Peter Attia MD367 - Tylenol, pregnancy, and autism: What recent studies show and how to interpret the data
CHAPTERS
- 0:01 – 5:34
Why Frame Tylenol and Autism Critically: Groundwork and Premises
Peter Attia explains why the episode exists and sets out premises: autism rates have risen with no single cause, science should be apolitical, and humans aren't wired for scientific thinking. He introduces the Bradford Hill criteria as the framework for epidemiology and notes the high bar for medications in pregnancy.
- •Autism rates have risen dramatically and a single cause is unlikely
- •Science is meant to be apolitical, though autism has become political
- •Humans are not naturally wired for scientific thought
- •Bradford Hill criteria are nine principles for judging whether an association is likely causal
- •Physicians generally advise stopping most medications and supplements during pregnancy
- 5:34 – 14:24
FDA Pregnancy Categories and a Three-Step Method for Evaluating Evidence
Attia walks through the older FDA letter categories A, B, C, D and X, noting the newer PLLR replaced them, and places Tylenol in category B. He then lays out the three questions he applies to any exposure-outcome link: is there an association, is it likely causal, and how large is the effect.
- •Categories A through X and the share of drugs in each
- •Tylenol sits in category B; ibuprofen and NSAIDs move to D in the third trimester
- •Step one: confirm a statistical association exists
- •Step two: judge the probability of causality, since biology has no proofs
- •Step three: understand effect size; science means updating priors
- 14:24 – 15:41
What Is the Claim About Acetaminophen and Autism?
With co-host Nick Stenson joining, Attia states the claim that maternal acetaminophen use in pregnancy is associated with autism and describes the government's response of asking the FDA for warnings and label changes. He notes the FDA and scientific community agree causality has not been established.
- •Claim is an association between prenatal acetaminophen use and autism
- •Government asked the FDA to issue warnings and change labels
- •No authoritative source claims causation has been shown
- •Some argue causality is plausible or likely and urge avoidance or strong precaution
- 15:41 – 22:45
The Multiple Comparisons Problem and Spurious Correlations
Attia explains that autism has risen fivefold by CDC figures, which drives a search for culprits. He uses a coin-flip psychic example and the Spurious Correlations website to show how testing many variables yields false hits, and why such links are hard to disprove in epidemiology.
- •CDC prevalence rose from 6.7 to 32.2 per 1,000 children
- •Testing many variables guarantees some significant findings by chance
- •Coin-flip psychic example illustrates the p-value problem
- •Spurious Correlations site shows absurd correlations with plausible-sounding stories
- •Epidemiology can't disprove links the way repeated RCTs can
- 22:45 – 28:53
The August Review Paper and the Pooled Risk Ratios
Attia describes the systematic review in BMC Environmental Health that triggered the recent alarm and notes it was not a meta-analysis. He disputes its claim of consistent results and dose-response, and walks through a forest plot his team built showing a pooled 5% relative risk increase.
- •Review was published in late August and was not a meta-analysis
- •Two of six studies showed no significant association
- •Dose-response in the largest study vanished in the fully adjusted model
- •His team pooled the data, showing about a 5% relative risk increase
- •How to read unity lines and confidence intervals
- 28:53 – 36:30
Flaws in the Xie Study and Why the Pooled Estimate Matches the Swedish Study
Attia explains why the small Xie study looks like a smoking gun but is unreliable: single cord blood samples, mostly undetectable metabolites, and biased retention in a cohort with 11% autism. He notes the pooled estimate equals the Swedish study's because that study dwarfs the others combined.
- •Acetaminophen clears within 24 hours, so cord blood reflects only recent use
- •Seventy percent of samples had no detectable metabolites
- •Selection and retention bias concentrated autism cases in the Boston Birth Cohort
- •Xie study is very small relative to the others
- •The Swedish study had more than ten times the participants of all others combined
- 36:30 – 45:22
The Swedish and Japanese Sibling-Controlled Cohort Studies
Attia details the Swedish cohort of about 2.5 million children, where a 5% relative increase (0.09% absolute) disappeared in sibling analysis, and rebuts the review's description of the design. He then covers a criticism about low Swedish usage rates and a new Japanese cohort that reached the same result.
- •Swedish hazard ratio of 1.05 with a 0.09% absolute risk increase
- •Sibling comparison abolished the association, suggesting confounding
- •Review's double-discordance description would introduce collider bias
- •Critics note only about 7.5% of Swedish mothers used acetaminophen
- •Japanese cohort with about 40% exposure showed the same abolition in sibling analysis
- 45:22 – 48:36
Why Observational Data Can't Prove Causality and the Bradford Hill Criteria
Attia uses the ice cream and drowning example to show how unseen confounders undermine observational data. He says only randomized trials can show causation, which is not feasible here, and introduces the nine Bradford Hill criteria.
