This article is more than 5 years old. Information may no longer be current.
OS, surrogate endpoints and the sins of Galileo
Click Here to Manage Email Alerts
The church of the 1600s held the position that Earth was the center of the universe, with the sun revolving around it. In 1633, Galileo was condemned for having championed the Copernican hypothesis that Earth instead revolved around the sun.
In the evaluation of anticancer therapies, the outcome of OS is the earth at the center of the clinical research universe. In their recent JAMA publication assessing multiple meta-analyses that correlated surrogate endpoints with OS, Prasad and colleagues concluded that surrogate endpoints were inadequate because they often correlated only weakly with OS.
OS is a highly valued measure of drug benefit and it can be measured very precisely. However, there are problems. The first is that OS requires much longer follow-up than PFS or response, and this can delay answers by years. If a therapy is effective, many life-years can be lost and patients will suffer while the drug is awaiting approval.
The second issue is that one needs large, expensive and time-consuming randomized clinical trials to determine OS benefit, although much smaller patient numbers may be adequate with a PFS outcome, and a still much smaller nonrandomized trial may suffice for single-agent response.
The third issue is that, although OS is a precise endpoint, it is a dirty endpoint that can be affected by many things unrelated to the experimental therapy. This includes crossover to the experimental therapy (as discussed by Prasad and colleagues), but it also includes any therapy — including palliative care — that prolongs survival after progression on the new agent.
Long post-progression survival (PPS) reduces the probability that a therapy will be associated with a statistically significant OS improvement. For example, if ratios of median survival times are taken as a rough estimate of HRs and a therapy increased PFS from 3 months to 6 months, the HR would be 0.5. If both groups then survived 12 months after progression, the survival in the experimental group would be 18 months, and in the control group 15 months. The OS HR would now be 0.83, and the probability of this achieving significance would be substantially less than with the PFS HR of 0.5, despite the absolute gain being the same.
This would be similar to what is seen in the example Prasad and colleagues give of everolimus (Afinitor, Novartis) in breast cancer. The difference in median PFS was 4.6 months (7.8 months vs. 3.2 months; HR = 0.45; P < .0001), whereas the difference in median OS was 4.4 months (31 months vs. 26.6 months; HR = 0.89).
Second-, third- and fourth-line therapies for breast cancer generally tend to be more effective than similar therapies for other solid tumors. Hence, PPS is likely to be longer in breast cancer than in most other malignancies. If long PPS is associated with a lower probability of PFS correlating with OS, then one might expect a worse correlation of PFS with OS in breast cancer than in other tumor types. In fact, our calculations based on data from Table 1 of Prasad and colleagues’ manuscript suggest that surrogate endpoints correlated with OS significantly better for other types of cancers than they did for breast cancer (median coefficient, 0.81 vs. 0.57, P = .002).
So the question then arises, what are the better endpoints to determine a drug’s true biological and clinical effectiveness and to serve as the center of the clinical trials universe? Is it PFS or response — for which measurements are somewhat less precise but which are less influenced by a host of other factors — or is it the precise but dirty and inefficient endpoint of OS? Some might conclude from the analysis we give above that in breast cancer, in particular, it would be important to use OS as the outcome because PFS does not correlate with it, although our conclusions would differ: It is the OS variable that misleads about the value of the drug, and not the opposite.
References:
Broglio KR and Berry DA. J Natl Cancer Inst. 2009;doi:10.1093/jnci/djp369.
Piccart M, et al. Ann Oncol. 2014;doi:10.1093/annonc/mdu456.
Stewart DJ. J Clin Oncol. 2012;doi:10.1200/JCO.2012.44.1220.
Stewart DJ and Kurzrock R. BMC Cancer. 2013;doi:10.1186/1471-2407-13-193.
Stewart DJ, et al. Clin Cancer Res. 2015;doi:10.1158/1078-0432.CCR-14-3246.
Temel JS, et al. N Engl J Med. 2010;doi:10.1056/NEJMoa1000678.
Yardley DA, et al. Adv Ther. 2013;doi:10.1007/s12325-013-0060-1.
For more information:
David J. Stewart, MD, FRCPC, is professor and head of the division of medical oncology at University of Ottawa. He can be reached at University of Ottawa, 501 Smyth Road, Ottawa, ON K1H 5H6, Canada; email: dstewart@toh.on.ca.
Disclosure: Stewart reports no relevant financial disclosures.