Paxlovid Trial Curiosities
Paxlovid trial results appear to demonstrate only half of participants were truly "at risk" of severe outcomes.
A few strange features of the Paxlovid trial design appear to unintentionally shed light on who is really at risk from the virus. In particular, almost no placebo participants who were already seropositive at recruitment went on to experience severe outcomes.
So, we’re giving Paxlovid to basically everyone, even though there appear to have been some recent “customer satisfaction” issues regarding whether some portion of recipients experience a rebound in viral replication and symptoms.1 This doesn’t matter. The White House revamped promotion of the drug on Monday, highlighting increased availability.2 Let’s get these pills into people’s tummies! After all, we’re in an emergency! (Paxlovid is not FDA-approved, only Emergency-Authorized.)
Since the theme of this post is “curiosities,” let’s pull over for a moment to behold the fine print on the FDA’s Paxlovid fact-sheet for healthcare providers. Among the conditions which precipitate Paxlovid’s Emergency-Use-Authorization is that, despite the Covid vaccines, the government asserts that:
Since the Pfizer/BioNTech Covid vaccine was approved by the FDA before the Paxlovid EUA (and Moderna immediately after), this statement implies the same organization does not consider either substance “adequate” to prevent Covid-19.
Could this assertion have certain legal implications? Or should the surgeon general be targeting this rogue outfit for misinformation? At the least the FDA should open a substack to share their opinion on the matter. I’m sure it would be very popular.3 Wait, scratch that; the FDA definitely considers both mRNA Covid vaccines adequate at preventing Covid-19.
Well, that’s a mess. In all likelihood the inclusion of “prevention” in the fact sheet is part of the generic description of the list of conditions for EUA, and is not automatically inclusive, or was only left in as an oversight. Who doesn’t make typos every now and then? It’s not like any lives are on the line or anything.
Moving on. Who will be receiving these revamped pills? The same FDA fact-sheet is vague. Patients already positive for the virus, and at “high risk” of Covid-19, but not yet hospitalized for Covid-19; the healthcare provider is further instructed to consult the hyperlinked CDC broadsheet to define risk. Qualifying conditions, which apply to anyone over 12 (even though the trial was only among adults 18 and over), turn out to include everything from being physically inactive to not using cocaine in an orderly manner:
In other words; being an American. Special “just go for it” bonus if you are a pregnant American, since this category was specifically excluded from the trial, even though, per the main EUA document, that same trial is the entire basis of the EUA; and since the “more likely to get very sick” claim is a patent lie.4 To put it succinctly, the fact that the FDA has EUA’d Paxlovid based on a trial with a given inclusion criteria, and yet outsources EUA-inclusion criteria to a separate organization (the CDC) that is not limited by the same standard, is fully bananas. The EUA is explicitly based on the trial.
The primary data supporting this EUA for Paxlovid are from EPIC-HR, a randomized, double-blind, placebo-controlled clinical trial studying Paxlovid for the treatment of non-hospitalized symptomatic adults with a laboratory confirmed diagnosis of SARS-CoV-2 infection. Patients were adults 18 years of age and older with a prespecified risk factor for progression to severe disease or were 60 years and older regardless of prespecified chronic medical conditions. All patients had not received a COVID-19 vaccine and had not been previously infected with COVID-19. The main outcome measured in the trial was the proportion of people who were hospitalized due to COVID-19 or died due to any cause during 28 days of follow-up.
“Prespecified,” “totally random,” what’s the difference? Risk is risk, right?
The Paxlovid trial inclusion criteria was also weirdly broad.
The Paxlovid Trial Inclusion Criteria was Weirdly Broad
The FDA’s deferral to the CDC for defining Paxlovid eligibility only compounds an original error, which was not scrutinizing whether the trial’s eligibility criteria was over-broad to begin with. Even though the trial report authors go out of their way to obscure the raw data, the signal for this over-broadness screams out from the charts in the Supplementary Appendix.
First, a review of those over-broad inclusion criteria. From the now-published EPIC-HR trial protocol:5
≥60 years of age;
Current smoker (cigarette smoking within the past 30 days) and history of at least 100 lifetime cigarettes;
Immunosuppressive disease (eg, bone marrow or organ transplantation or primary immune deficiencies) OR prolonged use of immune-weakening medications:
Has received corticosteroids equivalent to prednisone ≥20 mg daily for at least 14 consecutive days within 30 days prior to study entry.
Has received treatment with biologics (eg, infliximab, ustekinumab), immunomodulators (eg, methotrexate, 6MP, azathioprine) or cancer chemotherapy within 90 days prior to study entry.
