Parametric Survival Analysis of Neonatal Mortality in a Tertiary Care Setting: A Comparative Proportional Hazard Model Approach

Abstract

Background: Neonatal mortality represents a major public health concern characterised by complex survival dynamics and substantial aetiological heterogeneity. Appropriate survival modelling of time-to-event data is essential for elucidating disease-specific risk profile during the critical first 28 days of life. Neonatal mortality in intensive care settings exhibits an early-peak hazard profile inadequately captured by semiparametric methods alone. Comparative evaluation of parametric proportional hazard (PH) models that explicitly parameterise the baseline hazard provides both superior fit diagnostics and direct hazard quantification.
Objectives: To compare parametric survival models and identify the optimal distributional fit for neonatal mortality data, and to determine prognostic factors using the best-fitted model for neonatal outcome.
Methods: A prospective study design was used to collected the data of 686 neonates admitted to the Neonatal Intensive Care Unit (NICU) and Special Newborn Care Unit (SNCU) at Dr. Ram Manohar Lohia Institute of Medical Sciences (RMLIMS), Lucknow, was followed from admission until death, discharge, transfer, or day 28. Four survival models were fitted and compared by Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The proportional hazards (PH) assumption was assessed via Schoenfeld residuals with global test. Discriminative ability was evaluated using Harrell's C-concordance statistic
Results: Among 686 neonates, 155 (22.59%) died during follow-up. The Weibull PH model achieved the best fit (AIC = 827.77; BIC = 927.44). Key independent predictors included extreme prematurity, congenital malformations, perinatal asphyxia, neonatal sepsis, induced labour, and multiparity. The global Schoenfeld test supported the proportional hazards assumption (p = 0.0976). Harrell's C = 0.7948 indicated good discriminative performance.
Conclusions: The Weibull PH model provides the best parametric characterisation of neonatal mortality hazard dynamics. Extreme prematurity, critical clinical diagnoses, and induced labour are dominant independent hazard determinants, consistent across all model specifications.