A preprint published in 2025 claims Pfizer’s COVID vaccine kills more people than Moderna’s. Before you retweet it, let’s talk about why scientists and doctors with fancy credentials really, truly should know better.
There is a special kind of frustration that comes from watching people with impressive degrees — MDs, PhDs, MIT titles, Harvard pedigrees — produce work that most first-year epidemiology graduate students would flag in a methods critique. That frustration is what the preprint “Twelve-Month All-Cause Mortality after Initial COVID-19 Vaccination with Pfizer-BioNTech or mRNA-1273 among Adults Living in Florida” delivers in abundance. The authors are Retsef Levi, Fahad Mansuri, Melissa Jordan, and Florida Surgeon General Joseph Ladapo. Let’s go through exactly what they did wrong — and why they should have known better.
So, What Does the Paper Actually Say?
The study examined nearly 2 million Florida adults who received their first COVID-19 shot — either Pfizer (BNT162b2) or Moderna (mRNA-1273) — between December 2020 and August 2021. The researchers matched Pfizer recipients with Moderna recipients based on age, sex, race, ethnicity, where and when they were vaccinated, and the neighborhood they lived in. Then they checked who died within 12 months.
Their headline finding: Pfizer recipients had about 40% higher odds of dying from any cause compared to Moderna recipients (OR 1.407). That includes cardiovascular deaths, COVID deaths, and deaths from everything else. The authors concluded this finding has “implications for public health recommendations”.
Translation: Pfizer is killing people.
Reader, the data does not actually show that.
Problem #1: This Paper Has Not Been Peer Reviewed, and It Shows
The very first page of the document carries a disclaimer: “This preprint has not been certified by peer review and should not be used to guide clinical practice”. That’s a genuine warning that independent scientists have not yet checked whether the methods hold up. (Spoiler alert: The methods don’t hold up.)
Three of the four authors — Mansuri, Jordan, and Ladapo — work for the Florida Department of Health. This is the same department run by the same Ladapo, who, in 2023, was reported by Politico to have personally altered a state-funded study on COVID vaccine cardiac risk to make the results appear scarier than the original data. Documents obtained through a public records request showed that an earlier version of that study found “no significant risk,” and that “Dr. L’s Edits” later changed the language to say young men were at high risk. An internal complaint accused him of “scientific fraud.” He called the allegations “factually incorrect”.
The data underlying this new paper “is prohibited by statute from sharing in its raw form,” meaning outside researchers cannot check the numbers themselves. A Harvard-trained physician who has already been caught with his hands in the data jar probably shouldn’t be the only one with access to the data jar.
Problem #2: One of the Lead Authors Is a Supply Chain Professor
Retsef Levi, the study’s corresponding author and the person who gets the credit or blame for the findings, holds a PhD from MIT’s Sloan School of Management… in operations management. He is not a physician, not a public health researcher, not an epidemiologist. He teaches supply chain risk analysis.
None of that would be disqualifying if Levi came to the question with an open mind. But on social media, he has called COVID vaccines “the most failing product in the history of products” and, in 2023, publicly called for them to be pulled from the market. He was appointed by RFK Jr. to the CDC’s vaccine advisory committee (ACIP) and then handed the job of leading the COVID-19 vaccine review subgroup — despite having no formal training in vaccine science or epidemiology.
When the lead author has already announced his verdict before running the study, scientists call that confirmation bias. People with PhDs in research methods know that confirmation bias poisons results. Levi has a PhD in research methods. You do the math.
Problem #3: They Forgot to Match on the Most Important Thing
The authors matched their Pfizer and Moderna groups on seven criteria: age range, sex, race, ethnicity, vaccination site type, month of vaccination, and neighborhood census tract. That sounds thorough until you notice what is completely absent from that list: health conditions.
No diabetes. No heart disease. No obesity. No cancer. No record of whether someone was already seriously ill before they ever got a shot. The authors acknowledge this in the limitations section and then essentially shrug it off, citing a prior study suggesting that omitting health conditions “does not appear to increase the risk of bias”. The problem? That cited study was comparing vaccinated people to unvaccinated people. That is a very different question than comparing two vaccine types in a population where underlying health heavily shaped which vaccine someone was likely to receive in the first place.
In early 2021, Moderna doses were heavily routed through hospitals and county health departments, while Pfizer was dominant in pharmacies. People vaccinated at hospitals in January 2021 were not the same population as people vaccinated at Walgreens in August 2021, and no amount of census-tract matching fixes that. Any clinician or epidemiologist who has spent time in a hospital knows this immediately. These authors apparently did not find it worth pausing over.
Problem #4: Their “Sanity Check” Isn’t Very Sane
A standard practice in observational research is to use a negative control outcome, which is something that shouldn’t be affected by the exposure you’re studying, used to check whether hidden bias is warping your results. If your negative control shows a difference between the two groups, that’s a red flag that something other than the vaccine is explaining your findings.
The authors chose suicide and homicide as their negative control. Their logic: vaccine type shouldn’t affect whether someone gets shot or dies by suicide. And indeed, the odds ratio for the Pfizer group vs. the Moderna group on suicide/homicide was 1.218 and not statistically significant. The authors called this a clean bill of health for their study.
