Longitudinal associations between cardiovascular-kidney-metabolic stage and cognitive trajectories among middle-aged and older Chinese adults

Cardiovascular–kidney–metabolic (CKM) dysfunction and cognitive decline frequently coexist, but the temporal direction and potential biomarker pathways linking stage-based CKM burden to cognitive change remain uncertain. We analysed 5547 adults aged 45 years or older from the 2011–2015 China Health and Retirement Longitudinal Study. An adapted AHA-informed CKM staging algorithm (Stages 0–4) was mapped to available CHARLS variables. Cognition was represented by a study-specific 0–31 raw-component sum. Three-wave cross-lagged panel models (CLPMs) estimated bidirectional longitudinal associations; a random-intercept CLPM (RI-CLPM) assessed within-person robustness. Directed structure learning and single-biomarker decomposition were exploratory. In 4186 participants with baseline cognition > 7, an IPTW-adjusted standardized association analysis compared low (Stages 0–1), medium (Stage 2), and high (Stages 3–4) CKM risk using multinomial generalized propensity scores, stabilized inverse-probability weights, and parametric g-formula standardization. Stage frequencies were 488 (8.8%), 782 (14.1%), 3,450 (62.2%), 138 (2.5%), and 689 (12.4%) for Stages 0–4, respectively. Baseline cognition showed no monotonic stage gradient (ordinal trend β=−0.093, P = 0.174). In the CLPM, higher CKM stage predicted lower subsequent cognition (Wave 1→2: standardized β=−0.112, 95% CI −0.167 to −0.057; Wave 2→3: β=-0.165, 95% CI −0.226 to −0.104), while lower cognition weakly predicted higher subsequent CKM stage (β=−0.018 and − 0.024, respectively). Model fit was χ 2 =187.362 (42 df), CFI = 0.967, TLI = 0.943, RMSEA = 0.038 (90% CI 0.032–0.044), and SRMR = 0.024. In the RI-CLPM, the forward path remained (β=−0.108, 95% CI −0.171 to −0.045), whereas the reverse path was not statistically robust (β=−0.016, 95% CI −0.034 to 0.002). In the weighted analysis, cognitive-decline risks were 15.1%, 20.4%, and 34.1%; versus low risk, medium risk had RR = 1.35 (95% CI 1.14–1.60) and RD = 5.3% (95% CI 2.8–7.8), and high risk had RR = 2.26 (95% CI 1.82–2.80) and RD = 19.0% (95% CI 14.6–23.4). C-reactive protein yielded the largest eligible exploratory indirect association (7.1%; FDR q < 0.004); single-biomarker proportions were not summed. Higher CKM stage was prospectively associated with lower cognitive function, whereas evidence for the reverse within-person pathway was weaker. Covariate-standardized risk contrasts were graded across CKM categories, but the observational design, residual confounding, and adapted wave-specific exposure measurement preclude causal conclusions.