Discover more from Inside Medicine
Covid antivirals not shown to decrease household spread. Here's why they still might work.
At the moment, there is no known effective prophylaxis against Covid-19. The only way to avoid infection is to avoid exposure. Previously, an infusion called Evusheld was effective in preventing infections, but newer variants rendered that treatment (which was primarily for the immunocompromised) useless.
But what about the antivirals like Paxlovid (Pfizer) and Molnupiravir (Merck)? Might these drugs reduce SARS-CoV-2 infection among people with a known recent exposure to a Covid-positive patient, (also known as “post-exposure prophylaxis” or “PEP”)?
Maybe. And I say this despite the fact that there are now two randomized controlled trials that appear to show failure for PEP.
Let’s go through the reasons for my continued cautious optimism.
Inside Medicine goes into the data where others fear to tread. If you find this valuable, please consider becoming a paid subscriber to help keep it going. Thanks!
The data. Last year, Pfizer announced that its PEP trial failed. By day 14, investigators found that the rate of infection among participants who received 5 days of Paxlovid was 32% lower than the placebo group, and those who received a 10-day course had a 37% lower rate than the placebo group. Unfortunately, neither finding was statistically significant, and these numbers failed to meet the pre-defined goal of the study.
This week, Merck announced that its PEP trial also failed. Same story. In this case, participants who took Molnupiravir for days 5 days had a 24% lower rate of infection by day 14 than those who received placebo. But again, statistical significance was not reached. So it was also chalked up as a failure.
Why I’m still optimistic. Usually, I tend to say things like, “If an industry-funded study fails, it has to be really bad news because these studies are often subtly designed to succeed.” In this case, I think the investigators have a fundamental misunderstanding of coronavirus virology. As a result, the study designs were not adequate to detect what I think may be hidden benefits here (or rather, the detected lower rates of infection are meaningful, even though the statistics say they’re not distinguishable from noise).
In both studies, household contacts (who were confirmed not to have Covid-19 at the time of entry) were randomized to get either the antiviral or placebo after exposure to a person with a newish Covid-19 diagnosis (called the “index patient.”) I italicized newish because the household contacts were eligible for the study if the index patient had had a positive Covid-19 test and at least 1 symptom within 5 days.
The gaping hole is that symptoms generally come a few days after infection and don’t necessarily say anything about contagiousness. We know that Covid-19 patients who are asymptomatic (or pre-symptomatic) can and do have viral loads that are similar to those of symptomatic patients. That means in this study, the horse had left the barn in a great number of cases before the study even began.
A large fraction of coronavirus spread occurs early on (a few days after infection). The fact that these studies found (statistically insignificant) 24-37% reductions in infections despite enrolling household contacts of patients who were already symptomatic (for up to 5 days even) is actually remarkably good, in my view.
The reason these numbers were seen as failures is that the researchers were probably aiming for a 50% reduction (I can’t seem to find information on what researchers call “power calculations,” though I am trying to find out).
Why? Like I said: fundamental misunderstandings of coronavirus virology. If the researchers had realized how much transmission happens in the pre-symptomatic period, they might have aimed for a lower reduction as the bar for success. Also, the researchers looked at the number of participants who were positive by day 14. If a bunch of people in both arms of study were infected within 24-48 hours of the first dose of Paxlovid or Molnupiravir, those cases would tend to dilute the power of the study to detect any meaningful difference. (An analogy comes to mind to explain this statistical situation: imagine timing a stove’s performance boiling water. If the starting temperature is actually lower than you believed, you’d conclude the stove stinks. But in fact, your assumption about the baseline condition fooled you.)
Indeed, it turns out that Covid-19 patients are probably contagious for around 8 days on average (some shorter, some longer). And the peak viral load (which corresponds to peak contagiousness) in symptomatic cases has been found to occur on or around day 3 of symptoms. If household contacts were enrolled that day or the next day, the rate of infection in both arms of the study would be high. This is exactly the kind of thing that could take real 24-37% reductions and render them statistically insignificant, even though there might actually be a protective effect “in real life.” Meanwhile, asymptomatic patients may (may) have shorter windows of contagion, but their peak viral loads (which is likely when most spread occurs) appear similar to symptomatic cases.
All told, these studies show that if you enroll household contacts 1-5 days after exposure to a symptomatic Covid-positive patient, the reduction in infections is not statistically significant. But if the studies had aimed for less lofty goals (or had been designed to detect patients before they became symptomatic), the headlines may have have said that PEP worked! It’s amazing how something as subtle as study design can change the outcomes of a study.
Why it matters. Right now, there are no monoclonal antibodies (or any treatments) for reducing the likelihood of infection among patients with high risks of developing a serious case of Covid-19. PEP with Paxlovid or Molnupiravir may turn out to be the best option. If these drugs reduce infection among the immunocompromised even by a quarter or a third (and honestly, it could be much more than that, given the above), it would be worthwhile. Currently, however, that is not routine practice, largely because these studies were designed with an inadequate understanding of how and when this virus spreads from a source to a target.
It’s me, folks. Look, I’m not exactly the world’s greatest optimist when it comes to the Covid-19 antivirals. They work, but probably not as well as most people think. Certainly in the pre-vaccine era, Paxlovid and Molnupiravir showed real benefit in high-risk people. Post-vaccine, though, the benefits appear to be far smaller for Paxlovid, with positive data thus far limited to older people, especially those with significant medical problems (although I have reason to suspect it helps younger people with very serious medical problems as well) and almost no residual benefit for most Molnupiravir takers.
And yet, I remain an optimist on PEP for Covid-19. Given that these drugs decrease viral loads dramatically within a few days, it seems to me that these studies should be repeated. Next time, investigators should provide the participants the drugs (or placebo) weeks/months before anyone in a home gets Covid. Routine asymptomatic testing should be done (with rapid tests, or even better with PCR, though that’s less practical) and as soon as one person tests positive, the household contacts should start their pill packs (blinded to whether they’ve been given the drug or placebo).
I’d wager that the outcomes would be favorable, especially if the power calculations are done with a modern understanding of this virus’s transmission dynamics. Of course, unless someone does the science (correctly this time, with a proper understanding of the virology), we’ll never know.