Age trajectories of disability in middle and older age: a machine learning approach

Abstract

Abstract Background The study employed machine learning methods to explore age trajectories of Instrumental Activities of Daily Living (IADL) disability in a representative cohort of middle- and older-aged individuals. This is crucial for understanding the impact of disability onset on healthy life expectancy (HLE) across sub-populations. Methods We utilized data from the Taiwan Longitudinal Study on Aging, which included 5,334 individuals aged 50 and over from 1996 to 2011. To investigate the change in physical functions, we focused on the 6 IADL items assessed over five waves of surveys. Each IADL item was dichotomised and summed to calculate a disability score ranging from 0 to 6 based on its difficulty level. We used K-means clustering to identify IADL disability trajectories and applied 10-fold cross-validation, repeated ten times, to select the optimal classification. We transformed the identified groups into age trajectories based on the mean IADL score at each age within each group. The age trajectories were then smoothed using locally estimated scatterplot smoothing and generalized additive models. Results We identified four trajectory groups based on changes in IADL scores over 15 years: (1) individuals who remained consistently able, (2) individuals with persistent and slower decline, (3) individuals with persistent and faster decline, and (4) individuals with persistent disability. The proportions of the four groups were 56.1%, 20.0%, 16.1% and 7.8%, respectively. The average posterior probabilities for all groups were above 83.3%. The results of this study were comparable to those of traditional regression-based models. We also analyzed the associations between trajectory group membership and participants’ characteristics at baseline. Conclusions This machine learning approach based on K-means clustering is valuable for investigating diverse disability trajectories within the population and can inform future research to explore factors that may enhance HLE. Key messages • Machine learning enables investigating disability trajectories across sub-populations. • The study identified four distinct disability trajectory groups using K-means clustering, providing insights into factors for enhancing healthy life expectancy.