Surrogate endpoints in neoadjuvant rectal cancer trials: Statistical evaluation using data from the FFCD 9203 trial

Abstract

4086 Background: Neo-adjuvant rectal cancer trials use overall survival (OS) and local control as standard endpoints. Using earlier endpoints as primary outcomes in phase III trials requires them to be validated as surrogates for OS or local control. The aim of this study was to evaluate the statistical validity of specimen sterilization, pT and pN stages and circumferential resection margin as surrogate endpoints for OS and local recurrence free survival (LRFS). Methods: We used data from the FFCD 9203 trial (Gérard et al. J Clin Oncol 2006): preoperative radiotherapy (RT) vs preoperative chemoradiation (CT-RT) in T3–4 rectal cancers. Based on logistic regressions and Cox proportional hazards regression models we applied single-trial validation methods: Prentice criteria, proportion of treatment effect explained (PTE), relative effect, adjusted association and the proportion of variation explained (R²). OS was defined as time interval between randomization and death (all causes) and LRFS as time interval between randomization and local recurrence. Survival curves were estimated using Kaplan-Meier method and compared using Log-Rank tests. Results: Our analyses were based on the 742 eligible patients of the trial (367 patients in the RT arm and 375 in the CT-RT arm). Complete sterilization of the operative specimen was more frequent in the CT-RT arm (11.4% vs 3.6%, p<.0001) and the 5-year local recurrence rate was significantly lower (16.5% vs 8.1%, p=0.004) while the 5-year OS did not differ (67.9% vs 67.4%, p=0.69). None of the variables fulfilled the four Prentice criteria whatever final endpoint. The most surrogate variable was specimen sterilization categorized as pT0 or few residual cells (S1) vs evolutive residual cells (S0). It explained 12% of the effect of treatment on LRFS (PTE=0.119) but the lower bound of the 95% CI was under the critical limit of 0.5. After adjusting for treatment, relative risk of local recurrence (S1 vs S0) was 0.56 (95% CI [0.28–1.10], p=0.09, R²=0.05). With OS as final endpoint, PTE for sterilization was not interpretable (PTE=-249.3%). Conclusions: Our data did not allow validating relevant surrogates for long term outcomes in neo- adjuvant rectal cancer trials. To increase power, an individual meta-analysis is planned. No significant financial relationships to disclose.