Integrative regression modeling of insulin sensitivity, resistance, and beta-cell dysfunction in predicting female infertility: a cross-sectional NHANES study

Abstract

Insulin sensitivity, resistance, and pancreatic beta-cell function play key roles in metabolic and reproductive health. Despite the known role of glucose-insulin homeostasis in female reproductive health, the association and regression modeling of estimated glucose disposal rate (eGDR), homeostatic model assessment of insulin resistance (HOMA-IR), beta-cell function (HOMA-β), and insulin disposition index (Insulin-DI) with female infertility remain unclear. This cross-sectional study aimed to assess the associations between insulin-related indices and female infertility. This study examined data from 2832 women aged 18–45 years who participated in the National Health and Nutrition Examination Survey (NHANES) 2017–2023 cycles. To evaluate the associations between four insulin-related indices and infertility, multivariable logistic regression, generalized additive models, and threshold effect analyses were applied, while controlling for relevant sociodemographic, lifestyle, and metabolic confounders. Furthermore, integrated meta-analysis across models, distributional analysis, subgroup and interaction analyses were conducted to explore potential effect modifications. eGDR showed a strong inverse association with infertility across all models (OR range: 0.636–0.85, p < 0.001), with a significant nonlinear pattern (p = 0.003) and a threshold at 4.59. HOMA-IR was positively associated in unadjusted models (OR = 1.147, p < 0.001) but lost significance after full adjustment. HOMA-β showed a weak positive association (OR = 1.003, p = 0.0227), while Insulin-DI was inversely related to infertility only in unadjusted models. Subgroup analyses revealed significant effect modifications by BMI and age, especially for eGDR and HOMA-IR. Reduced insulin sensitivity, as indicated by lower eGDR, is strongly linked to higher infertility risk. HOMA-IR and Insulin-DI also show threshold-based nonlinear associations.