Morgenstern, J. Paxlovid doesn’t work. Surprise?, First10EM, May 18, 2026. Available at:
https://doi.org/10.51684/FIRS.145957
Despite incredibly underwhelming evidence from the outset, a massive amount of money was spent (or wasted) on Paxlovid during the COVID pandemic. Some of the issues are endemic to medical science. When we allow companies to test their own products with billions of dollars on the line, the results are horrendously biased from the outset. Unfortunately, despite incredibly weak and biased data, Paxlovid (among many other COVID therapeutics) was pushed forward using the incredibly flawed logic that we should lower our scientific standards during a pandemic. Now, years later, when Pfizer has already made its billions, and no one cares or is prescribing Paxlovid any more, we finally get some high quality independent data, and the results are just what you would expect.
The question
Does nirmatrelvir–ritonavir (Paxlovid) reduce the incidence of hospitalization or death among vaccinated higher-risk participants with SARS-CoV-2 infection?
The paper
Butler CC, Pinto AD, Harris V, Holmes J, Rahman NM, Cureton L, Hayward G, Richards DB, Lowe DM, Standing JF, Breuer J, Hood K, Png ME, Petrou S, Dorward J, Patel MG, Thomas NPB, Evans P, Hart ND, Jani BD, Hosseini B, Murthy S, McBrien K, Condon A, McDonald EG, Daley P, Greiver M, da Costa BR, Selby P, Jüni P, Lee TC, Shi H, Detry MA, Saunders CT, Fitzgerald M, Berry NS, Saville BR, Khoo SH, Nguyen-Van-Tam JS, Hobbs FDR, Yu LM, Little P; PANORAMIC Trial and CanTreatCOVID Trial Collaborative Groups. Oral Nirmatrelvir-Ritonavir for Covid-19 in Higher-Risk Outpatients. N Engl J Med. 2026 Apr 23;394(16):1583-1594. doi: 10.1056/NEJMoa2502457. PMID: 42019019
The methods
This publication actually covers 2 separate trials:
PANORAMIC was a national, multicenter, open-label adaptive clinical trial based out of 65 general practice hubs in the United Kingdom.
CanTreatCOVID was a national, multicenter, primary care, open-label adaptive clinical trial based out of Canada.
Patients
Higher-risk adults (≥50 years of age or ≥18 years of age with relevant coexisting conditions) in the community who had had symptoms of SARS-CoV-2 infection for 5 days or less and had a positive polymerase-chain-reaction or rapid antigen SARS-CoV-2 test.
Intervention
Paxlovid (nirmatrelvir 300 mg plus ritonavir 100 mg) twice daily for 5 day, in addition to usual care.
Comparison
Usual care alone.
Outcome
The primary outcome was nonelective hospital admission for any cause or death from any cause within 28 days after randomization.
The results
PANORAMIC
They included 3516 patients in this part of the trial, although 25,779 were randomized to other therapies, and 126,421 were screened. The adaptive multi-therapy aspect of the trial makes it a lot harder to assess the potential of selection bias. The mean age was 55, 70% were female, median length of symptoms was 3 days, and 98% had 2 or more COVID vaccinations. 8 patients (0.5%) in the usual care group were prescribed paxlovid outside of the context of the trial.
There was no difference in the primary outcome of hospitalization or death, which occurred in 0.8% of the paxlovid group and 0.7% of the usual care group.
“Early sustained recovery”, defined as symptom improvement by 14 days, was better with paxlovid (33% vs 22%), although I wouldn’t consider 14 days early recovery.
Adverse events occurred in 96% of the paxlovid group! For some reason they don’t report a number in the usual care group, which seems odd. Serious adverse events were only about 1%.
CanTreatCOVID
721 patients were randomized out of 1997 screened. The mean age was 55, 66% were female, median length of symptoms was 2 days, and 98% had 2 or more COVID vaccines. 11 patients (3%) in the usual care group were prescribed paxlovid outside of the context of the trial.
There was no difference in the primary outcome of hospitalization or death, which occurred in 0.6% of the paxlovid group and 1.2% of the usual care group.
“Early sustained recovery”, defined as symptom improvement by 14 days, was again better with paxlovid (69% vs 53%).
Adverse events were much higher with paxlovid (36% vs 6%), although the trend was flipped for serious adverse events (1% vs 3%).