- •Ice cream and drowning are linked by the confounder of heat
- •Not all confounders can ever be identified
- •RCTs are the only route to near-certain causality but aren't possible here
- •Bradford Hill proposed nine criteria in 1965 for when RCTs are unavailable
- 48:36 – 55:16
Bradford Hill Criteria: Strength, Consistency, Specificity and Temporality
Attia applies the first criteria to acetaminophen and autism. Strength is weak at 1.05 versus a 1.5 threshold for pharmacoepidemiology, consistency is moderate, specificity is weak, and temporality holds but with disagreement on the critical exposure window.
- •Smoking and lung cancer has about a 10x effect versus 1.05 here
- •Red meat and diabetes associations are stronger yet likely confounded
- •Consistency is reasonable in general-population studies but fades with sibling control
- •Many other stronger autism risk factors exist, so the link is nonspecific
- •Baccarelli and Parker disagree on the critical exposure window
- 55:16 – 1:02:22
Bradford Hill Criteria: Gradient, Plausibility, Analogy, Experiment, Coherence, and Verdict
Attia covers the remaining criteria: inconsistent dose dependency, hand-wavy plausibility via prostaglandins, aspirin analogy arguing against, no experimental evidence, and inconsistent animal data. He summarizes the ratings and concludes the probability the association is causal is very low, and that it cannot explain the rise in autism.
- •Dose dependency is modest and inconsistent
- •Mechanism is poorly understood; prostaglandin pathway is plausible but speculative
- •Aspirin showed a 13% risk reduction in the Swedish sibling analysis, arguing against
- •Animal studies are inconsistent and sometimes use extreme doses
- •Probability of causality judged very low; essentially zero chance it drives the rise
- 1:02:22 – 1:08:31
Genetics and Heritability: The Dominant Driver of Autism Risk
Attia explains that genetics account for an estimated 80 to 90% of variability in autism risk, using twin studies and height and weight as examples. He argues that maternal genetic predisposition could confound the acetaminophen link, as temperature does for ice cream and drowning.
- •Heritability estimated at 80 to 90% for autism
- •Monozygotic versus dizygotic twins form a natural experiment
- •Height is about 80% heritable and body weight about 60%
- •Maternal genetic predisposition might drive both Tylenol use and child autism risk
- 1:08:31 – 1:19:20
Explaining the Rise in Diagnoses: Diagnostic Expansion, Awareness, Age and Environment
Attia says genetics can't explain the multi-fold rise in diagnoses and traces how DSM criteria broadened. He attributes 40 to 60% to expanded criteria and 20 to 30% to awareness, with parental age, maternal obesity, preterm birth and air pollution contributing, and acknowledges acetaminophen could be a small environmental factor.
- •Expanded diagnostic criteria account for 40 to 60% of the increase
- •Increased awareness accounts for 20 to 30%
- •Advancing parental age explains about 5 to 15%
- •Maternal obesity, preterm birth and PM2.5 air pollution have risen
- •Acetaminophen could play a role but any contribution would be very small
- 1:19:20 – 1:24:12
Medication Use in Pregnancy: Balancing Risks and Benefits Across FDA Categories
Asked what he'd advise pregnant women, Attia says the general rule is to stop medications but that maternal health matters too. He illustrates with ApoB-lowering drugs, thyroid hormone and GLP-1s, then revisits the FDA categories with example drugs.
- •Mother's health is part of a healthy in utero environment
- •Lipid-lowering medication can be paused in pregnancy, thyroid hormone should not
- •GLP-1 decisions require nuanced thinking; he says he lacks enough data to comment
- •Examples by category, including Tylenol and metformin in B and statins in X
- 1:24:12 – 1:27:00
Considerations for Using Tylenol in Pregnancy: Pain and Fever
Attia discusses weighing the risk of not taking Tylenol. Minor aches may be worth powering through, but debilitating pain and especially fever carry their own risks, and fever and maternal infection are themselves linked to birth defects and autism, possibly confounding the acetaminophen signal.
- •For minor discomfort, skipping Tylenol may be reasonable
- •Severe pain that incapacitates the mother can itself pose a risk to the child
- •Fever is linked to higher risk of birth defects and autism
- •Acetaminophen is the safest option since NSAIDs and opioids are category D
- •Infection may be the signal that acetaminophen use is picking up
- 1:27:00 – 1:30:56
Final Thoughts: Thinking Critically Instead of Seeking Soundbites
Attia closes by explaining why he took the long route rather than offering a soundbite, urging listeners to practice critical thinking for future health scares. He sums up that risks must be weighed in context: skip for minor aches, judicious use for bad pain, and lean toward acetaminophen for fever.
- •Soundbites don't teach people to think for themselves
- •Similar exposure-versus-disease questions will keep recurring
- •No single risk factor should be considered in isolation
- •Associations here are very small and often vanish with rigorous corrections
- •Use with physician oversight is reasonable for significant pain; the balance favors treating fever
