Predicting the Survival Prognosis of Ovarian Cancer Patients in a Broad Learning Approach

Ovarian cancer remains a significant global health challenge and its survival prognosis plays an important role in clinic practice. In this paper, we design a broad learning model for survival prognosis (SPBL) to enhance the accuracy and ob-jectivity of survival predictions for ovarian cancer patients using machine learning techniques. Prognoses often rely exclusively on lesion information; however, our SPBL model integrates multi-source data. We also apply an interpretable SHAP model, to provide transparent and understandable explanations for the 3-year survival prognosis generated by the SPBL model. Based on the SEER database, we evaluate the propose models. The SPBL model exhibited notable improvements in performance, achieving increases in AUC, F1 score, and accuracy by 1.34%, 7.51%, and 6.08% compared with Random Forest, AdaBoost, GBDT, and XGBoost. The significance analysis suggest that our model can significantly enhance the prognosis process, leading to better-informed clinical decisions and optimized treatment strategies.