Association of the triglyceride-glucose (TyG) index and its composite indices with new-onset cataract: a longitudinal cohort study in southwestern China

Background Cataract-induced visual impairment is a major global public-health concern, with metabolic dysfunction and insulin resistance contributing to its pathogenesis. The triglyceride-glucose (TyG) index and its composites are validated insulin-resistance markers, yet their longitudinal links with incident cataract remain understudied, especially anthropometric- and inflammatory-derived composites. Methods This retrospective longitudinal cohort included 3 127 participants aged ≥ 40 years from a southwest Chinese health-checkup cohort (2013–2021), excluding those with baseline cataract. Cox proportional-hazards regression, receiver operating characteristic (ROC), and restricted cubic spline (RCS) analyses assessed associations, discriminatory performance, and dose-response relationships of TyG-related indices with incident cataract. Results During the entire study period, 576 (18.4%) participants developed incident cataract (incidence: 54.02 per 1000 person-years). In fully adjusted Cox models, upper quartiles of baseline TyG and its anthropometric composites (TyG-WHtR, TyG-ABSI, TyG-BMI) correlated independently with higher cataract risk (Q3/Q4, all FDR-adjusted q < 0.05); Q4-versus-Q1 HRs ranged 1.93 (TyG-ABSI)-2.44 (TyG-BMI), with significant linear trends ( P < 0.001). ROC showed baseline-index AUC 0.788–0.793 versus 0.780 for the covariate-only model. Exploratory K-means-derived Mean Exposure-Based Subgroups exhibited dose-dependent risk elevation (high vs. low subgroup fully-adjusted HR = 1.98–2.34, all P < 0.01; AUC = 0.615–0.645). For inflammatory composites TyG-NLR and TyG-MLR, quartile-level Cox associations turned non-significant after full adjustment; subgroup analyses showed null findings for TyG-NLR, while TyG-MLR lost significance after FDR correction ( q = 0.064). DeLong tests showed that anthropometric indices had significantly better discrimination than inflammatory indices (all FDR-adjusted q < 0.05). For the Mean Exposure-Based Subgroup analyses, inflammatory indices exhibited consistently poorer discrimination (all pairwise P < 0.001). RCS revealed significant overall dose-response for all six indices; five lacked nonlinearity, with marginal nonlinearity only for TyG-BMI. Conclusions TyG and its anthropometric composites are promising dose-dependent biomarkers for incident-cataract risk stratification, though their incremental discriminatory value over the covariate-only model is modest. Baseline indices showed moderate discrimination; exploratory long-term subgroup results require cautious interpretation due to look-ahead bias. Inflammatory-derived TyG-NLR and TyG-MLR have limited independent risk-stratifying capacity. These findings support further investigation of insulin-resistance-related metabolic markers for cataract prevention; causal inference needs prospective external validation.