CSA Home
Central Surgical Association

Back to 2022 Abstracts


Use Of Machine-learning To Establish A Survival Decision Tree For Radiation Therapy In Patients With Retroperitoneal Sarcoma
*Ryan Zeh1, *David Konieczkowski2, *Diamantis Tsilimigras2, *Valerie Grignol2, *Carlo Contreras2, Allan Tsung2, Timothy Pawlik2, Raphael Pollock2, Joal D Beane2
1The University of Pittsburgh, Pittsburgh, PA;2The Ohio State University, Columbus, OH

OBJECTIVE-- The addition of radiation therapy (RT) to surgery for retroperitoneal sarcoma (RPS) remains controversial. Improved patient selection may help identify optimal candidates for multimodality treatment. The aim of this analysis was to determine which patients who receive RT and surgery may derive the most and least benefit. METHODS-- Patients who received RT and underwent curative-intent resection for RPS between 2004 and 2016 were identified from a national cohort in the United States National Cancer Database. A machine-based classification and regression tree (CART) model was used to generate similar groups of patients relative to overall survival (OS) based on preoperative factors. RESULTS-- A total of 1443 patients received RT in addition to surgery. Median age was 61 years old and 55.0% were female. Most patients (66%) received care at an academic or integrated network cancer program. With a median follow-up of 84 months, receipt of RT was not associated with improved survival (p=0.81). CART analysis revealed a significant association between survival and American Joint Committee on Cancer (AJCC) stage, age, tumor histology, and Charlson comorbidity score (Figure). Application of these parameters via machine learning stratified patients into five cohorts with distinct survival outcomes. In the most favorable cohort (Cohort 1: AJCC≤2, Age≤61, Histology including fibrosarcoma, well differentiated liposarcoma, myxoid liposarcoma, and leiomyosarcoma), the 5-year OS was 81.7% and median OS was not reached; in the least favorable cohort (Cohort 6: AJCC stage >2, age >68) where the 5-year survival was 41.3% and median OS was 45.2 months (p< 0.001 vs Cohort 1). CONCLUSIONS-- In the absence of a defined survival benefit, patients with advanced AJCC stage, age, and significant medical comorbidities stand the least to gain from the addition of RT to surgery. In contrast, younger patients with good performance status and conventional RPS histologies, may reap the greatest benefit from RT. The survival decision tree established in this study may help inform optimal patient selection for the receipt of RT in RPS.


Back to 2022 Abstracts