Major Causes of Perinatal and Paediatric Mortality in Sub-Saharan Africa and South Asia: Adjustment for Selection Bias in the CHAMPS Network

Background: Studies of child mortality that employ minimally invasive tissue sampling (MITS) produce highly accurate cause of death data; however, selection bias may render these as non-representative of their underlying populations. Objectives: Estimate cause-specific mortality fractions and rates for the five most frequent causes—underlying and others in the chain of events leading to death—among stillbirths, neonatal, infant and child deaths—in Sub-Saharan Africa and South Asia, adjusted for any identified selection biases. Methods: The Child Health and Mortality Prevention Surveillance (CHAMPS) Network collects standardised, population-based, longitudinal data on causes of death among stillbirths and under-five children in 12 catchments in seven countries in Sub-Saharan Africa and South Asia. Cause-specific mortality fractions and rates were calculated for the five most frequent causes among stillbirths, neonatal, infant and child deaths, and for the five most frequent maternal conditions among perinatal deaths; all estimates were subsequently adjusted for selection bias. Selection probabilities were estimated from membership in subgroups defined by factors hypothesised to affect selection. Results: In 2017–2020, of 10,122 deaths ascertained, 5847 (57.8%) were enrolled in CHAMPS and 2654 (26.2%) additionally consented to MITS. Estimates were calculated for 265 and 65 site/age-specific causes of death and maternal conditions, respectively; five (1.9%) and four (6.2%) required adjustment, respectively, but they did not meaningfully change. Estimates were calculated for 34 site-specific causes of death among all stillbirths and under-five deaths combined; 28 (82.4%) required adjustment (all included age at death), and change-in-estimates demonstrated considerable variability. Conclusions: Selection bias is not a concern in the CHAMPS Network. Deaths where MITS were performed accurately represent the distribution of causes of death in their respective target populations, specifically when stratified by age or adjusted accordingly. Future studies of child mortality that employ MITS should consider adjusting for age at death for their measures of frequency.