Article Text
PDF
Abstract
Introduction Existing risk assessment tools to identify children at risk of hospitalised pneumonia-related mortality have shown suboptimal discriminatory value during external validation. Our objective was to derive and validate a novel risk assessment tool to identify children aged 2–59 months at risk of hospitalised pneumonia-related mortality across various settings.
Methods We used primary, baseline, patient-level data from 11 studies, including children evaluated for pneumonia in 20 low-income and middle-income countries. Patients with complete data were included in a logistic regression model to assess the association of candidate variables with the outcome hospitalised pneumonia-related mortality. Adjusted log coefficients were calculated for each candidate variable and assigned weighted points to derive the Pneumonia Research Partnership to Assess WHO Recommendations (PREPARE) risk assessment tool. We used bootstrapped selection with 200 repetitions to internally validate the PREPARE risk assessment tool.
Results A total of 27 388 children were included in the analysis (mean age 14.0 months, pneumonia-related case fatality ratio 3.1%). The PREPARE risk assessment tool included patient age, sex, weight-for-age z-score, body temperature, respiratory rate, unconsciousness or decreased level of consciousness, convulsions, cyanosis and hypoxaemia at baseline. The PREPARE risk assessment tool had good discriminatory value when internally validated (area under the curve 0.83, 95% CI 0.81 to 0.84).
Conclusions The PREPARE risk assessment tool had good discriminatory ability for identifying children at risk of hospitalised pneumonia-related mortality in a large, geographically diverse dataset. After external validation, this tool may be implemented in various settings to identify children at risk of hospitalised pneumonia-related mortality.
- Pneumonia
- Paediatrics
Data availability statement
Data are available upon reasonable request. Data may be made available upon reasonable request to the corresponding author.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
What is already known on this topic
In external validation of existing risk assessment tools to identify children at risk of hospitalised pneumonia-related mortality in varied settings, only the Respiratory Index of Severity in Children–Malawi score demonstrated fair discriminatory value (area under the curve (AUC) 0.75, 95% CI 0.74 to 0.77), while the Respiratory Index of Severity in Children score and a modified Pneumonia Etiology Research for Child Health score had limited discriminatory value (AUC 0.66, 95% CI 0.58 to 0.73, and AUC 0.55, 95% CI 0.37 to 0.73, respectively).
What this study adds
Using data from 27 388 children in 20 low-income and middle-income countries, the Pneumonia Research Partnership to Assess WHO Recommendations (PREPARE) risk assessment tool was developed to identify children aged 2–59 months at risk of hospitalised pneumonia-related mortality across various settings.
The PREPARE risk assessment tool had good discriminatory value when internally validated (AUC 0.83, 95% CI 0.81 to 0.84) and incorporates practical and commonly recorded clinical parameters to identify children at risk of hospitalised pneumonia-related mortality in various settings (ie, patient age, sex, weight-for-age z-score, body temperature, respiratory rate, unconsciousness or decreased level of consciousness, convulsions, cyanosis and hypoxaemia at baseline).
How this study might affect research, practice and/or policy
After external validation, the PREPARE risk assessment tool may be implemented in various hospital settings in low-income and middle-income countries to identify children at risk of hospitalised pneumonia-related mortality.
The impact of the implementation of existing risk assessment tools to identify children at risk of hospitalised pneumonia-related mortality must be compared with routine clinical care.
Introduction
Pneumonia is the leading cause of mortality among children 1–59 months of age, causing more than 800 000 deaths in this age group every year worldwide.1–3 Four risk assessment tools have been developed to identify children at risk of hospitalised pneumonia-related mortality in sub-Saharan Africa, South Asia and Southeast Asia.4–7 These risk assessment tools have important limitations, including the use of variables that are not routinely collected in clinical practice,5 being limited to single sites4–6 and the reliance on auscultatory findings with variable inter-rater reliability.4 6 Additionally, these risk assessment tools have not been widely implemented; thus, their potential to reduce hospitalised pneumonia-related mortality among children is unclear.8
We recently validated three of these risk assessment tools in a large, globally representative dataset using the demographic and clinical features of children at the time of admission.9 In that external validation, only the Respiratory Index of Severity in Children–Malawi (RISC-Malawi) score demonstrated fair discriminatory value (area under the curve (AUC) 0.75, 95% CI 0.74 to 0.77), while the Respiratory Index of Severity in Children (RISC) score and a modified Pneumonia Etiology Research for Child Health (PERCH) score had limited discriminatory value in identifying hospitalised children at risk of pneumonia-related mortality (AUC 0.66, 95% CI 0.58 to 0.73%, and AUC 0.55, 95% CI 0.37 to 0.73, respectively). The suboptimal performance of these prediction rules when applied externally raises questions on whether a novel tool incorporating different combinations of widely available clinical indicators could perform better across broad settings. Such a risk assessment tool should incorporate practical and commonly recorded clinical parameters to facilitate broad use and improved recognition of children at risk of hospitalised pneumonia-related mortality.
