Rethinking Molnupiravir

I was quite surprised by the efficacy that Merck reported for the viral polymerase inhibitor molnupiravir when those interim trial results were announced in October. But the FDA just held an advisory committee meeting on the drug yesterday (here’s Matthew Herper’s writeup), and the data that have come in since then are, frankly, unimpressive. Which makes a person think again about the whole risk/benefit balance for the drug.

It’s the benefit side that’s changed – the risk (as far as we can tell) is about the same. Let’s look at three components of it. First off, I do not think that molnupiravir is carcinogenic, as I detailed in this post. Its mutagenic potential has been studied thoroughly, and I believe that it’s safe to use as intended (both in its dosages and in its duration of therapy). There is another worry, of course: in Merck’s trial they deliberately excluded pregnant women, because (as the world knows) molnupiravir works by disrupting the RNA copying of the coronavirus. Now, DNA copying in human cells is different, but you wouldn’t want to find out the hard way that it’s not as different as you believed, so I think that caution is warranted. Embryogenesis is exactly where something like that could have the worst effects, and we know how to avoid that risk even though we’re not sure how much of a risk it even is.

The third risk factor is that you could cause mutations in the coronavirus, but not enough to make it non-viable, and perhaps even give it some mutations that make it worse as a pathogen. This issue came up during the advisory committee hearing several times. It’s a serious question, but for what it’s worth, I think odds of this happening are very low indeed. If you look at the sequence maps of where we’ve been seeing coronavirus mutations so far in the pandemic, you’ll note that there are many parts of its genome that have been very quiet. That’s not because things don’t mutate there – there’s no reason to think that there’s any intrinsically higher mutation rate in any particular part of the viral sequence. No, that is because mutations in those regions are actually disadvantages, and the resulting viruses don’t spread.

It’s similar to the famous realization that Abraham Wald came to while looking at the data for military aircraft damage during World War II. Famously, one recommendation had been to add armor to the areas on bomber aircraft that showed damage from fighters and flak, but Wald pointed out the survivorship bias implications: these planes were the ones that actually returned, and the damage they picked up indicated the places where you didn’t need armor, because it showed that planes could be hit there and survive. The armor needed to be in the other parts of the planes, because those were the ones where damage sent them down in flames. That’s what you’re seeing when you look at viral mutation maps: pure, 200-proof survivorship bias. The coronavirus is constantly throwing off mutations of every kind, and the overwhelming majority of those disappear and are never heard from again. I’ll have more to say on this in an upcoming post on the omicron variant, so I’ll leave this part of the discussion for now.

So those are the risks, as I see them. With the benefit shown in the interim analysis from the MOVe-OUT trial (whose enrollment was stopped for efficacy, remember), molnupiravir looked like a good bet. But what if that benefit isn’t as high as it looked like? That’s what made the advisory committee vote close (13-10), and it very much complicates the FDA’s decision about whether to approve the drug (and under what conditions). Herper’s article quotes a committee member saying “I think we need to stop and acknowledge that the whole reason we’re having this discussion is because the efficacy of this product is not overwhelmingly good”. The initial results were a 50% relative decrease in hospitalizations in the study’s unvaccinated high-risk patients (on an absolute basis, a decrease from 14% hospitalized down to 7%). But the final data showed only a 30% relative decrease, and because the numbers also changed in the control group, the absolute change was only 3% in the end. The advisory committee kept trying to work out how this happened, and it appears that Merck’s team really didn’t have a good answer, either. But it really makes you think that the interim read was about the rosiest view available, and that the real-world effects are going to be less impressive. I mean, the way those numbers were going, who’s to say that this is the floor? And remember, these were patients at higher risk of developing severe disease – what would the benefit look like if you opened the drug up to use by the general population as soon as anyone tested positive?

No, from the way things look, I would say that if the FDA approves the drug that it will only be for high-risk patients, and that there should be a REMS (risk evaluation and mitigation strategy) in place to make sure that it doesn’t get into any woman who might become pregnant during the course of treatment. I now think that molnupiravir is far less likely to be useful in (for example) a vaccinated patient with a breakthrough infection, unless they have other risk factors (advanced age, other diseases, and the like). That leaves me pinning more hopes on Pfizer’s protease inhibitor – and it also leaves me wondering why there hasn’t been a trial started with these two drugs in combination. A lower dose of molnupiravir has the potential to add to the efficacy of the Pfizer drug and perhaps to extend its utility in the face of possible resistance. But it’s not happening, and it should.

Writing about COVID-19

As well as building up a resource of information and analysis on COVID-19, we want to ensure that we pass on any tips about what can go wrong when writing about this subject: and how to get it right! If you have experience in writing about this area and feel you have advice that would help others, please contact us at:

We’d also like to hear from you at that address if you would like to regularly contribute links to the site. If you just want to suggest links on an occasional basis, please send them to: