Environment and pregnancy: what holds up

Most things said to cause autism before birth, from a mother's diet to the air she breathes, have been studied many times. Almost all of the findings are associations: two things that occur together, not one causing the other. A third factor, called a confounder, can drive both. Age, income, health, and the reasons a pregnancy was treated all move together, and a study that does not account for them can report a link that is not there.

Random assignment is the design that balances confounders between groups whether or not the researchers measured them, and no one can randomly assign a pregnancy exposure. Researchers use two other designs instead. Large cohorts follow many people and can show an association. Family designs compare siblings who differ in whether they were exposed, which subtracts the factors the siblings share, such as their genes and the home they grew up in. A sibling comparison removes the shared factors but not the ones that differ between the siblings, and it cannot fix an exposure that was measured the same way for both. A finding that survives a family comparison is stronger than one that does not, and a finding that disappears in one is usually explained by the family rather than by the exposure.¹

Held to that standard, one exposure stands out. Gestational valproate has evidence pointing toward a causal effect, and it is the only exposure in this entry that does. Parental age shows a consistent but modest association. The rest are weak, inconsistent, or explained by shared family factors. Prenatal acetaminophen, one of the most studied, is covered in Contested causes, where family-controlled studies find no association.

Valproate is an antiseizure and mood drug. In a US study using two health care databases covering 2000 to 2020, children exposed to valproate during pregnancy were diagnosed with autism more often than unexposed children: 10.5% against 4.2% among the children of mothers with epilepsy, and 1.9% in the general population. After adjustment, valproate carried a hazard ratio of 2.67 (95% confidence interval 1.69 to 4.20). In that study valproate was the positive control, and topiramate, the drug under study, was null (0.96, 0.56 to 1.65), as was the negative control, lamotrigine (1.00, 0.69 to 1.46).²

A 2025 meta-analysis of eight cohort studies, pooling more than six million people, found gestational valproate associated with autism (adjusted hazard ratio 3.10, 2.24 to 4.28) and with ADHD (1.62, 1.30 to 2.01). The risk was larger at higher doses: above about one gram a day, the hazard ratio was 6.32 (3.12 to 12.80). In an analysis of siblings who differed in whether they were exposed, the hazard ratio was 6.42 (2.02 to 20.42).³

The pattern points to the drug. Two other antiseizure drugs tested in the same design did not show it, which rules out a factor shared by the whole family, and both a dose pattern and a sibling comparison point away from the family and toward the drug. It does not rule out the reasons different drugs are prescribed, which differ between them.

The US drug label treats valproate as a known risk. As of October 2026, the label for divalproex sodium warns that valproate can cause birth defects and lower IQ, and states that "multiple large, epidemiological studies suggest that exposure to valproate monotherapy in utero may be associated with an increased risk" of autism, with adjusted estimates across studies ranging from 1.6 to 6.4. The label lists "confounding by indication" and "residual confounding" among the studies' limitations.⁴

Antidepressant use in pregnancy has been linked to autism and ADHD in many cohort studies. A 2026 meta-analysis of 37 studies, covering about 649,000 exposed pregnancies, found a small overall association and a larger one for autism (risk ratio 1.69, 1.24 to 2.30), with wide disagreement between the studies. The review found the association weakened or disappeared once it accounted for a mother's mental disorder, family and genetic factors, and measurement error. Its negative control, a father's antidepressant use during pregnancy, was itself associated with autism (1.28) and ADHD, which points to shared family factors rather than to the drug acting in the womb. One signal survived the checks: two older antidepressants, amitriptyline and nortriptyline, kept an association with autism (2.02, 1.32 to 3.10), while the newer drugs did not. The review concluded that treatment should be continued for pregnant women with moderate to severe depression.⁵

The two family studies disagree. In a Swedish study of 1.58 million children, first-trimester antidepressant exposure was associated with autism at the population level (hazard ratio 2.02), but when siblings who differed in exposure were compared the association fell to 0.83 (0.62 to 1.13), which is no association.⁶ A Stockholm study of 254,610 people found an adjusted odds ratio of 1.45 (1.13 to 1.85) against mothers with a psychiatric history who did not take the drugs. Its sibling comparison pointed the same way, though the estimate was imprecise, and its negative control, fathers prescribed antidepressants during the pregnancy, showed no increase in autism (1.13, 0.68 to 1.88); its authors concluded that the association might not be only confounding.⁷ In that study the risk concerned autism without intellectual disability, and the absolute risk was small.

