A large pooled analysis measured two limited urinary biomarker constructs, averaged the results across a heterogeneous population, and assessed parent-reported traits instead of autism. Newsweek enlarged that narrow null result into a verdict the study could not deliver.
By James Lyons-Weiler, PhD
Popular Rationalism
September 2, 2026
No one is saying that environmental exposures cause autism in everyone.
But Newsweek took a narrow epidemiological finding and expanded it until it appeared to answer a much larger question.
The study, published August 24, 2026, in JAMA Pediatrics, examined whether two urinary biomarker measures collected during pregnancy were associated with later scores on the Social Responsiveness Scale, or SRS. The investigators reported no statistically significant association after adjustment.
Newsweek converted that result into the proposition that the study posed a “significant challenge” to the idea that pesticides could contribute to autism. It went further, telling readers that “low-level pesticide exposures many people encounter through ordinary diets are unlikely to explain rising autism prevalence.”
The study did not examine rising autism prevalence. It did not comprehensively measure pesticide exposure. It did not use autism diagnosis as its outcome. It did not establish that vulnerable subgroups were unaffected.
Those are not peripheral limitations. They identify the precise epidemiological problem that has made so much autism research less informative than its headlines: a population-average result obtained from a narrow exposure measure is repeatedly presented as though it resolved every biologically plausible version of the question.
What the study measured
The study by Haleigh Cavalier and many, many, many colleagues included 3,339 mother-child pairs drawn from 20 sites in the National Institutes of Health Environmental influences on Child Health Outcomes program. Participants had been recruited between 1997 and 2019. Children were assessed between ages 2 and 19, with a mean assessment age of 6.4 years.
The exposure definition was remarkably limited:
Prenatal OP exposure using urinary concentrations of 3,5,6-trichloro-2-pyridinol (TCPy) and the sum of 6 dialkyl phosphate (ΣDAP) metabolites.
The outcome was a parent-reported SRS T score. It was not a clinical autism diagnosis.
The adjusted estimate for TCPy was a 0.13-point reduction in SRS score for each doubling of the urinary concentration, with a 95 percent confidence interval from a 0.57-point reduction to a 0.31-point increase. For ΣDAP, the adjusted estimate was a 0.42-point reduction per doubling, with a confidence interval from a 1.03-point reduction to a 0.19-point increase.
Higher SRS scores indicate more autism-related traits. The adjusted point estimates therefore remained in the apparently protective direction, although the confidence intervals included zero.
Under the study’s log-linear model, a tenfold increase in TCPy corresponds to an estimated SRS change of approximately −0.43 points, with a confidence interval from −1.89 to +1.03. A tenfold increase in ΣDAP corresponds to approximately −1.40 points, with a confidence interval from −3.42 to +0.63.
That is a reasonably precise null finding for the association the investigators estimated. It argues against a moderately large average linear relationship between these two urinary measures and SRS score in the pooled study population.
The authors were considerably more careful than Newsweek. Lead author Cavalier said the results did not rule out autism-related risk “at higher exposures or among more vulnerable groups.” The official ECHO summary called for better exposure measurements, examination of specific pregnancy stages, mixtures, high-exposure settings and genetic susceptibility.
Newsweek acknowledged some of those limitations and then wrote a conclusion that crossed them.
Large studies can erase effects concentrated in small groups
A large sample is valuable when the statistical question matches the biological question. Here, the reported primary estimate was a marginal effect across thousands of participants with different exposure patterns, geographic settings, enrollment periods, genetic backgrounds and baseline autism risks.
That estimate answers a population-average question. It does not determine whether a smaller susceptible group experienced a substantial effect.
Popular Rationalists already know this lesson well.
The arithmetic is simple. Suppose, for illustration, that 5 percent of children experienced a ten-point increase in SRS score following a particular exposure while the remaining 95 percent experienced no change. The pooled average would be one-half of one SRS point. A large study could estimate that small average very precisely while leaving the ten-point susceptible-group effect unidentified.
The larger sample would not be defective. The estimand would be nonresponsive to the subgroup question.
