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

Area of Science:

• Neuroscience
• Medical Informatics
• Biostatistics

Background:

• Multiple System Atrophy (MSA) is a fatal neurodegenerative disorder with unpredictable progression and poor outcomes.
• Accurate survival prediction is crucial for managing MSA patients.

Purpose of the Study:

• To characterize survival patterns in MSA patients.
• To identify key prognostic factors influencing survival.
• To develop an interpretable machine learning model for personalized survival prediction in MSA.

Main Methods:

• A multicenter longitudinal cohort study involving 391 MSA patients.
• Survival analysis using Kaplan-Meier and Cox regression.
• Development and validation of six machine learning survival models, including random survival forest, with SHAP for interpretability.

Main Results:

• Median survival was 6.9 years over a 4.9-year follow-up, with 149 deaths.
• The random survival forest model demonstrated superior predictive performance (C-index 0.769, AUC 0.815) using 11 predictors.
• An interactive web tool and risk score were created for clinical use.

Conclusions:

• An interpretable machine learning model accurately predicts individual survival in MSA.
• Key prognostic factors for MSA survival were identified.
• The developed tool can aid in clinical decision-making and patient counseling.