| Sumario: | Objectives Cognitively, physically, and socially active lifestyles are important factors of healthy aging. Our objective is to leverage novel machine learning techniques in an ethnoracially diverse population to yield powerful, explainable models of modifiable lifestyle factors contributing to memory in older adults. Methods Cross-sectional episodic memory was the outcome in the gradient-boosted tree models and SHAP (SHapley Additive exPlanations). Participants in the Kaiser Healthy Aging and Diverse Life Experiences (KHANDLE) and the Study of Healthy Aging African Americans (STAR) made up the cohort (n = 2,245). Inputs included income, sex, age, education, ethnoracial groups, and 25 modifiable lifestyle factors, which were grouped into a triad of categories: leisure time activities (LTAs), physical metrics, and socialization. SHAP values indicated global feature importance and individual variability. Global importance, aggregated in each triad category, was leveraged for comparison of relative contributions to memory. Results The global feature importance for predicting episodic memory over each of the triad categories was as follows: LTAs (0.225), physical metrics (0.169), and socialization (0.104). Each category outweighed income (0.042), ethnoracial groups (0.070), and education (0.093). LTAs had a greater importance value than sex (0.224) and age (0.203) for predicting episodic memory. Explanatory analysis depicted varying importance rankings on an individual basis. Discussion The results confirm the importance of lifestyle variables and highlight the relative contributions of the triad categories to episodic memory in aging. This suggests that these categories may equal or surpass the importance of factors like sex, age, ethnoracial groups, and education, which are known to affect memory.
|
|---|