OBJECTIVE  To develop a machine learning model based on clinical markers within 24 hours of admission to predict the risk of adverse outcomes in patients with severe fever with thrombocytopenia syndrome （SFTS） during hospitalization, and enable early risk stratification.

METHODS  A retrospective cohort study was conducted, enrolling 430 SFTS patients （346 in the improved group and 84 in the adverse outcome group） admitted to the First Affiliated Hospital of Nanjing Medical University from Apr. 2022 to May 2025. Clinical characteristics at admission and laboratory markers within 24 hours were collected. LASSO regression was applied for characteristic selection in the training set, and five algorithms （extreme gradient boosting, gradient boosting machine, random forest, support vector machine and logistic regression） were used to construct prediction models, with performance evaluated on an independent test set. SHAP method was employed to improve the model interpretability.

RESULTS  LASSO regression identified seven core predictors: age, SFTS viral nucleic acid load （log-transformed）, ferritin （Fer）, prothrombin time （PT）, serum creatinine （Scr）, lactate dehydrogenase （LDH） and procalcitonin （PCT）. The support vector machine （SVM） model demonstrated optimal overall performance in the test set （AUC=0.865, 95%CI: 0.781–0.948）, while other models achieved AUC of 0.832–0.858. SHAP analysis revealed age and lgSFTSV as the two most influential characteristics in model prediction.

CONCLUSIONS  This study develops and validates an interpretable machine learning model based on the SVM algorithm, which effectively predicts the risk of adverse outcomes in SFTS patients based on 24 h admission markers. The model integrates specific markers such as viral load and confirms the predictive value of age, lgSFTSV, coagulation and organ injury markers.