Development of Prediction Models for Healthy Ageing in Community-Dwelling Middle-Aged and Older Adults: A...

Healthy ageing is a central issue in public health; however, there is a lack of consensus regarding its determinants. We developed machine learning (ML) models to predict healthy ageing based on the characteristics of community-dwelling middle-aged and older adults.

A retrospective study.

This cohort study included participants aged 50 years or older from the English Longitudinal Study of Ageing.

We selected sociodemographic, health, lifestyle, and psychosocial characteristics at baseline. The outcome was healthy ageing at a 4-year follow-up, assessed based on functional status, preserved mobility, preserved muscle strength, absence of elevated depressive symptoms, absence of chronic diseases, and preserved cognitive function. We used the decision tree, logistic regression, neural network, and random forest algorithms to develop ML models and applied the SHapley Additive exPlanations algorithm to determine the contribution, positive or negative, of each predictor to the outcome.

Of the 6332 participants at baseline (median age 64 years), 27.9% were ageing healthily after 4 years. The ML model based on the random forest algorithm achieved the best performance on the test data set (area under the curve = 0.78, 95% CI 0.76-0.80). Normal physical performance, greater household wealth, chronological age, and self-perceived age between 50 and 59 years positively contributed, whereas physical inactivity, abdominal obesity, and not using the internet or email at baseline negatively affected the outcome.

ML models can help predict healthy ageing based on the characteristics of community-dwelling middle-aged and older adults. The available evidence can provide the basis for health strategies to promote active aging.