Objective: This study aimed to investigate factors that relate to the Activities of Daily Living (ADL) at discharge in patients with acute stroke using machine learning.
Methods: The study included 246 patients admitted to five acute hospitals. The medical characteristics and clinical assessment sub-items of the patients were evaluated, and we used eXtreme Gradient Boosting (XGBoost) to predict whether the patients would be independent in ADL at discharge. The contributing factors were examined using SHapley Additive exPlanations (SHAP).
Results: Prediction accuracy was high, with an area under the curve of at least 0.85 for both training and test data. The following contributing factors were highly ranked: Functional Ambulation Category, Brünnstrom Recovery Stage of the lower limb, the “turn over from supine position” basic movement of Ability for Basic Movement Scale-II (ABMS-II), the “dressing” item of Barthel Index, and the “remain standing” basic movement of ABMS-II.
Conclusion: The study suggests that gait, lower limb function on the paralyzed side, and movement ability are the most important contributing factors to ADL at discharge in patients with acute stroke.