Merck’s Keynote falls flat, and that’s just fine

With yesterday’s failure of Keytruda in head and neck cancer comes yet more evidence that not all anti-PD-1 agents were created equal. It also raises fresh questions about checkpoint inhibitor trial designs, and the precise role the US FDA now plays in overseeing the approval of these agents.

This is because Keytruda is already conditionally available for head and neck cancer, and the failed study in question, Keynote-040, might have served as the confirmatory trial to formalise approval; yet its failure is set to change nothing, at least if the signals Merck & Co has sent the sellside are anything to go by.

This might not necessarily be down to a regulatory view that the use of anti-PD(L)-1 agents across a vast swathe of cancer types should be strongly encouraged. But to strip a drug like Keytruda of an approval would be unusual for an FDA that seems to be loosening the straitjacket of approvability requirements, for cancer drugs at least.

To pull or not to pull?

A stark parallel exists with Roche’s Tecentriq, which in bladder cancer got conditional US approval – before flunking Imvigor-211, its sole confirmatory trial. Yet there is no indication that either Tecentriq or the three other checkpoint blockers with conditional approvals in bladder cancer are about to be pulled (Tecentriq failure puts accelerated approval in a spin, May 10, 2017).

Still, Merck, which told the sellside that the FDA remained comfortable with the current accelerated approval, has a stronger case for Keytruda in head and neck than Roche with Tecentriq in bladder cancer. For one thing, a second study, Keynote-048, is ongoing that could serve the confirmatory purpose, despite being in the first-line setting; Keynote-040, and the current conditional approval, are for second-line use.

Then there is the fact that Keynote-040 does not seem to have failed by much. Full data are not available yet, but Merck said risk of death was cut by 18% for the Keytruda arm versus chemo or Erbitux, with a p value of 0.03.

For reasons that are not clear Merck applied to this analysis a one-sided test – this considers significance only in the direction of interest, and normally doubles the chances of a successful outcome versus the more common two-sided test. And it said the 0.03 p value was not significant – presumably because multiple looks at the data had raised the bar.

Whatever the statistical minutiae, Keynote-040’s failure stands in contrast to the success of Bristol-Myers Squibb’s Opdivo in the Checkmate-141 head and neck trial. This makes it the third anti-PD-(L)1 surprise after bladder cancer – where Keytruda worked in Keynote-045 while Tecentriq failed – and first-line lung cancer of course (Checkmate-026 versus Keynote-024).

Are these inconsistencies due to differences in the properties of these agents, or is it down to trial design? The jury is still out, though it is known that the different antibodies do have varying binding affinities for their ligand, and that hitting PD-1 can theoretically exert subtly different effects versus targeting PD-L1.

An important factor might be the baseline characteristics of the patients who happen to be recruited into these studies. Merck has yet to disclose the PD-L1 expression levels in Keynote-040, for instance, while Bristol earlier said that in Checkmate-141 Opdivo worked better in high versus low expressers, Evercore ISI’s Umer Raffat points out.

Moreover, it is less than a year since Astrazeneca briefly suspended the Kestrel and Eagle studies of Imfinzi plus tremelimumab in head and neck cancer because of risk of bleeding. It could just be that this cancer will, for some checkpoint blockers at least, be a tough nut to crack.