Utilizing Consumer-grade Wearable Sensors for Unobtrusive Rehabilitation Outcome Prediction

Rehabilitation outcome prediction can be useful for clinicians providing therapy services to patients undergoing inpatient medical rehabilitation. Machine learning models trained with medical record information available at admission can predict rehabilitation outcomes fairly well. In our previous work, we found rehabilitation outcome prediction accuracy can be improved by also including inertial sensor-based features that objectively quantify patient movement abilities during therapy tasks. In this paper, we extend our prior work by unobtrusively and continuously collecting minute-by-minute movement data from 15 patients throughout their stay of inpatient rehabilitation using inexpensive, consumer-grade fitness trackers, specifically the Fitbit Charge with heart rate. From the Fitbit time series data, we extract features related to physical activity, heart rate, and sleep quality. We use these features as inputs to machine learning models to predict the discharge Functional Independence Measure (FIM) rehabilitation outcome. We also utilize patient similarity techniques to improve prediction accuracy. Results indicate prediction accuracy with the consumer-grade sensor data is close to the same accuracy as prior work using research-grade inertial sensor data. Using consumer-grade fitness devices to obtain highly accurate FIM predictions can help clinicians plan therapy activities during the inpatient stay, as well as assist with discharge to an appropriate setting.