Development of a Prognostic Nomogram for Severe Fever With Thrombocytopenia Syndrome Using Machine Learning: A Multicenter Retrospective Study

Severe fever with thrombocytopenia syndrome (SFTS), an emerging tick-borne infectious disease, is marked by rapid progression and high mortality. Current machine learning (ML)-based prognostic models for SFTS are not only scarce but also lack external validation. This multicenter retrospective cohort study aimed to develop a clinical risk prediction model for SFTS. It involved 1,215 SFTS patients from three Chinese hospitals and utilized multiple machine learning algorithms. ML algorithms were applied for variable selection and model construction. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC), accuracy, recall, and F1 score. The logistic regression (LR) algorithm-constructed predictive model achieved the best performance in both training and validation cohorts. Multivariate logistic regression was then used to build nomograms. The clinical utility of the optimal model was assessed via decision curve analysis (DCA) based on net benefit. Five independent risk factors for SFTS-related death were identified: age, consciousness disturbance, activated partial thromboplastin time (APTT), serum creatinine (SCr), and log- transformed viral load (lg viral load). Nomograms based on these factors exhibited high accuracy, with an AUC of 0.881 in the training group and 0.886 in the validation group. In conclusion, ML proves effective for identifying high -risk SFTS patients. ML - based prediction models and nomograms can accurately predict SFTS prognosis, and the five key features serve as valuable early - prognosis indicators deserving clinical attention.