In a study claiming to assess “4-year mortality” and “support the safety” of vaccines, they deliberately excluded the first six months—the exact window when vaccine adverse events would occur.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2842305
A deep dive into why observational vaccine studies can produce impossible results—and what that tells us about the research itselfThe Headline
Earlier this month, JAMA Network Open published a study with a reassuring conclusion: COVID-19 mRNA vaccination showed “no increased risk of 4-year all-cause mortality” among 28 million French adults aged 18-59. The authors declared their findings “further support the safety of mRNA vaccines.”
Major media outlets will likely report this as definitive evidence that COVID vaccines are safe long-term. But buried in the study’s own data tables is a finding that should give any critical reader pause:
According to this study, COVID-19 vaccination reduced drowning deaths by 27%.
Let that sink in.
The Impossible Results
The French study, conducted by researchers at EPI-PHARE (a joint epidemiology group of French health agencies), compared mortality between 22.7 million vaccinated and 5.9 million unvaccinated individuals over approximately 4 years. They found vaccinated people had significantly lower mortality across virtually every cause of death.
Here’s a selection from their Table 2:
| Cause of Death | “Protection” from Vaccination |
|---|---|
| COVID-19 | 74% lower mortality |
| Cancer | 15% lower mortality |
| Heart disease | 24% lower mortality |
| Drowning | 27% lower mortality |
| Traffic accidents | 26% lower mortality |
| Falls | 20% lower mortality |
| Suicide | 12% lower mortality |
The first three might be arguable. COVID-19 protection is expected. Perhaps COVID infection damages hearts and somehow accelerates cancer (though this wouldn’t explain a 4-year effect in people who may never have been infected).
But drowning? Traffic accidents? Falls?
There is no biological mechanism by which an mRNA vaccine injected into your deltoid muscle can prevent you from drowning in a swimming pool, crashing your car, or falling down stairs.
These impossible results are not a minor footnote. They are a flashing red warning sign that tells us exactly what’s happening in this study.
The Study’s Central Flaw: Who Gets Vaccinated?
The fundamental problem with any observational vaccine study is simple: people who choose to get vaccinated are systematically different from people who don’t.
This is called “healthy vaccinee bias,” and it’s not subtle. Consider who was most likely to get vaccinated in France in 2021:
- People who trust medical institutions
- People who follow public health recommendations
- People who have regular contact with doctors
- People with stable employment (easier to schedule appointments)
- People with health insurance and transportation
- People who are health-conscious generally
- People who wear seatbelts, don’t drink and drive, fence their pools
- People who are not currently dying of something else
This last point is crucial. If you’re in the ICU, receiving hospice care, or actively dying of cancer, you’re probably not going to a vaccination center. The very act of getting vaccinated selects for people healthy enough to do so.
The French researchers knew about this problem. They attempted to address it by statistically adjusting for 41 different health conditions, socioeconomic status, and demographic factors. They used sophisticated “propensity score weighting” to make the groups comparable.
It wasn’t enough.
We know it wasn’t enough because vaccinated people still had 27% fewer drowning deaths. Unless Pfizer has discovered a secret anti-drowning property of lipid nanoparticles, this can only mean one thing: the statistical adjustments failed to eliminate confounding.
The Six-Month Memory Hole
But there’s an even more fundamental problem with this study, one that goes beyond healthy vaccinee bias.
The researchers excluded the first six months after vaccination from their analysis entirely.
Read that again. In a study claiming to assess “4-year mortality” and “support the safety” of vaccines, they deliberately excluded the first six months—the exact window when vaccine adverse events would occur.
Their justification was technical: they wanted to avoid “immortal time bias” (a statistical issue where the vaccinated group gets credit for time when they couldn’t have died because they hadn’t been vaccinated yet). This is a real methodological concern.
But their solution created an even worse problem: survivorship bias.
Here’s how it works:
- Person A gets vaccinated in June 2021
- Person A dies of myocarditis in July 2021 (within 6 months)
- Person A is excluded from the main mortality analysis
- The study concludes “no increased mortality” among vaccinated people
- But Person A is dead
The study only counts vaccinated people who survived long enough to be counted. It’s like measuring parachute safety by only surveying people who landed successfully.
