Accuracy Comparison Between Artificial Intelligence and Multiple Regression Models for Predicting Motor Functional Independence Measure in Patients with Stroke in a Recovery-phase Rehabilitation Ward

Abstract

Objective: Accurate prognostication after stroke is essential for rehabilitation planning and discharge support. Predictive models based on artificial intelligence (AI) have recently gained attention as alternatives or complements to conventional statistical approaches. This study compared an AI model with a multiple regression model for predicting discharge motor Functional Independence Measure (FIM) scores in patients with stroke in a recovery-phase rehabilitation ward.

Methods: We included 498 patients with stroke. An AI model and a multiple regression model were developed to predict discharge motor FIM scores, and model performance was evaluated using cross-validation.

Results: The AI model achieved a significantly lower median absolute error than the regression model (p<0.001). In addition, approximately 40% of AI predictions were within five points of the observed discharge motor FIM score, approximately 10% higher than the regression model.

Conclusion: These findings suggest that AI models, which demonstrated good accuracy in this study, may provide superior predictive performance in this setting by capturing nonlinear relationships and complex interactions among variables. However, interpretability and transparency remain important limitations. AI and conventional statistical models should therefore be regarded as complementary tools, depending on the clinical purpose and context.