Predictors of mortality among neonates in Lusaka, Zambia: a comparative analysis of machine learning and traditional survival analysis techniques.

Introduction: Neonatal mortality remains a critical global health issue, with 2.3 million deaths in 2022. Sub-Saharan Africa bears 57% of under five deaths despite only 30% of global births, with Zambia ranking fourth highest in terms of neonatal mortality among neighboring countries. While traditional survival analysis has identified neonatal mortality risk factors, machine learning-based prediction remains underexplored. This study aimed to identify factors associated with neonatal mortality and compare the predictive performance of traditional survival analysis and machine learning models among neonates in Lusaka, Zambia (January2018-September 2019). Methods: Demographic and clinical data from 1,018 neonates were analyzed using seven models: Weibull, Lasso, Ridge, Elastic Net (regularized Cox), Random Survival Forests, DeepSurv neural networks and Gradient Boosting Machines. Model performance was evaluated using nested cross-validation with five outer folds and three inner folds for hyperparameter tuning. Predictive accuracy was assessed using the concordance index, time dependent area under the curve at 7, 14, and 28 days, brier scores, and calibration plots. Kaplan-Meier plots illustrated survival probabilities over time. Results: Of the 1,018 neonates, 757 (74.3%) died. Hypoxic-ischemic encephalopathy (TR = 0.71, 95% CI: 0.63-0.81) was associated with reduced survival, while higher birthweight was protective (TR = 1.88, 95% CI: 1.60-2.20). Sepsis demonstrated a paradoxical association with longer survival (TR = 1.16, 95% CI: 1.04-1.30), which persisted in sensitivity analyses. Among predictive models, the Random Survival Forests achieved the highest discrimination (C-index = 0.731) and consistently low Brier scores, outperforming Weibull (C-index = 0.622) and penalized Cox models (≈ 0.620). Gradient Boosting Machines were most miscalibrated, and DeepSurv showed low discrimination (C-index = 0.553). Feature importance analysis from Random Survival Forest identified birth weight as the dominant predictor, followed by sex, sepsis, and necrotizing enterocolitis. Discussion: While traditional Weibull models remain valuable for interpretability, machine learning approaches provide enhanced predictive accuracy. Hybrid modeling strategies may improve early risk identification and inform neonatal care in resource-limited settings.