Using machine learning algorithms to guide rehabilitation planning for home care clients

Area of Science:

• Gerontology
• Health Informatics
• Rehabilitation Science

Background:

• Identifying clients eligible for rehabilitation in home care settings presents a significant clinical challenge.
• Older adults in home care require effective rehabilitation strategies to improve daily functioning and facilitate discharge.

Purpose of the Study:

• To evaluate the effectiveness of machine learning algorithms, specifically Support Vector Machine (SVM) and K-Nearest Neighbors (KNN), in guiding rehabilitation planning for home care clients.
• To compare the predictive performance of SVM and KNN with the existing Activities of Daily Living Clinical Assessment Protocol (ADLCAP).

Main Methods:

• Secondary analysis of data from 24,724 longer-term home care clients in Ontario, utilizing the RAI-HC assessment system.
• Definition of rehabilitation potential based on improvement in Activities of Daily Living (ADL) functioning or successful discharge home.
• Comparison of SVM and KNN predictive performance against the ADLCAP using identical functional and health status indicators.

Main Results:

• Both KNN and SVM algorithms demonstrated substantially improved performance in predicting rehabilitation potential compared to the ADLCAP.
• Despite improvements, both machine learning algorithms exhibited notable false positive and false negative rates (KNN/SVM: FP > .18, FN > .34; ADLCAP: FP > .29, FN > .58).

Conclusions:

• Machine learning algorithms offer superior predictive capabilities for identifying rehabilitation potential in home care clients compared to the current ADLCAP.
• While machine learning predictions are less interpretable, they provide valuable insights for refining and improving clinical assessment protocols.