Given the limited discriminatory ability of prior risk assessment tools for pneumonia-related mortality when applied externally, our objective was to derive and validate a novel, widely applicable, risk assessment tool to identify children aged 2–59 months at risk of hospitalised pneumonia-related mortality. A risk assessment tool that is not region-specific may be useful to guide clinical care for children at greatest risk of hospitalised pneumonia-related mortality across settings.
Methods
Study design
We used the World Health Organization’s (WHO) Pneumonia Research Partnership to Assess WHO Recommendations (PREPARE) dataset to derive and validate a novel risk assessment tool for hospitalised pneumonia-related mortality among children 2–59 months of age across many global settings. We have described the details of the construction of the PREPARE dataset previously.10 Briefly, the PREPARE dataset includes primary data of individual patients from 30 study groups comprising 41 datasets of children evaluated for pneumonia in studies conducted from 1994 to 2014 in over 20 low-income and middle-income countries in Asia, Africa and Latin America, as well as the USA and Australia. We adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis guidelines.11
Patient and public involvement statement
The development of the research question was informed by the large burden of pneumonia-related mortality among children worldwide. Patients neither were advisers in this study nor were involved in the design, recruitment or conduct of the study. Results of this study will be made publicly available through open-access publication where study participants may access them.
Study population
As pneumonia is the leading cause of mortality among children aged 1–59 months and pneumonia is defined differently in infants aged <2 months,12 13 we restricted the derivation and validation of our risk assessment tool to infants and children aged 2–59 months. We also restricted our analysis to studies that included hospitalised patients as our outcome was in-hospital mortality in children with suspected pneumonia (ie, pneumonia-related mortality). We excluded community-based studies, hospital-based studies that did not report survival data and any deaths that occurred outside the hospital. Pneumonia was defined according to the 2013 WHO Pocket Book of Hospital Care for Children (ie, based on age-adjusted tachypnoea, presence of lower chest indrawing, general danger signs or signs of respiratory distress including head nodding/bobbing, nasal flaring or grunting in children with a cough or difficulty breathing).14
Candidate variables
All candidate variables were selected a priori. As hypoxaemia has been shown to be highly predictive of pneumonia-related mortality among children15–17 and to avoid potential selection bias, we prioritised the inclusion of studies within the PREPARE dataset that had oxygen saturation (SpO2) measurements in at least 70% of he participants. Other candidate variables including age, sex, weight-for-age z-score, temperature, respiratory rate, presence of chest indrawing, unconsciousness/decreased consciousness, convulsions and cyanosis were selected based on the results of a systematic review evaluating risk factors for pneumonia-related mortality in children <5 years of age,18 prior clinical prediction models,4–7 as well as availability of data in the PREPARE dataset. Furthermore, hypoxaemia, presence of chest indrawing, unconsciousness or decreased consciousness, convulsions and cyanosis are signs and symptoms of severe pneumonia or danger signs according to the 2013 WHO Pocket Book of Hospital Care for Children.14 Studies in the PREPARE dataset with >25% missing data points were excluded from our analysis to reduce selection bias. We did not include apnoea, gasping, grunting, nasal flaring, head nodding or stridor, which are general danger signs in the 2013 WHO Pocket Book of Hospital Care for Children14 due to high levels of missing data.
We defined tachypnoea as 0–9, 10–19 and >20 breaths/min above age-specific cutoffs (ie, >50 breaths/min for children aged 2–11 months and >40 breaths/min for children aged 12–59 months).12 14 We categorised weight-for-age z-scores as <−3 for severe malnutrition, −3 to −2 for moderate malnutrition and >−2 for normal weight19; temperature as <35.5°C for hypothermia, 35.5°C to 37.9°C for normothermia and ≥38°C for fever12; and SpO2 as <90% for severe hypoxaemia, 90%–92% for mild hypoxaemia, and 93%–100% as normal.20 These continuous variables were converted into categorical variables based on recommended thresholds before developing our model to facilitate the use of this risk assessment tool in clinical practice. All variables included in our analysis were recorded at enrolment or as baseline data in each of the included studies in the PREPARE dataset. All deaths included in this analysis occurred during the hospitalisation from which baseline data were collected.
