Recursive partitioning analysis to identify risk stratification for pediatric rhabdomyosarcoma: A real-world data retrospective cohort study based on the SEER database.

Abstract

e22005 Background: For pediatric rhabdomyosarcoma (RMS), IRSG risk stratification is based on two distinct staging systems that differ according to whether surgery is performed. Its complexity hinders its effective clinical application to a certain extent. We aimed to develop a simple, practical, and reproducible risk stratification model to offer treatment guidance for RMS. Methods: We conducted a retrospective study based on the Surveillance, Epidemiology, and End Results (SEER) database obtained from SEER*Stat software, version 8.4.0. Multivariate Cox regression analysis was used to identify independent prognostic factors in RMS. Later, prognostic factors were integrated to construct an OS prediction nomogram. Recursive partitioning analysis (RPA) was performed to stratify the risk of the patients. Results: A total of 223 patients with pediatric RMS were ultimately included in this study. All patients received chemotherapy, and most received radiotherapy (98.2%). The median follow-up time was 93.0 months, and the overall median survival time was not reached. Multivariate Cox analysis revealed that histology (hazard ratio [HR] = 2.2374, P = 0.031), tumor size (HR = 3.403, P = 0.008), and M classification (HR = 2.060, P = 0.028) were independent factors for OS in RMS. A nomogram was constructed to predict accuracy toward individual OS with a training C-index of 0.698 and a validation C-index of 0.781. Compared to the IRSG risk stratification, the risk groups stratified by RPA allowed for significant distinction between survival curves in our cohort (for 3-y OS, IDI = 0.04454772, P = 0.022; for 5-y OS, IDI = 0.05452665, P = 0.015). Conclusions: The risk stratification model constructed by RPA could help clinicians to identify patients with poor outcomes and assist them in making treatment and surveillance decisions.