Recursive partitioning analysis of pretreatment variables of 416 patients with locoregional esophageal cancer treated with definitive concomitant chemoradiotherapy on Intergroup and Radiation Therapy Oncology Group trials

Purpose To analyze the relative contributions of uniformly collected pretreatment patient- and tumor-related variables to survival and to identify the terminal nodes via recursive partitioning analysis (RPA) that could be used as a stratification variable for future Phase III trials. Methods and materials From two Intergroup trials (85-01, n = 130; and 94-05, n = 218) and one Radiation Therapy Oncology Group trial (92-07, n = 68), we identified 416 patients who were treated with definitive concomitant cisplatin and 5-FU-based chemoradiotherapy and analyzed their data for survival by RPA to define prognostic classes. The following pretreatment factors were evaluated: histologic type, age, weight loss, Karnofsky performance status, gender, race, T stage, tumor location, tumor size, N stage, and degree of dysphagia. The entire data set was considered as the initial node. The criterion for split points was the smallest p value less than unadjusted 0.05. Results Of the 416 patients, 336 (81%) were dead at the time of the analysis. The RPA identified only one significant split: pretreatment weight loss in the prior 6 months of <10% vs. ≥10%. After adjusting for multiple comparisons, no other split approached statistical significance. Conclusion Unlike our experience with malignant glioma, brain metastases, and locally advanced non-small-cell lung cancer, RPA failed to identify novel prognostic information that could be incorporated into the stratification scheme of future chemoradiotherapy trials for esophageal cancer. Furthermore, our analysis validated the percentage of weight loss as a stratification variable for esophageal cancer.