Effectiveness of artificial intelligence algorithms in predicting progression-free survival in epithelial ovarian cancer patients

Aims: This study aimed to assess the predictive performance of artificial intelligence–based models in estimating progressionfree survival (PFS) in patients with epithelial ovarian cancer and to compare various interpretable machine learning approaches. Methods: Between January 2015 and December 2020, a total of 167 patients who underwent surgical intervention at the Gynaecological Oncology Department of Antalya Training and Research Hospital were retrospectively included in the study if their data were complete. Clinical data were analysed, and the dataset was randomly divided into a training group (n=117; 75%) and a validation group (n=42; 25%). A machine learning (ML) analysis was conducted using the eight most relevant and widely applied algorithmic models for this study design. Model development time, mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) were evaluated. Results: Random Forest demonstrated the highest accuracy (MAE=16.45, CC=0.571, RMSE=20.98, time=0.03) and thus became the focus of subsequent analyses. Other algorithms included Linear Regression, Bootstrap Aggregating, Additive Regression, Random Committee, and Regression by Discretization (CC=0.533, 0.492, 0.449, 0.408, and 0.382, respectively). For Random Forest, a moderate correlation was observed between actual and predicted PFS values (CC=0.4–0.6), indicating moderate predictive performance. Conclusion: The findings of this study demonstrate that machine learning models, particularly Random Forest, can achieve moderate yet clinically relevant prognostic performance based on routinely collected clinical data. In particular, Random Forest demonstrates potential clinical value in guiding patient follow-up strategies and supporting individualized management in ovarian cancer, although further research is required to enhance its clinical validity and applicability.