My thoughts
As I said at the outset, I am not surprised by the results. I think we rushed the use of Paxlovid in the first place, and have never been convinced there was real benefit. EPIC-HR was a statistically positive trial, but when you combined the low pretest probability that we were going to find the first antiviral effective in respiratory viral infections with the many methodology problems and the fact that trial has run entirely by Pfizer, I did not think the results were strong enough to warrant widespread use of the drug. (Hammond 2022) However, the 1.3% mortality benefit definitely seemed important enough to warrant follow-up, independent research. Despite being completed at the same time, Pfizer sat on the results of the negative EPIC-SR for a full 2 years, presumably to drive up sales before tempering enthusiasm for Paxlovid. (Hammond 2024) It is nice to see some independent data, but at this point I think it is too late: Pfizer ran away with their bags full of cash, and basically no one is prescribing Paxlovid anymore. (Although, much like Tamiflu, I still see a large number of patients presenting with clear medication side effects after being prescribed these drugs from walk-in clinics, so I guess there are still some believers out there.)
Both of these trials were negative for their primary outcome, and I believe that is the result that will be most consistent with reality, but these trials don’t definitively prove that Paxlovid provides zero benefit.
Both trials ended up significantly under powered. Both trials are smaller than their calculated sample size. The Canadian trial just stopped because of slow recruitment. The UK trial realized that the incidence of the primary outcome was significantly lower than expected, ran a new sample size calculation, and somehow decided that despite the fact that they were seeing far fewer events, they actually needed a much smaller study (decreasing from their original plan of 10,600 patients all the way to 2,876.) I don’t follow their logic, but it does mean that the trial is going to be underpowered for small but real differences between the groups. In an under powered trial, I suppose the 0.6% difference seen in the Canadian trial might be clinically important.
Of course, I don’t know exactly what that 0.6% means. When you use a composite outcome as your primary outcome, it is essential that you present the data on the components on the composite. For some reason, they don’t do that here. They don’t present the data individually for death or hospitalization, only the combined number, which seems like a big editorial failure.
There are also the positive secondary outcomes, which some people will talk about. Of course, positive secondary outcomes in trials with negative primary outcomes are at best hypothesis generating, and will turn out to be wrong way more often than they are right. Also, the outcomes are just weird. First of all, 14 days is not early recovery. Second of all, how were 80% of the usual care patients still symptomatic at 14 days in the UK trial? Even the 50% in the Canadian trial seems way too high to me. I don’t know exactly what this is measuring. However, when you factor in the adverse event rates, it doesn’t seem like there could be a net benefit in overall symptoms. More importantly, these were unblinded trials looking at self reported symptoms, so I think the outcome should be mostly ignored in favor of the more objective primary outcome.
Unlike a lot of the trials we saw during the pandemic, these trials used all cause hospitalization or death as their outcome. This came up time and again during the pandemic, because trials kept using the inherently biased “COVID-related hospitalization” rather than just “hospitalizatoin”,. It was nice to see better methodology here. That being said, the Canadian study only got this information based on patient self report, and so accuracy is questionable. (If a patient was held overnight in the emergency department for a next day ultrasound completely unrelated to COVID, are they reporting that as a night spent in the hospital or not?)
People will probably make a big deal about changes in the population over time – the idea being that unvaccinated or previously unexposed patients were way higher risk, but that population doesn’t exist any more. Although that could be true, I think the much simpler explanation as to why Paxlovid doesn’t work in these two trials is that it never worked. My take on the initial research was that this drug probably did nothing, especially when you factored in the significant conflicts of interest. If all you care about is this single clinical decision, this distinction doesn’t matter. The answer is clear. No one should be using Paxlovid in 2026. However, I think it would be a mistake to whitewash the scientific mistakes made during the pandemic with retrospectively fit explanations based around shifting population risks. We do this over and over again in medicine. We rush to new therapies based on very low quality evidence, and then act shocked when high quality research demonstrates what was pretty obvious all along. (Think about therapeutic hypothermia, for example.)
Bottom line
These two open-label RCTs show no benefit from Paxlovid in preventing death or hospitalization when used in higher risk out patients with COVID-19. I see little to no reason for Paxlovid to be used in emergency department patients.
