Relationship between overall survival and progression-free survival for recent NCCTG glioblastoma multiforme trials.
2004 Background: With the approval of novel agents for glioblastoma (GBM) treatment, reliable historical outcome data for the common phase II endpoint of progression free survival (PFS) is not available. In particular, the association between progression and overall survival (OS) has not been evaluated after temozolomide (TMZ) became the standard therapy for newly diagnosed GBM. Methods: Datasets from 5 TMZ-including NCCTG clinical trials in newly diagnosed GBM (n=325, 2005-2011, 1 phase I only and 4 phase I and single arm phase II trials) were pooled and analyzed to evaluate the individual level correlation between PFS and OS using correlation coefficient, landmark analyses, and time-dependent variable analyses. A historical control group was constructed by pooling datasets from 12 pre-TMZ era NCCTG clinical trials in the same patient population (n=1359, 1980-2004). Each of the 5 newer trials was compared to the control group to estimate the treatment effect (hazard ratio) on PFS and OS, which were analyzed to evaluate the trial level correlation between PFS and OS using correlation coefficient and weighted linear regression. Results: The estimated correlation coefficient between PFS and OS at individual patient level is 0.75 (95%CI, 0.7 to 0.8). The estimated hazard ratios of death in the landmark analysis at 4 and 6 months are 2.15 (95%CI, 1.47 to 3.15) and 2.17 (95%CI, 1.57 to 3.12), respectively, and the estimated hazard ratio of death is 10.1 (95%CI, 6.6 to 15.4) when progression is treated as a time-dependent variable. For trial level correlation (against the historical control group), the estimated correlation coefficient between HRos and HRpfs is 0.51 (95%CI, -0.67 to 0.96), with moderate prediction: the predicted HRos is very close to the observed HRos in 4 of 5 cases, however the predicted HRos is significantly different from 1 in only 2 out of 5 cases due to small sample size in other 3 cases. Conclusions: PFS and OS are highly associated at the individual patient level but only moderately at the trial level, the latter possibly due to small sample size in some of the trials. Additional validation in other trials could support use of PFS as a primary endpoint for randomized phase II trials in GBM.