 1.52) Other −0.45* (−0.84 to –0.04) 0.30 (−0.09 to 0.70) *P<0.05; **P<0.001. †Unadjusted model taking into account clustering at local authority level, multilevel models fitted with weighted design weights, quadratic function added to reading. IMD, Index of Multiple Deprivation. In the multivariable models adjusted for covariates (table 3), poorer well-being was associated with multiple substances use and multiple unhealthy eating habits in a dose-dependent fashion. Being physically inactive, longer screen time and experiencing bullying were both associated with decrements in well-being in both sexes, with the association being stronger in girls than in boys. Table 3 Gender-stratified partially adjusted and fully adjusted multilevel modelling for well-being and explanatory variables

In the multivariable models adjusted for covariates (table 3), poorer well-being was associated with multiple substances use and multiple unhealthy eating habits in a dose-dependent fashion. Being physically inactive, longer screen time and experiencing bullying were both associated with decrements in well-being in both sexes, with the association being stronger in girls than in boys. Table 3 Gender-stratified partially adjusted and fully adjusted multilevel modelling for well-being and explanatory variables Variables Boys Girls Model† Model‡ Model† Model‡ B 95 % CI B 95 % CI B 95 % CI B 95 % CI Substance use None Reference Reference Reference Reference One −0.40** (−0.61 to –0.18) −0.14 (−0.36 to 0.08] −1.51** (−1.73 to –1.30) −0.77** (−0.97 to –0.57) Two −2.16** (−2.57 to –1.75) −1.05** (−1.42 to –0.67) −4.84** (−5.22 to –4.46) −2.67** (−3.01 to –2.33) Three −3.50** (−4.11 to –2.88) −1.63** (−2.16 to –1.09) −5.80** (−6.36 to –5.23) −2.79** (−3.35 to –2.24) Eating habits None Reference Reference Reference Reference One −1.63** (−1.85 to –1.41) −0.89** (−1.18 to –0.60) −2.63** (−2.92 to –2.34) −1.37** (−1.69 to –1.06) Two −3.31** (−3.55 to –3.08) −1.84** (−2.14 to –1.54) −4.84** (−5.12 to –4.55) −2.29** (−2.64 to –1.96) Three −4.95** (−5.34 to –4.56) −2.44** (−2.82 to –2.06) −6.46** (−6.79 to –6.10) −2.61** (−3.04 to –2.18) Sleeping hours (>8 hours) Not in the past 7 days Reference Reference Reference Reference Some days 2.98** (2.50 to 3.45] 2.69** (2.10 to 3.28) 4.69** (4.36 to 5.01) 3.71** (3.34 to 4.08) Most days 5.55** (5.11 to 5.99] 4.30** (3.78 to 4.82) 8.35** (8.01 to 8.69) 6.64** (6.28 to 7.00) Everyday 7.45** (7.03 to 7.86] 5.79** (5.29 to 6.28) 10.65** (10.29 to ,11.01) 8.16** (7.72 to 8.60) Bullying No Reference Reference Reference Reference Yes −4.64** (−4.86 to –4.42) −3.78** (−4.09 to –3.48) −5.77** (−5.98 to –5.56) −4.01** (−4.23 to –3.78) Physical activity Physically active Reference Reference Reference Reference Inactive −3.78** (−4.02 to –3.55) −2.63** (−2.95 to –2.30) −2.77** (−3.04 to –2.50) −1.70** (−2.01 to –1.39) Screen time About 2 hours/day Reference Reference Reference Reference ≤1 hours/day 0.09 (−0.57 to 0.75] 0.34 (−0.58 to 1.26) 0.09 (−0.69 to 0.87) −0.38 (−1.49 to 0.72) About 3–4 hours/day −1.22** (−1.60 to –0.83) −0.61* (−0.99 to –0.23) −1.26** (−1.70 to –0.83) −0.54 (−1.14 to 0.05) About 5–6 hours/day −2.15** (− 2.50 to – 1.81) −0.82** (−1.27 to –0.37) −3.04** (−3.49 to –2.58) −1.21** (−1.75 to ,–0.67) ≥7 hours/day −3.67** (−4.06 to –3.27) −1.20** (−1.65 to –0.75) −5.32** (−5.73 to –4.90) −1.81** (−2.29 to –1.33) Reading None Reference Reference Reference Reference About Half an hour/day 1.59** (1.38 to 1.80) 0.57** (0.26 to 0.88) 2

o – 1.81) −0.82** (−1.27 to –0.37) −3.04** (−3.49 to –2.58) −1.21** (−1.75 to ,–0.67) ≥7 hours/day −3.67** (−4.06 to –3.27) −1.20** (−1.65 to –0.75) −5.32** (−5.73 to –4.90) −1.81** (−2.29 to –1.33) Reading None Reference Reference Reference Reference About Half an hour/day 1.59** (1.38 to 1.80) 0.57** (0.26 to 0.88) 2 .13** (1.84 to 2.42) 0.56** (0.23 to 0.90) About 1 hour/day 2.24** (2.04 to 2.43) 1.04** (0.79 to 1.29) 2.51** (2.24 to 2.78) 0.60** (0.26 to 0.93) ≥2 hours/day 2.28** (2.05 to 2.54) 0.90** (0.55 to 1.25) 2.34** (2.03 to 2.64) 0.33* (0.02 to 0.64) IMD scores High deprivation – – Reference – – Reference Average deprivation – – −0.12 (−0.38 to 0.14) – – 0.16 (−0.16 to 0.49) Least deprivation – – 0.17 (−0.06 to 0.40) – – 0.36* (0.06 to 0.66) Ethnicity White – – Reference – – Reference Mixed – – −0.06 (0.81 to –0.59) – – 0.44 (−0.08 to 0.96) Asian – – −0.09 (0.81 to –0.84) – – 0.35 (−0.53 to 1.23) Black – – 0.99* (0.01 to 0.25) – – 1.75** (1.20 to 2.31) Other – – −0.59 (0.02 to –1.09) – – −0.19 (−0.88 to 0.49) Mode of questionnaire delivery Online – – – – paper – – 0.24 (−0.03 to 0.51) – – 1.26** (0.99 to 1.53) Multilevel models fitted with weighted design weights, quadratic function added to reading. *P<0.05; P<0.005. †Each predictor variable adjusted for ethnicity, mode of questionnaire delivery and IMD. ‡ All variables mutually adjusted for each other. IMD, Index of Multiple Deprivation.