Decision Tree Model for Predicting the Overall Survival of Endometrial Cancer Patients with Adenomyosis

Yeliz Cecen Donmez,1 Esra Keles,2 Fatih Şanlıkan,2 Sahra Sultan Kara,1 Ismail Baglar,1 Ayşe Demirden,3 Sinem Durmus,4 Hafize Uzun,5 Umut Erkok61Department of Obstetrics and Gynecology, University of Health Sciences, Kartal Dr. Lütfi Kırdar City Hospital, Istanbul, Turkey; 2Department of Gynecologic Oncology, University of Health Sciences, Dr. Lutfi Kırdar Kartal City Hospital, Istanbul, Turkey; 3Department of Obstetrics and Gynecology, Bağcılar Education and Research Hospital, Istanbul, Turkey; 4Department of Medical Biochemistry, Izmir Katip Celebi University, Faculty of Medicine, Izmir, Turkey; 5Department of Biochemistry, Istanbul Atlas University, Faculty of Medicine, Istanbul, Turkey; 6Department of Obstetrics and Gynecology, University of Health Sciences, Mogadishu Somalia-Turkey Recep Tayyip Erdoğan Training and Research Hospital, Mogadishu, SomaliaCorrespondence: Umut Erkok, Department of Obstetrics and Gynecology, University of Health Sciences Turkey, Mogadishu Somalia-Turkey Recep Tayyip Erdoğan Training and Research Hospital, Mogadishu, Somalia, Email [email protected]: This study aimed to evaluate the association between coexisting adenomyosis and clinicopathological features in endometrial cancer (EC) patients and to develop an exploratory decision tree model for survival prediction.Methods: A retrospective analysis was conducted on 400 patients who underwent primary surgery for histologically confirmed EC between 2008 and 2018 at a tertiary academic center. Patients stratified by histopathologically verified adenomyosis status. Clinical and pathological features were compared. Overall survival (OS) and disease-free survival (DFS) were assessed using Kaplan-Meier estimation and multivariable Cox proportional hazards regression, with formal testing of the proportional hazards assumption. As an exploratory adjunct, we trained an interpretable decision tree classifier for vital-status prediction using clinically established prognostic variables.Results: Among 400 women, 69 (17.3%) had adenomyosis and more often presented with early stage disease (97.1% vs 88.5%; OR 0.23, 95% CI 0.05– 0.98), less LVSI (10.1% vs 26.3%; OR 0.31, 95% CI 0.13– 0.71), and shallower myometrial invasion (OR 2.04, 95% CI 1.15– 3.62). In multivariable Cox models, adenomyosis was not independently associated with OS (HR 0.62, 95% CI 0.32– 1.20, p=0.153) or DFS (HR 0.78, 95% CI 0.43– 1.40). Older age, CA-125 > 35 U/mL, and non-endometrioid histology were independent predictors. The decision tree selected age, LVSI, and deep myometrial invasion ≥ 50% as primary splitters; adenomyosis was not selected. Test-set performance: accuracy 0.81, balanced accuracy 0.72, AUC 0.76.Conclusion: Coexisting adenomyosis in EC is associated with favorable clinicopathological features but does not independently predict OS or DFS after adjustment for established prognostic factors. The exploratory decision tree model identified age, LVSI, and deep myometrial invasion as the primary determinants of survival, while adenomyosis was not selected as a discriminating variable. These findings suggest that adenomyosis reflects a less aggressive disease phenotype but should not serve as a standalone prognostic marker. External validation of the decision tree model on independent cohorts is warranted before clinical application.Keywords: endometrial cancer, adenomyosis, overall survival, decision tree model, prognostic factors, CA-125