Short Birth Intervals, Maternal Care, and Neonatal Mortality in Rwanda: A Facility-Based Retrospective Cohort Study Using Hierarchical Modelling.

BACKGROUND

Neonatal mortality remains a major public health concern in sub-Saharan Africa. Evidence from facility-based studies may be affected by selection and referral biases. This study examined the association between short birth intervals, maternal care, and other maternal, reproductive, and healthcare factors at different causal levels and neonatal mortality in a referral hospital in Rwanda, using a hierarchical modelling approach.

METHODS

A retrospective cohort study was conducted among 1042 mother-neonate pairs admitted between 2022 and 2023. Variables were grouped into distal (socioeconomic), intermediate (reproductive), and proximal (clinical) levels. Sequential logistic regression models were applied. Due to substantial non-random missingness in birth interval data, complete-case analysis was used for intermediate model, with sensitivity analyses to assess robustness. A Directed Acyclic Graph was used to evaluate potential biases.

RESULTS

Among 1,042 neonates, 32 (3.07%) died. Lower maternal education was associated with higher odds of neonatal mortality (aOR 2.18; 95% CI 1.08-4.54), although this may reflect selection bias. Short birth interval showed an inverse association with neonatal mortality, but this finding was unstable and not interpretable due to non-random missing data and bias. Proximal factors showed consistent associations: inadequate antenatal care (<4 visits) was associated with higher odds (aOR 3.12; 95% CI 1.29-7.54), and preterm birth was strongly associated (aOR 11.67; 95% CI 3.74-36.35).

CONCLUSION

Proximal clinical factors, particularly preterm birth and inadequate antenatal care, were consistently associated with neonatal mortality. Associations with distal and intermediate factors were affected by bias and missing data, limiting their interpretability. These findings underscore the need for caution when interpreting hospital-based data and highlight the value of hierarchical modelling with causal frameworks.