What percent of autism cases are attributable to vaccines?
We now have an answer
Editor’s note: Steve Kirsch and I independently arrived at the same estimate for the Population Attributable Fraction (PAF) of autism cases caused by vaccines. I previously missed his groundbreaking work on this topic (see 1, 2, and 3). That being said, this is an extremely important facet of the autism causation conversation so I want to shine a light on it and build the case piece by piece so that readers will understand the revolutionary implications of this finding.
Population Attributable Fraction: A Primer
I. What is it?
The Population Attributable Fraction (PAF) answers a specific and important question: if we could eliminate a particular exposure from the entire population, what fraction of all cases of a particular disease would disappear?
It bridges the gap between two different ways of thinking about causation:
Individual risk: In this case, “How much more likely is a vaccinated child to develop autism than an unvaccinated child?” (In prospective studies this is usually expressed as the risk ratio, RR, and in retrospective studies this is usually expressed as the odds ratio, OR.)
Population burden: “How much of the total autism in the population is attributable to vaccination?” (This is the PAF.)
The formula is:
PAF = Pe(RR − 1) / [Pe(RR − 1) + 1]
Where:
Pe = prevalence of exposure in the population (what fraction of the population is exposed)
RR = relative risk, also known as the risk ratio, (or one can use the odds ratio as an approximation if the incidence in the population is low)
II. The history of PAF
1953: Smoking and lung cancer
The intellectual foundation was laid by Morton Levin, an epidemiologist at Roswell Park Memorial Institute in Buffalo, in a 1953 paper in Acta Unio Internationalis Contra Cancrum. Levin was trying to answer a practical public health question: we know smokers get lung cancer at much higher rates, but how much of the total lung cancer burden in the population could we eliminate by ending smoking? He called his measure the “attributable risk percent” and derived the initial formula.
1960s: Refinement and naming
Brian MacMahon and colleagues at Harvard helped formalize the framework during the 1960s. The terminology was inconsistent for decades — in the literature one sees “attributable risk,” “attributable fraction,” “etiologic fraction,” and “population attributable risk percent” — all referring to essentially the same concept. This terminological chaos may have slowed its adoption but this was clearly a valuable concept for epidemiology and public health.
1974: Miettinen’s contribution
Olli Miettinen, a Finnish-American epidemiologist, published a highly influential paper refining the formula and clarifying the distinction between the “attributable fraction in the exposed” (how much of the risk in exposed individuals is due to the exposure) versus the “population attributable fraction” (how much of the disease in the whole population is due to the exposure). Attributable fraction in the exposed is used in toxic tort litigation today (see Schachtman, 2015).
1980s onward: Public health adoption
PAF became standard in chronic disease epidemiology — tobacco, obesity, alcohol, occupational exposures, air pollution, etc. The CDC and WHO began routinely publishing PAF estimates for major risk factors. The massive Global Burden of Disease (GBD) project at the Institute for Health Metrics and Evaluation (IHME), which began in 1990 and is now updated annually, uses PAF as its central tool for ranking risk factors by their contribution to the disease burden worldwide. When you see headlines that say “alcohol causes X% of all cancer deaths” — that’s PAF.
III. Who uses PAF today?
Epidemiologists use it to prioritize which risk factors to target for public health intervention. If Risk Factor A has a high OR but low exposure prevalence, and Risk Factor B has a lower OR but near-universal exposure, PAF reveals that B may be the more important target.
Health economists use it to estimate the cost savings from eliminating a risk factor — e.g., how many hospitalizations and how much spending disappears if smoking rates drop by 20%.
Regulatory agencies use it to justify interventions — the EPA, for instance, uses PAF-type calculations when setting air quality standards to estimate how many deaths would be prevented.
IV. We can calculate the PAF for vaccines and autism
Three key studies help us create a range of plausible estimates of the PAF for vaccines and autism.
Mawson et al. (2017) in a study of homeschool children in Florida, Louisiana, Mississippi, and Oregon found that vaccinated children had 4.2 times higher odds of an autism diagnosis compared to unvaccinated children (OR = 4.2; CI: 1.2, 14.5).