Statistical analyses
We restricted our analyses to patients with no missing values for any candidate variables. We calculated the effective sample size required for the development of a new clinical prediction model21 based on the inputs R2 of <0.05, shrinkage factor of 0.9, 26 parameters and outcome (pneumonia-related mortality) prevalence of 3%. The minimum sample size required was 23 270, with 699 events (events per candidate predictor parameter of at least 26.8).
For the derivation of the PREPARE risk assessment tool, we constructed a multivariable backward regression model, including all candidate variables to assess the strength of the association of each candidate variable on hospitalised pneumonia-related mortality. Associations with 95% CI for adjusted ORs (aORs) that did not cross 1 were considered significant. Then, to determine the weighted points assigned to each candidate variable, we calculated the adjusted log coefficient of each candidate variable from the multivariable model, rounded it to the nearest 0.5 and then doubled the rounded log coefficients to form an integer.4 7 22 23 As the PREPARE risk assessment tool is intended to be used by clinicians in settings with all levels of resources, weighted points were assigned to each candidate variable to create a user-friendly risk assessment tool that can be simply calculated without the use of a computer or an application.24
To assess the discriminatory ability of the PREPARE risk assessment tool to identify children at risk of hospitalised pneumonia-related mortality, we internally validated the risk assessment tool using bootstrapping methodology with 200 repetitions and calculated the area under the receiver operating characteristic (ROC) (AUC).21 25 26 As pulse oximetry is not available in all settings, we conducted a sensitivity analysis of the performance of the PREPARE risk assessment tool excluding pulse oximetry. We used the Hosmer-Lemeshow test to assess the goodness of fit of the PREPARE risk assessment tool by testing the null hypothesis that the fitted values from the model were the same as observed.
We created an ROC curve for the PREPARE risk assessment tool and repeated these analyses without pulse oximetry. We created a risk predictiveness curve to demonstrate the cumulative percentage of children at risk of hospitalised pneumonia-related mortality by predicted risk. We created a calibration plot of the agreement between estimated and observed probabilities for hospitalised pneumonia-related mortality. We conducted decision curve analysis including all candidate variables to estimate the clinical utility of the PREPARE risk assessment tool. We used descriptive statistics to describe characteristics among children who were misclassified (ie, deemed low risk at optimal PREPARE risk assessment scores but died). All analyses were conducted using Stata V.16.1.
Results
Of 41 datasets with a total of 285 839 children in the PREPARE dataset, 11 studies with 27 388 children met our inclusion criteria (figure 1). Six of the included studies were randomised controlled trials; two were prospective cohort studies; two were retrospective cohort studies; and one was a prospective case series (table 1). The included studies were conducted in 20 different low-income or middle-income countries in Asia, Africa, Central and South America, the Caribbean and the Middle East. The mean age was 14.0±12.1 months and the case fatality ratio was 3.1%. Children included in the derivation and validation of the PREPARE risk assessment tool were slightly younger than those who did not meet the inclusion criteria (14.0±12.1 months vs 17.1±13.5 months) but did not differ substantially by sex (male 15 862 (57.9%) vs 100 023 (56.0%)), weight-for-age z-score (−1.12±2.00 vs −1.01±1.75) or mortality (n=856 (3.1%) vs n=6877 (3.8%)). Children with any missing parameter for any candidate variable were not included in the derivation or validation of the PREPARE risk assessment tool (online supplemental table 1). The risk predictiveness curve including all children who met the inclusion criteria demonstrated that most children were at low risk for mortality (online supplemental figure 1).
Supplemental material
Selection of hospital-based studies included in the derivation and validation of the novel PREPARE risk assessment tool to identify children 2–59 months of age at risk of hospitalised pneumonia-related mortality. *Candidate variables include age, sex, weight-for-age z-score, temperature, respiratory rate, oxygen saturation, presence of chest indrawing, unconsciousness/decreased consciousness, convulsions and cyanosis. PREPARE, Pneumonia Research Partnership to Assess WHO Recommendations.
