Statistical Advances in Child Survival Modelling: A Data-driven Nonlinear Analysis of Under-Five Survival Inequalities in Ghana

Abstract

Under-five survival remains a core indicator of population health and health system performance in Ghana; much of the existing evidence relies on linear or strictly parametric models that poorly capture complex risk dynamics. This study sought to provide a statistically robust assessment of the determinants of under-five survival by explicitly modelling nonlinear effects and improving predictive validation. The objectives included identifying demographic, socioeconomic, and health service factors associated with survival, assessing nonlinear effects of fertility and household factors on survival, and evaluating model performance using discrimination and calibration metrics. It analysed a nationally representative dataset of 34,663 under-five records through a generalised additive model with a logit link and penalised splines estimated via restricted maximum likelihood. Results indicate a highly significant nonlinear effect of fertility-related timing with 7.04 effective degrees of freedom and a test statistic of 530.9 (p < 0.001), revealing sharp survival declines at higher exposure to fertility. Postnatal care showed a protective association (β = -0.892, p = 0.029), while maternal education showed an overall significant quadratic effect (p = 0.022). The model showed moderate discrimination (AUC = 0.757) and good calibration across deciles. This study's contribution to novelty lies in the functional-form misspecification and validation gaps left by previous studies. Findings support fertility-spacing interventions, strengthened postnatal care, and region-specific strategies to accelerate progress toward Sustainable Development Goal 3.