To be fair, the researchers conducted a separate analysis of short-term mortality using a different statistical method (self-controlled case series). What did they find?
Vaccinated people had 29% lower mortality in the six months after vaccination—including 20% lower cancer deaths and 33% lower deaths from external causes (accidents).
This is biologically impossible. Vaccines don’t cure cancer in six months. Vaccines don’t prevent accidents.
The short-term analysis shows the same impossible protective effects as the long-term analysis, proving that healthy vaccinee bias pervades the entire study, not just the long-term follow-up.
The Data They Buried
The study contains another revealing pattern that received minimal attention: mortality differences by when people got vaccinated.
France implemented a “vaccine pass” (similar to vaccine mandates) on July 12, 2021, restricting access to restaurants, venues, and travel for unvaccinated people. This allows a natural experiment: comparing people who got vaccinated voluntarily (before July 12) versus those who got vaccinated under coercion (after July 12).
The results are striking:
| Period | Mortality Reduction (Vaccinated vs. Unvaccinated) |
|---|---|
| Before vaccine pass (voluntary) | 31% lower |
| After vaccine pass (coerced) | 7% lower |
When vaccination was voluntary, vaccinated people had dramatically lower mortality. When it became effectively mandatory, the difference nearly disappeared.
Why? Because mandatory vaccination diluted the healthy vaccinee effect. When only health-conscious people get vaccinated, vaccinated groups look healthier. When everyone has to get vaccinated regardless of their health consciousness, the groups become more similar.
This is powerful evidence that the mortality difference reflects who chooses vaccination, not what vaccination does.
The authors note this finding but don’t grapple with its implications: if the “protective effect” of vaccination drops by 77% (from 31% to 7%) when vaccination becomes mandatory, how much of the original effect was ever real?
What the Authors Admit (But Then Ignore)
To their credit, the researchers acknowledge some of these problems. In their Discussion section, they write:
“We also found that vaccinated individuals were more socioeconomically advantaged and likely benefited from better health care management, variables insufficiently captured in our data. These factors may partly explain the observed negative association between vaccination and mortality.”
This is a remarkable admission. They’re saying their statistical adjustments didn’t work—that unmeasured confounding “may partly explain” their findings.
But then look at their conclusion:
“In this national cohort study of 28 million individuals, the results found no increased risk of 4-year all-cause mortality in individuals aged 18 to 59 years vaccinated against COVID-19, further supporting the safety of the mRNA vaccines.”
How do you go from “confounding may partly explain our findings” to “our findings support vaccine safety”?
If confounding explains why vaccinated people have fewer drowning deaths, it might also explain why vaccinated people have fewer deaths from everything else. You can’t selectively invoke confounding for the results you don’t like while treating the rest as causal evidence.
The Cancer Problem
Let’s focus on cancer mortality specifically, because it’s particularly revealing.
The study found vaccinated people had 15% lower cancer mortality overall, with specific reductions of:
- 32% lower breast cancer mortality
- 15% lower lung cancer mortality
- 11% lower colorectal cancer mortality
These are extraordinary claims. A 32% reduction in breast cancer mortality would be one of the largest treatment effects in oncology. Actual cancer drugs that produce effects this large become blockbuster therapeutics.
But there’s no mechanism here. mRNA vaccines weren’t designed to fight cancer. The cancers killing people in 2023-2025 were already present or developing in 2021 when people were vaccinated. Cancer progression over 3-4 years simply cannot be influenced by a COVID vaccine.
So what explains this finding?
Screening and treatment access. People who get vaccinated are people who see doctors. People who see doctors get cancer screening. People with detected cancers get treatment. People with treated cancers live longer.
This isn’t about the vaccine. It’s about the kind of person who gets vaccinated.
What Would Proper Evidence Look Like?
If we wanted to actually know whether COVID vaccines affect long-term mortality, we would need:
1. Randomized controlled trials with long-term follow-up
The original Pfizer and Moderna trials were unblinded early and participants were offered vaccination, eliminating the control group. This was ethically defensible given the pandemic but scientifically costly. We gave up our ability to generate the strongest evidence.
2. Analysis starting from vaccination, not six months later
Any safety study must include the period when adverse events occur. Excluding this window makes safety assessment impossible.
