As Fortune writes, "BMS shed about 16% of its market value Friday after a clinical trial failure." Here. At FiercePharma, "The news swept away $25B in market value." Here. Matthew Herper at Forbes picked up on the important personalized medicine angle, here, and this blog will nerd-dive that, further below.
What happened in brief after the break, and why pathology, test standards, and personalized medicine play a very big role.
There is huge interest in targeting the immune system against cancer, mainly by interfering with the PDL-1 system that protects cancer from immune-mediated cell destruction. Several PDF-1 indications and drugs are in hand and results are sometimes dramatic.
In lung cancer, Merk's Keytruda and BMS's Opdivo are both approved in second-line therapy (after a first therapy has failed). Opdivo was being studied as monotherapy, in first-line patients, and in patients with only 5% of cells positive for PDL-1. 5% isn't very much.
There are other studies, and advocates, for cutting PDL-1 positive at higher values - 10%, maybe 25%, and for subgrouping data for patient impact when cells are 50% or even 75% positive for PLD-1. Matt Herper at Forbes picked up that this headline-making disaster says something good about precision medicine - the pony under the big pile of bad news (here). Matt hones in on the scientific decision here:
Merck chose to run its lung cancer trial in patients whose tumors had PD-L1 levels of at least 50%. But Bristol-Myers Squibb chose to test patients who had PD-L1 levels of more than 5%.... this difference seems the most likely explanation for its failure.
But here are a couple things to think about:
- Tests Vary Alot and it Matters Alot. FDA and AACR co-sponsored a workshop last year on the challenges of harmonization among biomarker test methods, e.g. if there are three kits with three antibodies for PDL-1. The 133-page deck is still online, here, and a four page white paper about harmonization methods was written, here. IHC sounds old-fashioned, but PDL-1 testing is intrinsically an immunohistochemical test today - you don't really want to grind-and-bind the tumor to measure fibroblasts or macrophages that are PDL-1 positive. Pathology test harmonization is really really hard. Add "X" many IVD tests and "Y" many LDT tests and the complexity multiplies. (see also [fn 1] ).
- CAP Today - Two Articles. CAP Today ran a very long, very detailed review article updating us on expert opinions and the ins and outs of PDL-1 testing and scoring - February 2016, here. It's so important, and complicated, CAP Today revisited it with a cover story just a few months later in July 2016, here.
- The financial incentives for (Barely) Personalized Medicine. Medical centers (or pharma's) eager to get patients on drug probably have all sorts of ways to boost sensitivity through antigen recovery, extra antigen amplification, and so on. So you get "better" PDL-1. But patients with the smallest natural expression seem to have the smallest real-world drug benefit. Note that this issue around ultra-sensitive biomarker dectection is still being ironed out in NGS tumor sequencing - what's the value in detecting mutations found in 5% of cells? (This topic also arose at an FDA-AACR liquid biopsy policy workshop in July 2016).
- Pricing by Indication Meets IHC? (Why not?) There's a lot of interest at PBMs, payers, Medicare, and journals about "pricing by indication." This is usually taken to mean different cancers and the same anti-cancer drug (e.g. Avastin in colon cancer versus lung cancer - pricing by indication.) However, the Opdivo results suggest that if you go down this road, it's just as rationale to pay more for Opdivo in the lung cancer cases that have 75% cell positivity for PDL1, and less, maybe way less, in the cases that have 5% cell positivity.
- Don't Believe Every Aspect of an RCT. Even if the trial had been, just barely, positive in the 5% population, it wouldn't mean that 5% expressers were responders. Maybe (in this hypothetical) there were a lot of 50% and 25% PDFL-1expressers with strong results, and if you average in some 5% expressers down at the bottom, they might have been zero-responders yet not have enough numbers to ruin the trial's average response over all patients studied (those with 5-75% positivity). Subgroup analysis would be helpful, but that has its own problems statistically. The question is worth asking since it's not intuitive that 5% expressors would be responders - maybe, but prove it hard.
Fn 1. For more on PDL1 variability, methods, and solutions, see In the Pipeline (here) and the Blueprint Industry Group / AACR initiatives ( here, here, here); for a six document cloud ZIP file (~500pp) of the 2015.03.24 AACR/FDA conference on PDL1 and CDX harmonization, here.