Nonlinear Association of Total Physical Activity with Cognitive Impairment in Older Chinese Adults: A Cross-Sectional Analysis of CHARLS

Background Cognitive impairment (CI) is a major challenge in China’s rapidly aging population. Although physical activity (PA) is generally considered beneficial, the dose–response relationship—particularly at high exposure levels—remains uncertain. Objective To characterize the association between total physical activity (TPA, MET-min/week) and CI in older Chinese adults, with emphasis on nonlinearity and subgroup heterogeneity. Methods We analyzed adults aged ≥ 60 years in the 2018 wave of the China Health and Retirement Longitudinal Study (CHARLS; n = 5,952). TPA was derived from the IPAQ-LF and examined as quartiles and as a log-transformed, z-standardized continuous variable. CI was defined as a validated composite score < 10. Associations were estimated using multivariable logistic regression with robust (HC3) SEs; dose–response was modeled with restricted cubic splines (RCS) and corroborated using a segmented (piecewise) logistic model. Prespecified covariates included sociodemographic, behavioral, and health factors. Sensitivity analyses included survey-weighted models, multiple imputation for missing covariates, alternative TPA/CI operationalizations, and FDR-controlled interaction tests. Results Relative to Q1 (< 1,732.5 MET-min/week), Q2 and Q3 showed no association with CI (OR 0.98, 95% CI 0.78–1.22; OR 1.19, 95% CI 0.96–1.49). Q4 (≥ 9,198.0 MET-min/week) was associated with higher CI odds (OR 1.41, 95% CI 1.11–1.78). Continuous log-TPA (per + 1 SD) was not associated with CI. RCS indicated a reverse J-shaped dose–response, with risk elevation confined to very high TPA; the segmented model located the inflection near the upper exposure tail (≈ 9,200 MET-min/week). Patterns were directionally stronger among women, younger-old adults, urban residents, and those with lower education or depressive symptoms, but no multiplicative interactions were significant after FDR correction. Findings were robust to weighting, multiple imputation, and alternative exposure/outcome definitions. Conclusions In this national sample of older Chinese adults, TPA exhibited a reverse J-shaped relationship with CI: low-to-moderate activity was not associated with increased risk, whereas very high TPA was linked to higher odds of impairment. Results support individualized, context-specific PA guidance that recognizes potential upper-limit risks in late life and should be verified in longitudinal and interventional studies.