Validity and efficiency of progression-free survival (PFS)-2 as a surrogate endpoint for overall survival (OS) in advanced cancer.

6518 Background: PFS is used widely as a primary endpoint in oncology trials and as a surrogate for OS. PFS-2, defined as time from randomization to progression on second line therapy, is potentially a more reliable surrogate than PFS but requires additional follow-up time. We evaluated the validity and efficiency of PFS-2 as a surrogate endpoint for OS and compared its performance with PFS. Methods: We performed an electronic search to identify randomized trials of advanced solid tumors reporting on PFS, PFS-2 and OS as pre-specified endpoints. We compared the correlations in the relative treatment effect for OS with PFS and PFS-2. We extracted data from Kaplan-Meier survival curves to create individual patient data to estimate time to statistical significance (TSS), defined as logrank P < 0.05. We further computed the samples size (person-year follow-up) required to reach statistical significance and assessed for the effect of survival post-progression (SPP). If trials failed to reach statistical significance for a particular endpoint, the maximal follow-up time was used. Results: Our study consisted of 39 analysis units with 20714 patients. Correlations of the OS treatment effect with the PFS and PFS-2 treatment effects were r= 0.12 (95% CI 0.00-0.13) and r= 0.67 (95% CI 0.08-0.69) respectively. The median differences in TSS between OS with PFS, OS with PFS-2, and PFS2 with PFS were 8.7, 8.4, and 3.2 months respectively. For trials with a SPP < 12 months, smaller differences in the median TSS as compared to SPP ≥12 months were seen. The median differences in follow-up required to reach statistical significance were 329, 151 and 122 person-years (PY) respectively. These differences were equivalent to a median 18.3% increase in sample size required for PFS-2 over PFS. The median increase in sample sizes were 18.2% and 22.3% for trials with SPP < 12 and ≥12 months respectively. Conclusions: PFS-2 has a stronger correlation with OS benefit than PFS. When PFS-2 is used as primary endpoint, trials will require an additional median 3.2 months to achieve statistical significance, and an additional 18.3% increase of sample size over PFS. PFS-2 balances between improved correlation with OS but modest increase in follow-up time and sample size. PFS-2 should therefore be considered as a primary endpoint in future trials of advanced cancers.[Table: see text]