Machine Learning–Based Prediction of Functional Disability: a Cohort Study of Japanese Older Adults in 2013–2019

Abstract

Abstract Background It is important to identify older adults at high risk of functional disability and to take preventive measures for them at an early stage. To our knowledge, there are no studies that predict functional disability among community-dwelling older adults using machine learning algorithms. Objective To construct a model that can predict functional disability over 5 years using basic machine learning algorithms. Design A cohort study with a mean follow-up of 5.4 years. Participants We used data from the Japan Gerontological Evaluation Study, which involved 73,262 people aged  ≥ 65 years who were not certified as requiring long-term care. The baseline survey was conducted in 2013 in 19 municipalities. Main Measures We defined the onset of functional disability as the new certification of needing long-term care that was ascertained by linking participants to public registries of long-term care insurance. All 183 candidate predictors were measured by self-report questionnaires. Key Results During the study period, 16,361 (22.3%) participants experienced the onset of functional disability. Among machine learning–based models, ridge regression (C statistic = 0.818) and gradient boosting (0.817) effectively predicted functional disability. In both models, we identified age, self-rated health, variables related to falls and posture stabilization, and diagnoses of Parkinson’s disease and dementia as important features. Additionally, the ridge regression model identified the household characteristics such as the number of members, income, and receiving public assistance as important predictors, while the gradient boosting model selected moderate physical activity and driving. Based on the ridge regression model, we developed a simplified risk score for functional disability, and it also indicated good performance at the cut-off of 6/7 points. Conclusions Machine learning–based models showed effective performance prediction over 5 years. Our findings suggest that measuring and adding the variables identified as important features can improve the prediction of functional disability.