But look at that number again. An OR (odds ratio, or the ratio of odds of exposure in the cases versus controls) of 1.218 with a confidence interval from 0.936 to 1.585 is close to — but not at — 1.0. That’s the very definition of a borderline result, and it actually hints at residual confounding rather than ruling it out. Suicide and homicide rates in Florida during 2020–2021 were shaped by poverty, housing instability, mental health access, and geographic factors. Many of these factors overlap with the same social conditions that drove vaccine type selection. The fact that the negative control is whispering “something’s off,” and the authors heard “all clear,” is a problem.
Problem #5: When and Where People Got Vaccinated Matters Enormously
Between December 2020 and August 2021, Pfizer and Moderna vaccines were not distributed to identical groups of people at identical times, as mentioned above. The rollout shifted dramatically: in January 2021, about 15% of Florida vaccines were given at hospitals and 10% at pharmacies. By August 2021, hospitals accounted for less than 1% and pharmacies for about 80%.
People vaccinated at hospitals in January 2021 — during the early priority phases — were largely older, sicker, and more vulnerable than people who walked into a Walgreens in August. Moderna was concentrated in those early, institutional settings; Pfizer spread more broadly into pharmacies as the rollout scaled up. That means the Pfizer group, on average, included more of the younger and healthier people who got vaccinated later in the rollout, while the Moderna group skewed toward the higher-risk early recipients.
If that’s true, then Moderna recipients should actually look sicker at baseline. And that would predict more deaths in the Moderna group, not fewer. The fact that the results show the opposite suggests the confounding is more complicated than the authors’ matching captured. Matching on the month someone was vaccinated does not fully capture whether they were a priority-tier patient or a healthy adult who just finally got their turn.
Problem #6: The Sensitivity Analysis Is Circular
The authors ran an additional analysis to estimate how powerful a hidden confounder would have to be in order to wipe out their findings. Their answer: an unmeasured variable would need to be associated with both vaccine type (OR 1.6) and mortality (OR 5.2) to make the results disappear. They took this as proof that no plausible confounder could explain away their results.
Here’s the catch: serious cardiovascular disease, cancer, diabetes with complications, and advanced frailty regularly produce mortality odds ratios well above 5.0 in real-world data. The authors chose not to match on any of these conditions. So what they’ve actually demonstrated is: we didn’t control for the most important confounders, and then we showed those confounders would have to be strong to matter. They are, and which we know they are, because that’s what the clinical literature says.
That’s a circle drawn very carefully around a hole more than a “robustness check”.
Problem #7: The Conclusions Are Way Bigger Than the Evidence
Here is what this study’s methods could honestly support: Among matched Florida adults who received a first COVID-19 vaccine dose in 2020–2021, Pfizer recipients had higher observed mortality over 12 months, and we couldn’t fully rule out confounding.
Here is what the authors actually write: the findings “may have implications for public health recommendations and evaluation frameworks for vaccines” and are “consistent with cumulative evidence from prior literature showing worse outcomes” with Pfizer.
Then look at how they plan to share these still-unreviewed results: not through peer review, but through “press releases by the institutions represented by the authors and through patient organizations”. This is not how rigorous science gets communicated. This is how unreviewed claims skip the checking process and go directly to people’s social media feeds.
Problem #8: Check the Acknowledgments
At the very end of the paper, the authors thank Tracy Beth Høeg, MD, PhD, “for assistance with study method development and writing during the early stage of this research”. Høeg — a sports medicine doctor who has publicly questioned the U.S. childhood vaccination schedule, questioned CDC guidance, and, as of late 2025, was appointed acting head of the FDA’s Center for Drug Evaluation and Research — helped design the methods of this study. At the FDA, she has reportedly been working to change COVID vaccine labels to say the risks outweigh the benefits for young men, based in part on an analysis with no supporting data released publicly.
The methods of this study — the same methods we’ve just spent several hundred words critiquing — were shaped in part by someone whose stated goal is to restrict access to COVID vaccines. That is not a conspiracy theory. It is in the acknowledgments section. They wrote it themselves.
“We gots the receipts,” as the kids would say.
The Bottom Line
Joseph Ladapo attended Harvard Medical School. Retsef Levi holds a PhD from MIT. Tracy Beth Høeg holds both an MD and a PhD. These are real credentials, and they matter. But credentials are not a substitute for intellectual honesty, and intellectual honesty requires following the evidence wherever it leads. It’s not about constructing a study, on data only you control, with methods shaped by ideological allies, published without peer review, and then issuing press releases before anyone can check your work.
The conclusion this paper wants you to walk away with — that Pfizer’s COVID vaccine is meaningfully more deadly than Moderna’s — is not supported by the methods used to reach it. The matching missed the most important confounders. The negative control hinted at bias rather than ruling it out. The sensitivity analysis set a threshold that known confounders can easily clear. And the people who designed, ran, and interpreted this study have publicly stated, before and after, that they believe COVID vaccines are dangerous and should be restricted or removed.
People with doctorates are supposed to be able to spot the difference between a finding and a predetermined conclusion. Here, unfortunately, the line was crossed — and in a paper that could genuinely mislead people about a vaccine that helped prevent millions of deaths.
This paper is a preprint, available at doi: 10.1101/2025.04.25.25326460. The disclaimer on its first page tells you not to use it to guide clinical decisions. The authors put that disclaimer there themselves, further proving their rutabaga-like self-awareness.