Both parents' age at the child's birth is associated with autism. In a study of 5.77 million children across five countries, fathers aged 50 or older had a relative risk of 1.66 (1.49 to 1.85) compared with fathers aged 20 to 29, and mothers aged 40 to 49 a relative risk of 1.15 (1.06 to 1.24) compared with mothers aged 20 to 29; very young mothers also had a raised risk. The effect was larger when the two parents' ages differed widely.⁸ A 2017 meta-analysis of 27 studies found the highest parental age category associated with autism (odds ratio 1.41 for mothers and 1.55 for fathers), and each ten-year increase in age with about a fifth higher risk.⁹ A 2024 meta-analysis of 41 articles reached similar estimates.¹⁰

The association is real but small, and its cause is not settled. The leading explanation is that older fathers pass on more new mutations in sperm. Direct sequencing of 78 Icelandic families found the number of new mutations rising by about two a year with the father's age, and, after accounting for random chance, the father's age explained nearly all of the remaining variation in the count.¹¹ That supports the mutation link, not the autism link. Paternal and maternal age move together, and the studies disagree on whether the paternal effect is independent of the maternal one: the five-country study adjusted for the other parent's age and both effects remained,⁸ while a single case-control study in Qatar found the paternal association losing significance once maternal age was taken into account, with maternal estimates larger than those in the large cohorts.¹²

Maternal diabetes during pregnancy is associated with a higher likelihood of autism in the child. Two umbrella reviews published in 2026 agree: gestational diabetes carries an odds ratio of about 1.29, preexisting diabetes a stronger and more consistent association, and the credibility of the evidence runs from suggestive to weak.^13,14 The same reviews find similar weak associations for obesity before pregnancy, excessive weight gain, high blood pressure and pre-eclampsia.¹³

Being born early or small is associated with autism more steadily than any of those conditions. Across 75 studies, autism was diagnosed in 2.8% of children born very preterm and 2.1% of those born preterm, against 0.6% of children born at term, with a similar gradient by birth weight and by size for gestational age.¹⁵ A meta-analysis of 52 studies put the odds of an autism diagnosis about 3.3 times higher in people born preterm than in the general population.¹⁶ Part of that gap is how autism is counted: in the same review, the pooled prevalence in preterm samples was 20% when studies used screening questionnaires and 6% when they used a diagnostic assessment, so some of the difference reflects the screening tools and the closer monitoring that preterm children receive.

These are associations, not causes. Preterm birth and autism can share the same upstream factors, such as an infection during pregnancy or family and genetic factors, and none of the studies here uses a sibling or family design that could separate them.

Infections during pregnancy have been linked to autism in many cohort studies. A meta-analysis of 36 studies found maternal infection or fever associated with autism (odds ratio 1.32, 1.20 to 1.46), with little variation by the type or site of infection, and some evidence that mothers of autistic children recalled infections differently.¹⁷ The 2026 pregnancy-complications review lists maternal autoimmune disease and asthma among the associated conditions, at the same weak level of evidence.¹³

The one study to test the association against family factors found it did not survive. In 549,967 children in Stockholm County, Sweden, an infection during pregnancy was associated with autism (hazard ratio 1.16, 1.09 to 1.23), but the same association appeared for an infection in the year before pregnancy, a control that should show nothing if the infection acted on the fetus. When siblings who differed in exposure were compared, the association fell to 0.94 (0.82 to 1.08), which is no association. The authors concluded that the autism association "does not appear to reflect a causal relationship" but is more likely explained by factors shared between family members, and that preventing infection is not expected to reduce autism.¹⁸

Air pollution during pregnancy is associated with autism in some studies and not others. A meta-analysis of 28 studies found an overall increase of 3.9% for all pollutants in its main model and 12.3% in its most favorable model, but no increase at all in its least favorable model. Each 5-microgram-per-cubic-meter increase in fine particles (PM2.5) raised the estimate in every model (global +7%, pessimistic +5%, optimistic +15%).¹⁹ A 2024 review of cohort studies found nitrogen dioxide associated with autism (risk ratio 1.20, 1.03 to 1.38) and phthalate metabolites among the pollutants linked to it.²⁰

Phthalates are plastic softeners that leach from vinyl, food packaging and some medical tubing. A 2020 systematic review of six phthalates found only slight or uncertain evidence for each phthalate and outcome pair it examined, including autism and social behavior, with the one exception of motor effects for one phthalate. The review warned that this should not be read as proof there is no effect, because exposure is hard to measure and mixtures and differences between the sexes may hide one.²¹ A 2026 study from the Barwon Infant Study in Australia examined one phthalate, DEHP, in 847 children. It found that a pattern of chemical tags on cord-blood DNA, across a network of 531 genes, explained part of the link between DEHP exposure and autism and ADHD symptoms at ages 2 and 4; the pattern held in independent blood and brain samples.²² The outcomes were symptoms on a questionnaire, not diagnoses, and the design cannot show that DEHP caused them.

