Physical Performance is Significantly More Important Than Health Condition When Considering Rehabilitation Outcome

Abstract

<h3>Research Objectives</h3> To determine the most important features in a rehabilitation prediction model using machine learning. <h3>Design</h3> Physical performance data were collected at evaluation and discharge from physical (PT) and occupational therapy (OT) in skilled nursing (SNF). These data along with the patient's prior living situation, medical diagnosis (grouped in broad categories), prior living situation, and age were used in a retrospective k-means cluster analysis. A profile of each group was developed using additional demographic and clinical process data. A predictive model was generated using an XGboost model with hyperparameters tuning. <h3>Setting</h3> Infinity Rehab is a post-acute therapy provider in over 150 skilled nursing facilities (SNF) treating approximately 13000 patients/year. <h3>Participants</h3> Data from 25000 older adult patients admitted to skilled nursing facilities were used for model development. <h3>Interventions</h3> The model was trained on 80% of the cases and tested on the remainder. <h3>Main Outcome Measures</h3> Time to rise from a chair 5 times (5xSTS), gait speed over 4 meters, 6-minute Walk Test (6MWT), grip strength (Grip), the Continuity Assessment Record Evaluation (CARE) mobility and self-care item set, and the St. Louis Mental Status Examination (SLUMS). <h3>Results</h3> The cluster analysis yielded 5 patient groupings who were different in baseline and discharge characteristics, change values, rehabilitation value, and prior and discharge living situation. 20 features were identified as contributing to the group assignment. The 3 most important features were initial 5xSTS, 6MWT, and gait speed, followed by measures of activity limitation. Of the 20 features, only 7 were medical diagnosis categories with the highest of these (non-stroke, tumor, or degenerative brain disorders) in the 13th position. Of the 7 medical diagnosis categories, 4 were neurological in nature. <h3>Conclusions</h3> When considering outcomes from rehabilitation in SNF, measures reflecting impairment in functional strength, activity tolerance, and speed are more important than activity limitations and much more important than medical diagnosis. Medical diagnosis impacted cluster group assignment the most when it was a neurological condition. <h3>Author(s) Disclosures</h3> The authors have nothing to disclose.