Development and Validation of a Predictive Model Using Logistic Regression and Machine Learning for Carotid Artery Atherosclerosis Risk in Postmenopausal Han Women from Northern China: A Retrospective Case-Control Study.

Background: Postmenopausal women are at elevated risk for carotid atherosclerosis due to estrogen decline, yet existing prediction models do not incorporate genetic markers. This study aimed to develop and validate a risk prediction model combining clinical variables and estrogen receptor 1 rs9340799 genotype for postmenopausal Han women in northern China using logistic regression and machine learning approaches. Methods: A total of 276 postmenopausal Han women in northern China were recruited from 1 December 2024 to 30 April 2025. They were divided into a case group (n = 162) and a control group (n = 114) based on the presence of carotid atherosclerosis (CAS). ESR1 rs9340799, a polymorphism in the estrogen receptor alpha gene, was selected for genotyping (MAF > 0.05 in Chinese population) using high-resolution melting PCR. Univariate and multivariable logistic regression identified risk factors, and a nomogram was constructed based on age, systolic blood pressure (SBP), glucose (GLU), and rs9340799 genotype. Six machine learning algorithms were compared; logistic regression was selected given its ease of interpretation and resistance to overfitting. The model was externally validated in 47 subjects in the same institution from 1 May 2025 to 30 June 2025. Results: In the case group, age, menopause duration, SBP, and GLU were significantly higher than in controls (all P < 0.01), and rs9340799 GG genotype showed a protective effect in univariate analysis (P = 0.045). Age, SBP, GLU, and rs9340799 were selected as predictors after excluding menopause duration due to collinearity with age. Among six ML algorithms, logistic regression achieved the highest AUC (0.709, 95% CI: 0.596-0.821) with acceptable calibration (Brier score = 0.227, HL P = 0.263) and the broadest DCA threshold range (0.30-0.75), and was selected as the final model. SHapley Additive exPlanations (SHAP) analysis identified age as the most influential predictor. External validation in 47 subjects yielded an AUC of 0.65 (95% CI: 0.48-0.82), sensitivity of 0.82, and specificity of 0.55; the wide CI crossing 0.50 reflects the limited sample size, and further validation is warranted. Conclusion: This study developed a logistic regression model that included age, systolic blood pressure, GLU and ESR1 rs9340799 genotypes. The model demonstrated moderate discrimination in predicting the risk of carotid atherosclerosis in postmenopausal Han women in northern China. This model should be considered preliminary and needs further validation in larger, independent, multicenter cohorts for identifying high-risk populations for early intervention.