Folic acid, taken before and during early pregnancy, is recommended to prevent neural tube defects, and some studies find it associated with a lower likelihood of autism. A 2025 umbrella review of 23 reviews found supplementation associated with a lower likelihood of autism (pooled odds ratio 0.66, 0.55 to 0.79), though it rated half the included reviews very low quality and the evidence overall of low certainty.²³ Individual studies report larger effects: in 45,300 Israeli children, mothers who took folic acid or multivitamins before pregnancy had a relative risk of 0.39 (0.30 to 0.50) for their child being autistic, and those who took them during pregnancy 0.27 (0.22 to 0.33).²⁴ A 2017 meta-analysis reached a smaller protective estimate (risk ratio 0.77, 0.64 to 0.93), with moderate disagreement between the studies it pooled.²⁵

The protective reading is not secure. Mothers who take supplements differ from those who do not in many ways a study may not capture, and the Israeli study's effect sizes are far larger than any pooled estimate, which suggests the comparison groups differ in more than the supplement. One study found the opposite pattern at high doses: in 1,257 mother-child pairs, both low and high supplementation were linked to more autism, and the highest blood folate levels were linked to 2.5 times the likelihood.²⁶

One hypothesis holds that some autism begins before birth when a mother's antibodies, made against her own tissues, cross the placenta and bind proteins in the developing brain. In a small study, 7 of 61 mothers of autistic children had antibodies that reacted to fetal brain proteins, against none of 102 comparison mothers, and reactivity to two protein bands went with a parent's report of regression in the child.²⁷ A follow-up identified seven target proteins and found the antibodies in 23% of mothers of autistic children and 1% of comparison mothers.²⁸ In a mouse model built around the same targets, offspring showed altered development and social behavior.²⁹

Later studies have repeated the finding beyond the first laboratory's clinic. One drew on blood collected from mothers before their children were born and found the antibody patterns in 10% of mothers of autistic children, 4% of mothers of children with intellectual disability, and 1% of comparison mothers.³⁰ A pilot at two clinics in other states found the patterns in 24% of mothers of autistic children, whose children scored higher on two measures of autism traits but did not differ in IQ or adaptive function.³¹ The first three studies come from one laboratory, which developed the test; the later studies were run elsewhere, though the test is still that laboratory's.

Related work finds antibodies against brain tissue in autistic children themselves. One small study found them in autistic children and also in their unaffected siblings, which weakens a clean autism-specific read.³² Others report brain-specific antibodies in the plasma of some autistic children,³³ and a postmortem study of 25 autistic and 30 comparison brains found more lymphocytes around blood vessels in about two thirds of the autistic brains, alongside a marker of astrocyte injury.³⁴

The direction of cause is unresolved. The antibodies may be a marker of an immune difference rather than its cause, and a mother's antibodies could be a response to a fetus that is already developing differently. That the patterns appear in blood drawn before the child was born argues that they come first, but it does not show they do harm. Two small studies of intravenous immunoglobulin, an antibody treatment, report improvements in some children, but neither was controlled, and neither is evidence of a treatment for lifelong autism.^35,36 The hypothesis applies to a subgroup, possibly with more severe presentations and, in one small study, parent-reported regression, not to autism in general or to autistic adults.

Autistic writers approach the question of causes with the history of blame in mind. The "refrigerator mother" theory, the idea that cold, unloving parents caused their children's autism, was discredited long ago, but it shaped a generation of families. Eric Garcia writes that the psychologist Bruno Bettelheim "smeared parents by saying they wished their autistic child did not exist," creating the stereotype of "refrigerator mothers."^c1 Steve Silberman, whose history of autism is a community source for context, records that Kanner's blame of parents "made his syndrome a source of shame and stigma for families worldwide while sending autism research off in the wrong direction for decades."^c2 Even after parents were cleared of causing autism, the pressure on mothers to fix it continued: Amy Pearson and Kieran Rose write of "a long tradition of parent (mother) shaming and blaming in the autism industry."^c3

For many autistic writers the deeper worry is where the question leads. Kayla Rodriguez writes that autism "isn't a bad 'disease' that needs to be 'cured' or 'prevented'."^c4 Steven Kapp writes that professionals and society "have unjustly blamed parents for causing autism."^c5 Luke Beardon states flatly that "autism is not caused by bad parenting."^c6 For these writers, a question about what causes autism turns quickly into a question about how to prevent it.