Comparative predictive value of the cholesterol-high-density lipoprotein-glucose index versus the triglyceride-glucose index for gestational dysglycemia: a two-cohort study

Background Early risk stratification for gestational dysglycemia is important for improving maternal and neonatal outcomes. Derived from fasting triglycerides and glucose, the triglyceride–glucose (TyG) index is widely used to approximate insulin resistance, whereas the cholesterol-high-density lipoprotein-glucose (CHG) index incorporates broader lipid metabolism. We compared the associations and discriminative performance of TyG and CHG in a national survey discovery cohort and an independent clinical validation cohort. Methods We analyzed a survey-weighted discovery cohort from NHANES 2007–2018, in which the primary outcome was self-reported GDM history. We further evaluated an independent validation cohort with clinically diagnosed GDM (n = 217). Associations and predictive performance were assessed using multivariable logistic regression, receiver operating characteristic (ROC) analysis, calibration analysis, and decision curve analysis (DCA). Additional analyses included adjustment for continuous fasting blood glucose in NHANES, supportive analyses restricted to currently pregnant NHANES participants from 2007–2012 using proxy-defined gestational fasting dysglycemia (fasting blood glucose ≥5.1 mmol/L), and gestational-week-adjusted sensitivity analyses in the validation cohort. Results In the NHANES discovery cohort, CHG showed a stronger association with self-reported GDM history than TyG in the primary adjusted models and yielded a numerically higher AUC than TyG. After additional adjustment for continuous fasting blood glucose, the association for TyG was attenuated, whereas CHG remained significantly associated. In the clinical validation cohort, CHG also showed numerically higher discriminative performance than TyG, and the overall findings remained directionally consistent after gestational-week adjustment. Supportive analyses in currently pregnant NHANES participants showed directionally similar but statistically imprecise estimates because of the limited sample size. Conclusion Both TyG and CHG are simple, low-cost indices associated with gestational dysglycemia/GDM. Across the discovery and validation cohorts, CHG generally showed stronger associations and numerically better discrimination than TyG; however, its overall discriminative performance remained modest and should be interpreted as that of a potential risk marker rather than a standalone clinical screening tool. Further prospective studies are needed to validate these findings.