A Pilot Study of Machine Learning–Based Algorithms to Assist Integrated Care for Older Community-Dwelling Adults

Abstract

As life expectancy increases, there is a growing consensus on the development of integrated care encompassing the health and daily activities of older adults. In recent years, although the demand for machine learning applications in healthcare has increased, only a few studies have implemented machine learning–based systems in integrated care for older adults owing to the complex needs of older adults and the coarseness of the available data. Our study aims to explore the possibility of implementing machine learning decision-support algorithms in the integrated care of older community-dwelling adults. Our experiment uses secondary data based on the community-based integrated service model. Such data were collected from 511 older adults through 162 assessment items in which tailored services were selected from 18 available services. We implemented four machine learning models: decision tree, random forest, K-nearest neighbors, and multilayer perceptron. The area under the receiver operating characteristic curve results of the four models were decision tree = 0.89, K-nearest neighbors = 0.88, random forest = 0.93, and multilayer perceptron = 0.88. The results suggest that machine learning–based decision-assisting algorithms can improve the quality of tailored services for integrated care with intensive involvement of face-to-face tasks by reducing the simple, repetitive tasks of care managers.