Machine learning enables early risk stratification of hymenopteran stings: evidence from a tropical multicenter cohort.

Background: Hymenopteran stings (from bees, wasps, and hornets) can trigger severe systemic reactions, especially in tropical regions, risking patient safety and emergency care efficiency. Accurate early risk stratification is essential to guide timely intervention. Objective: To develop and validate an interpretable machine learning model for early prediction of severe outcomes following hymenopteran stings. Methods: We retrospectively analyzed 942 cases from a multicenter cohort in Hainan Province, China. Questionnaires with >20% missing data were excluded. Mean substitution was applied for primary missing data imputation, with multiple imputation by chained equations (MICE) used for sensitivity analysis. Seven supervised classifiers were trained using five-fold cross-validation; class imbalance was addressed using the adaptive synthetic sampling (ADASYN) algorithm. Model performance was evaluated via area under the receiver operating characteristic curve (AUC), recall, and precision, and feature importance was interpreted using Shapley additive explanations (SHAP) values. Results: Among 942 patients, 8.7% developed severe systemic complications. The distribution by species was: wasps (25.5%), honey bees (8.9%), and unknown species (65.6%). The optimal Extra Trees model achieved an AUC of 0.982, recall of 0.956, and precision of 0.926 in the held-out validation set. Key predictors included hypotension, dyspnea, altered mental status, elevated leukocyte counts, and abnormal creatinine levels. A web-based risk calculator was deployed for bedside application. Given the small number of high-risk cases, these high AUC values may overestimate real-world performance and require external validation. Conclusion: We developed an interpretable, deployable tool for early triage of hymenopteran sting patients in tropical settings. Emergency integration may improve clinical decisions and outcomes.