Suitable survival models for analyzing infant and child mortality data

Abstract

Survival analysis offers a sophisticated framework for examining infant and child mortality, facilitating time-to-event analysis and the identification of critical risk factors. This study leverages data from the 2018 Nigerian Demographic and Health Survey (NDHS) to evaluate the appropriateness of various modeling approaches. It uncovers substantial violations of the proportional hazards assumption in the Cox model, underscoring the need for alternative strategies when this assumption fails. To address these issues, regularization techniques such as Lasso, Ridge, and Elastic Net are employed to refine model fit. The Lasso model, in particular, enhances interpretability by selectively eliminating less significant covariates, while Ridge and Elastic Net contribute marginally to model improvement. Among parametric survival models, the Lognormal model proves most effective for analyzing infant mortality, whereas the Weibull model surpasses both the Exponential and Lognormal models in fitting child mortality data, as evidenced by lower AIC, BIC, and superior log-likelihood values. These results highlight the efficacy of Lasso in variable selection and emphasize the importance of choosing appropriate parametric models for precise mortality analysis.