External validation of a machine learning model for classifying hospitalization outcomes in Asian elephants across three elephant hospitals in Thailand

Classification of hospitalization outcomes in wildlife hospitals may support clinical assessment and resource planning, but model performance can vary across institutions with different case profiles and care practices. This study evaluated a previously developed random forest model for classifying hospitalization outcomes in Asian elephants using temporally distinct 2025 data from three elephant hospitals in Thailand. The validation comprised temporal validation at the original development center and institutional validation at two additional hospitals. Because length of stay was included as a predictor, the model was evaluated as a hospitalization-course classification model rather than as an admission-time prognostic tool. The final validation cohort comprised 237 hospitalization episodes contributed by 215 elephants, including 17 deceased, 36 ongoing, and 184 recovered episodes. Overall accuracy was 0.793, compared with 0.776 for a recovered-only majority-class classifier. Macro-averaged balanced accuracy was 0.732, the prevalence-weighted one-vs-rest area under the receiver operating characteristic curve (AUC) was 0.854, and multiclass log loss was 0.617. Classification performance was stronger for ongoing and recovered episodes than for deceased episodes. The limited number of deceased episodes resulted in substantial uncertainty around mortality-focused performance and calibration estimates. Calibration patterns differed across outcome classes, with an overall multiclass Brier score of 0.323. In the binary deceased-versus-recovered sensitivity analysis, the AUC was 0.726 (95% confidence interval, 0.585–0.839), indicating moderate but imprecise discrimination. Overall, the model retained discriminatory information across the validation cohort, but its modest improvement over the majority-class benchmark, weaker deceased-class performance, and reliance on hospitalization-duration information warrant cautious interpretation. Further prospective evaluation is needed before routine clinical application.