Abstract
BACKGROUND: Insulin resistance proxy indicators are significantly associated with cardiovascular disease (CVD) and diabetes. However, the correlations between the estimated glucose disposal rate (eGDR) index and CVD and its subtypes have yet to be thoroughly researched. METHODS: 10,690 respondents with diabetes and prediabetes from the NHANES 1999-2016 were enrolled in the study. Three machine learning methods (SVM-RFE, XGBoost, and Boruta algorithms) were employed to select the most critical variables. Logistic regression models were established to evaluate the association between eGDR and CVD. We applied ROC curves, C-statistics, NRI, IDI, calibration curves, and DCA curves to assess model performance. Subgroup analyses were conducted to investigate the association among different subgroups. RESULTS: Participants in the higher quartile showed a decreased prevalence of CVD. Multivariate logistic regression models and RCS curves demonstrated that eGDR had an independently negative linear correlation with the likelihood of CVD[Q4 vs. Q1: OR 0.24(0.18,0.32)], CAD[OR 0.81(0.78,0.85)], CHF[OR 0.81(0.76,0.86)], and stroke[0.85(0.80,0.90)]. Model evaluation showed better performance in fully adjusted models than basic models[C-statistics(Model 3 vs. Model 1): CVD(0.683 vs. 0.814), CAD(0.672 vs. 0.807), CHF(0.714 vs. 0.839) and stroke(0.660 vs. 0.790)]. The AUCs of eGDR were significantly higher than the values of other IR surrogates in the unadjusted models, and slightly higher in the fully adjusted models. Subgroup analyses indicated that the results were robust. CONCLUSION: A lower eGDR was significantly associated with a heightened likelihood of CVD and its subtypes in diabetic and prediabetic populations. And eGDR exhibited better performance in evaluating the associations compared to other IR proxies encompassing TyG, HOMA-IR, QCUIKI, METS-IR, etc.