Do Surrogate Endpoints Better Correlate with Overall Survival in Studies That Did Not Allow for Crossover or Reported...

In previous studies, correlation between overall survival (OS) and surrogate endpoints like objective response rate (ORR) or progression-free survival (PFS) in advanced non-small cell lung cancer (NSCLC) was poor. This can be biased by crossover and postprogression treatments.

To evaluate the relationship between these two surrogate endpoints and OS in advanced NSCLC studies that did not allow for crossover or reported balanced post-progression treatments.

A systematic review in patients with advanced NSCLC receiving second- and further-line therapy was performed. The relationship between the absolute difference in ORR or median PFS (mPFS) and the absolute difference in median OS (mOS) was assessed using the correlation coefficient (R) and weighted regression models. The analysis was repeated in predefined data cuts based on crossover and balance of postprogression treatments. When the upper limit of R's 95% confidence interval (CI) was more than 0.7, the surrogate threshold effect (STE) was estimated.

In total, 146 randomized clinical trials (43,061 patients) were included. The mean ORR, mPFS, and mOS were 12.2% ± 11.2%, 3.2 ± 1.3 months, and 9.6 ± 4.1 months, respectively. The correlation coefficients of ORR and mPFS were 0.181 (95% CI 0.016-0.337) and 0.254 (95% CI 0.074-0.418), respectively, with mOS. Nevertheless, in trials that did not allow crossover and reported balanced postprogression treatments, the correlation coefficients of ORR and mPFS were 0.528 (95% CI 0.081-0.798) and 0.778 (95% CI 0.475-0.916), respectively, with mOS. On the basis of STE estimation, in trials showing significant treatment effect size of 41.0% or more ORR or 4.15 or more mPFS months, OS benefit can be expected with sufficient certainty.

Crossover and postprogression treatments may bias the relationship between surrogate endpoints and OS. Presented STE calculation can be used to interpret treatment effect on either ORR or PFS when used as primary endpoints.