Prediction of Responders to Post-stroke Rehabilitation Therapy Based on Section GG of the Inpatient Rehabilitation Facility-Patient Assessment Instrument

Section GG of the Inpatient Rehabilitation Facility (IRF) -Patient Assessment Instrument was recently introduced to monitor functional changes and to help with achievable goal-setting during rehabilitation. Accurately predicting how patients will respond to standard therapies early in the rehabilitation process could help clinicians create more targeted, personalized treatments to improve their patients’ outcomes. In this work, we developed a Random Forest classifier to predict therapy responders, as defined by the Section GG score. The model achieved an accuracy of 81.4% using simple features collected at IRF admission and surpassed the accuracy of predictions based on therapist goals. Such a model could support more effective therapy planning by facilitating new interventions for patients expected not to respond to standard-of-care therapies, thereby improving patient outcomes.