Why Acetaminophen Isn’t Autism’s Cigarette
Overlapping cohorts and statistical mirages prove something hard to prove: nothing
🍝 The Pasta Problem: How the Prada Review Boiled Science to Mush
I like to cook pasta. The good kind—bronze-cut, durum wheat, slightly rough so sauce clings like an affectionate cat.
But if you’ve ever made pasta, you know the single most dangerous mistake is overcooking. Al dente isn’t an affectation; it’s Italian for integrity. Once you pass the razor-thin threshold between tender and mush, every Italian grandmother feels it like a disturbance in the Force.
Now imagine a scientist walks into your kitchen claiming he can prove your pasta is perfect by measuring its internal viscosity and optical starch scattering. He doesn’t need to taste it—he has data.
That’s the Prada review—the paper that produced a White House briefing on acetaminophen and autism.
The setup: Timing, politics, and the illusion of discovery
The review dropped in August 2025, just days before the administration’s self-imposed “autism breakthrough” deadline. Within days, RFK Jr. stood at a podium flanked by Donald Trump and the heads of the FDA and NIH, citing the Prada paper as proof that prenatal Tylenol might cause autism.
Scary!
Except the Prada study wasn’t a study. It was a review of other studies—sort of. Not a systematic review that carefully extracts and reconciles evidence, but a collage: overlapping datasets, misinterpreted subgroups, and statistical déjà vu.
It examined ten studies. But only one found even a weak association. Yet the authors—including the dean of Harvard’s School of Public Health—declared the evidence “strong.”
Weird!
Exhibit A: The mirage of Ahlquist
The first case in this forensic file is a Swedish study by Ahlquist: 2.5 million births over 25 years—impressive on paper.
The problem? They missed almost everyone who actually took acetaminophen.
Only 7.5% of pregnant women reportedly used it—a number so absurdly low it had to be wrong. Other Swedish studies show 40–60%. Buried in a supplement, the authors admitted they counted only “regular” users, excluding “sporadic” ones. But that’s everyone who takes Tylenol like a normal human being. They take it for fevers or pains or headaches. Not every day like a blood pressure pill.
That’s not a small error; it’s a fatal one. When nine-tenths of exposures vanish, your “unexposed” group is full of exposed people.
Their result, a relative risk of 1.05 which disappeared when they compared siblings born to the same mother, is null and void. It cannot answer questions about acetaminophen exposure and autism.
Exhibit B: The meta-muddle
Next up, the Alemany meta-analysis—six European cohorts rolled into one. On paper, 74,000 mother–child pairs. In practice, five of six cohorts showed no association.
The sixth, a Danish dataset, was so large it mathematically drowned the others. Its tiny relative-risk bump (1.22) dragged the whole meta-analysis up to a meaningless 1.19.
And here’s where Prada’s team committed its first mortal sin: double dipping.
They cited the Alemany meta-analysis and three of the individual cohorts within it as separate studies—counting the same subjects twice. Spain’s INMA cohort appears under its original investigator and again in the meta-analysis. So do the British and Danish cohorts.
That’s not sloppy. That’s disqualifying.
The authors even wrote, with jaw-dropping confidence, that they “excluded duplicate datasets to avoid redundancy.” That’s like bragging you washed your hands before making the sandwich—then handing over the sponge.
Exhibit C: The Spanish split and the gender gimmick
The Avella-García study from Spain is the prototype for how wishful thinking becomes a “signal.”
In 2,000 mother–child pairs, the overall finding was clear: no association.
But when they split by sex, they found a tiny uptick in autism-like scores among boys and a mirror-image dip among girls—both statistically significant, opposite in direction.
Instead of concluding those random blips cancelled each other out, the authors—and Prada—claimed victory: acetaminophen causes autism in boys!
They ignored the equal and opposite protective effect in girls.
That’s not science. It’s delusion.
If acetaminophen truly increased autism risk, it wouldn’t reverse polarity when exposed to a Y chromosome. You might as well claim Tylenol changes its mechanism depending on whether you paint the nursery pink or blue.
Exhibit D: The cord-blood curiosity
Then there’s the Ji study—the only one with a biomarker instead of self-report. Umbilical cord sampling! Fancy.
Except every single sample tested positive for acetaminophen. Which means there was no control group. Because the drug’s half-life is only a few hours this necessarily means everyone got Tylenol (almost certainly during labor, when obstetricians routinely give it).
The authors then looked for a correlation between acetaminophen levels and autism and found a faint gradient. Prada called this “dose-response.” But dose-response only matters when you’ve shown a difference between exposed and unexposed groups.
If everyone’s exposed, that’s not dose-response—it’s astrology.
Exhibit E: The Danish domino
Finally, the one study that actually reported a positive association: Denmark’s 60,000-pair cohort. Relative risk: 1.22.
For comparison, smoking and lung cancer clock in at 30—a 3,000% increase. Tylenol and autism? Twenty-two percent.
And even that signal collapsed on inspection. The authors admitted the dose-response appeared “only with autism of the hyperkinetic subtype”—an outdated label for kids with both ADHD and autism traits. That’s not a subtype, and it’s not a link.
Prada’s review, however, elevated this throwaway line to “proof of dose-response.”
The bigger picture: Weak tea and confounding
Across all ten studies (or thirteen, minus the duplicates), exactly one showed a positive signal—and a weak one at that. The rest were null, negative, or methodologically broken.
Crucially, however, every known confounder tilts toward finding a false positive, because women who take acetaminophen are more likely to have fevers, infections, or psychiatric conditions (that’s why they take the acetaminophen), all independently linked to autism.
Yet most studies still found no association. That’s not absence of evidence. It’s evidence of absence.
Prada’s team didn’t just fail to prove a link; they assembled the strongest argument yet that no link exists.
Epilogue: Al dente reasoning
Overcooked pasta loses its shape because it’s been boiled too long. Prada et al. overcooked their data until all structure dissolved, then strained out the null findings and served up mush.
When nine of ten studies show no association—and the tenth shows a risk so tiny it wouldn’t move the needle on a home pregnancy test—the only honest conclusion is the opposite of the one they drew:
Acetaminophen doesn’t cause autism.
If this review came from a kitchen, I’d send it back.
Interesting question, thank you. That particular tidbit of advice is borne of concerns that pregnant women are immuno compromised in certain ways and for certain exposures. Listeria monocytogenes is a bacterium that perinatologists seem to feel pregnant women are more vulnerable to than others, and that has led to widespread recommendations to be careful about deli meat, which can be a carrier. I love your question because it’s not an issue that I have dived deeply into and maybe I will. So I can’t tell you a fact-based answer, though I wish I could, and I’ll let you know if I do. Thanks for the question.
Interesting…What do you make of telling pregnant women not to eat deli meat?
Neil