Interpretable Machine Learning Model for Individualized Survival Prediction of Multiple System Atrophy

Background Multiple system atrophy (MSA) is a fatal neurodegenerative disease with highly variable progression and poor prognosis. This study aimed to characterize survival patterns, identify prognostic factors, and develop an interpretable machine learning model for individualized survival prediction in MSA. Methods In this multicenter longitudinal cohort study, 391 MSA patients were analyzed. Prognostic factors were identified using Kaplan–Meier and Cox regression analyses. Six survival models were trained with five‐fold cross‐validation, with Shapley Additive Explanations (SHAP) being used for feature selection and interpretability. Results During a median follow‐up of 4.9 years, 149 deaths occurred, with a median survival of 6.9 years. The random survival forest model achieved the best performance (C‐index 0.769; mean time‐dependent area under the curve [AUC] 0.815) with 11 predictors. An interactive web‐based platform and a risk score were developed for clinical application. Conclusion This interpretable machine learning model accurately predicted individualized survival in MSA and identified key prognostic factors.