Sex-stratified analyses do little to solve this. Susceptibility could depend on PON1 genotype or enzyme activity, maternal and fetal metabolism, the developmental stage at exposure, coexposure to another chemical, exposure intensity, nutritional status, or a particular autism phenotype. A male-versus-female comparison cannot substitute for those interaction analyses.
The study pooled data from 20 sites and pregnancies spanning more than two decades. That increases geographic and demographic breadth, but it also increases heterogeneity. Exposure levels, pesticide formulations, residential and occupational circumstances, specimen timing, diet, diagnostic opportunity and SRS administration age may differ across sites and periods. A marginal estimate compresses those differences into one number.
This is the first central error in Newsweek’s interpretation: a pooled population average was treated as evidence against effects in biologically or environmentally defined subsets.
It is also a recurring problem in autism epidemiology. Autism is heterogeneous, yet studies frequently treat it as a single outcome and calculate one association across everyone. Environmental effects that require susceptibility, timing or a cofactor are diluted by design. The resulting average is then described as though it disproved the conditional effect.
Precision cannot rescue the wrong estimand.
Two urinary biomarker constructs cannot stand in for prenatal pesticide exposure
The study’s second major limitation concerns what counted as exposure.
TCPy is associated principally with chlorpyrifos and chlorpyrifos-methyl, but urinary TCPy can also reflect exposure to TCPy already present in food or the environment. It is not a clean measurement of the amount of parent chlorpyrifos that reached the mother, placenta or fetus.
ΣDAP combines six dialkyl phosphate metabolites. These are nonspecific metabolites shared by numerous organophosphate pesticides. The sum collapses parent compounds with different potencies, toxicological properties and exposure routes into one quantity. It cannot identify which parent pesticide produced the urinary result, and it excludes organophosphates that do not produce the measured DAP metabolites.
This is exposure substitution. A limited set of urinary breakdown products was labeled “prenatal organophosphate exposure,” and the public account allowed that label to expand into “pesticides.”
The substitution becomes more serious when timing is considered. Organophosphate metabolites are short-lived. They can vary sharply with recent diet, hydration, metabolism and the interval between exposure and urine collection. Newsweek reports that the women supplied one urine specimen during pregnancy. If that description is correct, a transient measurement was used to represent exposure across an entire pregnancy.
But only during pregnancy.
A reliability study in 120 pregnant women collected urine during three gestational periods and reported intraclass correlation coefficients from 0.14 to 0.38 for the six DAP metabolites. The ICC for total DAPs was 0.30. Those values indicate substantial within-person variability even with serial sampling.
Nondifferential exposure misclassification usually moves regression estimates toward the null. Increasing the sample size narrows the confidence interval around the association with the mismeasured exposure. It does not reconstruct the exposure that was missed.
This distinction matters. The ECHO confidence intervals may tightly bound the association between a spot urinary measure and SRS score. They do not provide an equally tight bound on the effect of cumulative exposure, peak exposure, exposure during a critical neurodevelopmental window, a particular parent pesticide, or fetal target-tissue dose.
The urinary DAP problem also complicates adjustment for diet. Consumption of fruits and vegetables can increase measured urinary metabolites while contributing beneficial nutrients and correlating with socioeconomic conditions associated with developmental outcomes. Preformed DAPs on food can raise the biomarker without establishing a proportionate internal dose of the active parent pesticide. The crude analyses in the ECHO study initially suggested inverse associations, which weakened after adjustment.
That pattern may reflect confounding by diet or socioeconomic advantage. It may also expose the weakness of the biomarker as a measure of the toxic exposure the study claims to represent. The accessible abstract does not provide enough information to reconstruct the complete covariate sequence or the causal assumptions behind it.
Autism-related traits are not the same endpoint as autism
The SRS is a legitimate quantitative instrument. A continuous scale can detect changes that would be missed by a binary diagnosis, and prior ECHO work found that short and full SRS versions produced similar estimates when used to study preterm birth.
It remains a parent-reported measure of social responsiveness and related behavior. Autism diagnosis requires a broader clinical assessment. Autism can include restricted and repetitive behaviors, sensory differences, language and cognitive features, regression, adaptive-function deficits and medical comorbidities that are not reducible to a single average social-trait score.