HIV infection with CD4 cell count <200 mm3 and a viral load less than 400 copies/mL
Chronic lung disease (if asthma, requires daily prescribed therapy);
Known diagnosis of hypertension;
CVD [cardiovascular disease], defined as history of any of the following: myocardial infarction, stroke, TIA, HF, angina with prescribed nitroglycerin, CABG, PCI, carotid endarterectomy, and aortic bypass;
Type 1 or Type 2 diabetes mellitus;
CKD [chronic kidney disease] provided the participant does not meet Exclusion Criterion 5;
Sickle cell disease;
Neurodevelopmental disorders (eg, cerebral palsy, Down’s syndrome) or other conditions that confer medical complexity (eg, genetic or metabolic syndromes and severe congenital anomalies);
Active cancer, other than localized skin cancer, including those requiring treatment as long as the treatment is not among the prohibited medications that must be administered/continued during the trial period;
Medical-related technological dependence (eg, CPAP [not related to COVID-19])
So, it’s not “pregnancy and TV watching broad,” but it’s broad, and exceeds criteria used in the Monulpiravir trial (no in for smokers, neurodevelopmental disorders; and solid organ transplant and sickle cell disease were removed in Part 2 to reflect changing WHO guidance).6 Note that both trials excluded the Covid-vaccinated or previously confirmed infected.
This is counter-intuitive, given that a better selection of the truly at-risk would improve the apparent effect of the intervention. In fact that seems to have played out. Despite the impressive, “overwhelming” efficacy demonstrated by Paxlovid, the results could have been effectively doubled had certain groups been left out. But what defines those “groups” turns out to be surprising.
Now let’s get to the figures.
Early seroconversion and low viral load protects against severe outcomes, even in a specifically “high risk” population:
For the 528 individuals randomized to the placebo group who were already antibody positive, severe outcomes were in the single digits. Likewise for the 387 who were deemed “<4” viral load, meaning that the cycle count for their PCR test was over 19 or so.7 Obviously these two groups would overlap to some extent, if not entirely.
Methods always matter. Everyone is seronegative before infection; so in what way can being “seropositive” define low risk? In other words, how can the trial criteria be translated to a real-world risk evaluation?
Samples for serology were taken when the trial “intervened” on the participant, which was after a positive test (usually PCR) for the virus, and the onset of at least one of the normal symptoms (i.e., being “sick”), but before 5 days had elapsed since that onset. This is when subjects in “MITT1” were randomized to either Paxlovid or placebo, and so it is also when the “fate” for the placebo group was determined. Seropositivity was determined based on both IgG and IgM antibodies, and included both anti-spike and anti-nucleocapsid antibodies:8
Two assays were utilized for serology testing. The first assay is designed to detect […] (Ig) G, IgA, and IgM antibodies to the SARS-CoV-2 spike (S) protein receptor binding domain (RBD).[…] The second assay is designed to detect host IgG and IgM against the viral nucleocapsid protein (N).
Once again, the trial excluded individuals with Covid vaccination (as reported during the medical assessment at enrollment) or previous confirmed infection. This high rate of antibody positives (51.84%; (540+528)/(540+487+528+505)) should thus be driven by early immune response to the current infection.
For comparison, the Monulpiravir trial, which excluded the same groups, only found 21% of antibody-screened participants to test positive for N antibodies during randomization ((136+146)/(136+146+541+520)); the trial did not add serology testing until partway through). Partly for this reason, the effect doesn’t stand out as much, or could be confused for a failure to screen the previously infected. All the same, among that 21% or 282 (136+146) participants, behold, the same protective effect:
(In fact, Molnupiravir seems if anything to sabotage the protective effect, driving a higher rate of severe outcomes among the seropositive vs the placebo group.9)
What seems to account for the incredibly high rates of seropositivity in the Paxlovid trial is the inclusion of the IgM assay. However, confounders, such as geographic location, may be at play.
The Paxlovid trial thus appears to have just casually, inadvertently experimentally discovered is a powerful new rubric for accurately determining a low/high risk profile among patients with alleged comorbidities:
The EPIC-HR “Accidental Trial Intervention Assay”
Test PCR positive, get an IgM-inclusive antibody test within 5 days of symptom onset:
Antibody positive: low risk (1.5% severe outcomes)
Antibody negative: high risk (11.5%! severe outcomes)
Or, test PCR positive, get another PCR within 5 days of symptom onset:
Above ~19 cycle count (or equivalent viral load proxy for given brand): low risk (.8% severe outcomes)
Below same: high risk (8.1% severe outcomes)
There may be confounders here; but even if there are confounders, the Accidental Trial Intervention Assay (ATIA) would still presumably translate to real world determination of risk (i.e., by spotting patients with the confounders). It may also be the case that a 3-day cutoff, which was used for 65% of placebo subjects, is even more protective (i.e., placebo patients who were seropositive on Day 3 might not have even contributed to the 8 severe outcomes, whereas the PCR-based assay might work for both Day 3 and 510). Both are difficult to decipher without access to the raw data.
Now, perhaps you could say that 1.5% is still a “high” risk, justifying treatment with Paxlovid (on the assumption that this rebound issue doesn’t lead to long term net negative outcomes). That’s fine. But more granular access to the raw data in the trial might reveal even more useful applications of the Accidental Trial Intervention Assay. For example, it may help decipher when older patients are not at high risk, and when younger patients are, in a way that would far exceed the accuracy of the list of comorbidities that drove inclusion into the trial. Something like 900 placebo-randomized subjects were included for a reason other than being over 60; something like 46 of them had severe outcomes; the Accidental Trial Intervention Assay could possibly have flagged all of them in advance, leading to a doubling or more of the accuracy of the high risk designation.