Other FOAMed
Paxlovid:
Paxlovid doesn’t work for long COVID either (The STOP-PASC trial)
EPIC-HR: Some underwhelming data on Paxlovid
EPIC-SR: The negative paxlovid data Pfizer has been sitting on
Paxlovid evidence: still very little reason to prescribe
EBM:
Evidence based medicine is easy
Evidence based medicine resources
References
Butler CC, Pinto AD, Harris V, Holmes J, Rahman NM, Cureton L, Hayward G, Richards DB, Lowe DM, Standing JF, Breuer J, Hood K, Png ME, Petrou S, Dorward J, Patel MG, Thomas NPB, Evans P, Hart ND, Jani BD, Hosseini B, Murthy S, McBrien K, Condon A, McDonald EG, Daley P, Greiver M, da Costa BR, Selby P, Jüni P, Lee TC, Shi H, Detry MA, Saunders CT, Fitzgerald M, Berry NS, Saville BR, Khoo SH, Nguyen-Van-Tam JS, Hobbs FDR, Yu LM, Little P; PANORAMIC Trial and CanTreatCOVID Trial Collaborative Groups. Oral Nirmatrelvir-Ritonavir for Covid-19 in Higher-Risk Outpatients. N Engl J Med. 2026 Apr 23;394(16):1583-1594. doi: 10.1056/NEJMoa2502457. PMID: 42019019
Hammond J, Leister-Tebbe H, Gardner A, Abreu P, Bao W, Wisemandle W, Baniecki M, Hendrick VM, Damle B, Simón-Campos A, Pypstra R, Rusnak JM; EPIC-HR Investigators. Oral Nirmatrelvir for High-Risk, Nonhospitalized Adults with Covid-19. N Engl J Med. 2022 Feb 16:NEJMoa2118542. doi: 10.1056/NEJMoa2118542. Epub ahead of print. PMID: 35172054
Hammond J, Fountaine RJ, Yunis C, Fleishaker D, Almas M, Bao W, Wisemandle W, Baniecki ML, Hendrick VM, Kalfov V, Simón-Campos JA, Pypstra R, Rusnak JM. Nirmatrelvir for Vaccinated or Unvaccinated Adult Outpatients with Covid-19. N Engl J Med. 2024 Apr 4;390(13):1186-1195. doi: 10.1056/NEJMoa2309003. PMID: 38598573
22 thoughts on “Paxlovid doesn’t work. Surprise?”
I remember people advocating for Paxlovid saying..” It’s like Tamiflu for COVID!” : /
Haha. Yeah. That was exactly my argument not to use it. Funny how the same words can mean dramatically different things to different people.
Sounds like you are coming around (several years ago you accused me of wearing a tin foil hat ☺). Keep fighting the good fight.
Someone once told me the more folks on the committee (school, church, office, government…..it doesn’t matter), the less you can count on their work. No idea if it’s true but worth a thought.
After the primary, and then the Delta, variants came and went, about the only patients I saw in the ER were unvaccinated. The vaccinated were either not getting ill, or not enough to be unable to wait to the next day to contact their PCP.
It looks like
Paxlovid prevented transmission as well as limiting illness. It is not clear that all the participantsin the studies cited were truly high risk. I think this report is biased against the med, although obviously we need to do better!
One should consider the fact that these studies were conducted well in the tail end of the pandemic when Covid-19 was rapidly evolving into a less pathogenic strain. This would account for the low incidence of hospitalizations (compared to early in the pandemic), and, as you point out, rendered both studies underpowered. Thus, it would be fair to say that Paxlovid is minimally effective against the milder strain circulating at the time of the study, but it does not mean that it was completely ineffective against more virulent strains. I agree there is little need for its use now, considering how Covid-19 is now a URI nuisance. But, I would certainly reach for it if a new virulent strain were to emerge.
That is a possible interpretation, but given the questions and especially significant conflict of interest with the original research, I think the more likely interpretation is that it never provided an appreciable net benefit. It is very similar to Tamiflu.
At this point, I would not prescribe it with a new virulent strain, but I would support an RCT at that point.
98% of the participants in both studies had at least 2 vaccines against Covid… of course Paxlovid isn’t going to provide benefit against hospitalization or death in such a population. The vaccines are highly effective against hospitalization and death, and most of the population has been exposed to the virus by now anyway. Your comparison of this drug to oseltamivir is ridiculous. Paxlovid reduced death to 0 in the initial study, in an unvaccinated population. That should be celebrated.
Influenza and Covid are different viruses, and anti-virals targeted at them should be expected to have different outcomes. Steroid increases mortality in respiratory failure caused by influenza, but reduces mortality in respiratory failure caused by Covid, as one example.
