To develop a machine learning (ML) algorithm to predict survival probabilities for patients with epithelial ovarian cancer (EOC).
Data were obtained from the SEER database for women diagnosed with EOC between 2004 and 2020. Clinical features, treatment regimens and overall survival (OS) were analyzed. Cox regression was conducted to identify prognostic factors associated with EOC. We employed 5-fold cross-validation to improve the accuracy of the model. Random Survival Forest (RSF), Gradient Boosting Survival Analysis (GBSA) and Support Vector Machine (SVM) were used to develop ML models, then compared with the Cox model. The predictive performance of these models was assessed using AUC, concordance index (C-index), and Brier scores.
A total of 12,949 EOC patients were selected from the SEER database. We identified 14 independent prognostic factors for OS and constructed predictive models. The GBSA model demonstrated superior AUC, C-index, and Brier scores at each time point, outperforming the Cox model. SHAP analysis showed that FIGO stage, grade, and lymph node dissection were the most important features in the GBSA model.
The GBSA model outperforms traditional methods in survival prediction, offering a valuable tool for clinicians to make informed decisions about patient prognosis.