Dr. Bhattacharya recently refused to allow CDC to publishe the results of a study that showed current mRNA COVID vaccines remain highly effective against severe disease leading to hospitalization.
That arbitrary decison has been widely condemned. Recently Bhattacharya opened up about why he blocked publication of the study. He objects to the study design and called it “crap”. The type of study he objecys to is called a test negative design study.
What exacrly is a “test negative study” anyway and why is this study method used?
The main reason the test-negative design (TND) is used is that it is a relatively fast, practical, and inexpensive way to estimate real-world vaccine effectiveness after a vaccine is already in widespread use.
Public health agencies like the Centers for Disease Control and Prevention and the World Health Organization use it because randomized placebo-controlled trials often become difficult or impossible once a vaccine is recommended broadly.
Here are the main reasons it became so popular:
1. It partially controls for healthcare-seeking behavior
This is the central idea behind TND.
Instead of comparing:
vaccinated people in the general population
vs.unvaccinated people in the general population,
researchers compare:
people who all became sick enough to seek testing or care.
That helps reduce one major bias:
some people are simply more likely to go to doctors or get tested than others.
The design tries to make the two groups more comparable.
2. It is much faster than randomized trials
RCTs are:
expensive,
slow,
logistically difficult,
and sometimes ethically controversial once vaccines are recommended.
During COVID, officials wanted rapid estimates for:
Delta variant effectiveness,
Omicron escape,
booster durability,
hospitalization protection,
waning immunity.
TND studies can often be completed in weeks using:
hospital records,
pharmacy data,
testing databases.
3. It works reasonably well for influenza surveillance
TND was widely developed in influenza vaccine research long before COVID.
Influenza changes yearly, so researchers needed a recurring system to estimate vaccine effectiveness each season without running a new placebo trial every year.
Over time, TND became institutionalized in respiratory virus epidemiology.
COVID researchers largely adapted the influenza framework.
4. Severe outcomes are relatively uncommon
To measure protection against:
ICU admission,
death,
hospitalization,
you often need huge populations.
Observational designs like TND allow analysis of millions of people from real-world datasets.
5. Public health agencies needed ongoing monitoring
COVID evolved rapidly:
variants changed,
immunity waned,
boosters rolled out,
prior infection became widespread.
TND studies were one of the few scalable tools for continuous monitoring.
Why critics object
Critics argue that the assumptions behind TND may break down during politically and behaviorally polarized situations like COVID.
For example:
vaccinated people may test more frequently,
unvaccinated people may avoid healthcare,
prior infection may differ dramatically,
masking/social behavior differs,
vaccine mandates affect who gets tested.
If those differences are large, the groups may not truly be comparable.
That can distort effectiveness estimates.
This criticism is taken seriously in epidemiology journals. It is not fringe to say TNDs have vulnerabilities.
Why agencies still use them anyway
Because the alternatives also have major problems.
The choices are often:
imperfect observational data now,
vs.no real-time estimates at all.
Most epidemiologists view TND as:
imperfect,
bias-prone,
but still informative if carefully interpreted alongside:
RCTs,
cohort studies,
mechanistic immunology,
hospitalization trends,
and international comparisons.
So the reason TND is used is not because researchers think it is perfect.
It is because many believe it is the most practical large-scale surveillance tool available for respiratory vaccine effectiveness in the real world.
In a test-negative design study:
Cases = symptomatic people who test positive for the target infection.
Controls = symptomatic people who test negative.
If vaccination works, you would expect:
fewer vaccinated people among the positive cases,
and relatively more vaccinated people among the negative controls.
The vaccine effectiveness (VE) estimate comes from comparing those vaccination odds.
Conceptually:
If vaccination rates are similar in both groups → VE is near zero.
If cases are much less vaccinated than controls → VE appears high.
The usual formula is:
VE=1−ORVE = 1 - ORVE=1−OR
where OR is the odds ratio of vaccination among cases versus controls.
So:
OR = 1 → VE = 0%
OR = 0.5 → VE ≈ 50%
OR = 0.1 → VE ≈ 90%
For example:
GroupVaccinatedUnvaccinatedCOVID-positive cases20%80%Test-negative controls60%40%
Vaccination is much less common among cases than controls, so VE would calculate as strongly protective.
But this is exactly where critics focus their concerns.
The method assumes the groups differ mainly because of vaccine protection — not because of:
different testing behavior,
different exposure risks,
prior infection,
healthcare access,
occupational mandates,
political/cultural behavior differences,
or differential likelihood of seeking care.
If those factors systematically differ between vaccinated and unvaccinated populations, the VE estimate can be biased upward or downward.
That is why TND studies require careful adjustment and why they are generally considered weaker evidence than randomized trials.
Whether a new placebo-controlled COVID mRNA vaccine RCT is “unethical” now is genuinely debated — but the answer depends heavily on:
who is being studied,
what vaccine is being tested,
what outcomes are being measured,
and what standard of care already exists.
The blanket statement “placebo-controlled COVID vaccine trials are unethical” is not universally accepted anymore in 2026.
Early pandemic: strong ethical argument against placebo
In late 2020–2021, once the original mRNA vaccines showed clear protection against severe COVID in large randomized trials, many ethicists argued it was unethical to:
knowingly leave high-risk participants unvaccinated,
when an authorized vaccine already existed.
That is a standard clinical research principle:
you generally do not withhold an effective therapy from high-risk subjects if serious harm may result.
So many placebo arms were stopped early or crossover vaccination was offered.
