Identifying risk factors for vasculogenic etiology in patients with erectile dysfunction based on clinical features and machine learning.

Background: Vasculogenic erectile dysfunction (ED) is an important subtype of organic ED, and its development and progression are closely related to endocrine, metabolic, and psychological factors. Identifying risk factors for vasculogenic ED may facilitate early recognition and targeted intervention. Methods: This study included 519 patients diagnosed with ED using penile color Doppler duplex ultrasonography (CDDU) as the gold standard. Clinical and laboratory indicators were collected. Feature selection was strictly performed within the training set using univariate logistic regression, the Boruta algorithm, and least absolute shrinkage and selection operator (LASSO) regression. Based on the selected key variables, five machine learning models-logistic regression, random forest, support vector machine, light gradient boosting machine (LightGBM), and extreme gradient boosting (XGBoost)-were constructed and compared. Model performance was evaluated using metrics including the area under the receiver operating characteristic curve (AUC), and the SHapley Additive exPlanations (SHAP) method was employed to interpret the optimal model. Results: Among the 519 patients, 235 were diagnosed with vasculogenic ED. Feature selection identified seven key risk factors: age, hypertension, smoking, diabetes, Hamilton Anxiety Scale (HAMA) score, total testosterone (T), and estradiol (E2). The random forest model performed best in the validation set, but its discriminative ability was only moderate (AUC = 0.682, 95% confidence interval [CI]: 0.598-0.768). SHAP analysis revealed that age contributed most to the model predictions, followed by hypertension, T, and smoking; the HAMA score also ranked highly. Testosterone levels exhibited a nonlinear U-shaped association with vasculogenic ED risk. Conclusion: Based on routine clinical indicators, this study identified seven key factors associated with vasculogenic ED. Among them, anxiety as measured by the HAMA score was recognized as a non-traditional factor, suggesting a complex interplay between psychological factors and vascular pathology; however, the specific direction of this relationship remains to be elucidated by prospective studies. The machine learning model constructed in this study showed moderate discriminative ability and is currently insufficient to support independent clinical decision-making. Future research should collect multicenter, large-sample data and adjust model parameters for further validation and optimization.