Andrology. 2026 Aug 14. doi: 10.1111/andr.70358. Online ahead of print.
ABSTRACT
BACKGROUND: Erectile dysfunction (ED) is a prevalent male health disorder and a recognized sentinel marker for cardiovascular disease. Current diagnostic reliance on subjective questionnaires or invasive examinations limits early screening.
AIM: We aimed to develop and validate a machine learning model based on routine blood test data to predict ED risk to facilitate its early clinical screening.
METHODS: Data from 4116 men in the NHANES database (2001-2004) formed the training/internal validation sets. An independent external validation set comprised 489 clinical patients with NPTR-confirmed ED. From 49 initial demographic and blood-based indicators, feature selection via univariate logistic, multivariate logistic, and LASSO regression identified nine key predictors. Seven machine learning models were constructed, optimized via grid search with fivefold cross-validation, and evaluated using ROC analysis, calibration curves, and decision curve analysis (DCA). The optimal model was interpreted via SHAP.
RESULTS: The random forest model achieved superior performance, with an external validation AUC of 0.934, accuracy of 0.918, and specificity of 0.986, significantly outperforming logistic regression (AUC = 0.743). SHAP analysis identified age, sex hormone-binding globulin (SHBG), testosterone, glucose, cholesterol, and creatinine as the most influential predictors.
DISCUSSION: The study established a concise, nine-feature blood test panel and validated a high-performing predictive model in an independent clinical cohort, demonstrating significant clinical net benefit. Meanwhile, this model provides an economical, non-invasive, and scalable screening tool suitable for health check-ups, facilitating early ED identification.
CONCLUSION: A machine learning model based on routine blood tests can effectively evaluate ED risk, offering a novel foundation for early screening and precision management.
PMID:42599050 | DOI:10.1111/andr.70358