The simple version of this week’s story is that a political appointee killed a CDC study showing COVID vaccines work. The version that survives contact with the documents is stranger, and more interesting: the blocked study leaned on a statistical shortcut whose central weakness had already been spelled out, in print, by the same CDC scientist who went on to lead it.
To see why that matters, you need to know one thing about the method at the center of the fight. It is called the test-negative design, and it was built two decades ago to solve a practical problem in flu surveillance. You want to know whether this year’s shot is working, in real time, without waiting on a year-long randomized trial. So you take everyone who shows up at a hospital or urgent care with respiratory symptoms, test them, and compare the vaccination rates of the people who test positive against the people who test negative. If the vaccine is doing its job, the positives should be the less-vaccinated group. The clever part is that everyone in the sample was sick enough to seek care, which quietly cancels out a lot of the differences between people who get vaccines and people who don’t. It is fast, it is cheap, and for twenty years it has been the workhorse of vaccine-effectiveness surveillance.
It also has one structural weakness: it can only adjust for what it measures. And the thing it has never been able to see clearly is who already had the disease.
That is the unglamorous core of a controversy now being sold as a morality play. This week a study of the 2025-2026 COVID-19 vaccines appeared in JAMA Network Open after acting CDC director Jay Bhattacharya pulled it from the agency’s own Morbidity and Mortality Weekly Report this spring, where it had been slated for the March 19 issue. The headlines wrote themselves. A political appointee, installed by an administration skeptical of COVID vaccines, killed a study showing the shots work. TrialSite News called it “the study that wouldn’t die.” Medscape ran it as the study “the CDC doesn’t want you to see.” The shape of the coverage is censorship, and the villain is a Stanford economist who supposedly let politics run over science.
Give that narrative its full strength, because it is not nothing. The study, led by CDC epidemiologists Ryan Wiegand and Ruth Link-Gelles, had cleared the agency’s scientific review and been approved by MMWR’s editors before it was pulled. Its numbers are not extreme: the updated shots came out 50% effective (95% CI, 42–57) against emergency-department and urgent-care visits and 55% effective (95% CI, 41–66) against hospitalization, drawn from more than 85,000 encounters across 253 facilities in seven states between September and December of last year. Those are modest, believable figures, the kind of incremental result surveillance produces every season. And the test-negative design is not fringe statistics. A separate analysis in the same journal tested it against five randomized COVID-19 vaccine trials and found a concordance correlation of 0.86 between the design’s estimates and the trial results. When the assumptions hold, the method works. Natalie Dean, the Emory biostatistician who wrote the accompanying editorial, is right to call it an important and practical tool.
So where is the problem? It is in a single sentence in the study’s methods, and it is the sentence that makes Bhattacharya’s objection something other than vandalism. The authors state plainly that their analysis “did not account for previous SARS-CoV-2 infection or COVID-19 vaccination.” They adjusted for age, sex, race, calendar time, and geography. They did not adjust for whether a person had already had COVID, in a country where the overwhelming majority of people have now been infected at least once, many of them more than once.
Here is the detail the censorship framing leaves out. The lead author of the blocked study, Ryan Wiegand, is also the lead author of a 2024 paper in Nature Communications whose entire purpose was to quantify how badly the test-negative design can mislead you when you fail to control for prior infection. That simulation found that unadjusted estimates are biased, and that in some scenarios, late after vaccination or with fewer doses, they can fall below zero, producing the false appearance that a vaccine is harmful when it is not. The man who ran the surveillance study had already published the mathematical case for why its central simplification matters.
It cuts both ways, and the direction is worth being precise about. Wiegand’s own paper found the bias usually runs toward underestimating the vaccine, with a mean of about 1.4 percentage points and an effect that rarely exceeds 8 points, which argues against any claim that the 55% figure was inflated. But that was never Bhattacharya’s claim. His was narrower and harder to dismiss: the estimate “could be an overestimate or an underestimate; it’s impossible to tell.” That is not a denialist talking. It is a restatement of the caveat the study’s own lead author published a year earlier.
Martin Kulldorff, the Harvard-trained biostatistician now on the federal vaccine advisory committee, raised the companion objection: why lean on a fast, assumption-heavy snapshot when longer-term studies could answer the question more cleanly? These are not the complaints of people who don’t understand epidemiology. Bhattacharya and Kulldorff co-authored the Great Barrington Declaration, and you can disagree with them sharply on policy and still see that “this design has a documented blind spot” is a statistical observation, not a conspiracy theory.
Now add what the establishment coverage tends to mention last, if at all. The study was funded by the CDC through contracts to Westat and Kaiser, and several authors disclosed research grants from Moderna, Pfizer, and Novavax, the same companies whose product was being evaluated. None of that makes the 55% figure wrong. But it is exactly the kind of entanglement that, in any other field, would be the lede rather than a disclosure footnote, and it is fair to ask why a methodological dispute over a manufacturer-adjacent study gets covered as a free-speech emergency.
So what is actually true here? Both things, uncomfortably. The vaccines very likely do provide modest added protection against severe COVID, and the test-negative design is a legitimate tool that has earned its place. And Bhattacharya had a literature-backed reason to say a flagship CDC bulletin should not stamp a single uncontrolled estimate as settled fact, when the agency’s own statisticians have published on the ways that estimate can drift. You can think pulling the paper was heavy-handed and still see that the “science versus politics” frame is doing a lot of unearned work, because the cleaner story, the one with a hero and a censor, skips the part where the method’s limits were documented by the methodologists themselves.
What’s worth watching is whether the CDC now produces the analysis this whole fight implies it should: vaccine-effectiveness estimates that adjust for prior infection and prior vaccination, the variables the field has known for years are load-bearing, and then shows readers what survives. In a population whose immunity now comes mostly from having already been sick, that is the number that would actually tell us how well these shots work. The study that “wouldn’t die” got published. The harder question, the one about the method underneath it, is still waiting for its answer.
Sources
- JAMA Network Open – Wiegand, Link-Gelles et al., interim 2025-2026 COVID-19 vaccine effectiveness, test-negative design (2026)
- Nature Communications / PMC – Wiegand et al., bias and negative values from test-negative design without controlling for prior SARS-CoV-2 infection (2024)
- JAMA Network Open – evaluating the test-negative design against five randomized trials (concordance 0.86)
- Science – after pulling the vaccine study, Bhattacharya criticizes a long-running CDC publication
- NBC News – COVID vaccine study the acting CDC director blocked is published in an outside journal
- Washington Times / AP – COVID-19 vaccine study blocked from CDC journal is published elsewhere (2026)
- Medscape – The COVID Vaccine Study the CDC Doesn’t Want You to See
- TrialSite News – Science, Politics, and the COVID Vaccine Study That Wouldn’t Die