Machine learning-based prediction of fall within 6 months of stroke onset: A prospective study.

OBJECTIVE: To develop and evaluate machine learning-based models for predicting fall risk within 6 months of stroke onset. METHODS: This prospective study enrolled acute stroke patients from three tertiary hospitals in Gansu Province, China. The study was conducted from December 2022 to October 2024. Participants were followed for 6 months. Two machine learning algorithms, decision tree (DT) and logistic regression (LR), were developed. Baseline data were collected within 72 h of admission. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, accuracy, sensitivity, specificity, precision, F1-score. RESULTS: Of the 388 patients in the final analytic cohort, 286 (73.7%) were male and 102 (26.3%) were female, with a mean age of 64.27 ± 11.30 years. The incidence of falls within 6 months after stroke was 28.1% (109/388). The LR model achieved an AUC of 0.859, accuracy of 79.3%, sensitivity of 0.547, specificity of 0.890, precision of 0.663, F1-score of 0.599, and Brier score of 0.133. The DT model achieved an AUC of 0.863, accuracy of 81.4%, sensitivity of 0.618, specificity of 0.892, precision of 0.692, F1-score of 0.652, and Brier score of 0.132. The LR model was visualized as a nomogram. CONCLUSIONS: The LR and DT models showed comparable predictive performance for identifying patients at risk of falls within 6 months after stroke onset. These interpretable models may support early risk stratification. However, external validation is required to confirm their generalizability.