New data: Paxlovid may help high-risk vaccinated adults ages 50 and under, but not patients with asthma or those without serious medical conditions.

Should vaccinated adults ages 50 and under who get Covid-19 take Paxlovid?

That was what I wanted to know.

Last year, I’d started to notice that young, healthy, and vaccinated people were taking the drug frequently, even though it had only ever been shown to help high-risk unvaccinated patients. As Covid became more common, young people reached out to me asking if they needed the drug.

I was pessimistic but I didn’t know the real answer. Pfizer had cancelled its own study of younger healthy people (including vaccinated patients) because things were not working out. Researchers in Israel found an association with good outcomes in older people who were vaccinated, but the effect was far less impressive than the original blockbuster pre-vaccine era study, and patients ages 40-64 did not appear to benefit.

While I was predisposed to think the drug was unlikely to help most younger, healthy, vaccinated adults, I thought it could help young people who had serious medical conditions. So, I gathered a team of researchers and we looked into it.

Today, we announced our findings. We conducted an analysis of a large database containing millions of possible patients, and thousands of Covid-19 patients who were similar enough to study. I’m proud that we were able to publish our results in Clinical Infectious Diseases, the most prestigious infectious diseases journal in the United States.

The upshot:

We found the Paxlovid was associated with decreases in serious outcomes in vaccinated adults ages 18-50 but that the effect was limited to those with serious comorbidities like cancer and heart disease. Notably, people with chronic respiratory conditions like asthma also did not appear to benefit from Paxlovid.

In our study “serious outcomes” were defined as emergency department visits (all causes), hospitalization, or mortality. The effect of Paxlovid in our study population of younger vaccinated people was either large or small, depending on how you view things. On one hand, the rate of the bad outcomes we tracked was 30% lower among Paxlovid recipients (4.9% vs 7.0%). But most of that was ER visits. In our dataset, you’d have to give 91 patients the drug to prevent 1 hospitalization—and we studied a remarkably sick cohort of people. For example, 33% of the study cohort had cancer. That’s hardly representative of the population of adults ages 18-50.

Still, 10 deaths occurred in the cohort we tracked, all of which occurred in people who had not received Paxlovid. While we tried to control for risks across the groups, it’s possible that the patients who did not get Paxlovid had higher risks than we were able to measure. That’s always a risk of observational studies and why a prospective randomized trial with blinding would be superior to our retrospective study. Our study does not prove causality. It is “hypothesis generating.” Still, right now, it’s the best we’ve got.

In the graphic below, we show how different patient groups did. The dark magenta vertical line (“No change”) is the line in the sand. Any dot to its left implies a lower rate of serious outcome among patients who took Paxlovid. However, if the horizontal line crosses the dark magenta line, the finding is not statistically meaningful. So, patients with both cancer and cardiovascular disease who took Paxlovid had an over 50% reduction in serious outcomes. But there was no change among patients with existing lung problems (“Chronic lower respiratory disease” including asthma), or those without other serious comorbidities.

Be in the room where it happens.

Working with my colleagues on this research was extraordinarily rewarding. First, I got to work with some exceptionally smart and thoughtful people. In particular, my Harvard colleagues Dr. Sarju Ganatra and Dr. Paul Sax sharpened my mind and made me a better researcher during this process. Second, and relatedly, I saw something fascinating play out about how existing leanings matter when doing work like this.

Early in the study, it became clear that some of my colleagues were “Paxlovid optimists” while I was something of a “Paxlovid pessimist.” This diversity of bias (which we affectionately refer to as our “priors”) actually made the paper far stronger than it otherwise would have been, I believe.

A Paxlovid optimist could look at our data and find ways to package this as a big win for the drug. “Paxlovid was associated with improved outcomes for the entire study cohort!” they could say.

“Aha, but wait!” a Paxlovid pessimist might say. “The positive findings were exclusively driven by patients with serious conditions like cancer and heart disease. When you look at patients who were about as healthy as the general population, the drug did nothing!”

Indeed, various iterations of just this kind of back-and-forth played out behind the scenes during the interpretation and writing phase of the project. The data were the data. But how we framed and contextualized the findings was up for discussion. Depending on how things went, the final result could have looked very different.

As it happened, I was tasked with writing most of the manuscript, and for creating the first draft. With each round of revisions, I would consider comments (and even bright line objections) from my collaborators. Sometimes I would find ways to compromise, and sometimes I pushed back and explained why. But ultimately, because it was I who “held the pen,” I had the most opportunity to gently guide where things headed in terms of how we presented our findings, and our interpretations. And yet, I was working with genuinely smart people, who made important points I could not ignore. The discourse was high-level and the lobbying was bi-directional.

The result is a paper that was written by someone like me, tempered by the responsibility of representing others. (Funny, that.) If you read the paper closely, you can perhaps hear that the “main voice” of this paper is not one of unbridled enthusiasm. It’s consciously cautious and centrist. Yes, the drug was associated with good results overall. But no, not everyone who is young and vaccinated should take it. The downsides of both under-prescribing and over-prescribing are discussed. The need for more data is acknowledged.

In the end, I’m proud not just of the collaborative effort and the result, but of the process. While it wasn’t a randomized, placebo-controlled, blinded, prospective study—which would be the gold standard for a drug like this—it was the scientific process at its best.

Update:

I’ve received a number of emails and Twitter comments asking about Long Covid and Paxlovid. This study did not study that. Another study (not yet peer-reviewed) that I was an author on (but far less involved with) found a possible smaller association of benefit, but I am much less confident about that because this database is very vague in its “capture” of Long Covid. The Long Covid and Paxlovid story is completely unresolved. We should not prescribe it for Long Covid until we have data—because there are always possible harms that we did not anticipate. Higher-quality studies are underway, and I await those results.

The best data we have is that the drug metformin (low-cost diabetes drug), may reduce the incidence of Long Covid. We do not know if it treats it, once diagnosed. While the prevention study still needs to be confirmed (especially in vaccinated patients), it’s the only drug I would even consider prescribing to a patient who has Covid in an effort to prevent Long Covid. I’ve written about that here in the past.