MADRID – Although pembrolizumab (Keytruda) was associated with a 19% reduction in the risk of death compared with the standard of care in patients with relapsed or metastatic head and neck squamous cell carcinoma (HNSCC) in the KEYNOTE 040 trial, the immune checkpoint inhibitor just missed meeting its primary efficacy endpoint of an improvement in overall survival.
The fault may lie in the confounding of overall survival results when patients who were initially assigned to standard of care were crossed over to subsequent therapy with a checkpoint inhibitor after the study ended, said lead investigator Ezra Cohen, MD, of Moores Cancer Center at UC San Diego Health Sciences in La Jolla, Calif.
“This was a trial that clearly did not meet its primary endpoint, but was felt to confer some benefit – at least in the opinion of this investigator – to pembrolizumab vs. standard of care,” he said at the European Society for Medical Oncology Congress.
The byzantine statistical design of the trial, while it may warm a mathematician’s heart, imposed stringent restrictions on the data that may also have led to the ultimate failure of the programmed death-1 (PD-1) inhibitor to meet the efficacy endpoint, he said.
In July 2017, Merck, which markets pembrolizumab, announced the failure in a, promising to present the data in a future medical meeting. Dr. Cohen’s ESMO presentation was the fulfillment of that promise.
In the trial, patients with squamous cell carcinomas of the oral cavity, oropharynx, hypopharynx, or larynx with disease progression after a platinum-containing chemotherapy regimen, or recurrence/disease progress within 3-6 months of multimodal therapy using platinum, were randomly assigned to therapy with either pembrolizumab 200 mg intravenously every 3 weeks for 2 years, or to standard of care at the investigator’s choice: either methotrexate 40 mg/m2 weekly, docetaxel 75 mg/m2 every 3 weeks, or cetuximab (Erbitux) 250 mg/m2 weekly after a loading dose of 400 mg per m2.
Some math required
Dr. Cohen explained that the statistical design of the trial involved a multiplicity strategy using a family-wise alpha strictly controlled at 0.025. The alpha was allocated in a stepwise fashion. The study hypothesis was that pembrolizumab would have an overall survival (OS) advantage with a one-side alpha, and if that was met, OS would then be looked at in specific cohorts according to expression of the PD ligand 1 (PD-L1) on cells, followed by evaluation of objective response rates and progression-free survival in each subgroup.
The final analysis was to be performed after 380 OS events had occurred among 495 randomized patients. The prespecified efficacy boundary for OS in the intention-to-treat (ITT) population was .0175, translating into a hazard ratio (HR) of 0.80.
Median OS in the ITT population was 8.4 months for the pembrolizumab arm, compared with 7.1 months for the standard-of-care arm. This translated into an HR of 0.81 (P = .0204), which do not reach the aforementioned efficacy boundary.
Rates of 1-year overall survival were 37% in the experimental arm, vs. 27% in the standard-of-care arm.
“When one looks at the biomarker-specific cohorts, we can see that these differences are further exaggerated in favor of pembrolizumab, Dr. Cohen said.
An analysis stratified by patients with PD-L1 expression on 1% or more of cells vs. less than 1% showed median overall OS of 8.7 months for those with higher levels of expression, vs. 7.1 months for expression levels below 1% (HR 0.75, P = .0078).
Similarly, an analysis comparing patients with a tumor proportion score (TPS; expression of PD-L1 in the membranes of 50% or more of tumor cells) with patients whose tumors had lower TPS levels showed a median OS of 11.6 months with pembrolizumab, vs. 7.9 months with standard of care (HR 0.54, P = .0017).
Objective response rates were also significantly higher with pembrolizumab in the patients with higher levels of both overall PD-L1 expression (17.3% vs. 9.9% with standard care), and in patients with 50% or more of tumor cells expressing membrane PD-L1 (26.6% vs. 65%, P = .0009).
In an exploratory analysis, the investigators also found that among patients in the standard of care arm who went on to receive a checkpoint inhibitor as subsequent therapy, the median OS was 20.1 months, compared with 9.8 months for patients who received other subsequent therapies, and 4.8 months for those who did not receive subsequent therapy, suggesting that crossover to a checkpoint inhibitor may have diluted overall survival differences between the trial arms, Dr. Cohen said.
Treatment-related adverse events of any grade were more frequent in the standard of care arm, except for hypothyroidism, which was substantially more frequent with pembrolizumab. The incidence of immune-mediated adverse events other than hypothyroidism was generally similar between the treatment arms.
“I think pembrolizumab, despite not meeting the primary endpoint of overall survival, showed evidence of activity and a [good] safety profile. So I think this study is borderline possible, since there is a 19% reduction in the risk of death, with a hazard ratio borderline to the statistical hypothesis that was initially planned,” said invited discussant Sandrine Faivre, MD, PhD, of Bichat-Beaujon University Hospitals in Paris.
Although pembrolizumab did not reach the primary efficacy endpoint in the ITT population, among patients with TPS of 50% or greater, the benefit with the PD-1 inhibitor was “exquisite” she said, adding that this subgroup of patients made up only 26% of the ITT population.
The study was funded by Merck. Dr. Cohen disclosed stock ownership in Human Longevity, Inc, and being an advisory board member for AstraZeneca, Bristol-Myers Squibb, Human Longevity, Merck, and Pfizer. Dr. Faivre disclosed unspecified relationships with Bayer Pharma, Bristol-Myers Squibb, Eli Lilly, Ipsen, Merck, Serono, and Novartis.