Exenterative Surgery in Advanced Pelvic Cancer: Survival Outcomes and an Exploratory Predictive Modeling Framework for Surgical Decision Support

Background/Objectives: Pelvic exenteration is among the most aggressive procedures in surgical oncology and remains associated with substantial morbidity and highly variable oncologic outcomes. Achieving negative resection margins (R0) is the strongest determinant of long-term survival, yet accurate preoperative prediction of resectability remains difficult. This study evaluated oncologic outcomes following pelvic exenteration and explored the feasibility of a predictive modeling framework for estimating the probability of achieving R0 resection using routinely available pretherapeutic variables. Methods: A retrospective single-center cohort study included 229 patients undergoing pelvic exenteration for advanced pelvic disease between 2008 and 2024. Survival outcomes were analyzed using Kaplan–Meier and Cox regression methods. An exploratory machine-learning framework based on Random Forest (RF) and Extreme Gradient Boosting (XGB) algorithms was developed using demographic, oncologic, radiologic, therapeutic, and biological preoperative variables to estimate the likelihood of achieving negative surgical margins. Results: Median overall survival (OS) was 40 months (95%CI: 37–42 months), while median progression-free survival (PFS) was 18 months (95%CI: 17–19 months). Recurrent disease was associated with significantly worse oncologic outcomes, whereas colorectal malignancies demonstrated superior survival compared with urologic and gynecologic tumors. Multivariable analysis identified resection-margin status as the only independent predictor significantly associated with both OS (HR 1.338, 95%CI 1.211–1.764, p = 0.0137) and PFS (HR 3.499, 95%CI 2.357–5.195, p < 0.0001). RF achieved higher apparent overall accuracy (83.6%) but failed to identify positive-margin cases. XGB demonstrated more balanced discriminatory performance, with an ROC-AUC of 0.792 and improved identification of incomplete resections. Absence of radiologically visible loco-regional lymphadenopathy emerged as the strongest predictor of achieving R0 resection. Conclusions: Pelvic exenteration continues to provide meaningful oncologic benefit in selected patients when R0 resection is achievable. Although the predictive models remain exploratory and are not suitable for direct clinical implementation, they demonstrate the feasibility of integrating multidimensional pretherapeutic variables into decision-support frameworks for patient selection.