Note that it’s not like researchers haven’t tried to improve diagnostic accuracy for severe outcomes in general; or that no one has thought to look into viral load as a assessment tool before.11 What this demonstrates is how many corners get cut in such study designs all the time. Even the second-hand, vendor-administered, and half-accidental serology screening experiment performed by the EPIC-HR trial exceeds the quality of most research into infection outcomes for SARS-CoV-2 which attempt to intercept patients before hospitalization. Were it otherwise, we could have had this data two years ago.
We could have a better, and frankly more intuitive understanding of why severe outcomes occur. I.e., that they occur when, and because, the immune response gets a late start;12 whereas by only looking at severe outcomes after they occur (including after death) we potentially conflate the later immune response to the out-of-control viral replication as having caused that replication to begin with.
Also note, that all of this is now almost useless since most of the world has either been injected or previously infected. Early seroconversion or high cycle count may no longer correspond to risk profile. The entire experiment would need to be repeated from scratch.
The EPIC-HR’s report of outcomes by comorbidity seems to fail to confirm the broad net of inclusion criteria used by the trial. For the most part, the samples for the more questionable discrete comorbidities were too small to yield any severe outcomes to begin with. For “smoking,” however, the sample was large enough to apparently refute that inclusion category:
Though not as powerful as the serology or PCR-based assay, “Yes” for smoking and “No” for hypertension are the only things that seem to confer a measure of protection to the placebo group. But here, even though the signal is weaker, the potential influence of confounders is even stronger. For one thing, there is no sub-sub-division by age, so there’s no way to tell if the universally-higher rates of outcomes in the “Yes” categories (except smoking and those censored for small sample sizes) is an affirmation of the inclusion criteria or just an artifact of the older subjects having many of the same comorbidities. Likewise for the higher outcomes in the “No” group for smoking.
What is clear is that both during the trial, and in real world use (including among the already Covid-vaccinated), Paxlovid is widely being distributed to groups in which it has not actually demonstrated a benefit. They are just “bonus” risk groups.
Just as with the Covid vaccines themselves, we need the raw data.13
If you derived value from this post, please drop a few coins in your fact-barista’s tip jar.
“As COVID cases rise, the U.S. is in a better place than before, Jha says.” (audio transcript) (2022, April 26.) npr.org
This is sarcasm. I have been arguing that there is strong evidence of severe efficacy since August (last substantially revisited in “Truth Bombs (As a Verb)”), and have been resoundingly vindicated on this front, to no discernible benefit to my subscriber count. However, I also have not been promoting boosters, unlike the government; so it is not surprising to see more contradiction in the official messaging here.
Hammond, J. et al. “Oral Nirmatrelvir for High-Risk, Nonhospitalized Adults with Covid-19.” N Engl J Med 2022; 386:1397-1408
Bernal, A. et al. “Molnupiravir for Oral Treatment of Covid-19 in Nonhospitalized Patients.” N Engl J Med 2022; 386:509-520
Per my reconstruction of the methods for determining viral load in “Unfinished Business, Pt 2.” However, it may be the case that a different PCR assay was used for randomization than for the data handed over for the FDA independent analysis. In either case there was a certain “non-lowness threshold” to the cycle count that conferred protection.
(Hammond, J. et al.) Supplemental Appendix, p. 10.
No surprise that sabotaging host gene expression would hinder an already-effective, underway immune response. For more, see “Doppelgänger.”
EPIC-HR modified the inclusion criteria midway, from patients captured within 3 days of symptom onset to those captured within 5 days; this standard (“MITT1”) is used for Fig 2.
Either of the 3 or 5 day post-onset standards could have broken the ATIA assay, but didn’t, which should mean the assay has been demonstrated to work for either time point (and so finding out if one cutoff drove higher rates of seropositivity or high PCR cycle counts, for example, is not critical, though it would be nice to know), though likely with different levels of sensitivity and selectivity at each.
Such as Brunet-Ratnasingham, E. et al. “Integrated immunovirological profiling validates plasma SARS-CoV-2 RNA as an early predictor of COVID-19 mortality.” SCIENCE ADVANCES 26 Nov 2021 Vol 7, Issue 48.
As has been speculated previously, including by Lin, C. et al., the authors of the “Prior coronavirus OAS, derp” paper. Although their results show that prior coronavirus antibodies in no way correlate to severe outcomes, their speculation that a late immune response of some cause is what leads to severe outcomes makes sense. See “Darmok and the Spike Protein at Tanagra.”
The “cutting room floor” of this post is a discussion of the trial report’s seemingly total obscuring of symptom persistence outcomes. Duration of symptoms is one of the secondary endpoints listed in the trial, and yet no reporting of this outcome seems to be provided. Thus, if the raw data of the trial contains evidence of symptomatic rebound for the Paxlovid-treated group, it is being hidden.