Longitudinal trajectories of estimated glucose disposal rate and risk of incident type 2 diabetes in Chinese adults: a prospective community-based cohort study

Insulin resistance is central to type 2 diabetes (T2D), and the estimated glucose disposal rate (eGDR) is a validated surrogate for it. Most evidence linking eGDR to incident T2D rests on a single measurement. We identified longitudinal eGDR trajectories in a community population and their associations with incident T2D. In this prospective community-based cohort, 2,140 adults free of diabetes at baseline underwent four biennial examinations between 2017 and 2023. A joint latent class mixed model with a Weibull baseline hazard simultaneously modelled longitudinal eGDR trajectories and time to incident T2D, with latent classes chosen from the Bayesian information criterion, entropy, posterior probability, and interpretability. Hazard ratios were adjusted for age, sex, education, income, body mass index, smoking, alcohol use, physical inactivity, and family histories of hypertension and diabetes. Sensitivity analyses excluded participants with baseline HbA1c ≥ 6.0%, excluded early events and applied a year-two landmark, tested classification robustness, redefined incident T2D by fasting glucose alone, and varied the handling of missing data and post-diagnosis measurements. Pre-specified subgroups were age, sex, BMI, baseline glycaemic status, and hypertension. Three eGDR trajectories were identified: High-Stable 69.2%, Low-Stable 20.5%, and Moderate-Curvilinear 10.3%. Compared with High-Stable, the adjusted hazard ratio was 1.72, 95% CI 1.21–2.44, for Low-Stable and 2.82, 95% CI 1.76–4.53, for Moderate-Curvilinear. The two associations differed in robustness. The Low-Stable association held after excluding participants with baseline HbA1c ≥ 6.0%, under a two-class model, and across subgroups, though it was attenuated to non-significance in the lag and fasting-glucose-only analyses. The Moderate-Curvilinear association carried the highest point estimate despite a baseline eGDR close to the reference, yet reversed direction when early events were excluded and should be regarded as preliminary. No significant effect modification, all P-interaction > 0.05. In this community-based cohort, eGDR exhibited three distinguishable longitudinal trajectory patterns. Within a joint modelling framework, a persistently low eGDR trajectory and a U-shaped trajectory were both independently associated with incident T2D, although the Moderate-Curvilinear association was concentrated in the early follow-up period and was not sustained after excluding early events, and should therefore be regarded as preliminary. These findings warrant validation in independent populations with longer follow-up, more diverse ethnic compositions, and interval-censored analytical frameworks before informing clinical risk stratification.