The Impact of Obesity and Frailty Indicators on Cognitive Decline in Older Adults: Findings from Machine Learning-Based Prediction

Background Cognitive decline is a major public health concern. While obesity and frailty are known risk factors, the relationships are often non-linear and complex. Traditional linear models may fail to capture the intricate interactions between diverse obesity phenotypes, physical function, and cognitive aging. This study aims to identify the non-linear patterns and key predictors of cognitive decline using machine learning across multinational cohorts. Methods We conducted a longitudinal analysis using data from three large aging cohorts: the China Health and Retirement Longitudinal Study (CHARLS), the Health and Retirement Study (HRS, USA), and the English Longitudinal Study of Ageing (ELSA). A total of 20,586 participants aged 60 years and older were included. Six machine learning models were trained to predict cognitive decline, with Random Forest (RF) demonstrating the highest predictive accuracy (AUC = 0.862). SHAP (SHapley Additive exPlanations) values and Generalized Additive Models (GAMs) were employed to interpret the black-box model and visualize non-linear relationships. Results The Random Forest model outperformed traditional logistic regression, highlighting the non-linear nature of the data. SHAP analysis identified waist-to-height ratio (WHtR) and grip strength as the top two modifiable predictors of cognitive decline (after age). Non-linear dependence plots revealed distinct U-shaped associations: both low and high values of WHtR and grip strength were associated with increased risk. Specifically, the lowest risk was observed at a WHtR of approximately 0.68 and a grip strength of 45.02 kg. Gait speed showed a V-shaped pattern, with a protective threshold below 0.84 m/s. These non-linear patterns were consistent across the three diverse cohorts. Conclusion Our findings demonstrate that the relationships between obesity, frailty indicators, and cognitive decline are fundamentally non-linear. Public health strategies should move beyond simple "one-size-fits-all" recommendations (e.g., just losing weight) towards personalized thresholds. Maintaining a waist-to-height ratio around 0.5–0.7 and preserving moderate muscle strength may be optimal for dementia prevention in older adults.