Vinod Nair co-wrote this blog post with Pranav Srivastava.
Since the first guidelines on surrogate endpoints came out in 2003—codifying the idea that endpoints like progression-free survival (PFS), time-to-progression (TTP) and objective response rate (ORR) could be used to gain special approval—they’ve become a mainstay in oncology trial designs, with a significant increase in the number of abstracts reporting PFS as an endpoint.
They’ve certainly helped expedite the trials and drug approval process, which once lasted decades but now takes around 12 years or less. Patients who have diseases with poor prognosis have benefited as new disease-altering drugs have become available quickly. One such case study is lung cancer or, more specifically, non-small-cell lung cancer (NSCLC), which makes up close to 80% of all lung cancer cases.
PFS has some advantages over overall survival. It matures earlier than overall survival (shorter follow-up), enabling a potential therapy to benefit patients sooner. Unlike overall survival, PFS remains unaffected by subsequent lines of therapy. This is more important in earlier line settings. If overall survival is the primary endpoint in these studies, a patient would have received multiple lines of therapy before death. This would result in subsequent lines of therapy having an effect on outcome, so the outcome can’t be cleanly attributed to the investigational agent. In the case of PFS, the clinical benefit of the drug isn’t affected by any subsequent lines of therapy.
BLOG POST: Is the CAR-T Upset?
However, there is a flip side. Achieving PFS has as much to do with trial design and patient populations as it does with the drug’s efficacy. Many clinical trials evaluating targeted therapies against chemo in select patient populations have demonstrated an improvement in PFS without a significant improvement in overall survival. Out of 18 pivotal trials (conducted after 1990) that reported both overall survival and PFS endpoints, only nine trials reported improvements in both endpoints, while six trials reported improvements only in PFS (such as the LUX-Lung 6 trial and AVAiL trial), and three trials reported improvement only in overall survival (the CheckMate 057 trial).
The challenge arises as a result of drugs like those in lung cancer, where there has been a discordance between PFS benefits and survival benefits. When is it okay to have no PFS benefit if there is a statistical overall survival benefit, or vice-versa? This question has been deliberated upon in many publications, and some reasons have been put forth to explain this discordance. PFS assessment may be subject to errors arising out of measurement methodology, subjectivity of interpretation and evaluator bias. While measuring PFS, progression is considered to have occurred on the date on which the new lesion is observed, which is usually as per a predefined schedule. Therefore, unlike the overall survival endpoint, which has a definite event date, there isn’t a singular or exact progression date that can be determined, introducing error in the measurement.
Specifically in NSCLC, another reason that has been put forth is the existence of genetic heterogeneity in the NSCLC patient population. This argument has been supported through biomarker-based subgroup analysis in some trials where concordance wasn’t observed in overall survival and PFS benefits during primary analysis.
In the LUX-Lung 3 and LUX-Lung 6 trials, PFS and overall survival benefit concordance was only seen during subgroup analysis with patients expressing del19 EGFR mutations. A similar trend was observed in the LUME-Lung 1 trial conducted in patients with advanced NSCLC. Improvement in PFS (the primary endpoint) was observed with nintedanib in all histologies, but the overall survival benefits were only observed in patients with adenocarcinomas.
In light of the biomarker-based argument, it becomes clear that even when a trial doesn’t meet the PFS endpoint, which occurs earlier than survival benefits, a genetic or tumor histology subgroup analysis is required in order to identify subgroups that could benefit from the drug. This would ensure a faster availability of the drugs for those patients, instead of patients having to wait for overall survival benefits to be demonstrated at a later point in time.
NSCLC is not an outlier with regard to this PFS/overall survival discordance. In the recently reported CheckMate-214 trial, the investigational drug combination (Opdivo and Yervoy) met two out of three co-primary endpoints (a significant effect in ORR and overall survival, but no statistical difference in PFS when compared to Sutent), therefore emphasizing the importance of patient characteristics in determining clinical efficacy in other tumors.
Can we do something more? There are already trials that track other endpoints such as PFS2 as well to help provide more guidance on the impact of the drug (not just when it’s being taken, but also what happens afterward). What this leaves companies, and the broader healthcare system, is a question of balance: expediency to market over life-lengthening benefit. Perhaps we really need to look at a triangulation of benefits—an improvement index—rather than just a singular measure that can factor in quality, quantity and duration of benefits over time.