Mawson & Jacob (2025) using Florida Medicaid data found that children with 11 or more visits to the doctor that included vaccinations were 4.4 times more likely to have been diagnosed with autism than those with no visit for vaccination (95% CI: 2.85, 6.84).
Hooker & Miller (2021) in a study of three medical practices in the U.S. found that vaccination increases the odds of developing autism 5.03-fold (95% CI 1.64, 15.5).
So we can now plug these into our formula:
PAF = Pe(RR − 1) / [Pe(RR − 1) + 1]
Where:
Pe = prevalence of exposure in the population (what fraction of the population is exposed)
RR = relative risk (I’m using odds ratio as an approximation which is justifiable in this case because autism prevalence is still considered somewhat “rare” — in the single digits)
We’ll use a Pe of 0.97 to reflect the fact that 97% of children are vaccinated.
And then we’ll calculate the PAF for each of the three key studies:
Mawson et al. 2017 — OR = 4.2, Pe = 0.97
PAF = 0.97(3.2) / (0.97 × 3.2 + 1) = 3.104 / (3.104 + 1) = 3.104 / 4.104 = 75.6%
Mawson & Jacob 2025 — OR = 4.4, Pe = 0.97
PAF = 0.97(3.4) / (0.97 × 3.4 + 1) = 3.298 / (3.298 + 1) = 3.298 / 4.298 = 76.7%
Hooker & Miller 2021 — OR = 5.03, Pe = 0.97
PAF = 0.97(4.03) / (0.97 × 4.03 + 1) = 3.909 / (3.909 + 1) = 3.909 / 4.909 = 79.6%
So that gives us our range — 75.6% to 79.6%. What this means is that, if the OR is real and causal, approximately 75.6% to 79.6% of current autism cases would not exist if no children were vaccinated. That is an extraordinarily large number.
V. Qualifications and limitations
1. A modest OR combined with near-universal exposure produces a huge PAF. As I showed in my last article, even a 4-5x individual OR becomes a massive epidemic in society when 97% of children are exposed, which is what is happening with vaccines and autism. One could choose a lower Pe (say, the percentage of children who are “fully up to date” on vaccines according to the CDC schedule) and that would bring down the PAF estimates somewhat.
2. A key limitation — PAF assumes the OR (or RR) is causal, not merely associational. If the association between vaccination and autism is confounded — for instance, if children who aren’t vaccinated are systematically different from children who are vaccinated in ways that independently affect autism risk — then the PAF calculation would overstate the true causal burden.
We can have that conversation, however, I don’t think it will change the results. The best available vaccinated vs. unvaccinated studies (Mawson et al. 2017, Hooker & Miller 2021, and Mawson & Jacob 2025) control for confounders and examine a wide range of study populations yet reach similar conclusions.
Furthermore, numerous independent scholars (e.g., Christopher Exley, Lucija Tomljenovic, Christopher Shaw, Romain Gherardi, Yehuda Shoenfeld, and Peter McCullough) have worked out plausible pathways by which vaccines can cause autism and several studies have been published that satisfy the Bradford Hill criteria for causation (see Tomljenovic & Shaw, 2011; Bjelogrlić, 2025; and Hooker et al. 2026).
3. The confidence intervals in Mawson et al. 2017 and Hooker & Miller 2021 are large. If one calculated a PAF using the lower bound of their 95% confidence intervals one would have a much lower PAF (roughly 16%). But the sample size and number of autism cases in Mawson & Jacob 2025 are large (resulting in a narrower confidence interval) and the results are consistent across the three studies which lends credibility to these estimates.
4. There is a way to convert OR to RR using a formula from Zhang and Yu (1998). It’s laborious and introduces other uncertainties in the process (namely whether the unexposed group is truly unexposed in previous studies). I ran the numbers anyway and this only reduces my upper-bound PAF estimate by two points, not enough to significantly change my conclusions.
VI. Putting this in context
Smoking and lung cancer. In the Norwegian Women and Cancer Study the PAF of lung cancer deaths attributed to smoking was 85.5% (Hansen et al. 2020). A 2011 meta-analysis found a PAF of lung cancer cases caused by smoking of 93.1% in men and 82.6% in women (Pesch et al. 2011). As smoking has declined in the U.S., the fraction of lung cancer deaths attributable to smoking decreased from 81.4% to 74.7% (Shiels et al. 2023).
