BLUE RIBBON POSTER THERAPY: An Innovative Approach Using Ensemble Model Improves Risk Stratification of Ovarian Cancer: A Real World Cohort with PARP Inhibitor Maintaenance Treatment from China
Abstract
Purpose
Ovarian cancer, recognized as the most lethal gynecologic malignancy, is frequently detected at advanced stages and is associated with an unfavorable prognosis and heterogeneous therapeutic responses. Poly (ADP-ribose) polymerase inhibitors (PARPi) have revolutionized ovarian cancer treatment, especially for BRCA /HRD-positive patients. However, PARPi efficacy exhibits variability and uncertainty, highlighting the need for improved risk stratification models to better predict patient outcomes and guide treatment decisions.
Methods
A retrospective analysis was conducted on 737 ovarian cancer patients. Clinical, genetic, pathological, and biochemical data were collected and analyzed. Feature selection was performed using univariate Cox regression and multicollinearity checks. Multimodal models were constructed using Cox proportional hazards model and a late-fusion approach, integrating diverse data types. Model performance was evaluated using Kaplan-Meier analysis and concordance indices(c-index) in training and test sets.
Results
For primary patients, the integrated model combining clinical and pathological features (CP model) achieved the best performance (c-index = 0.71), significantly outperforming individual models. In recurrent patients, the clinical and biochemical model (CB model) showed the highest accuracy (c-index = 0.558). Key prognostic factors differed between groups. For primary patients, significant factors included HRD status (P = 0.042), pathological type (P = 0.05), PARPi type (P < 0.001) and hospital type of surgery (P = 0.04). In recurrent patients, performance status(P = 0.005) , platinum sensitivity(P < 0.001) and CA125 (P = 0.03) were identified as critical factors. Multimodal integration demonstrated complementary value, improving predictive performance over single-modality models. Additionally, the CP model stratified primary patients into high- and low-risk groups with significant differences in PFS (P < 0.001), and the CB model achieved good stratification in recurrent patients (P = 0.054).
Conclusion
The ensemble model approach effectively stratifies ovarian cancer patients receiving PARPi therapy, providing valuable insights into prognosis and treatment planning.