Prospective classification of functional dependence: Insights from machine learning and the Canadian Longitudinal Study on Aging

Abstract

PurposeFunctional dependence is a multifactorial health condition affecting well-being and life expectancy. To better understand the mechanisms underlying this condition, we aimed to identify the variables that best prospectively classify adults with and without limitations in basic and instrumental activities of daily living.

MethodsA filtering approach was used to select the best predictors of functional status from 4,248 candidate predictors collected in 39,927 participants aged 44 to 88 years old at baseline. Several machine learning models using the selected baseline variables (2010-2015) were compared for their ability to classify participants by functional status (dependent vs. independent) at follow up (2018-2021) on a training dataset (n=31,941) of participants from the Canadian Longitudinal Study on Aging. The best performing model was then examined on a test dataset (n=7,986) to confirm sensitivity, specificity, and accuracy.

ResultsEighteen candidate baseline variables were identified as the best predictors of functional status at follow up. Logistic regression was the best performing model for classifying participants by functional status and achieved balanced accuracy of 81.9% on the test dataset. The absence of functional limitations at baseline, stronger grip strength, being free of pain and of chronic conditions, being a female, having a drivers license, and good memory were associated with greater odds functionally independence at follow-up. In contrast, older age, psychological distress, walking slowly, being retired, having one or more chronic conditions, and never going for walks were associated with greater odds of functional dependence at follow-up.

ConclusionFunctional status can be best prospectively estimated by health condition, age, muscle strength, short-term memory, physical activity, psychological distress, and sex. These predictors can estimate functional status over 6 years ahead with high accuracy. This early identification of people at risk of functional dependence allows sufficient time for the implementation of interventions aiming to delay functional decline.