Letter to the Editor: JCO: Before We Throw out Progression-Free Survival As a Valid End Point…

Before We Throw out Progression-Free Survival As a Valid End Point…

To the Editor:

Although it does not affect their argument against progression-free survival (PFS) as an end point, Booth and Eisenhauer¹ calculated incorrectly that a 20% increase in tumor diameter translates into a 44% increase in tumor volume. It actually translates into an increase in volume of 73%. They divided the volume increase by the final volume, not the initial volume.

They state that if PFS were an effective surrogate for overall survival (OS), then the hazard ratios (HRs) for PFS and OS should be similar.¹ However, this may often not be the case.
Having similar OS and PFS HRs could occur if the drug continued to have the same impact on tumor growth after tumor progression as it did before tumor progression. This might happen if one continued the drug after progression and it continued to slow tumor growth, as may occur with androgen ablation for prostate cancer.² It might also occur even if the drug were stopped at tumor progression if it had caused some permanent change in the tumor, such as killing off more rapidly growing cells, permitting outgrowth of slowly growing, resistant cells. Mathematical modeling suggests that this may sometimes occur.³ Postprogression survival might also be longer with an experimental therapy than in the control arm if the experimental therapy led to substantial tumor shrinkage (because a smaller tumor would require more cell doublings to cause death after progression than would a larger tumor). However, in many cases, one might expect that none of these factors would have a major impact on postprogression survival, and would anticipate the HR for survival from the time of progression to be less favorable than the HR for time to progression. Because the OS HR includes both pre- and postprogression intervals, one would expect that even with effective agents, the OS HR would usually be less favorable than would the PFS HR.

Booth and Eisenhauer¹ state that a gain in PFS should not be obscured by postprogression treatment. However, there are now several examples of effective second- and third-line therapies that affect OS. Of equal importance, OS is often quite variable even in the absence of effective therapy. The longer the postprogression OS, the lower the probability that a PFS difference will translate into a statistically significant OS difference.⁴ For example, if the PFS HR were roughly equivalent to the ratio of median PFS times in two study arms, and a new therapy increased median PFS from 3 months to 6 months, the PFS HR would be 0.5. If both groups then survived for 1 year postprogression, the median OS would be 18 months in the experimental group and 15 months in the control group. The OS HR would be roughly 0.83, and the probability of achieving statistical significance would be substantially less for OS than for PFS.

Booth and Eisenhauer use the example of paclitaxel in ovarian cancer to argue that cross-over to the experimental agent does not muddy the OS water in randomized trials.¹ However, although prior therapy may substantially increase resistance to some types of therapies, if may have much less impact for others. For example, non–small-cell lung cancer (NSCLC) patients with EGFR mutations or EML4-ALK translocations who are treated with EGFR tyrosine kinase inhibitors⁵ or crizotinib,⁶ respectively, may have dramatic responses even if heavily pretreated. Hence, if unselected patients crossed over from one chemotherapy regimen to another, cross-resistance might minimize the effect of cross-over on OS, whereas crossing over selected patients with a required target to an effective targeted agent might substantially impact OS.

The authors also point out that OS is a hard end point whereas PFS is not, and that PFS measurement is subject to bias.¹ Although this is true, OS can be affected by several factors that are unlikely to affect PFS, and this can also lead to erroneous conclusions. For example, early initiation of palliative care in advanced NSCLC significantly improves OS,⁷ whereas it would be less likely to impact PFS.

Although randomized trials assessing OS are currently the gold standard in oncology research, there are several reasons that such trials may give the wrong answer, particularly if the study is done in unselected patients.⁸ Hence, rather than concluding that PFS and response are flawed end points because they do not necessarily agree with OS outcomes, it is important to carefully evaluate the specifics of the relevant studies, particularly before using negative OS results to discard an agent that has promising PFS or response outcomes.