A metabolic marker-based diagnostic model for precancerous and malignant endometrial lesions in insulin-resistant PCOS women with sonographically suspected endometrial polyps.

BACKGROUND

Women with insulin-resistant polycystic ovary syndrome (PCOS-IR) and sonographically suspected endometrial polyps carry an elevated risk of endometrial premalignant and malignant lesions. This study aimed to characterize clinical and metabolic profiles across pathological subgroups and develop an exploratory risk-stratification model for endometrial neoplasia in this high-risk population, with a specific focus on insulin resistance as a core pathophysiological driver.

METHODS

A total of 185 PCOS-IR patients with ultrasound-detected endometrial polyps were retrospectively enrolled and stratified into benign (n=125), atypical hyperplasia (AH, n=34), and endometrial carcinoma (EC, n=26) groups. For modeling, AH and EC were combined into an endometrial neoplasia endpoint. A two-stage strategy combining LASSO regression and stepwise logistic regression was used for variable selection and model construction. Model performance was assessed via ROC curve, calibration curve, decision curve analysis, and multicollinearity diagnostics using the Variance Inflation Factor (VIF).

RESULTS

Compared with the benign group, the endometrial neoplasia group exhibited significantly higher HOMA-IR, fasting plasma glucose, 2-hour OGTT glucose, and fasting insulin, and significantly lower HDL-C (all P< 0.05). The final model incorporated age, high-density lipoprotein cholesterol (HDL-C), free androgen index (FAI), and homeostasis model assessment of insulin resistance (HOMA-IR), achieving moderate discrimination (AUC = 0.767) with high sensitivity (0.913) and specificity of 0.500. All variables had VIF values<5, confirming no significant multicollinearity. Good calibration and net clinical benefit were demonstrated in internal validation.

CONCLUSION

Insulin resistance is linked to early metabolic alterations in PCOS-IR patients with endometrial polyps, and is a core component of the final predictive model. This four-variable model shows moderate discriminatory performance as an exploratory adjunctive tool for pre-hysteroscopy risk stratification, with high sensitivity to minimize missed diagnoses of endometrial neoplasia. Given the lack of external validation and the heterogeneity of the combined neoplasia endpoint, the model remains hypothesis-generating. Further external validation in independent cohorts is required before clinical application.