Evaluation of pathological complete response as a surrogate endpoint for overall survival in resectable oesophageal cancer: integrated analysis of individual patient data from phase III trials.

Abstract

Overall survival (OS) is the standard endpoint for oncological treatment efficacy, but requires long follow-up. The aim of this study was to evaluate pCR as a surrogate for OS in oesophageal cancer. An integrated analysis of individual patient data (IPD) from phase III trials comparing perioperative therapies for resectable oesophageal and gastro-oesophageal junction cancer was conducted. Individual-level surrogacy between pCR and OS was assessed using Kendall's rank correlation coefficient (τ). A τ of 0.8 was considered a threshold for a good surrogate. As no method estimating τ between an ordinal endpoint and OS has been reported, a new method was proposed using the inverse-probability-of-censoring weighted estimator adjusted for tied data. Of 22 eligible trials, 10 provided IPD for 1641 patients, including 624 who received neoadjuvant chemotherapy (NAC; 45 (7.2%) achieved pCR) and 1017 who received neoadjuvant chemoradiotherapy (NACRT; 299 (29.4%) achieved pCR). In the NAC subgroup, patients with pCR had an HR for OS of 0.12 (95% c.i. 0.05 to 0.33), the C-index was 0.54 (95% c.i. 0.52 to 0.56), and τ was 0.256. In the NACRT subgroup, the HR was 0.57 (95% c.i. 0.47 to 0.70), the C-index was 0.56 (95% c.i. 0.54 to 0.58), and τ was 0.174. Hypothetical data suggested that achieving strong surrogacy (τ of 0.8) required an HR of 0.09 (95% c.i. 0.07 to 0.11). Although pCR was correlated with OS, no evidence of individual-level surrogacy with OS was demonstrated, making it inappropriate to consider pCR as a surrogate endpoint for OS in resectable oesophageal cancer.