Association of insulin resistance surrogate indices with multimorbidity among postmenopausal women: cross-sectional and longitudinal evidence from NHANES and CHARLS with machine learning-based evaluation.

BACKGROUND: Insulin resistance (IR) is closely linked to multiple chronic diseases, but the comparative associations and discriminatory performance of different IR surrogate indices for multimorbidity in postmenopausal women remain unclear. METHODS: We analyzed postmenopausal women from NHANES and CHARLS. Eleven IR surrogate indices were evaluated using weighted logistic regression in NHANES and Cox proportional hazards models in CHARLS. Dose-response relationships were examined using restricted cubic splines. Discriminatory performance was assessed using ROC analyses, including time-dependent ROC analyses for CHARLS. Machine learning models were additionally developed using NHANES data to classify prevalent multimorbidity. RESULTS: In NHANES, most IR surrogate indices were positively associated with prevalent multimorbidity, whereas estimated glucose disposal rate (eGDR) showed a strong inverse association (OR = 0.28, 95% CI 0.21-0.37, p < 0.001). In CHARLS, eGDR was also inversely associated with incident multimorbidity in both baseline (HR = 0.78, 95% CI 0.73-0.84, p < 0.001) and cumulative analyses (HR = 0.67, 95% CI 0.58-0.77, p < 0.001). eGDR showed the highest discrimination among the evaluated indices in NHANES (AUC = 0.809) and in the CHARLS cumulative analysis (AUC = 0.634). Machine learning analysis in NHANES identified eGDR as the most important feature, and the optimal support vector machine model achieved an AUC of 0.835. Sensitivity analyses excluding hypertension, diabetes, and dyslipidemia attenuated the eGDR associations in CHARLS. CONCLUSIONS: Multiple IR surrogate indices were associated with multimorbidity among postmenopausal women, with eGDR showing the most consistent inverse association across analyses. These indices may assist population-level risk stratification, although their clinical utility requires further external validation.