An FDA advisory committee on Thursday said the confirmatory trial data for the KRAS inhibitor sotorasib (Lumakras) in previously treated non-small cell lung cancer (NSCLC) cannot be reliably interpreted in the CodeBreaK 200 study.
Most members of the Oncologic Drugs Advisory Committee (ODAC) agreed with FDA staff that multiple sources of systemic bias during the conduct of the trial presented challenges in interpreting the study’s primary endpoint of progression-free survival (PFS). Only two members voted “yes” that the trial data can be reliably interpreted, with the remaining 10 panelists voting “no.”
“No one expects a perfect [randomized control trial], but what we hope for is a small number of issues in trial conduct, and an effect large enough to withstand uncertainties caused by those issues,” said Mark Conaway, PhD, of the University of Virginia in Charlottesville, in explaining his “no” vote. “For this trial, we seem to have the opposite — a large number of issues that cloud the interpretation of a small observed effect.”
As to what that means for sotorasib, the FDA’s Harpreet Singh, MD, emphasized that ODAC was not voting on whether the drug should be recommended to receive regular approval.
“It is not our intent to immediately withdraw a drug that has a ‘failed’ confirmatory trial,” she said. “It is under accelerated approval and there are multiple pathways available to us, and we are not making this move to withdraw the drug from the market based on these results.”
The phase III CodeBreaK 200 trial met its primary endpoint, showing superior PFS per blinded independent central review with sotorasib versus docetaxel (5.6 vs 4.5 months; HR 0.66, 95% CI 0.51-0.86). However, overall survival (OS) was no different between the two arms (10.6 vs 11.3 months, respectively; HR 1.01, 95% CI 0.77-1.33).
However, Singh noted that while the trial met its primary endpoint of PFS, it did so “with a small or incremental effect against single-agent docetaxel,” a standard second-line option in NSCLC.
Furthermore, she said that initial signs of potential bias, such as a high rate of patient dropout in the docetaxel arm relative to the sotorasib arm “led us to further investigate the potential of systemic and open-label bias in CodeBreaK 200.”
Specifically, Singh pointed out that 23 patients randomized to the docetaxel arm dropped out of the study, or withdrew consent shortly after they were made aware of their treatment assignment, compared with just two patients in the sotorasib arm.
She suggested that early, well-publicized data on sotorasib may have increased patient and investigator awareness of the drug, and their desire to access it. Thus, the high rate of early dropout in the docetaxel arm was a signal that CodeBreaK 200 “may have had a perceived lack of equipoise by both patients and providers,” she said.
In addition to the issue of the asymmetric early dropout rate, FDA staff also expressed concerns about investigator assessment of progressive disease favoring the sotorasib arm, and patient crossover from docetaxel to sotorasib before assessment of disease progression by the blinded independent central review.
In explaining his “no” vote, Daniel Spratt, MD, of Case Western Reserve University in Cleveland, noted that both PFS1 and PFS2 favored sotorasib, “but there was no difference in overall survival, and no further explanation, which leads me to believe that, with all the discussion we had, that there is likely bias, or inaccuracies in the PFS assessment. So, unfortunately, I lack the confidence and reliability in the PFS endpoint.”
The FDA granted accelerated approval to sotorasib in 2021 based on objective response data from the single-arm CodeBreaK 100 trial, making it the first targeted agent for treating tumors with KRAS 512C mutations, which account for roughly 13% of all mutations in nonsquamous NSCLC.
“Drugging KRAS 512C is certainly a landmark scientific discovery, and there is little question about the activity of sotorasib,” commented Neil Vasan, MD, PhD, of Columbia University Medical Center in New York City. “But, we were asked if the PFS can be reliably interpreted in CodeBreaK 200 and I felt the answer was no. The magnitude of effect is small … statistically significant, but not clinically significant.”
He added that the oncology community needs to come up with strategies to mitigate the perception of equipoise, “which may have led to biases in this trial, and I think we in the community have to address this, so we can balance hope with hype for new therapies for our patients.”
Jorge J. Nieva, MD, of the Keck School of Medicine of the University of Southern California in Los Angeles, was one of the two “yes” votes, and reasoned that the study met its primary endpoint based on the intent-to-treat analysis.
“Ultimately, we have to take the statistical plan as it is written and analyze things according to what was planned,” he said. “The post-hoc analyses are informative, but they don’t change the benefits that were observed, and I don’t think a type I error occurred here. Given the corroborating evidence, I have confidence the drug does have a PFS benefit over the comparator in this case.”
Both the FDA and Amgen — sotorasib’s developer — conducted sensitivity and “tipping point” analyses to address the patients in the study who were censored early due to dropouts or crossovers.
Amgen representatives argued that their analyses showed that CodeBreaK 200 can be reliably interpreted to confirm the clinical benefit of sotorasib, while the FDA analysis suggested that the primary endpoint of PFS “may not be sufficiently robust or able to withstand variability in patient outcomes.”
Singh said the fact that both Amgen and the FDA conducted these different analyses confirmed FDA’s concerns about the reliability of the study’s results.
“We have a heavily clinical committee that is trying to understand complex statistical concepts,” she pointed out. “I think that understanding that both the FDA and the applicant performed various statistical analyses to interpret the robustness of the results — and different assumptions can be used with different results — highlights our overall challenge in interpreting these results.”
Richard Pazdur, MD, director of the FDA’s Oncology Center of Excellence, said the focus on these PFS analyses shows the value of using OS as a primary endpoint.
Docetaxel’s approval was based on an OS advantage, he noted, “and we wouldn’t be discussing all of these complexities of biases and sensitivity analyses if we had been dealing with an inferiority in overall survival or noninferiority in overall survival.”