Predicting rehabilitation treatment helpfulness to stroke patients: A supervised learning approach

Abstract

Stroke (Cerebral vascular accident, CVA) is a common and serious disease. Most of the survivals would be disabled after their illness recovery, causes serious burden on caregivers. It is said that rehabilitation could help functional recovery of stroke patients, regain independence after stroke. Due to the long course of stroke, how to prevent survivals from recurrence is an important issue. This study attempts to examine the relationship between stroke recurrence and strength of rehabilitation, and build a stroke recurrence prediction model utilizing a number of supervised learning techniques to assist physicians with making clinical decisions.

In the past, most of the related work used the samples from a single hospital as a sample, but it cannot fully catch all the clinic information of the patients. Therefore, this study used the Longitudinal Health Insurance Database 2010 of the NHIRD as the data source, to examine the effectiveness of rehabilitation.

In terms of accuracy rate of all classifiers, we get the best effectiveness (78%) while adopting the inpatient admission dataset and C4.5 to predict recurrence. We also find physical therapy, occupational therapy and speech therapy treatments during inpatient admission are the key factors to decrease the chance to recrudesce in the rehabilitation periods. The higher strength and frequency rehabilitation treatment is also the key influence variables in our high accuracy prediction model which means that is useful to lower the recurrence rate of stroke patients.