We're all a bit worried about the two neurological disorders seen in the AstraZeneca vaccine trial. In the media, one is being called multiple sclerosis; the other is being called transverse myelitis. These diseases can happen spontaneously, can happen as a result of a viral
infection (we think) and could possibly be an immune response to a vaccine. With the whole world watching and hoping, I have only sympathy for the trialists and sponsors and of course the participants. I'm seeing calls for "opening the books" and breaking blinds and increasing
transparency. I'm seeing people call for statistical tests to let us know the truth. But here's the thing: you can't make meaningful statistical claims about two events that may not even be the same. These serious adverse events are almost never so simple to figure out. Just as
a ballpark guess it would probably take 10 or more clear cases that fall only in the vaccine group to have real statistical confidence of causation rather than random luck. And breaking the blind publicly will likely make the trials difficult or impossible to perform in a
way that gives us meaningful information. We need large, well-conducted trials. We need safety boards working overtime on those trials. We need those safety boards to be able to do their jobs, which they can't do if they're having to unblind the entire world every time they
review an SAE. I know that some people feel like they can't trust federal agencies, and I know that those agencies are under intense political pressure right now. But I think our most rational move forward is to allow FDA and DSMBs to do their work, to strongly discourage
premature termination of trials over soft "efficacy" when we know that it's efficacy AND safety that are crucial to the success of a vaccine, and to continue to support good, focused science while we debate philosophically amongst ourselves about politics.
One specific thing that might help immediately would be to make freely/widely accessible prior vaccine trial datasets (flu etc) with annotated SAE rates in vaccine vs placebo groups. That's something that could be done immediately and would help the DSMBs to do their job.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Just a couple thoughts from the realm of COVID-19 trials. 1. We are honestly very excited that fewer patients are afflicted by COVID-19. We will work as hard as we can to make sure trials are available to patients wherever they are in the world AND we will celebrate the fact that
this is becoming a rarer disease.
2. This really looks to be a disease of unvaccinated people now. It's striking how little ARDS we are seeing among vaccinated people. Not sure whether any specific line of evidence is helpful, but, wow. In the big COVID ARDS trial we're
running, about 90% of the patients are unvaccinated. There are the public health reasons to be vaccinated as well, but the clinical side of things -- preventing life-threatening disease -- is quite persuasive as well.
In case the story about full-dose anticoagulation in critically ill patients is confusing, a quick comment about the design and the reason what looks like harm is being called futility. Recall that in the prespecified Bayesian model, they used a threshold of 99% probability
that the OR was > 1 or < 1. This is an appropriate move. If you drop the probabilities lower, the Bayesian approach starts to look a lot more like an error-inflation machine. So if you're being rigorous, you want for 99% probability before you declare superiority or inferiority.
At the relevant interims, they met the 99% probability for superiority in the moderate (no organ failure) group, but they only had 98.5% probability of inferiority/harm in the organ failure/severe group. However, they had a futility bound that was clearly met, so they were able
I'm thinking more about the Pfizer vaccine (and grateful for those who have clarified some key points). First, I agree that early efficacy (right after the second injection) prevents infection in this patient population. That seems statistically certain.
Second, the population is those the investigator deems to be at high risk. I know that we've all wanted "wiggle room" in our inclusion criteria and that's not wrong. But for generalizability we want to know what "high risk" really means. In the event, it looks like "high-risk"
meant ~40/10,000 in the placebo group got COVID in the first month, versus ~4/10,000 (having to estimate because not enough detail in the news release). Note that (contra some bizarre innumeracy in Vox), this doesn't mean it fully protects 9 out of every 10 people who receive it
I've been hearing and thinking more about the Pfizer mRNA vaccine this morning. It does seem like this signal is likely to be real (there's not enough data in the news release to know, but most credible back calculations suggest that the efficacy signal is real). So I've been
trying to think what bothers me about it. And it's the ways that the culture of interim analyses seems to be shifting. Instead of the clear boundaries keeping the businesspeople from interfering with trials while they're being run, we're now seeing premature breaking of
that confidentiality for reasons that aren't entirely clear to me. I would hate for one of the legacies of the COVID pandemic to be that pharma gets to break the protections of interim analyses in order to beat their competitors in competition or sway the public. I can't wait
HAHPS trial also published--atsjournals.org/doi/abs/10.151…. I'm incredibly proud of @Research_Inter ability to run a pragmatic controlled trial in context of ORCHID to allow us to move more efficiently through candidate agents. There's an early suggestion that azithromycin may be
useful (looking forward to RECOVERY results in that regard). If anyone can remember back to March, there was a plan to provide HCQ with essentially no oversight, consent, or monitoring. With retrospect, HAHPS (and its sister trial HyAzOUT) were exactly the right response.
Based on the suggestion of a harm signal in trials and environments with minimal safety monitoring and open eligibility criteria and the entirely neutral effect seen in ORCHID, I believe that having HAHPS as a response to a public impulse to use unproven therapies outside of
Once a year I give a lecture to the 2d year medical students on humanizing intensive care. It's fun to be with these bright young people moving toward careers as physicians. One asked me an interesting and important question that seems like it's worth reflection.
Specifically, how do we engage people in positive and respectful ways if we don't have an authentic connection with the patient/family member? We know from RCTs that rote condolence letters after a death seem to raise and then disappoint expectations of intimacy, so how do we
support people without being ingenuine (and/or raising expectations we can't meet)? These questions resonate especially loudly in our modern cultural moment where authenticity is prized, and our radar is tuned to detect inauthenticity, especially among those with power.