A chemical exposure might influence one autism phenotype more strongly than another. It could increase the probability of crossing a diagnostic threshold in a susceptible subset without producing a large shift across the full SRS distribution. It could affect cognition, attention or another developmental domain without appearing as a strong social-responsiveness signal.
Mothers might intentionally clean up her diet during pregnancy. Or maybe during her first, but not second, pregnancy.
The study itself recognized that organophosphates are established neurotoxicants and that previous work has associated prenatal exposure with cognitive and attention outcomes. A null SRS result cannot be transferred to every neurodevelopmental endpoint.
Newsweek made a second transfer when it connected this individual-level association study to rising autism prevalence. The investigators did not estimate autism incidence by birth year, relate changes in pesticide exposure to changes in autism prevalence, or model whether measured exposure could account for the secular trend. Recruitment dates extending from 1997 to 2019 do not turn the study into a time-trend analysis.
An average SRS association and an explanation for national prevalence are different propositions.
The prior human evidence remains mixed
The ECHO result deserves a place in the literature. It does not erase the conflicts already present there.
In the agricultural CHAMACOS cohort, prenatal urinary DAP concentrations were measured at approximately 13 and 26 weeks. A tenfold increase in DAP concentration was associated with a 2.7-point increase in parent-reported SRS T score at age 14, with a 95 percent confidence interval from 0.9 to 4.5 points. Residential proximity estimates in the same study were largely null.
The HOME study measured six DAP metabolites in two pregnancy urine specimens. Its adjusted estimate was a 1.2-point reduction in SRS score per tenfold increase in ΣDAP, with a confidence interval from a four-point reduction to a 1.6-point increase.
Generation R measured DAP metabolites at three pregnancy stages and found no association with autistic traits at age six. Its authors still warned that measurement error was inherent in the biomarkers and called for more urine samples.
The MARBLES study used repeated pregnancy measurements and gold-standard clinical instruments in 203 children at elevated familial risk. The overall result was null, while the estimate for one metabolite among girls suggested increased risk but remained imprecise.
In the French PELAGIE cohort, DAP measures were not associated with autistic traits, while detected chlorpyrifos or chlorpyrifos-oxon in maternal urine was associated with higher Childhood Autism Spectrum Test scores. That analysis was based on a small urine subset and requires replication.
These studies differ in population, exposure intensity, number and timing of specimens, pesticide markers, outcome instruments and susceptibility structure. Their findings are genuinely inconsistent. The ECHO study adds a large, relatively precise marginal null at generally low exposure levels. It does not make the conflicting studies disappear, and it does not show which design best captured biologically relevant exposure.
The epidemiological reason the literature misleads
The central failure is repeated substitution.
Specific urinary metabolites become “pesticides.” A short-lived specimen becomes “prenatal exposure.” A pooled marginal coefficient becomes a conclusion about vulnerable subsets. An SRS score becomes “autism.” An individual-level association becomes an explanation of national prevalence. Statistical nonsignificance becomes evidence that the proposed causal contribution is unlikely.
Each substitution broadens the claim while the underlying evidence remains narrow.
Newsweek’s article is useful because it makes the process unusually visible. The report accurately acknowledged that the study did not diagnose autism and could not exclude high exposure or genetic susceptibility. Yet its final interpretation treated those exclusions as though they had little bearing on the conclusion.
They have everything to do with the conclusion.
The responsible finding is direct: among the pooled ECHO participants, the measured concentrations of TCPy and six aggregated DAP metabolites were not associated with a higher average parent-reported SRS score after adjustment. That result weighs against a large average association with those measured biomarkers at the observed exposure levels.
Newsweek wanted the study to answer whether pesticides may contribute to autism and whether environmental exposures help explain the rise in autism prevalence. Answering those questions would require pesticide-specific and repeated exposure measurements, explicit modeling of critical windows and mixtures, clinical and phenotype-specific outcomes, and analyses designed around susceptible groups.
The epidemiological record cannot answer a question that the study design averaged away.
The correct study is: Can we predict, using family history, genetic risk factors, and environmental exposures determine is could autism in children, from preconception to age 8, be prevented by finding predictors of environmental and genetic factors that interact.
