Machine Learning Models Show High Discriminatory Value in Predicting Postdischarge Mortality in Neonates and Young Children

Researchers from Muhimbili University of Health and Allied Sciences in Dar es Salaam, Tanzania have made significant progress in developing machine learning models to identify neonates and young children at risk for postdischarge mortality. The study, which analyzed data from a prospective observational cohort at Muhimbili National Hospital in Dar es Salaam, Tanzania and John F. Kennedy Medical Center in Monrovia, Liberia, used six machine learning algorithms to develop risk assessment tools. The findings suggest that machine learning models can have greater discriminatory value than traditional logistic regression in predicting postdischarge mortality in these vulnerable populations.

Key Takeaways:

• The study enrolled a total of 2310 neonates and 1933 young children, of which 71 (3.1%) neonates and 67 (3.5%) young children died after hospital discharge.
• The machine learning models, including XGBoost, Hist Gradient Boost, and Neural Network, yielded the greatest discriminatory value (area under the receiver operating characteristic curves range: 0.94-0.99) and fewest features, which included six features for neonates and five for young children.
• For neonates, discharge against medical advice, low birth weight, and supplemental oxygen requirement during hospitalization were predictive of postdischarge mortality.
• For young children, discharge against medical advice, pallor, and chronic medical problems were predictive of postdischarge mortality.
• The study's findings suggest that machine learning models can have excellent discriminatory value in predicting postdischarge mortality among neonates and young children, and external validation of these tools is warranted to assist in the design of interventions to reduce postdischarge mortality in these vulnerable populations.

Statistics:

• 71 (3.1%) neonates and 67 (3.5%) young children died after hospital discharge.
• Machine learning models yielded the following area under the receiver operating characteristic curves: XGBoost (0.94), Hist Gradient Boost (0.97), Neural Network (0.99).
• The models included six features for neonates: discharge against medical advice, low birth weight, supplemental oxygen requirement, chronic medical problems, pallor, and age.
• The models included five features for young children: discharge against medical advice, pallor, chronic medical problems, low birth weight, and supplemental oxygen requirement.

Sources:

• (2025) Machine learning approaches to identify neonates and young children at risk for postdischarge mortality in Dar es Salaam, Tanzania and Monrovia, Liberia. BMJ Paediatrics Open, 9(1).
• Emory Pediatric Research Alliance Junior Faculty Focused Award; Boston Children's Hospital Global Health Program; National Institutes of Health; Palfrey Fund For Child Health Advocacy; Emory University Pediatric Biostatistics Core.