Unpacking the null: a post-hoc analysis of a cluster-randomised controlled trial of the WHO Safe Childbirth Checklist in Uttar Pradesh, India (BetterBirth)

Abstract

Background: A coaching-based implementation of the WHO Safe Childbirth Checklist in Uttar Pradesh, India, improved adherence to evidence-based practices, but did not reduce perinatal mortality, maternal morbidity, or maternal mortality. We examined facility-level correlates of the outcomes, which varied widely across the 120 study facilities.

Methods: We did a post-hoc analysis of the coaching-based implementation of the WHO Safe Childbirth Checklist in Uttar Pradesh. We used multivariable modelling to identify correlations between 30 facility-level characteristics and each health outcome (perinatal mortality, maternal morbidity, or maternal mortality). To identify contexts in which the intervention might have had an effect, we then ran the models on data restricted to the period of intensive coaching and among patients not referred out of the facilities.

Findings: In the multivariable context, perinatal mortality was associated with only 3 of the 30 variables: female literacy at the district level, geographical location, and previous neonatal mortality. Maternal morbidity was only associated with geographical location. No facility-level predictors were associated with maternal mortality. Among facilities in the lowest tertile of birth volume (<95 births per month), our models estimated perinatal mortality was 17 (95% CI 11·7-24·8) per 1000 births in the intervention group versus 38 (31·6-44·8) per 1000 in the control group (p<0·0001).

Interpretation: Mortality was not directly associated with measured facility-level indicators but was associated with general risk factors. The absence of correlation between expected predictors and patient outcomes and the association between improved outcomes and the intervention in smaller facilities suggest a need for additional measures of quality of care that take into account complexity.

Funding: Bill & Melinda Gates Foundation.

Figure 1

Variation in perinatal mortality (A), maternal mortality (B), and prevalence of maternal morbidity (C) at the facility level Each column represents one of the 120 facilities included in the BetterBirth trial, sorted by outcome.

Figure 2

Estimated perinatal mortality by group and tertile of birth volume Mean (95% CI) estimates were based on multivariable models, adjusted for the proportion of female literacy in the district, geographical location in central region, previous neonatal mortality, proportion of referrals, and proportion of caesarean-section deliveries; the analysis was restricted to the intensive-coaching phase, excluding referrals. Each tertile includes 40 facilities.