Modelling Mortality Risk in Malaria Patients Using Logistic Regression Model: A Case Study of Nakuru Level 6 Hospital

Malaria remains a major cause of preventable illness and death in sub-Saharan Africa, yet routinely collected hospital data are not consistently integrated into objective tools for identifying patients at greatest risk of death. This study developed and internally evaluated a five-predictor logistic regression model for in-hospital mortality among malaria patients admitted to Nakuru Level 6 Hospital, Kenya. A retrospective observational design was used, drawing on electronic medical records from 1,500 patients admitted between 2020 and 2026. In-hospital death was the binary outcome, with age, sex, parasite density, haemoglobin level, and platelet count included as predictors. Fifty-four patients died, corresponding to a mortality rate of 3.6%. The original prespecified linear model showed strong discrimination (AUC = 0.922), but diagnostic testing identified significant nonlinearity for age, haemoglobin level, and platelet count. These three predictors were therefore refitted using restricted cubic splines while retaining the same five clinical predictors. The corrected nonlinear model significantly improved fit and discrimination, with an apparent AUC of 0.962. Bootstrap validation produced an optimism-corrected AUC of 0.955 and a Brier score of 0.0275, while repeated stratified cross-validation produced an AUC of 0.954 and a Brier score of 0.0273. Calibration intercepts were close to zero, although slopes below one indicated some residual optimism. Parasite density remained positively associated with mortality, sex remained non-significant, and nonlinear relationships were observed for age, haemoglobin, and platelet count. The conventional 0.5 threshold showed poor sensitivity, while lower thresholds improved detection but were not considered clinically established cut-offs. The findings support use of routine hospital data for mortality risk stratification, but external validation, possible recalibration, and prospective evaluation of clinical thresholds and utility are required before implementation in practice.