Interpretable Machine Learning for Stroke Recovery: Predicting Discharge and 3-Month Functional Outcomes.

Area of Science:

• Neurology
• Data Science
• Medical Informatics

Background:

• Stroke is a primary cause of long-term disability globally.
• Understanding factors influencing recovery is crucial for effective patient management.
• Modified Rankin Scale (mRS) is a key outcome measure for stroke survivors.

Purpose of the Study:

• To investigate factors influencing modified Rankin Scale scores post-stroke using Machine Learning.
• To analyze the evolution of these influencing factors over time (discharge vs. 3 months post-discharge).
• To interpret the significance of various factors using SHAP (Shapley Additive Explanations).

Main Methods:

• Analysis of data from 116 stroke patients.
• Application of four predictive models: Logistic Regression, Support Vector Machine, Random Forest, and Extreme Gradient Boosting (XGB).
• Utilized SHAP for interpreting the significance of predictive factors.

Main Results:

• The XGB model demonstrated strong predictive performance (AUC 79% at discharge, 87% at 3 months).
• National Institutes of Health Stroke Scale was most critical at discharge.
• Post-discharge destination became more significant at three months, alongside age, time metrics, thrombolysis, and long-term health management.

Conclusions:

• Stroke recovery is a dynamic process with evolving influencing factors.
• Early clinical interventions are vital, but long-term health management gains importance.
• Findings support tailored rehabilitation strategies and informed discharge decisions based on evolving patient needs.