Development and Validation of a Predictive Model for Severe Malaria Outcomes in Nigerian Children Using Machine Learning Approaches: Evidence from a Multi-Site Cohort

Malaria severity prediction in children at clinical presentation remains a critical challenge in resource-limited settings, where triage decisions directly affect outcome. Existing severity scores lack validation in West African paediatric populations. This study develops and validates a machine learning predictive model for severe malaria outcomes using a multi-site prospective cohort of children aged 6 months to 12 years admitted with falciparum malaria across six tertiary hospitals spanning Nigeria's six geopolitical zones. A total of 3,240 children were enrolled over 24 months from 2022 to 2023. Standardised data collection captured 48 clinical, laboratory, and sociodemographic variables at admission. Outcome was defined as a composite of death, neurological sequelae, or ICU-level care requirement at 72 hours. Five machine learning algorithms were compared: logistic regression, random forest, gradient boosting, support vector machine, and a deep neural network. The gradient boosting model outperformed all others with an area under the receiver operating characteristic curve (AUROC) of 0.91, sensitivity of 86.4%, and specificity of 88.2% in external validation. The top predictive features included Glasgow Coma Scale score, blood glucose concentration, haemoglobin level, respiratory rate, and peripheral parasitaemia density. The model demonstrated superior performance to the Lambaréné Organ Dysfunction Score across all sites. An open-source clinical decision support tool was developed and piloted in two sites. This study makes a major original methodological and clinical contribution by delivering the first Nigeria-validated machine learning severity prediction model for paediatric malaria. Keywords: machine learning, severe malaria, predictive model, multi-site cohort, Nigeria.