Investigators excluded targeted therapies from the analysis that are associated with known safety concerns, such as dihydropyrimidine dehydrogenase gene testing and capecitabine, as well as any drug-indication pairs that were first approved in biomarker-unselected populations. In addition, the researchers excluded pertuzumab from the investigation because the trial conditions for the non-RCTs and RCTs for this drug were not similar enough to warrant direct comparison. Also, because it was first tested in all-comers, rucaparib was also removed from the analysis of biomarker-based drugs.
The researchers noted that they excluded almost all programmed cell death ligand-1 (PD-L1) drugs from their analysis, as some immunotherapies focused on checkpoint inhibition have been approved for use and indicated based on the existing levels of this biomarker, while other similar drugs focused on this target have been approved for use in patients regardless of existing PD-L1 levels. Wrote the study authors, “the PD-L1 level cutoffs are not always uniform, and the consensus on PD-L1 as a biomarker and the correct method of quantifying PD-L1 levels are still being debated.”
However, Dr Gyawali provided some additional context for why the group included programmed cell death receptor-1 (PD-1)–blocking antibodies pembrolizumab and nivolumab in the analysis: “We focused on targeted drugs, and PD-L1 or PD-1 inhibitors aren’t considered targeted therapies in the usual meaning of the term. PD-L1 aren’t classical targets … But we did include PD-1 inhibitors that were approved based on [microsatellite instability] status because that is a targeted treatment.”
The researchers found that although the RRs in non-RCTs were not “systemically” higher than the responses seen in RCTs (63% higher in non-RCTs), the DoRs for these targeted therapies were consistently higher (ie, 87% higher) in single-arm studies compared with controlled studies.
These findings suggest that the DoRs seen in early trial phases may decrease with time in later phases of study. Thus, DoR perhaps should not be used as a metric in treatment decision making, the authors reasoned, and single-arm trials with poor preliminary results should probably be abandoned: “Resources could be better channeled by conducting RCTs of drugs that showed responses in non-RCTs impressive enough to remain useful even if they might be expected to decrease by a certain percentage in RCTs,” the authors wrote.
The study authors concluded that while accelerated approval is a tool that the US Food and Drug Administration should still employ, the agency should not provide full approval to cancer drugs on the basis of durable responses in single-arm trials. “Caution must be exercised when approving or prescribing targeted drugs based on data on durable responses derived from non-RCTs, because the responses could be overestimates and poor predictors of survival benefit,” they reported.
Something else to consider about pembrolizumab and nivolumab, specifically, is that no companion diagnostics were approved contemporaneously when they were given the green light for the treatment of patients with microsatellite instability-high (MSI-H) and deficient mismatch repair (dMMR) defects. If different next-generation sequencing tools are used to assess MSI-H or dMMR status, this could mean that the results could be variable depending on the sequencing method used. The clinical implication of the use of different diagnostic tests to interrogate for MSI-H or dMMR status is that “if the drug doesn’t work in a particular patient, we will not know whether the test was defective and showed a false positive, or whether the drug didn’t work,” noted Dr Gyawali, who added: “I don’t understand the haste in approving a targeted drug that doesn’t even have a companion diagnostic based on responses from a single-arm trial!”
And, when asked if the team flagged any serious adverse events for the specific drug-indication pairs included in the analysis that got approval based on phase 1 trials — ceritinib and crizotinib — Dr Gyawali said that while the team didn’t explicitly look at the adverse event profiles across the drugs in the study, the aforementioned 2 drugs, specifically, have already been tested in RCTs for other indications — so any safety signals associated with ceritinib and crizotinib likely “haven’t stayed hidden,” and any other future safety red flags associated with these drugs may reveal themselves in postmarketing studies. “But this question is very important for other drugs like larotrectinib that have received their first approval based on single-arm early small trials, because important safety signals may remain hidden due to small sample size.”
Gyawali B, D’Andrea E, Franklin JM, Kesselheim AS. Response rates and durations of response for biomarker-based cancer drugs in nonrandomized versus randomized trials [published online January 7, 2019]. J Natl Compr Canc Netw. doi: 10.6004/jnccn.2019.7345