The broader point was we have never had a single antiviral effective against any respiratory virus, so our priors going into all the COVID trials should have been very low.
The initial Paxlovid trials showed promise, but had many issues, and were run entirely by the company that stood to make billions. They needed replication. There is absolutely no certainty of any benefit.
We now may never see benefit, because, as you say, vaccination has removed a large amount of the risk. But there is still a very good chance that all the potential benefit seen in the original trial was purely bias, not reflective of a true benefit.
Replying more generally to the vaccination claims made in this thread, and separately, I agree with Dr. Morgenstern that the Paxlovid discussion still suffers from uncertainty around external validity and replication:
COVID vaccines were effective against COVID hospitalization and death in older and high-risk populations, especially pre-Omicron and soon after dosing. However, the broader claim that “the vaccines are highly effective against hospitalization and death” is too broad and, as a general statement, misleading. The evidence does not support broadly applying it across age, baseline risk, prior infection, variant era, time since dose, or repeated boosting.
Importantly, all-cause mortality remained elevated, and in many places increased, during the vaccination era, which is difficult to reconcile with a broad claim of strong population-wide overall mortality protection. The evidence is strongest for COVID-specific severe outcomes in older/high-risk groups and weaker for blanket claims across all populations or overall mortality.
More specifically to this discussion, if 98% of participants were vaccinated and previously exposed, these studies may tell us more about Paxlovid in a low-risk, hybrid-immunity population than about its efficacy in earlier immunologically naïve populations. That limits external validity. At the same time, as Dr. Morgenstern notes, uncertainty cuts both ways: the absence of benefit in today’s lower-risk population does not prove lack of benefit in earlier higher-risk populations, but neither do earlier industry-run trials eliminate the need for replication and independent confirmation.
Your commentary sounds heavily biased. You gloss over the early days” when we did not have vaccines and people were dying. This gave us something to try. The studies seem to include too many low risk patients to give us a decent result. And having treated many long COVID patients, does this prevent it? There is a lot more nuance than ”It doesn’t work”.
Thanks for the comment. The mentality of “it gave us something to try” is exactly the kind of mentality I am arguing against, and I think was so dangerous during the early pandemic. Most medical interventions fail. Everything causes harm. We are searching for interventions in which the benefits outweigh the harms. The idea of rushing past the normal scientific process just to “have something to try” will almost certainly cause more harm than good.
Could Paxlovid provide some value, such as in long COVID? Of course, but based on the data we have, it is extremely unlikely. Researchers should continue to probe, but the appropriate clinical conclusion for now is that it shouldn’t be used.
So you admit that both trials were significantly underpowered, yet make a sweeping conclusion anyway.
Opinion discarded.
Thanks for the engagement
There is no mention about the efficacy of Paxlovid for reducing the occurrence of “long COVID”. Has that aspect been studied?
It has, and it does not work. That research has been covered elsewhere, and was not the focus of this study, so there isn’t much to say.
I thought we knew this already,!
How do we know the “adverse effects” were not related to the Covid strain the patient had? I took Paxlovid and it was amazing, I felt like I wasn’t sick once I started it. I did take it on day 1 of my infection and that I believe is a huge factor. Once the virus replicates, I don’t see a lot of benefit in its use, but if you catch it early, it seems to work great. I had no adverse issues. This article seems very biased against the medication, but I really wonder what strain participants had, was this early in the pandemic or later on when the strains were less virulent? Having almost died with Covid the first time I had it and being out of work for 5 months after it, with NO previous underlying issues, I would not hesitate to use it again if I needed it. I will say if it was not prescribed early enough, it did not appear to have any impact on symptoms from what I saw in patients.
There is a reason we value science over anecdote
Here’s a less passionate counterpoint to your thesis. It seems like you haven’t conceded much, and indeed with statements like “Paxlovid never worked” you clearly don’t understand RR reductions seen in the original trials. I will concede that hospitalization and death are much less common in 2026, but are still taking a toll greater/equal to flu, and we should not discount tens of thousands of deaths a year… nor trivialize reductions in viral load as a paradigm.
https://mccormickmd.substack.com/p/would-you-still-take-paxlovid-when
You are assuming the RR reductions are real. I think that is a very bad assumption, both because you should have had low priors entering the study, and because the study is heavily biased by financial conflict of interest. More likely than not that those point estimates were bias and not reality.