Characteristics of studies included in the derivation and validation of the PREPARE risk assessment tool
Derivation of the PREPARE risk assessment tool
Weight-for-age z-score of <−3 (aOR 5.16, 95% CI 4.37 to 6.09), body temperature of <35.5°C (aOR 4.80, 95% CI 3.05 to 5.57) and SpO2of <90% (aOR 2.99, 95% CI 2.51 to 3.58) were most strongly associated with hospitalised pneumonia-related mortality among all included children (table 2). The PREPARE risk assessment tool had a score ranging from 0 to 17 and incorporated patient age, sex, weight-for-age z-score, body temperature, respiratory rate, unconsciousness or decreased level of consciousness, convulsions, cyanosis and hypoxaemia at baseline (table 3).
Multivariable regression model for hospitalised pneumonia-related mortality among all included children aged 2–59 months (n=27 388)
Components of the PREPARE risk assessment tool including all children who met the inclusion criteria (n=27 388)
Our sensitivity analysis excluding pulse oximetry demonstrated a weight-for-age z-score of <−3, body temperature of <35.5°C and cyanosis were most strongly associated with hospitalised pneumonia-related mortality among all children (online supplemental table 2). The PREPARE risk assessment tool including all children and excluding pulse oximetry had a score ranging from 0 to 20 (online supplemental table 3).
Validation of the PREPARE risk assessment tool
The internal validation of the PREPARE risk assessment tool using the bootstrap method demonstrated an AUC of 0.83 (95% CI 0.81 to 0.84) among all children (figure 2) and 0.81 (95% CI 0.79 to 0.82) when excluding pulse oximetry. The PREPARE risk assessment tool had maximised sensitivity with concurrent maximised specificity at a score of ≥5 (72.6% sensitivity, 76.5% specificity; +likelihood ratio [LR] of 3.09 (95% CI 2.89 to 3.30) and −LR of 0.36 (95% CI 0.31 to 0.42)) in identifying children at risk of hospitalised pneumonia-related mortality. A score of ≥4 had 81.8% sensitivity, 65.4% specificity, +LR of 2.37 (95% CI 2.25 to 2.49) and −LR of 0.28 (95% CI 0.23 to 0.34), and a score of ≥6 demonstrated 61.3% sensitivity and 84.4% specificity with a +LR of 3.93 (95% CI 3.61 to 4.29) and a −LR of −0.46 (95% CI 0.41 to 0.52).
Receiver operating characteristic curve for the PREPARE risk assessment tool for children at risk of hospitalised pneumonia-related mortality among children 2–59 months of age (n=27 388). Area under receiver operating curve: 0.83 (95% CI 0.81 to 0.84). PREPARE, Pneumonia Research Partnership to Assess WHO Recommendations.
Goodness of fit, calibration and decision curve analysis
The Hosmer-Lemeshow test to assess the goodness-of-fit demonstrated a p value of <0.001, meaning that the observed and expected proportions were not the same across all groups. The calibration plot of observed against expected probabilities for the assessment of prediction model performance demonstrated reasonably good model calibration (online supplemental figure 2). The PREPARE risk assessment tool had higher net benefit than individual candidate variables in predicting pneumonia-related mortality (online supplemental figure 3).
Misclassified patients in validation of the PREPARE risk assessment tool
At a PREPARE risk assessment tool score of <4, 75 children (16.9% of all deaths) were classified as low risk but died (online supplemental table 4). At a score of <5, 113 children died (25.5% of all deaths) and were misclassified for their risk of hospitalised pneumonia-related mortality. At a score of <6, 160 children died (36.1% of all deaths) and were incorrectly classified for their risk of hospitalised pneumonia-related mortality. Misclassified children commonly had a weight-for-age z-score of ≥2, normothermia and lower chest indrawing; were of normal consciousness; and were not cyanotic. No children with hypothermia were misclassified for their risk of hospitalised pneumonia-related mortality at a score of <4, <5 or <6.
Discussion
Early identification of children at risk of mortality during hospitalisation may allow for the allocation of resources to potentially prevent such deaths.27 The PREPARE risk assessment tool was derived from the largest and most geographically diverse patient population of all existing risk assessment tools for hospitalised pneumonia-related mortality and included variables that are routinely assessed in clinical practice. Our novel risk assessment tool demonstrated good discriminatory ability when internally applied to patients from a range of settings both with and without use of pulse oximetry. After external validation, the PREPARE risk assessment tool may be used to identify children at risk of hospitalised pneumonia-related mortality and could be used for monitoring of children hospitalised with pneumonia.