Asbestos and mesothelioma. Drawing on the Global Burden of Disease 2019 dataset across 204 countries, Chen et al. (2024) found that 91.7% of mesothelioma deaths were attributable to occupational asbestos exposure. Only about 4% of lung cancer cases in U.S. men are attributable to asbestos exposure (McCormack et al. 2012).
Radon and lung cancer. Radon is a naturally occurring radioactive gas that seeps up from soil and rock into buildings. It is the second leading cause of lung cancer after smoking and the leading cause among non-smokers. Gaskin et al. (2018) estimated that residential radon causes approximately 16.5% of global lung cancer deaths (that was the median PAF across the 66 countries surveyed).
So the PAF for vaccines and autism cases is about the same as the PAF for smoking and lung cancer deaths, somewhat less than the PAF for asbestos and mesothelioma deaths, and significantly higher than the PAF for radon and lung cancer deaths.
Seen in this light, vaccines are one of the most harmful exposures in the world and pose a massive health burden on the population — particularly since the exposures are so widespread.
VII. Conclusion
Steve Kirsch was the first person to calculate a PAF for vaccines and autism but I was not aware of his work on this topic until after I calculated it myself. We both reached the same conclusions — Mr. Kirsch found a PAF of 75% to 80% and I found a PAF of 75.6% to 79.6%. This is an astonishingly high number and it shows that vaccines are one of the largest preventable harms to the health of the public.
If history is any guide, supporters of the status quo will launch waves of ad hominem attacks in response to these findings. But these results are difficult to dismiss. The best vaccinated vs. unvaccinated studies show that vaccines increase autism risk by a factor of 4.2 to 5.03. All I’ve done here is to plug the best available data into a widely used epidemiological equation to calculate the percentage of autism cases in the population caused by vaccines. You likely haven’t seen these calculations before because widespread regulatory and epistemic capture prevents mainstream scholars from asking questions they know might end their careers.
While the pharmaceutical industry tries to figure out how to protect its profits from the devastating and possibly criminal implications of these findings, for everyone else, this is extremely good news — we now know how to prevent up to 79.6% of autism cases — stop vaccinating.
Given these straightforward findings and the high quality of the underlying data I am calling on President Trump, Secretary Kennedy, NIH Director Bhattacharya, and FDA Commissioner Makary to immediately halt and permanently end the CDC Child and Adolescent Vaccine Schedule and the CDC Pregnancy Vaccine Recommendations.
I am also calling on all states to immediately halt and permanently end vaccine mandates for school or any form of employment.
And I am calling on all parents who currently use mainstream allopathic pediatricians to fire that provider and instead find a functional, integrative, or naturopathic medical doctor who doesn’t vaccinate in their practice.
Autism is just ONE harm caused by vaccines. The next task is to calculate the PAF for vaccines and a wide range of additional harms including ADHD, allergies, arthritis, asthma, autoimmune disorders, Alzheimer’s, childhood cancers, diabetes, ear infections, eczema, epilepsy/seizure disorders, Gulf War Syndrome, and hundreds of other chronic conditions.
Blessings to the warriors. 🙌
Prayers for everyone fighting to stop the iatrogenocide. 🙏
Huzzah for everyone building the parallel society our hearts know is possible. ✊
In the comments, please let me know what’s on your mind.
As always, I welcome any corrections.
This is super interesting. The Baddies want the number to be zero. Apparently, Pharma got so desperate last night that they even spent money on bots to go after me over on Facebook. Many people on our side want the number to be 99.9%. I think that 79.6% is likely the lower bound, the actual number could be higher. We will need much more sophisticated (multifactorial) models to figure this out. I wanted to publish this because it's the conventional way of calculating this figure (which the mainstream will never do) and one has to start somewhere. And in a sane world this would be more than enough to stop the vaccine program altogether.
From cigarettes to childhood vaccines. Profiting off the systematic poisoning of children. What a world we live in. Thank you, Toby, for this important work and for not giving up.