Introduction: Ovarian cancer (OC) is a leading gynecological malignancy with poor prognosis due to late diagnosis. Traditional models overlook nonlinear interactions among clinicopathological factors. This study developed an interpretable eXtreme Gradient Boosting (XGBoost)-Cox model to predict overall survival (OS) and progression-free survival (PFS) in OC.
Methods: We conducted a retrospective analysis of 313 OC patients treated at a single center between 2012 and 2025. An XGBoost model with a Cox proportional hazards objective was used to generate a machine learning–derived risk score (ML-score), which was subsequently incorporated into Cox regression analysis. Model interpretability was enhanced using Shapley additive explanations (SHAP). A nomogram was constructed based on the ML-score, and model performance was evaluated using Harrell’s concordance index (C-index), time-dependent receiver operating characteristic (ROC) analysis, and decision curve analysis (DCA).
Results: SHAP highlighted key factors: targeted therapy, histological type, tumor size, age, postoperative ICU admission for OS; age, tumor size, stage, targeted therapy, bleeding, histological type, omental/lymph node metastasis for PFS. ML-score stratified high-/low-risk groups (OS: HR = 4.67, 95% CI 3.14–6.96; PFS: HR = 4.05, 95% CI 2.85–5.75; p < 0.0001). ML model showed superior C-index and area under the ROC curve (AUC) with increasing long-term accuracy and better DCA net benefit.
Conclusion: This ML model provides robust, interpretable prognostication, capturing nonlinear effects of factors like targeted therapy and tumor burden.
Supplementary Information: The online version contains supplementary material available at 10.1186/s12885-026-15921-7.
Keywords: Machine Learning; Ovarian Cancer; SHAP Interpretability; Survival Prediction; XGBoost-Cox Model.