Project Details

Practical Prediction of Inpatient Newborn Mortality in Nigeria and Kenya with Machine Learning

Shiraz Badurdeen, MA, MBBChir, PGDip, MRCPCH, FRACP, PhD

Summary

BACKGROUND: Newborn mortality in low- and middle-income countries (LMICs) remains unacceptably high, particularly in sub-Saharan Africa. Early identification of newborns at highest risk of death could enable more efficient use of scarce clinical resources, but existing prediction tools are outdated and impractical for frontline care. GAP: There is currently no accurate, context-appropriate, and implementable model to predict inpatient neonatal mortality using routinely collected data in LMIC hospital settings. This limits clinicians’ ability to prioritize care for the most vulnerable newborns. HYPOTHESIS: We hypothesize that routinely collected clinical data at hospital admission can be used to develop and validate prognostic models that accurately predict inpatient neonatal mortality in Nigeria and Kenya. METHODS: We will develop and validate machine-learning models using two large multicenter datasets: the Nigerian Oxygen Implementation Project (12 hospitals; n~7,000) for model development and the Kenyan Clinical Information Network (22 hospitals; n~100,000) for external validation. Models including XGBoost, Lasso regression, and decision trees will be evaluated using internal–external cross-validation, with performance assessed by discrimination, calibration, and clinical utility in accordance with TRIPOD-AI guidelines. RESULTS: Pending. IMPACT: This project will deliver practical risk-stratification tools, including a web-based application and simplified flowcharts, to support frontline clinicians in identifying high-risk newborns for targeted care. The findings will inform WHO risk stratification efforts and provide a foundation for future implementation and impact studies to improve newborn survival in resource-limited settings. Website Link: https://www.mcri.edu.au/researcher-details/shiraz-badurdeen