Opinion versus practice regarding the use of rehabilitation services in home care: an investigation using machine learning algorithms.

Abstract

BackgroundResources for home care rehabilitation are limited, and many home care clients who could benefit do not receive rehabilitation therapy. The interRAI Contact Assessment (CA) is a new screening instrument comprised of a subset of interRAI Home Care (HC) items, designed to be used as a preliminary assessment to identify which potential home care clients should be referred for a full assessment, or for services such as rehabilitation. We investigated which client characteristics are most relevant in predicting rehabilitation use in the full interRAI HC assessment.MethodsWe applied two algorithms from machine learning and data mining ― the LASSO and the random forest ― to frequency matched interRAI HC and service utilization data for home care clients in Ontario, Canada.ResultsAnalyses confirmed the importance of functional decline and mobility variables in targeting rehabilitation services, but suggested that other items in use as potential predictors may be less relevant. Six of the most highly ranked items related to ambulation. Diagnosis of cancer was highly associated with decreased rehabilitation use; however, cognitive status was not.ConclusionsInconsistencies between variables considered important for classifying clients who need rehabilitation and those identified in this study based on use may indicate a discrepancy in the client characteristics considered relevant in theory versus actual practice.