Comparative survival analysis of ovarian cancer using conventional statistical models and neural networks

Ovarian cancer remains one of the most lethal gynecological malignancies, with accurate survival prediction being critical for personalized care and treatment planning. This study aims to compare the performance of conventional survival analysis methods with advanced neural network–based approaches in predicting survival outcomes for ovarian cancer patients. A retrospective dataset of 980 ovarian cancer patients obtained from The Cancer Genome Atlas (TCGA) and other public repositories was analyzed. The dataset included clinical (age, stage, histological subtype), pathological (tumor characteristics), and treatment-related variables (chemotherapy response, surgical resection status). Conventional survival analysis methods, such as the Cox proportional hazards model and Kaplan–Meier estimations, were compared with neural network–based approaches, including DeepSurv and other time-to-event models. Model performance was assessed using metrics such as concordance index (C-index), integrated Brier score (IBS), and calibration curves. Advanced neural network models demonstrated superior predictive performance compared to conventional methods, with DeepSurv achieving the highest C-index (0.82) and lowest IBS (0.16). Neural networks provided better calibration with observed survival outcomes but required higher computational resources. While conventional methods remain widely used for their interpretability, incorporating neural network–based models into clinical workflows could enhance personalized treatment planning and survival prediction in ovarian cancer.