Association of surgical resection with survival in retroperitoneal leiomyosarcoma based on SEER propensity score matching and machine-learning models

Abstract

Abstract

Retroperitoneal leiomyosarcoma (RLS) is a rare and aggressive subtype of soft tissue sarcoma with limited population-level evidence guiding surgical decision-making. This study aimed to assess the prognostic value of surgery in patients with RLS using a large real-world cohort and advanced analytical methods. Patients diagnosed with RLS between 2000 and 2019 were identified from the Surveillance, Epidemiology, and End Results (SEER) database. Propensity score matching (PSM) was used to balance baseline variables. Overall survival (OS) and cancer-specific survival (CSS) were analyzed using Kaplan–Meier curves and Cox proportional hazards models. Random survival forests (RSF) were applied to evaluate variable importance and model robustness. A total of 1041 patients were included, of whom 817 (78.5%) underwent surgery. Before matching, significant imbalances were observed in age, grade, and SEER stage. After 1:1 PSM (159 matched pairs), covariate balance was substantially improved. Surgery was associated with significantly improved survival (OS: HR = 0.34, 95% CI: 0.26–0.45; CSS: HR = 0.34, 95% CI: 0.25–0.46; both
P
 < 0.001). High-grade tumors and advanced SEER stage remained independent adverse prognostic factors. RSF consistently ranked surgery, stage, and grade as the most important predictors of survival. Surgical resection status was strongly associated with survival in SEER-based analyses, but this association is subject to substantial unmeasured confounding by resectability, anatomic extent, and patient fitness; therefore, results should be interpreted as prognostic rather than causal and highlight the need for multidisciplinary assessment in high-volume sarcoma centers.