SEVMALrisk: An Externally Validated Clinical Risk Score for Early Risk Stratification of Severe Imported Malaria

Early recognition of severe imported malaria is essential because treatment decisions frequently need to be made before reliable parasite density measurements become available. Objective tools to support early risk stratification during this interval are lacking. This study developed and externally validated SEVMALrisk, an explainable clinical risk score derived from a machine-learning prediction model using routinely available admission laboratory parameters.

Adult patients with confirmed malaria from the Rotterdam Malaria Cohort (n = 832) were used for prediction model development. Eight supervised machine-learning algorithms were compared. The best-performing model was optimized, interpreted using SHapley Additive exPlanations (SHAP) and translated into the SEVMALrisk clinical risk score. Independent external validation was performed in a multicenter cohort (n = 124).

Random Forest showed the highest discriminative performance and was selected for model development. The optimized prediction model achieved an area under the receiver operating characteristic curve (AUROC) of 0.962 with good calibration (Brier score 0.048). SHAP analysis identified eight routinely available laboratory parameters (total bilirubin, ASAT, leukocytes, thrombocytes, erythrocytes, creatinine, CRP and urea) for inclusion in the clinical risk score. The predefined decision threshold (26 points) remained unchanged during external validation. Internal validation yielded likelihood ratios of 9.39 (LR+) and 0.19 (LR-), corresponding to post-test probabilities of 51.9% and 2.1%, respectively. External validation demonstrated comparable performance (LR+ 8.02; LR- 0.18), with post-test probabilities of 59.2% and 3.2%.

SEVMALrisk is an explainable and externally validated clinical risk score for early risk stratification of severe imported malaria. Rather than replacing parasite density assessment or WHO severity criteria, it supports evidence-based treatment decisions before reliable parasite quantification becomes available.