Disease-free survival as a surrogate endpoint for overall survival in adults with resectable esophageal or gastroesophageal junction cancer: A correlation meta-analysis

Objectives: To evaluate disease-free survival (DFS) as a surrogate endpoint for overall survival (OS) using aggregate-level data from resectable esophageal or gastroesophageal junction cancer (EC/GEJC) trials assessing therapies in (neo)adjuvant and perioperative settings.

Methods: A systematic literature review was conducted to identify trials reporting OS and DFS, or compatible progression-free survival (PFS). Bivariate random-effects meta-analysis was used to estimate correlation between the treatment effects on DFS/PFS and OS, and weighted linear regression models assuming trial sample sizes as weights were used to estimate surrogacy equations. The primary analysis consisted of trials across all treatment settings, and secondary analysis consisted of trials only in the adjuvant setting. Leave-one-out cross-validation (LOOCV) was performed to measure the stability and predictive accuracy of the surrogacy equations while surrogate threshold effects (STE)-the minimum treatment effect on DFS/PFS that would translate into a positive OS benefit-were derived to measure their usefulness.

Results: The primary analysis included 26 trials. The estimated correlation coefficient between the hazard ratio (HR) of DFS/PFS (HR_DFS/PFS) and HR of OS (HR_OS) was 0.83 (95% confidence interval [CI]: 0.70-0.90). The estimated surrogacy equation was log(HR_OS) = 0.80 × log(HR_DFS/PFS) with a corresponding STE of 0.82. Reported HR_OS was within the 95% prediction interval of the predicted HR_OS from the model for more than 95% of the trials in the LOOCV, indicating a valid model. Secondary analysis included 7 trials with an estimated correlation coefficient of 0.76 (95% CI: 0.18-0.95). Through LOOCV, the surrogacy equation in the adjuvant setting was deemed valid.

Conclusions: Our meta-analysis suggests that HR_DFS/PFS -where DFS/PFS is defined as time from resection to disease recurrence (local, locoregional, or distant) or death-is correlated to HR_OS, and a valid and useful surrogate predictor for HR_OS in the neoadjuvant, perioperative, or adjuvant settings.