The associations of the triglyceride–glucose index and estimated glucose disposal rate with incident cardiometabolic multimorbidity vary across obesity phenotypes: a longitudinal cohort study

Abstract

Background Cardiometabolic multimorbidity (CMM) is increasingly common in older adults and is closely linked to insulin resistance (IR). However, whether IR surrogates perform similarly across heterogeneous obesity phenotypes remains unclear. Using the English Longitudinal Study of Ageing (ELSA), we examined the phenotype-specific associations of the triglyceride–glucose (TyG) index and the estimated glucose disposal rate (eGDR) with incident CMM and compared their predictive utility. Methods We included 4,198 participants from the ELSA in our analysis. Cox proportional hazards models and restricted cubic splines were used to evaluate the association between IR markers (the TyG index and eGDR) and the risk of CMM, as well as to assess potential nonlinearity. We then quantified the IR–related disease burden across obesity phenotypes using the population attributable fraction (PAF) and compared the predictive importance of variables across obesity subgroups using Shapley additive explanations (SHAP) values derived from machine learning models. Results Over a median follow-up of 6.8 years, there were 547 incident CMM events. The association between IR and CMM was strongest in the predominantly isolated central obesity group, surpassing those observed in the no obesity-risk and dual obesity-risk groups. In this subgroup, each 1-standard deviation (SD) increase in the TyG index was linked to a 60.8% higher risk of incident CMM, whereas each 1-SD increase in eGDR was linked to a 60.1% lower risk. Restricted cubic spline analyses revealed monotonic, linear associations for both TyG and eGDR across all three obesity phenotype groups. PAF analyses further showed that the predominantly isolated central obesity group had the highest attributable burden for both TyG and eGDR (37 and 65%, respectively). In internal validation, the machine learning models demonstrated promising discriminative ability, with areas under the curve of 0.827, 0.945, and 0.881. SHAP analyses identified eGDR and TyG as the leading contributors to model predictions. Conclusion IR surrogates were associated with incident CMM and showed potential utility for CMM risk prediction, with the strongest gradients and greatest population impact observed in the predominantly isolated central obesity phenotype. eGDR may be particularly useful for phenotype-tailored CMM risk stratification.