A progressive prediction model towards home-based stroke rehabilitation programs

More than 795,000 people suffer from stroke each year in the United States. Rehabilitation is crucial to help restore mobility function in activities of daily life. Many research works manifest that effective prediction of rehabilitation outcomes can significantly improve personalized rehabilitation strategies and maximize health outcomes. Existing studies on predicting stroke rehabilitation outcomes are primarily based on clinical or inpatient data, and there are few in-depth investigations in home-based rehabilitation prediction. In this work, we perform a retrospective study on mRehab (mobile rehabilitation), a six-week home-based rehabilitation program, where we investigated the in-program user data across 12 different rehabilitation activities performed by 16 stroke survivors. We proposed a new progressive prediction framework that constructs the multiple linear regression (MLR) and the random forest (RF) models, and found that combining the clinical and demographic data of stroke survivors with movement phenotyping data will significantly improve the prediction model's performance. Specifically, our proposed model can accurately predict the outcome only with the first two weeks’ data with the root mean square error (RMSE) of 0.1050 in the MLR model and 0.0011 in the RF model. Moreover, we identify the features that significantly contribute to the prediction outcomes in home-based rehabilitation programs through feature correlation analyses. To the best of our knowledge, this work is the first study towards the rehabilitation outcome prediction in a home-based program.

Duke Scholars

Author Ming-Chun Huang DKU Faculty