Supporting public health efforts in India and Nepal with probabilistic child death modelling

Abstract

Background: Child mortality remains a major public health concern in South Asia, shaping population dynamics and affecting family well-being. Understanding household-level mortality patterns is essential for identifying high-risk groups and developing effective interventions. This study analyzes child mortality data from households in Eastern Uttar Pradesh, India, and Nepal, where deaths are rare but occasionally clustered within families.

Methods: Four probabilistic models were applied to the observed number of child deaths per household: the Geometric distribution, the Inflated Geometric distribution to accommodate excess zeros, and the Himanshu compounded distribution. Model parameters were estimated using Maximum Likelihood Estimation (MLE). Model adequacy was evaluated through Chi-square goodness-of-fit tests comparing expected and observed household mortality counts.

Results: All models showed strong consistency with the empirical data. Chi-square test results produced high p-values (>0.95), indicating that each model successfully captured the zero-heavy structure and the infrequent higher mortality events present in the datasets from both regions.

Conclusions: The findings demonstrate that zero-inflated and compounded probabilistic models provide reliable representations of household-level child mortality in South Asia. These modeling approaches can support better identification of vulnerable households and improve the predictive accuracy of mortality assessments, contributing to more targeted public health strategies aimed at reducing child deaths.