Today is different
Now the ethical landscape is much more nuanced because:
population immunity is widespread,
prior infection is extremely common,
current variants are less virulent on average than early strains,
effectiveness against infection is modest and transient,
and the risk-benefit balance varies greatly by age and health status.
So the ethics depend on the specific population.
Examples where placebo RCTs might still be considered ethical
Healthy low-risk adults
A placebo-controlled booster trial in healthy young adults could plausibly be considered ethical because:
baseline severe COVID risk is relatively low,
many already have hybrid immunity,
uncertainty exists regarding incremental booster benefit.
Updated formulations
If studying:
a new variant-targeted vaccine,
a nasal vaccine,
or a next-generation platform,
a placebo arm may be ethically defensible if there is genuine uncertainty (“clinical equipoise”) about whether the new product adds meaningful benefit.
Low-risk outcomes
Trials studying:
symptom duration,
viral shedding,
mild illness,
have fewer ethical concerns than trials where severe disease prevention is the endpoint.
Situations where placebo may still be viewed as unethical
Many ethicists would still object to placebo-controlled trials in:
frail elderly populations,
severely immunocompromised patients,
nursing home residents,
or high-risk medical groups,
if an existing vaccine is believed to provide meaningful protection against hospitalization or death.
Why regulators often prefer non-inferiority trials now
Instead of placebo, newer vaccine studies commonly compare:
new vaccine vs old vaccine,
rather than:vaccine vs nothing.
These are called non-inferiority or active-comparator trials.
The logic:
you do not deny participants an existing intervention thought to help.
Critics of this approach
Some scientists — including people like Jay Bhattacharya and others skeptical of current evidence quality — argue that avoiding placebo arms can make it harder to:
accurately measure absolute benefit,
detect harms,
or determine whether repeated boosting still provides meaningful clinical benefit.
They argue that “ethical” concerns are sometimes overstated and may shield products from rigorous evaluation.
Others strongly disagree and believe withholding vaccination from vulnerable groups would violate research ethics.
Bottom line
A fair summary is:
In 2020–2021, placebo-controlled COVID vaccine RCTs had a strong ethical challenge once efficacy against severe disease was demonstrated.
In 2026, the ethical case against placebo is much weaker in some populations and stronger in others.
The ethics now depend heavily on the population studied and the specific scientific question being asked.
There is legitimate debate among scientists and ethicists about where the line should be drawn today.
No — calling the entire “test-negative design” (TND) approach “crap” is not the mainstream scientific view. But it is also true that TND studies have important limitations and can produce biased estimates if poorly designed or interpreted.
The fair scientific answer is:
Test-negative designs are widely accepted and commonly used in vaccine effectiveness research, especially for influenza and COVID vaccines.
They are considered useful observational tools, particularly when randomized placebo-controlled trials are impractical or unethical.
However, they are not as strong as well-conducted randomized controlled trials (RCTs) and can be vulnerable to several kinds of bias.
Jay Bhattacharya has been among the critics arguing that TND studies during COVID may have overstated vaccine effectiveness because of selection bias, healthcare-seeking behavior differences, prior immunity differences, and testing behavior. Those concerns are not invented; they are legitimate methodological debates. But saying the entire design is worthless is a stronger claim than most epidemiologists would support.
What is a test-negative design?
In a typical TND study:
People who seek care for respiratory symptoms are tested.
Those who test positive become “cases.”
Those who test negative become “controls.”
Researchers compare vaccination rates between the two groups.
The basic vaccine effectiveness calculation is:
VE=1−OR
where OR is the odds ratio of vaccination among cases vs. controls.
The appeal is that both groups were sick enough to seek testing, which partially controls for healthcare-seeking behavior.
Why many epidemiologists like TNDs
Supporters argue that TND studies:
are practical and fast,
use real-world data,
reduce certain biases compared with ordinary observational studies,
have worked reasonably well for influenza vaccine monitoring for years,
can track waning immunity and variant performance.
That is why CDC, WHO, and many academic groups have used them extensively.
The criticisms are real
Bhattacharya and others are correct that TNDs can be biased. The scientific literature openly discusses this. For example, statistical papers note risks from:
differential testing behavior,
prior infection,
healthcare utilization differences,
collider bias,
residual confounding,
reasons for testing.
One major issue during COVID:
vaccinated and unvaccinated people often behaved differently socially, medically, and politically. That can distort observational comparisons.
Another issue:
if vaccinated people are more likely to test frequently, seek care earlier, or differ in prior infection exposure, effectiveness estimates can shift substantially.
But the existence of bias does not make the method useless
That is the key point.
All observational epidemiology has limitations. The question is whether:
the biases are understood,
investigators try to adjust for them,
results are consistent across multiple methods and datasets.
For COVID vaccines specifically:
early RCTs already demonstrated substantial protection against symptomatic disease and severe illness,
later TND studies were mainly used to estimate durability, booster effects, and variant-specific performance.
So the entire evidence base did not rest solely on TND studies.
Where critics have a stronger argument
Critics probably have their strongest case when:
TND studies are used to estimate very small effects,
studies are highly politicized,
prior infection is poorly measured,
populations differ greatly behaviorally,
or observational estimates conflict with randomized data.
For example, late-pandemic estimates of repeated booster benefit became increasingly difficult to interpret because prior infection histories became extremely heterogeneous.
Bottom line
A balanced summary would be:
Test-negative design studies are not “crap.”
They are a recognized observational epidemiology method with strengths and important weaknesses.
Bhattacharya is correct that they can produce misleading results if biases are not carefully handled.
Most epidemiologists would still say TNDs provide valuable evidence when interpreted cautiously and combined with other study designs, rather than dismissing them outright.
