Getting it first versus getting it right: weighing the value of and evidence for progression-free survival as a surrogate endpoint for overall survival in glioblastoma.

Abstract

Han et al claim that progression-free survival (PFS) may be an appropriate surrogate endpoint for overall survival (OS) in glioblastoma (GBM) trials.1 Unfortunately, the data presented do not support such a broad claim. In their current response, Han et al state that there were 2 major findings supporting the surrogacy conclusion in their original manuscript: (i) lead time using PFS over OS and (ii) the strong correlation of PFS hazard radio (HR) to OS HR in non–bevacizumab-containing comparative studies. First, lead time for PFS over OS suggests utility rather than supporting a claim for surrogacy. It answers a question of why a surrogate for OS should be used, not whether PFS actually is a surrogate. Consequently, the only finding that supports a surrogacy claim is the strong correlation between PFS HR and OS HR in non–bevacizumab-containing studies. Correlation between PFS HR estimates and OS HR estimates is not a sufficient argument to show or suggest a surrogacy relation in isolation. An absence of any treatment effect on PFS and OS, for example, does not imply a low correlation between PFS HR estimates and OS HR estimates. Furthermore, “non–bevacizumab-containing studies” is simply too broad a characterization given the data and implies a similar correlation for future agents. The original findings by Han et al, particularly with respect to correlation of positive treatment effects, are driven almost entirely by the EORTC 26981/NCIC CE.3 results. It would then be more accurate to assert a strong correlation between the positive effects of one specific agent, temozolomide (TMZ), on PFS and subsequent positive effects on OS. The example that we mentioned of anti-VEGF agents was offered as a counterexample to caution more widespread application of the TMZ findings without careful analysis. In contrast to the authors' claim that the applicability of the strong TMZ-based PFS/OS HR correlations to anti-VEGF agents may require only “further validation,” the available evidence suggests that there is in fact no correlation.2,3 Furthermore, whether the lead time gained is significant is a debatable point. No one would disagree that making correct decisions earlier may have some benefit, but a potential benefit in expediency must be weighed against the cost and chance of making incorrect decisions. There are now 2 large randomized studies in which using a PFS effect to predict an OS effect would have resulted in incorrect decision-making.2,3 Our editorial was trying to make the point that the consequences of incorrect decision-making may outweigh small increases in lead time. Particularly given the “decade-long unmet medical need,” we cannot afford to waste time and resources chasing false signals. In response to the disagreement with our argument concerning the ability of survival post progression (SPP) to dilute PFS effects. Strong correlations between PFS effect and OS effect can be interpreted as a limited dilution by SPP. Broglio and Berry4 discussed in detail how SPP modulates the relationship between PFS effect signal and OS effect signal. Although we agree that their results are based on modeling assumptions, they remain appropriate for interpreting the correlation between PFS effect and OS effect estimates, including the results by Han et al1 and the cited findings by Tang et al.5 In their response, Han et al concluded by stating that the implication of their analysis was “not to replace OS with PFS,” a somewhat confusing statement given the commonly held definition of a surrogate endpoint as one that is intended to substitute for a clinical endpoint.6 If the intention was only to claim that PFS might provide important information for decision-making in some contexts, there is no disagreement. There is evidence to support a claim that a positive PFS effect predicts a positive OS effect for TMZ7 and evidence to support a claim that positive PFS effects do not predict a positive OS effect for bevacizumab.2,3 Future trials investigating novel therapies may have further unpredictable, complex, and varied relationships between PFS and OS effects. These relationships should be evaluated using available data during trial development and perhaps be modeled during the course of the trial,8 rather than assuming a specific relationship. The degree of uncertainty around therapy-specific PFS/OS effect correlations and the consequences of incorrect decision-making should be considered along with the potential benefit from earlier information. PFS may give some valuable early information but falls well short of any reasonable bar for assumed surrogacy.