2020 has been a weird year by any standard. So, I shouldn’t be surprised that the bamlanivimab story is so weird. But still. Anyhow, if you’re ready to go down the rabbit hole, here we go…
derivation and early testing
The story begins with Eli Lilly designing a monoclonal anti-spike antibody to neutralize COVID-19. The antibody was tested in rhesus macaques, where it caused a substantial reduction in viral titers (figure below).1 Sounds good…
One minor detail though – the antibody was given before the macaques were exposed to virus. So the drug seems to work, but you would need to go back in time and give it before the patient was exposed to the virus. Details, details.
ACTIV-3 clinical trial
This was a trial of bamlanivimab among hospitalized patients. The trial seems to have been halted prematurely due to futility. The only information currently available about the trial is a press release:
BLAZE-1 trial: initial design and results
BLAZE-1 is an adaptive, phase II trial to evaluate the use of bamlanivimab among outpatients with COVID.2 The trial is designed, written, and performed largely by Eli Lilly. The initial component of the trial randomized 452 patients to placebo or to three different doses of bamlanivimab. The study was powered to detect a reduction of 0.9 logs of virus (87% absolute reduction).
Patient characteristics are shown below. Patients were relatively young and at low risk of severe disease. In fact, one of the exclusion criterion was literally if the patients “have any serious concomitant systemic disease, condition, or disorder that, in the opinion of the investigator, should preclude participation in this study.” It appears that the study was looking for low-risk patients in whom they could prove a reduction in viral load, without encountering any worrisome toxicity.
The primary endpoint was a change in viral load at 11 days. This basically wasn’t met (table below). You may recall from above that the study was powered to detect a reduction in viral load of 0.9 logs. In all cases, the differences in viral load were well below 0.9 logs. At 11 days, one of the groups did have a reduction in viral load, with a p-value barely below 0.05. However, when corrected for the multiple comparisons performed here, that isn’t actually statistically significant (e.g., a Bonferroni correction for four comparisons would drop the p-value cutoff to 0.05/4 = 0.01). Furthermore, there’s no sensible pharmacological rationale for why the intermediate-dose group would show the greatest effect – so this seems more like the play of chance than causality.
So I would regard this primary endpoint as negative. The authors (more than half of whom are employed by Eli Lilly) try to spin this as positive. Seriously, kudos to them. Trying to spin a one-in-four co-primary endpoint as positive takes some serious chutzpa and linguistic acrobatics. To achieve this, they ended up transcribing the above table directly into the abstract, thereby creating one of the more confusing and awkward abstracts of modern medicine:
Let’s move on to the secondary endpoints. There’s a creative flex that goes on here. In order to make a tortured argument that the primary endpoint was positive, the authors needed to imply that there was something uniquely better about the the 2800 mg dose. However, in order to twist the secondary endpoints towards positivity, they switched their tune – and started combining all the treated patients (suddenly, it seems, all doses of drug are equivalent ??).
Patients treated with bamlanivimab had marginally lower symptoms at some timepoints (figure below). The difference is statistically marginal and of dubious clinical significance:
The most notable secondary endpoint ended up being hospitalization rates. 6.3% of the control group was hospitalized, compared to 1.6% of patients treated with bamlanivimab (p=0.02)(table below). This is provocative, but given the numerous comparisons among a variety of secondary endpoints, it’s significance is questionable.
There’s also a misleading sleight of hand here where the authors use the term “hospitalization” to refer to either hospitalization or emergency department visits. Equating these very, very different things is justified as follows:
Redefining the term “hospitalization” appears to be a post hoc modification of the secondary endpoints. In the original study protocol reproduced below, hospitalization was defined in a more reasonable way:
I can’t find anywhere, in the manuscript or supplemental appendix, how many patients were legitimately admitted to the hospital. This may seem like a quibbling detail, but with such small numbers a tiny numerical shift could cause the difference in “hospitalization” rate to lose statistical significance.
BLAZE-1 trial: next phase of the trial
BLAZE-1 is an adaptive trial – so Eli Lilly is shaping the trial as it goes along. That’s fine – that’s the way an adaptive trial is supposed to work. In fact, the way the trial is being adapted speaks volumes.
So what does the first phase of the trial show? The primary endpoint was essentially negative. There were nonetheless some signals of potential benefit in some of the secondary endpoints. So perhaps bamlanivimab works – but not as well as was initially hoped.
So, Eli Lilly has decided to combine bamlanivimab with a second antibody into a cocktail. The next phase of the study will test this two-antibody cocktail, compared to bamlanivimab alone or placebo. This seems like an entirely logical next step.
And here’s where things get weird
So bamlanivimab may show a bit of promise in a Phase II clinical trial, but it’s very very far from a slam-dunk therapy for COVID. Eli Lilly itself seems to be pivoting away from using bamlanivimab monotherapy – implicitly admitting that bamlanivimab isn’t very effective.
Yet somehow, bamlanivimab has received an Emergency Use Authorization (EUA) from the FDA and is currently being used in some centers. How did that happen? Why? I don’t understand.
The NIH guidelines admit that there isn’t enough data to know whether bamlanivimab works:
The IDSA guidelines likewise recommend against using bamlanivimab (see below). However, they add an interesting phrase: people may want it anyway, if they put a high value on “uncertain benefits” and a low value on “uncertain adverse events.” So, if you’re really into experimenting with drugs, maybe this is the right therapy for you?
- Bamlanivimab showed no benefit in the ACTIV-3 trial involving hospitalized patients with COVID-19.
- Bamlanivimab failed to achieve the primary endpoint of viral load reduction in a trial involving outpatients with COVID-19 (BLAZE-1). It may have caused some debatable improvements in secondary endpoints.
- Eli Lilly is now testing a cocktail of bamlanivimab plus another antibody. The company seems to realize that bamlanivimab might have some efficacy, but it’s far from a wonder drug.
- The IDSA and NIH guidelines recommend against generally using bamlanivimab.
- Somehow, bamlanivimab has nonetheless achieved an Emergency Use Authorization via the FDA and is being used at some centers.
- 2020 is weird.
more on bamlanivimab
- NEJM Journal Watch – Bamlanivimab for COVID-19 — Hard to Pronounce, Even Harder to Give.
- 5 things to know about bamlanivimab – ID stewardship, by Timothy Gauthier
- Expert advice on pronouncing bamlanivimab by Stephan Colbert and Jimmy Fallon:
- 1.Jones B, Brown-Augsburger P, Corbett K, et al. LY-CoV555, a rapidly isolated potent neutralizing antibody, provides protection in a non-human primate model of SARS-CoV-2 infection. bioRxiv. Published online October 1, 2020. doi:10.1101/2020.09.30.318972
- 2.Chen P, Nirula A, Heller B, et al. SARS-CoV-2 Neutralizing Antibody LY-CoV555 in Outpatients with Covid-19. N Engl J Med. Published online October 28, 2020. doi:10.1056/NEJMoa2029849