The PREPARE risk assessment tool includes the assessment of a patient’s age, sex, weight-for-age z-score, body temperature, respiratory rate, level of consciousness, presence of convulsions, cyanosis and SpO2. Weight-for-age z-score, level of consciousness and SpO2 of <90% were associated with mortality in the RISC-Malawi and PERCH scores.6 7 Younger age was also associated with mortality in the PERCH score.7 We additionally identify findings of hypothermia (ie, <35.5°C), tachypnoea and convulsions associated with mortality among children hospitalised with pneumonia. Hypothermia was strongly associated with mortality, second only to malnutrition. No children with hypothermia were misclassified for their risk of hospitalised pneumonia-related mortality at PREPARE risk assessment tool scores of <4, <5 or <6. Most variables included in the PREPARE risk assessment tool can be easily assessed by providers of many training levels in various settings. However, the assessment of SpO2 requires a pulse oximeter with paediatric probes and additional training in the accurate use of pulse oximetry. Prior studies demonstrate that community level health workers and first-level health facility workers can be trained to accurately use pulse oximetry.28 29 The PREPARE risk assessment tool also had good discriminatory value in our sensitivity analysis excluding pulse oximetry. Therefore, the PREPARE risk assessment tool may be useful in identifying children at risk of hospitalised pneumonia-related mortality in settings with limited access to pulse oximetry.
Due to missing data in the included datasets, we were not able to assess if the presence of wheezing was potentially protective. Wheezing has been associated with lower mortality rates among children,30 is more common in viral pneumonia and bronchiolitis than bacterial pneumonia,31–33 and serves as a protective variable in other clinical prediction models.4 6 However, accurate auscultation requires a stethoscope and experienced clinicians. Moreover, prior studies demonstrate variable inter-rater reliability for the detection of wheezing in children.34–36 Thus, the exclusion of wheezing from the PREPARE risk assessment tool may allow providers in various settings and with varying training levels to more easily use this risk assessment tool.
The WHO Integrated Management of Childhood Illness guidelines recommend that HIV-uninfected children 2–59 months of age who have lower chest indrawing and no danger signs be treated with oral amoxicillin at home without hospital referral.12 This recommendation has been criticised with some calling for its revision.37 38 The presence of lower chest indrawing was included in both the RISC and modified Respiratory Index of Severity in Children (mRISC), but not in the RISC-Malawi or PERCH risk assessment tools for hospitalised pneumonia-related mortality.4–7 Lower chest indrawing was not associated with mortality in our risk assessment tool. The RISC and mRISC scores demonstrated an association between the presence of chest indrawing and mortality.4 5 Accordingly, the PREPARE risk assessment tool should be externally validated prior to implementation with careful attention paid to the presence of lower chest indrawing, given the differing findings in prior models.
Our novel risk assessment tool demonstrated good discriminatory ability in this within-sample validation in children from 20 low-income and middle-income countries in Asia, Africa, Central and South America, the Caribbean and the Middle East. Prior clinical prediction models developed in single countries (ie, RISC, mRISC and RISC-Malawi) have demonstrated AUC from 0.80 to 0.92 when internally validated,4–6 while the PERCH score, derived from five countries in sub-Saharan Africa as well as Thailand and Bangladesh, had an AUC of 0.76 on internal validation.7 However, when externally applied to patients in various settings, only the RISC-Malawi score had fair discriminatory ability.9 The PREPARE risk assessment tool transcends region-specific issues, such as differing epidemiology of causative viruses or bacteria and variations in availability of measures such as chest radiography, supplemental oxygen, antibiotic availability and unmeasured variables that can contribute to morbidity, given its derivation and validation from a widely representative patient population. However, the PREPARE risk assessment tool must be externally validated prior to implementation. Though the PREPARE risk assessment tool had maximised test characteristics at a score of ≥5, some patients were misclassified for their risk of hospitalised pneumonia-related mortality at that score. Higher scores had fewer misclassified children but in turn had lower sensitivity. Thus, clinicians may use higher PREPARE risk assessment tool scores to identify children at high risk of hospitalised pneumonia-related mortality, but scores above 5 should not be reliably used to rule out the possibility of hospitalised pneumonia-related mortality. Future studies assessing the impact of the implementation of the PREPARE risk assessment tool must be compared with routine clinical care.