3. Dose-response analysis
If vaccines protect against mortality, more doses should mean more protection. The French study lumps together people with 1, 2, and 3 doses, hiding any dose-response pattern.
4. Time-to-event curves by cause
We need to see day-by-day mortality by specific causes, not just aggregated statistics. When do deaths occur? What’s the pattern in the first days, weeks, and months?
5. Negative control outcomes that actually work
The researchers used trauma hospitalizations as “negative controls”—outcomes that shouldn’t be affected by vaccination. When vaccinated people had fewer traumas too, this should have been the headline, not a footnote. It proves the entire analysis is confounded.
Why This Matters
I want to be clear about what I’m not saying.
I’m not saying COVID vaccines are dangerous. I’m not saying they don’t prevent COVID deaths. I’m not saying people shouldn’t get vaccinated.
What I’m saying is that this study cannot tell us whether COVID vaccines are safe long-term, despite claiming to do exactly that.
The impossible findings (protection against drowning, traffic accidents, falls, cancer) prove that unmeasured confounding dominates the results. The six-month exclusion window removes the exact period when vaccine harms would be most likely to appear. The dramatic attenuation when vaccination became mandatory shows that selection effects, not vaccine effects, explain most of the mortality difference.
This matters because:
- People deserve honest information. If we don’t know something, we should say we don’t know. Claiming certainty we don’t have erodes trust.
- Methodologically flawed studies crowd out better ones. When JAMA publishes a study claiming to resolve the question, funders and researchers may feel less urgency to pursue rigorous investigation.
- Real safety signals could be hidden. If the healthy vaccinee effect creates a large “protective” association, it could mask smaller real harms. A true 5% increase in cardiac mortality, for example, would be invisible against a background of 25% “protection” from confounding.
- Science requires self-correction. Peer reviewers and editors should have caught these issues. The fact that impossible results (drowning prevention) are published without prominent discussion suggests quality control failures.
The Bottom Line
The French study tells us one thing reliably: in France, people who chose to get COVID vaccines were systematically healthier and more safety-conscious than people who didn’t.
It tells us almost nothing about whether the vaccines themselves affect long-term mortality.
That doesn’t mean vaccines are harmful. It doesn’t mean vaccines are safe. It means we still don’t know with the rigor we should demand for an intervention given to billions of people.
The next time you see a headline claiming to definitively prove vaccine safety or danger based on an observational study, ask yourself one question:
Did vaccinated people also have fewer drowning deaths?
If the study doesn’t report this, they’re hiding something. If they do report it and vaccinated people were “protected” from drowning, you know the study’s fundamental limitation. And if they report it and the rates were equal—well, you might actually have a well-designed study worth reading.
Until then, we’re drowning in confounded data and calling it science.
This analysis was conducted using INGA314.AI, a systematic approach to detecting logical errors, scope violations, and methodological biases in scientific research. The original study is: Semenzato L, et al. “COVID-19 mRNA Vaccination and 4-Year All-Cause Mortality Among Adults Aged 18 to 59 Years in France.” JAMA Netw Open. 2025;8(12):e2546822.
Technical Appendix: Key Numbers
For those who want the specifics:
| Finding | Data |
|---|---|
| Study population | 22.8 million vaccinated, 5.9 million unvaccinated |
| Overall mortality HR | 0.75 (95% CI: 0.75-0.76) |
| COVID-19 mortality HR | 0.26 (95% CI: 0.22-0.30) |
| Drowning mortality HR | 0.73 (95% CI: 0.57-0.99) |
| Traffic accident HR | 0.74 (95% CI: 0.67-0.83) |
| Cancer mortality HR | 0.85 (95% CI: 0.83-0.88) |
| Before vaccine pass HR | 0.69 (95% CI: 0.67-0.69) |
| After vaccine pass HR | 0.93 (95% CI: 0.91-0.95) |
| Six-month short-term mortality RI | 0.71 (95% CI: 0.69-0.73) |
| Short-term external causes RI | 0.67 (95% CI: 0.61-0.72) |
All of these show “protection” from vaccination. All of them are biologically implausible at these magnitudes for causes unrelated to COVID-19.
System D 