Cross-national analysis of social determinants of frailty among middle-aged and older adults: a machine learning study in the USA, England, and China

Frailty has become a growing global health concern and is associated with social determinants of health (SDoH). However, the relative importance and cumulative contribution of multidomain SDoH to frailty, and whether these relationships differ across countries, require further investigations. We included participants aged ≥45 years from the USA (N = 5792), England (N = 3773), and China (N = 5016). SDoH (n = 121 for the USA, n = 125 for England, and n = 94 for China) were selected across seven domains. Frailty was assessed by the frailty index (FI). We developed Extreme Gradient Boosting to predict frailty at the 4-year follow-up and used SHapley Additive exPlanations to quantify variable-wise and domain-wise contributions of SDoH, and to explore nonlinear relationships between SDoH and frailty. Our models explained 0.242 (95% confidence interval [CI]: 0.203–0.281), 0.258 (95% CI: 0.190–0.324), and 0.172 (95% CI: 0.126–0.215) of the variance in FI among all participants from the USA, England, and China. Health behaviors and social connections or stressors were the most important domains in the USA and England, while material circumstances contributed largely in China. We observed several common important SDoH predictors across countries, such as body mass index and sleep duration, whereas their nonlinear relationship with frailty showed differences, and other country-specific risk factors were also identified. Our findings reveal the priorities of SDoH domains for addressing aging disparities and promoting healthy aging, especially region-specific risk factors for tailored public health prevention strategies.