Association of lipid-frailty composite indices with cardiovascular disease risk in individuals with cardiovascular-kidney-metabolic syndrome stages 0–3: evidence from the UK Biobank and CHARLS

Abstract

Non-traditional lipid indices have been increasingly investigated as predictors of cardiovascular disease (CVD) risk, and frailty reflects cumulative physiological decline. Whether integrating these complementary dimensions improves CVD risk prediction remains unexplored. We aimed to construct lipid-frailty composite indices (LFCIs), examine their associations with incident CVD, and evaluate their predictive performance among individuals at cardiovascular-kidney-metabolic (CKM) syndrome stages 0–3. We analyzed 6143 participants from the China Health and Retirement Longitudinal Study (CHARLS) and 235,263 participants from the UK Biobank (UKB) classified as having CKM syndrome stages 0–3 at baseline. The primary outcome was incident CVD among participants free of clinical CVD at baseline. Six LFCIs were constructed by integrating six non-traditional lipid indices with the frailty index (FI). Associations between LFCIs and incident CVD were evaluated using Cox proportional hazards models, Kaplan-Meier curves, and restricted cubic splines. Mediation analyses examined whether high-sensitivity C-reactive protein mediated these associations. The predictive performance of the LFCIs was evaluated using five machine-learning models. Subgroup and sensitivity analyses assessed the robustness of the findings. During median follow-up periods of 9.0 years in CHARLS and 17.7 years in UKB, 1,142 (18.6%) and 28,450 (12.1%) participants developed incident CVD, respectively. In the fully adjusted models, each 1-standard deviation increase in CRI-I-FI (Castelli Risk Index I-FI) was associated with a higher hazard of incident CVD in both CHARLS (HR 1.18, 95% CI 1.12–1.23) and UKB (HR 1.30, 95% CI 1.29–1.32). In analyses by CRI-I-FI tertile, compared with participants in the lowest tertile, those in the middle and highest tertiles had HRs of 1.60 (95% CI 1.35–1.88) and 1.89 (95% CI 1.59–2.24) in CHARLS. Similar graded associations were observed in UKB. Models incorporating CRI-I-FI showed the best performance, achieving AUCs of 0.634 in CHARLS and 0.717 in UKB, showing higher AUCs than models incorporating the corresponding lipid indices alone. Several LFCIs showed evidence of nonlinear associations with incident CVD. Higher LFCI values were associated with increased CVD risk. Compared with models incorporating lipid indices alone, models integrating both lipid indices and FI showed greater discrimination and may help identify individuals at elevated CVD risk.