Overall survival is a critically relevant end point in randomized oncology clinical trials, but its interpretation is often not as straightforward as some may wish to acknowledge.
Maurie Markman, MD
Oncology clinical and research communities and the FDA have increasingly recognized that progression-free survival (PFS) may have a critical role serving as a primary end point in randomized trials designed for regulatory approval. However, this helpful development has not prevented some academics from continuing to suggest that using disease progression is considered a poor surrogate and that overall survival (OS) is objectively more meaningful than PFS.
A recent example of the rigidity of this line of thought is a recent manuscript published as a research letter.1 In an examination of the price of oncology products in the United States, the authors suggest that a demonstrated improvement in OS in a randomized trial for regulatory approval indicated superior efficacy compared with evidence of superior PFS in the approval process.
There can be no legitimate debate regarding the relevance of data generated from randomized trials in the development of beneficial oncologic therapeutics. Commonly used primary end points in these studies include both OS and PFS. Although the importance of OS to patients and their families is understandable, demonstrating a statistically significant improvement in this metric is often difficult due to the availability of multiple therapeutic options in an increasing number of clinical settings after an individual has progressed or failed to respond to the assigned approach in a randomized trial.2
The disassociation in the direct relationship between PFS and OS because of the evolution in patient care was highlighted in a provocative report examining this issue in metastatic melanoma.3 When a new generation of highly effective targeted therapeutics was introduced into clinical trials, there was a not-surprising correlation between these 2 survival metrics. Patients progressing on the “control study arm” and being unable to subsequently receive the experimental agent or agents experienced both shorter PFS and reduced OS compared with individuals randomly assigned to receive treatment on the investigative study arm.3 However, once cross-over was permitted, the earlier reported correlation between PFS and OS (correlation coefficient, 0.96) in the absence of study cross-over was substantially reduced (OS correlation, 0.55).
Of note, individuals randomly assigned to the experimental and control arms benefited from having received the novel agents and experienced improved OS from either the administration of the investigative agent from the start or after initial progression or failure to respond to treatment in the control arm.3
An older, yet critical example of the disassociation between PFS and OS is provided by the randomized trial, which compared the investigative treatment imatinib (Gleevec) with standard-of-care interferon-alfa (IFN-α) plus low-dose cytarabine as first-line therapy for chronic myeloid leukemia.4 This landmark study demonstrated a statically significant improvement in time to disease progression at 12 months, with 96.6% of patients who received imatinib having no progression vs 79.9% in control group (P < .001). No difference in OS was observed between the experimental and standard-of-care regimens (P = .23).4
This failure to document an OS benefit was surely because of the study’s ethically mandatory cross-over design, with 90% of patients randomly assigned to IFN-α plus cytarabine receiving imatinib within 9 months of initiating study treatment.5
However, based on the metrics used in the manuscript highlighted at this beginning of this commentary, the failure of imatinib to demonstrate improved OS compared with the control regimen should suggest this strategy not be considered to provide a high level of efficacy.6 This would be an absurd conclusion.
A more recent example of the complexity of evaluating the influence of multiple therapeutic options on OS outcomes among individuals who have completed, failed to respond, or progressed following an initial response to therapy on a randomized trial is the provocative data from trials evaluating single-agent PARP inhibitors as later-line therapy for epithelial ovarian cancer.7 The safety and substantial clinical benefit of this class of agents in ovarian cancer when employed earlier in course of disease, particularly in the presence of homologous recombination deficiency, has been well documented.7
However, limited data from several phase 3 studies have revealed possible inferior OS for individuals randomly assigned to PARP inhibitors compared with cytotoxic chemotherapy.7 Once an individual has exited a clinical trial, posttreatment options are appropriately optimized, and clinically beneficial options, such as alternative agents, radiation therapy, or surgery may be employed. As a result, one can only speculate as to the reasons for this quite unexpected and disconcerting observation related to OS.
Is it possible that despite the reasonable suggestion that the cancers of patients in these later-line chemotherapy trials are “relatively chemoresistant,” individuals randomly assigned to receive a PARP inhibitor may have been less likely to receive additional chemotherapy after leaving the trial compared with patients treated on the chemotherapy arms? Further, is it possible that although generally well tolerated, the adverse effects of PARP inhibitors (eg, low-grade nausea, fatigue, anemia, etc) experienced at this point in a patient’s illness may have been more likely to result in a decision to discontinue further treatment compared with those randomly assigned to cytotoxic therapy? Finally, is it possible that failure to receive a drug, for example, carboplatin, in a patient with even a modest degree of persistent chemosensitivity within the tumor may have led to a shorter OS?8
This discussion is pure speculation but highlights a plausible explanation for the observation of reduced survival. And yes, OS is a critically relevant end point in randomized oncology clinical trials, but its interpretation is often not as straightforward as some may wish to acknowledge.
Maurie Markman, MD, is president of Medicine & Science at Cancer Treatment Centers of America, a part of City of Hope.