Children in an observational Medicaid cohort were more likely to receive an autism spectrum disorder diagnosis when their estimated PM2.5 exposure was higher across pregnancy, the researchers found. The association was expressed as a hazard ratio of 1.17, with a 95% confidence interval from 1.00 to 1.37. A hazard ratio is the study's comparison of diagnosis rates over time. The same analysis found a negative association with attention-deficit/hyperactivity disorder, or ADHD: a hazard ratio of 0.90, with a 95% confidence interval from 0.83 to 0.99. The findings describe associations, not proof that PM2.5 causes autism or protects against ADHD.
The cohort included 1,547,244 Medicaid-enrolled mother-child pairs whose conceptions occurred from 2001 through 2013. Researchers examined PM2.5 assigned at residential zip-code level over 12 weeks before conception, 37 weeks of pregnancy and the first three years after birth. After exclusions for missing data, the final analytic sample included 1,526,817 singleton births, or 99% of the sample.
Exposure estimates came from a 1 km2 ensemble model covering 2000 through 2016. When a mother's prenatal zip code changed, each differing zip code received 50% of the pregnancy exposure duration, and postnatal zip codes were updated annually. Outcomes were identified through validated claims algorithms using ICD-9 codes, age at diagnosis and medication records. The reported positive predictive values, a measure of how often an identified case is a true case, ranged from 82% to 98%.
Pregnancy did not show one simple pattern
To examine whether timing mattered, the researchers used distributed-lag models, which estimate associations for different exposure windows. A 10 μg/m3 increase in PM2.5 during gestational weeks 23 to 31 was associated with an ASD estimate of 1.09, with a 95% confidence interval from 1.02 to 1.17. ADHD estimates were 1.03, with a 95% confidence interval from 1.01 to 1.06, for weeks 16 to 22 and 0.90, with a 95% confidence interval from 0.85 to 0.94, for weeks 1 to 15. Multiple window testing and observational bias limit how confidently these timing patterns can be interpreted.
Looking at cumulative exposure before conception and across pregnancy, the study found no overall associations for the other neurodevelopmental disorders it examined. The prenatal ADHD estimate below 1 should not be read as evidence that PM2.5 was protective; the analysis was observational and does not establish a protective effect.
Signals after birth varied, too
After adjustment for prenatal and preconception exposure, cumulative postnatal PM2.5 exposure was associated with developmental coordination disorder, or DCD, at an estimate of 1.81, with a 95% confidence interval from 0.94 to 3.47. The interval was wide and included the null, and no association was detected for other cumulative postnatal neurodevelopmental disorders. Age-specific analyses suggested an ADHD estimate of 1.17 for exposure at age 0, with a 95% confidence interval from 0.94 to 1.46, and a DCD estimate of 1.72 for exposure at age 1, with a 95% confidence interval from 0.97 to 3.07. Estimates for intellectual disability at ages 1 and 2 were 1.91 and 1.76, respectively, and were described as suggestive.
The pattern varied by race or ethnicity. For prenatal exposure and ADHD, the reported estimates were 1.12 among Hispanic enrollees, 0.89 among Black enrollees and 0.79 among White enrollees; the interaction p-value was less than 0.001. Prenatal exposure was also associated with intellectual disability among Hispanic enrollees at 1.57, with a 95% confidence interval from 1.02 to 2.43, but not among White enrollees at 0.94, with a 95% confidence interval from 0.61 to 1.48. For postnatal exposure and ADHD, estimates were 1.10 among Black enrollees, 1.28 among Hispanic enrollees and 0.90 among White enrollees, with an interaction p-value of 0.001. The subgroup differences could reflect structural, diagnostic, exposure or follow-up differences, rather than biological differences alone.
Why the results need caution
Several checks produced similar patterns. Results were similar when the analysis used maximum lags of 26 or 30 weeks, restricted the outcome to diagnoses at age 3 or later, adjusted for NO2, applied censoring weights or used three- or four-degree-of-freedom specifications. The timing of the ASD windows was preserved. These checks do not remove residual confounding or uncertainty in how exposure was measured.
The study's design also set important limits. The main prenatal distributed-lag analysis required complete weekly exposure information through 37 gestational weeks, so the prenatal findings depend on that eligibility rule. PM2.5 was assigned by residential zip code, while outcomes came from claims algorithms rather than standardized clinical assessments. The evidence is therefore most directly about the Medicaid-enrolled population studied and does not establish individual personal exposure levels, absolute risks or cause and effect.
Paper data and sources
Original title: Preconception, Prenatal and Postnatal PM2.5 Exposure and Neurodevelopmental Disorder Risk among Medicaid Recipients
Authors: Shupler M, Qiu X, Huybrechts K et al.
Journal/Repository: Environmental science & technology
Status: Peer-reviewed
First online: 2026-10-19
DOI: Not available
Original paper · Full text