Visualized machine learning models combined with propensity score matching analysis in single PR-positive breast cancer prognosis: a multicenter population-based study.

The characteristics of single PR-positive (ER-PR+, sPR+) breast cancer (BC) and its prognosis are not well elucidated due to its rarity and conflicting evidence. There is a lack of an accurate and efficient model for predicting survival, thereby rendering treatment challenging for clinicians. Whether endocrine therapy should be intensified in sPR+ BC patients was another controversial clinical topic. We constructed and cross-validated XGBoost models that showed high precision and accuracy in predicting the survival of patients with sPR+ BC cases (1-year: AUC=0.904; 3-year: AUC=0.847; 5-year: AUC=0.824). The F1 score for the 1-, 3-, and 5-year models were 0.91, 0.88, and 0.85, respectively. The models exhibited superior performance in an external, independent dataset (1-year: AUC=0.889; 3-year: AUC=0.846; 5-year: AUC=0.821). Further, intensified endocrine therapy did not provide a significant overall survival benefit compared to initial or no endocrine therapy (P=0.600, HR: 1.46; 95% CI: 0.35-6.17). Propensity-score matching (PSM)-adjusted data showed that there was no statistically significant difference in the prognosis between ER-PR+HER2+ and ER-PR-HER2+ BC. Patients having the ER-PR+HER2- subtype had a slightly worse prognosis than those with the ER-PR-HER2- subtype. In conclusion, XGBoost models can be highly reproducible and effective in predicting survival in patients with sPR+ BC. Our findings revealed that patients with sPR-positive BC may not benefit from endocrine therapy. Patients with sPR+ BC may benefit from intensive adjuvant chemotherapy compared to endocrine therapy.