A Machine Learning Approach for Estimating Intrinsic Capacity Age and Its Associations with Multimorbidity and Geroprotective Agents.

Abstract

Background and objectives: Aging is associated with functional decline and multimorbidity, highlighting the need for holistic biomarkers to monitor healthy aging. The aim of this study was to validate intrinsic capacity age (IC-age) as a biomarker of aging and to examine its association with multimorbidity and geroprotective agents.

Research design and methods: A cross-sectional study was conducted with data from 48,068 participants aged ≥60 years from the 9th wave of the Survey of Health, Ageing and Retirement in Europe (SHARE, 2021-2022). Random forest regression was used to train a model predicting IC-age based on biomarkers (cognitive, psychological, sensory, vitality, locomotion) and demographic factors.

Results: IC-age showed a prediction error of 5.3 years (r = 0.55). Biomarkers for vitality (handgrip strength), cognitive (verbal fluency and memory), and sensory (hearing aid use) domains were important contributors. General linear models assessed associations with multimorbidity, physical activity, and smoking. Intrinsic capacity age was significantly higher in individuals with multimorbidity and smokers compared with healthy individuals. Physical activity exhibited a protective effect on IC-age, with vigorous activity showing a particularly pronounced benefit in women.

Discussion and implications: This model demonstrates that IC domains can estimate biological age and distinguish individuals based on their comorbidities. It also underscores the role of physical activity as a key geroprotective factor, with vigorous physical activity in females with comorbidities showing the most pronounced protective effect on IC-age. These results validate the concept of IC-age as a comprehensive measure of aging and highlight its potential to inform personalized interventions and public health strategies.