Surrogate endpoints for overall survival (OS) in advanced pancreatic cancer (APC): Analysis of randomized clinical trials (RCTs) exploring gemcitabine (G)-based combinations

Abstract

4575 Background: G-based combinations often result in an improved overall response rate (ORR) in APC; this improvement in ORR, however, seldom translates into a significant OS advantage. We therefore evaluated the appropriateness of ORR and progression-free survival (PFS) as surrogate endpoints for OS in RCTs comparing single-agent G and G-based combinations as first-line treatment for APC. Methods: Phase III trials reporting ORR or PFS and OS data were considered eligible. Potential correlations were explored according to a linear regression model considering both the actual outcome (ORR or PFS and OS) for each single arm and the calculated relative risk (RR) for each outcome in paired comparisons. Correlation was estimated according to Pearson’s (r) and R2 coefficients (parametric) and Spearman’s (Rho) coefficient (non-parametric). A model to calculate the target sample size to correctly identify 0.4, 0.7, and 1.5 mos benefits in OS was derived as well. Results: Nineteen (6,288 pts) and 17 (4,882 pts) RCTs were identified for the ORR/OS and PFS/OS correlation, respectively. When considering ORR rates and medians, ORR did not significantly correlate with OS (r=0.23, R2=0.06, p=0.14; Rho=0.16, p=0.33), while PFS showed a strong linear correlation (r=0.75, R2=0.56, p<0.0001; Rho=0.90, p<0.0001). Similarly, when considering RRs, ORR did not significantly correlate with OS (r=-0.17, R2=0.03, p=0.46; Rho=-0.18, p=0.44), while PFS showed a strong linear correlation (r=0.91, R2=0.82, p<0.0001; Rho=0.59, p=0.01). Based on these data, we derived a sample size model (beta-coefficient=0.75) to calculate how many patients would be necessary to demonstrate a significant OS advantage of 0.4, 0.7 and 1.5 mos, respectively, using PFS as a surrogate endpoint; according to this model, 0.5, 1, and 2 months improvements in PFS would be necessary to translate into the target OS advantages, requiring 2,370, 678 and 222 pts, respectively. Conclusions: In APC, ORR does not seem to correlate with OS. Conversely, PFS could be a reliable surrogate endpoint for survival in this setting, although the detection of relatively small differences in PFS would require a remarkably large sample size. No significant financial relationships to disclose.