That is the state of the science. It is, as I advised Sec. of HHS Robert F. Kennedy, Jr., unethical to study environmental factors without genetics or to study genetic factors without environmental factors.
If, as I have proposed, autism is an acquired cellular detoxification deficiency syndrome, the total exposure to environmental toxicants should be studied from all sources: food, water, in-home exposures, and medicine.
Anything else is treading water.
Sources
Cavalier HM, Volk HE, Kivumbi A, et al. “Prenatal Organophosphate Exposure and Autism-Related Traits in Children.” JAMA Pediatrics. Published online August 24, 2026. doi:10.1001/jamapediatrics.2026.3785. PubMed record.
Environmental influences on Child Health Outcomes. “Prenatal Exposure to Organophosphate Pesticides Not Associated with Autism-Related Traits in Children in Large NIH Study.” August 24, 2026.
NYU Langone Health. “Researchers Find No Connection Between Pesticide Exposure and Social Difficulties Common in Autism.” August 24, 2026.
Spaan S, Pronk A, Koch HM, et al. “Reliability of Concentrations of Organophosphate Pesticide Metabolites in Serial Urine Specimens from Pregnancy in the Generation R Study.” Journal of Exposure Science & Environmental Epidemiology. 2015;25(3):286–294.
Sagiv SK, Harris MH, Gunier RB, et al. “Prenatal Organophosphate Pesticide Exposure and Traits Related to Autism Spectrum Disorders in a Population Living in Proximity to Agriculture.” Environmental Health Perspectives. 2018;126(4):047012.
Millenson ME, Braun JM, Calafat AM, et al. “Urinary Organophosphate Insecticide Metabolite Concentrations During Pregnancy and Children’s Interpersonal, Communication, Repetitive, and Stereotypic Behaviors at 8 Years of Age: The HOME Study.” Environmental Research. 2017;157:9–16.
van den Dries MA, Guxens M, Pronk A, et al. “Organophosphate Pesticide Metabolite Concentrations in Urine During Pregnancy and Offspring Attention-Deficit Hyperactivity Disorder and Autistic Traits.” Environment International. 2019;131:105002.
Philippat C, Barkoski J, Tancredi DJ, et al. “Prenatal Exposure to Organophosphate Pesticides and Risk of Autism Spectrum Disorders and Other Non-Typical Development at 3 Years in a High-Risk Cohort.” International Journal of Hygiene and Environmental Health. 2018;221(3):548–555.
Lizé M, Monfort C, Rouget F, et al. “Prenatal Exposure to Organophosphate Pesticides and Autism Spectrum Disorders in 11-Year-Old Children in the French PELAGIE Cohort.” Environmental Research. 2022;212:113348.
Patti MA, Ning X, Hosseini M, et al. “A Comparative Analysis of the Full and Short Versions of the Social Responsiveness Scale in Estimating an Established Autism Risk Factor Association in ECHO.” Journal of Autism and Developmental Disorders. 2025;55(6):2050–2058.
Thanks for this very fine scientific perspective. We've known for decades that autism entails defective detoxification pathways, so yes, the total spectrum of environmental toxicant exposures should be studied. I'd suggest adding a sentence at the end of your article to the effect that susceptible subgroups of children should be identified using a molecular epidemiology approach--e.g., via blood SNP panels, red-cell glutathione (and related GSH enzyme pathways), urinary organic acids, etc. This has been done in smaller studies, and it's of course more labor-intensive and expensive, but we'll obtain more robust outcomes and bring more awareness to the fact that these kids are the proverbial canaries in the coal mine with respect to today's toxic world.
"They have everything to do with the conclusion." The only sentence that the "average" American even has a chance of understanding.
Dr. Weiler is usually the smartest guy in the room. Those who have ruined our "environment", be it by injecting heavy metals via IM into infants or poisoning the ground, air and water around us, would never allow any other narrative to emerge, than the one they control.
So, I will dumb it down, perhaps that a 6th grader might understand. It has been the Phrma vaccine playbook for 40 years:
YOU WILL NEVER FIND WHAT YOU ARE NOT LOOKING FOR.
Just keep repeating the lie: "safe and effective."