Longitudinal trajectories of the triglyceride-glucose index and risk of incident type 2 diabetes in adults undergoing repeated health examinations.

Background: The triglyceride-glucose (TyG) index is an established surrogate marker of insulin resistance. However, most evidence has relied on a single baseline measurement and may not fully capture long-term metabolic exposure. This study examined whether longitudinal TyG trajectories are independently associated with subsequent incident type 2 diabetes mellitus (T2DM). Methods: This retrospective longitudinal cohort study used repeated health examination data from adults without diabetes at baseline between 2011 and 2025. TyG trajectories were identified using latent class mixed models based on repeated TyG measurements. Follow-up for incident T2DM began after the third eligible TyG measurement to preserve temporal ordering between exposure assessment and outcome occurrence. Multivariable Cox proportional hazards models with multiple imputation were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Restricted cubic splines and time-dependent area under the curve (AUC) analyses were used for dose-response and predictive performance assessments. Results: Among 16, 751 participants included in trajectory modeling, three TyG trajectory groups were identified and ordered by mean TyG level: low-ascending (G1), moderate-ascending (G2), and high-persistent (G3). A total of 10, 163 participants contributed follow-up after the third-visit landmark, among whom 430 incident T2DM events were recorded. Incidence rates were 2.90, 13.47, and 37.25 per 1, 000 person-years in G1, G2, and G3, respectively. In the main adjusted model, compared with G1, both G2 (HR = 2.27, 95% CI: 1.64-3.14) and G3 (HR = 5.40, 95% CI: 3.87-7.53) were associated with higher risks of incident T2DM (both P < 0.001). A graded association was observed across trajectory levels (HR per level increase = 2.34, 95% CI: 2.01-2.72; P < 0.001). Adding TyG information provided modest incremental discrimination beyond the base clinical model, with limited additional improvement over baseline TyG alone. Conclusions: Longitudinal TyG trajectories identified distinct diabetes risk strata among adults undergoing repeated health examinations. A high-persistent TyG trajectory was associated with increased risk of incident T2DM after the landmark time point, supporting the potential value of monitoring TyG dynamics for early risk stratification in routine care.