Abstract
BACKGROUND: Polycystic ovary syndrome (PCOS) is a common endocrine-metabolic condition in reproductive-aged women, linked to infertility and long-term cardiometabolic risk. Early identification remains challenging because current diagnosis relies on hormone testing and imaging. This research sought to develop and evaluate an interpretable machine learning (ML) model and a simplified nomogram for the early detection of PCOS. METHODS: Data from 1,600 women at the First People's Hospital of Jiashan were used for model training, with 283 external cases from Jiaxing Hospital of Traditional Chinese Medicine for validation. Twenty-three routine laboratory indicators were analyzed. After LASSO feature selection, seven ML algorithms were compared. The best-performing XGBoost model was interpreted using Shapley Additive exPlanations (SHAP). A logistic regression-based nomogram was developed from the key predictors. RESULTS: The XGBoost model showed excellent discrimination (AUC = 0.919 internal; 0.923 external). SHAP identified DHEAS, AMH, TG, and age as key contributors. The nomogram also performed well (AUC = 0.901 train; 0.887 test). CONCLUSIONS: This interpretable "XGBoost + SHAP" and nomogram framework provides an accurate, transparent, and practical tool for early PCOS screening and individualized management.