Meta-analyses of randomized controlled trials show suboptimal validity of surrogate outcomes for overall survival in advanced colorectal cancer

Abstract

Objectives: To quantify and compare the treatment effects on three surrogate end points, progression-free survival (PFS), time to progression (TTP), and tumor response rate (TR) vs. overall survival (OS) based on a meta-analysis of randomized controlled trials (RCTs) of drug interventions in advanced colorectal cancer (aCRC). Study Design and Setting: We systematically searched for RCTs of pharmacologic therapies in aCRC between 2003 and 2013. Trial characteristics, risk of bias, and outcomes were recorded based on a predefined form. Univariate and multivariate random-effects meta-analyses were used to estimate pooled summary treatment effects. The ratio of hazard ratios (HRs)/odds ratios (ORs) and difference in medians were used to quantify the degree of difference in treatment effects on the surrogate end points and OS. Spearman rho, surrogate threshold effect (STE), and R-2 were also estimated across predefined trial-level covariates. Results: We included 101 RCTs. In univariate and multivariate meta-analyses, we found larger treatment effects for the surrogates than for OS. Compared with OS, treatment effects were on average 13% higher when HRs were measured and 3% to 45% higher when ORs were considered; differences in median PFS/TTP were higher than on OS by an average of 0.5 month. Spearman rho ranged from 0.39 to 0.80, mean R-2 from 0.06 to 0.65, and STE was 0.8 for HRPFS, 0.64 for HRTTP or 0.28 for ORTR. The stratified analyses revealed high variability across all strata. Conclusion: None of the end points in this study were found to achieve the level of evidence (ie, mean R-trial(2) > 0.60) that has been set to select high or excellent correlation levels by common surrogate evaluation tools. Previous surrogacy relationships observed between PFS and TTP vs. OS in selected settings may not apply across other classes or lines of therapy. (c) 2015 Elsevier Inc. All rights reserved.