Development and validation of a machine learning-based model for assessing coronary artery disease risk in postmenopausal women: a dual-center retrospective study.
Source: PubMed, NCBI / U.S. National Library of Medicine
To investigate the risk factors of coronary heart disease (CHD) and develop a risk assessment model for CHD in postmenopausal women. General information, medical history, and laboratory test results of the patients were collected from postmenopausal women with CHD admitted to two medical centers in Yangzhou (Jiangsu Province, China) from November, 2018 to November, 2023. After excluding cases with incomplete medical records, 1197 patients were included, who were divided into the training cohort (=821) and validation cohort (=376) based on the hospital of admission. In the training cohort, the risk factors for CHD in postmenopausal women were identified using Lasso regression, multivariate logistic regression analysis, and machine learning algorithms including Light GradientBoosting Machine (LGBM), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), and Naive Bayes (NB). Risk assessment models were constructed using these algorithms, and their performance was evaluated using ROC curves, decision curve analysis (DCA), and calibration curves. Lasso regression suggested body mass index (BMI) classification and glycated hemoglobin were independent risk factors for CHD in postmenopausal women, whereas age at menopause and high-density lipoprotein cholesterol (HDL-C) were independent protective factors (<0.05). Among the machine learning models, XGBoost demonstrated the best assessment performance in b
Abstract
To investigate the risk factors of coronary heart disease (CHD) and develop a risk assessment model for CHD in postmenopausal women. General information, medical history, and laboratory test results of the patients were collected from postmenopausal women with CHD admitted to two medical centers in Yangzhou (Jiangsu Province, China) from November, 2018 to November, 2023. After excluding cases with incomplete medical records, 1197 patients were included, who were divided into the training cohort (=821) and validation cohort (=376) based on the hospital of admission. In the training cohort, the risk factors for CHD in postmenopausal women were identified using Lasso regression, multivariate logistic regression analysis, and machine learning algorithms including Light GradientBoosting Machine (LGBM), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), and Naive Bayes (NB). Risk assessment models were constructed using these algorithms, and their performance was evaluated using ROC curves, decision curve analysis (DCA), and calibration curves. Lasso regression suggested body mass index (BMI) classification and glycated hemoglobin were independent risk factors for CHD in postmenopausal women, whereas age at menopause and high-density lipoprotein cholesterol (HDL-C) were independent protective factors (<0.05). Among the machine learning models, XGBoost demonstrated the best assessment performance in both the training set (AUC: 0.912; sensitivity: 0.892; specificity: 0.766; recall: 0.892; F1-score: 0.899) and the validation set (AUC: 0.891; sensitivity: 0.836; specificity: 0.921; recall: 0.837; F1-score: 0.877). Calibration curve and DCA curve analyses indicated good consistency between the predicted and actual outcomes. A nomogram and SHAP summary plot were used to visualize and interpret the logistic regression model and the XGBoost model, respectively. The risk assessment model for CHD in Chinese postmenopausal women established in this study demonstrates good accuracy and applicability to allow early identification of high-risk patients.