Limitations
Our study is subject to several limitations. As many studies in the PREPARE dataset used tachypnoea as an entry point sign to diagnose pneumonia, tachypnoea as a predictor of mortality may have been overestimated. However, we assessed the contribution of varying severity of tachypnoea to mortality, which may reduce potential bias introduced by including tachypnoea as a predictor. Furthermore, the WHO manual on oxygen therapy for children recommends oxygen therapy for children with severe lower chest wall indrawing, respiratory rate of ≥70 breaths/min and head nodding in settings where pulse oximetry is not available.20 As we conducted an analysis of previously collected data, not all candidate variables for the novel risk assessment tool were available in large numbers in the PREPARE dataset. Specifically, we were not able to assess HIV infection status as a candidate variable, which has been associated with mortality in other studies.4 39 40 Future studies incorporating HIV infection status into the PREPARE risk assessment tool in endemic regions may be needed. Furthermore, given the retrospective nature of this study, we were not able to assess several clinical variables that have been associated with mortality in other studies, such as grunting, duration of illness7 or signs such as apnoea, gasping, nasal flaring or head nodding,14 or anaemia and pallor.41 42 Apnoea, gasping, nasal flaring and head nodding have not been incorporated in other risk assessment tools.4–7 Future studies incorporating these signs into risk assessment tools may be warranted. The case fatality rate was higher among children with missing data than among those who had complete data. This may have been due to deaths that occurred early in the hospitalisation before time was granted to fully collect clinical data. Additionally, six of the datasets included in our analysis came from randomised controlled trials, which tend to be highly selective of patients and may explain part of the lower case fatality rate in our derivation and validation populations compared with children who were excluded from our analysis. An additional limitation is the possibility of confounding by disease severity and management variation across regions. For example, some variables included in the PREPARE risk assessment tool, such as SpO2, may be modified through interventions (eg, the use of supplemental oxygen). Lastly, our analysis did not assess the role of the quality of care, supplemental oxygen availability, regional variations in clinical care or antibiotics administered.
Conclusions
The PREPARE risk assessment tool is a novel tool that had good discriminatory ability at hospital admission to identify children at risk of hospitalised pneumonia-related mortality when applied to >27 000 children in 20 low-income and middle-income countries. The PREPARE risk assessment tool may help direct resources to children at highest risk of hospitalised pneumonia-related mortality in resource-limited settings. This novel tool includes routinely collected variables that can be assessed by healthcare providers of all levels. External validation of the PREPARE risk assessment tool and a comparison of its impact on hospitalised pneumonia-related mortality among children across various settings compared with standard clinical care may be warranted.
Data availability statement
Data are available upon reasonable request. Data may be made available upon reasonable request to the corresponding author.
Ethics statements
Patient consent for publication
Ethics approval
All studies included in this deidentified dataset were previously granted clearance by ethical review boards from each participating site.
Acknowledgments
We thank the following Pneumonia REesearch Partnership to Assess WHO REecommendations dataset collaborators: Angela Gentile, Bradford D Gessner, Tabish Hazir, Luis Marinez Arroyo, Rumina Libster, Kerry-Ann O'Grady, Raul O. Ruvinsky, Claudia Turner and Syed Mohammad Akram uz Zaman. We have provided detailed information on author contributions and positionality in our reflexivity statement in online supplemental table 5.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Handling editor Sanni Yaya
Twitter @CarinaTKing, @tinylungsglobal
Deceased Deceased
Contributors SQ secured the funding. CAR, MIN, YBN and SQ conceptualised and designed the study, interpreted the data and wrote the first draft of the manuscript. CAR and YBN are responsible for the overall content and take full responsibility for the work, had access to the data, and made the decision to publish this article. YBN verified the underlying data and conducted the statistical analyses. TC, CK, NL, EDM, CM, CC, SAM, MN, JM, EA-Y, NC, MH, PH, PMJ, JML, WBM, AP, DMT, NTVN, CBK, ML, SA, AB, MC, PN, JWP, GP-B, VSP, MR-A, VR, GR, MS, PV, JW, RA, SBan, II, IM-R, GM, SS, MS, SKS, SBas, TS, SBh, NW, AWC, HC, HN and JF oversaw data collection, verified the underlying data, assisted with the interpretation of the data, and reviewed and provided input to the final draft. CAR and YBN had final responsibility for the decision to submit for publication.
Funding This study was funded by the Bill & Melinda Gates Foundation (#INV-007927) through a grant to the WHO to SQ. The funders had no role in the study design or in the collection, analysis or interpretation of the data. The funders did not write the report and had no role in the decision to submit the paper for publication.
Disclaimer The expressed views and opinions do not necessarily represent the policies of the WHO.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.