Data-driven glycolipid network phenotypes reveal a graded risk spectrum for cardiometabolic multimorbidity: a prospective study in middle-aged and older Chinese adults

Abstract Background Glycolipid dysregulation is a core driver of cardiometabolic multimorbidity (CMM), yet its role as an integrated network in CMM pathogenesis remains unexplored. Objective To explore the association between individual differential glycolipid covariance networks (IDGCNs)—pairwise glycolipid correlation matrix from six glycolipid biomarkers—and incident CMM. Methods We utilized 9-year longitudinal data (2011–2020) from 12,267 middle-aged and older Chinese adults. IDGCNs were constructed from six glycolipid biomarkers-fasting plasma glucose, glycated hemoglobin A1c, total cholesterol, triglycerides, high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C)-and quantified as 6 × 6 partial-correlation matrices, with 15 off-diagonal elements transformed into feature vectors. K-means clustering (k = 3) identified network phenotypes. Fine-Gray subdistribution hazards regression models were used to estimate hazard ratios (HRs) for incident CMM across clusters. The effects of glycolipid-coupling mixtures were evaluated using quantile-based g-computation. The incremental predictive utility was assessed using the C-statistic, net reclassification improvement, integrated discrimination improvement, and likelihood-ratio test. Results During a median follow-up of 9 years, 2,040 participants developed CMM. Relative to the normal group, multivariable-adjusted HRs (95%CI) for incident CMM escalated across clusters: Cluster 1 = 2.321 (1.400–3.850), Cluster 2 = 5.385 (3.155–9.192), and Cluster 3 = 5.845 (3.027–11.284). HDL-related couplings were protective, while glucose–LDL coupling was deleterious factors. The addition of the glycolipid-coupling mixture to conventional risk factors improved predictive metrics significantly. Conclusions Glycolipid network topology independently predicts incident CMM, supporting a shift from single-biomarker thresholds to network-based risk phenotyping.