Integration of coagulation parameters Enhances deep Learning-Based survival prediction in High-Grade serous ovarian...

High-grade serous ovarian cancer (HGSOC) remains a leading cause of gynecologic cancer mortality, with heterogeneous clinical outcomes necessitating improved prognostic models. This study aimed to develop and validate a comprehensive survival prediction model integrating traditional clinicopathological factors with novel molecular and coagulation parameters.

We retrospectively analyzed 216 HGSOC patients treated between 2012 and 2017. A comprehensive machine learning framework incorporating 88 algorithms was developed to predict survival outcomes, integrating conventional prognostic factors with coagulation parameters, particularly D-dimer levels. Model performance was evaluated using time-dependent AUC, concordance index, and calibration curves. External validation was performed using an independent cohort of 108 patients from three institutions.

The machine learning model demonstrated excellent discriminative capability (AUC 0.771, 95% CI 0.709-0.832), with improving predictive accuracy from 1-year to 5-year follow-up. Multivariate analysis identified five independent prognostic factors: p53 expression, lymphadenectomy, TNM stage, hypercoagulability, and Ki67 expression. The model showed robust performance in external validation.

Our novel machine learning-based survival prediction model demonstrates superior prognostic accuracy and temporal stability compared to conventional approaches. The integration of coagulation parameters provides new insights into disease progression. This model could facilitate personalized treatment decisions in clinical practice, though further prospective validation is warranted.