Key Takeaways
- A Kaiser Permanente Northern California cohort of 224,353 children found that associations between recorded maternal diagnoses and offspring autism were generally similar across maternal racial and ethnic groups.
- Four of 25 recorded maternal diagnoses, overweight or obesity, depressive episode, asthma, and other maternal disease complicating pregnancy, showed statistically significant interaction terms with maternal race and ethnicity after correction for multiple testing.
- Point estimates for these conditions were often higher among White mothers than in other racial and ethnic groups, but the pattern was not uniform across every comparison. The study did not determine why the estimates differed.
- Autism was diagnosed during follow-up in 2.4% of children overall, ranging from 2.0% among children of White mothers to 3.2% among children of Black mothers.
- Earlier studies have also associated maternal asthma, depression, and some forms or timings of diabetes with offspring autism, but these observational findings do not establish that the maternal conditions themselves cause autism.
- The new study does not show that race biologically changes autism susceptibility or that treating any maternal condition would alter a child’s likelihood of autism.
Introduction
Research involving more than 224,000 children examined whether associations between recorded maternal health diagnoses and offspring autism differed across racial and ethnic groups.
Recorded maternal diagnoses including obesity, diabetes, asthma, depression, and anxiety have been associated with offspring autism in observational research. What those associations mean, however, remains difficult to determine.
A diagnosis recorded before childbirth can reflect more than the underlying condition. Disease severity, treatment, genetics, environmental exposures, socioeconomic circumstances, access to health care, and diagnostic practices can all influence what researchers observe. Studies also use different exposure windows, which may not establish whether a condition was active throughout pregnancy.
A large study published October 5, 2026, in JAMA Network Open examined another layer of that problem: whether associations between recorded maternal diagnoses and offspring autism are similar across racial and ethnic groups.
The retrospective cohort study included 224,353 children born within Kaiser Permanente Northern California. Researchers found that most maternal diagnoses showed broadly similar patterns across groups. Four diagnoses, however, showed statistical evidence that the magnitude of their association with autism varied according to maternal race and ethnicity (Ben Messaoud et al., 2026).
The results should not be interpreted as evidence that race biologically determines how maternal illness affects autism likelihood. Instead, they raise questions about whether the same recorded diagnosis may capture different combinations of disease severity, treatment, environmental exposure, health care access, and other factors across populations.
What the Study Examined
Researchers studied children born within Kaiser Permanente Northern California from January 1, 2010, through December 31, 2017.
The final analysis included 224,353 children linked to maternal health records. Because siblings could be included, this should not be interpreted as 224,353 independent mothers. The statistical analyses accounted for sibling clustering.
Mothers were required to have been members of the health system for at least three months in each of the two years before delivery.
Maternal diagnoses were identified from inpatient and outpatient electronic health records using ICD-9 and ICD-10 diagnostic codes.
Importantly, the researchers were not simply measuring conditions diagnosed during pregnancy.
For nonchronic conditions, children were considered exposed when their mother had the diagnosis recorded within 12 months before childbirth. Chronic conditions were assessed over a 48-month period before delivery because those illnesses may persist even when they are not repeatedly documented.
A chronic diagnosis could therefore have been recorded several years before childbirth. The exposure definitions do not establish that every condition was active throughout pregnancy.
Researchers initially considered 42 maternal diagnoses. Twenty-five occurred frequently enough across the racial and ethnic groups to be included in the principal analyses. Diagnoses were not mutually exclusive, meaning a mother could have more than one condition.
The health system categorized maternal race and ethnicity as Asian and Pacific Islander, Black, Hispanic, White, or other or unknown.
Children were followed from birth until the earliest of an autism diagnosis after 12 months of age, disenrollment from the health system, death, or February 29, 2020.
Among children diagnosed with autism, 85% had been evaluated at a specialized Kaiser Permanente autism assessment center using standardized procedures that included the Autism Diagnostic Observation Schedule. The remaining diagnoses were made by qualified clinicians.
Researchers used Cox proportional hazards models and adjusted for maternal age at delivery, insurance type, child sex, birth year, and maternal health care utilization. The analyses also accounted for sibling clustering.
Insurance type served as a proxy for socioeconomic status, while maternal health care utilization helped account for differences in contact with the health system. Neither measure can fully capture socioeconomic circumstances or differences in access and quality of care.
Researchers then tested whether associations between each maternal diagnosis and autism differed according to maternal race and ethnicity. Because numerous comparisons were performed, they used a false discovery rate correction to reduce the likelihood of chance findings (Ben Messaoud et al., 2026).
What Researchers Found
During a median follow-up of 3.7 years, 5,448 children received a recorded autism diagnosis after age 12 months, representing 2.4% of the cohort.
Among children diagnosed with autism, the median age at diagnosis was 3.0 years.
The proportion receiving an autism diagnosis during follow-up differed according to maternal race and ethnicity:
- Black mothers: 3.2%
- Asian and Pacific Islander mothers: 2.9%
- Hispanic mothers: 2.5%
- Other or unknown race and ethnicity: 2.4%
- White mothers: 2.0%
These percentages describe diagnoses recorded during this study’s follow-up. They are not estimates of lifetime autism prevalence or evidence that maternal race caused differences in autism likelihood.
Race and ethnicity can correlate with environmental exposures, socioeconomic circumstances, health care access, screening practices, referral patterns, and other factors. Differences in autism identification can also affect observed diagnosis rates.
Four Diagnoses Showed Evidence of Different Association Strengths
The central result was that most recorded maternal diagnoses did not show statistically significant evidence that their associations with autism differed across racial and ethnic groups.
Four of the 25 diagnoses had significant interaction terms after correction for multiple comparisons:
- overweight or obesity
- depressive episode
- asthma
- other maternal disease complicating pregnancy
Point estimates for these conditions were often higher among White mothers than in other racial and ethnic groups, but the pattern was not uniform across every comparison. The study did not determine why the estimates differed.
A statistically significant interaction provides evidence that estimated associations vary across groups. It does not explain the reason for the difference, establish that the difference is clinically important, or identify race as its biological cause.
Confidence intervals for most race- and ethnicity-specific estimates also overlapped. Overweight or obesity was the exception specifically highlighted by the researchers, with a higher association estimate among White mothers than among Black or Asian and Pacific Islander mothers (Ben Messaoud et al., 2026).
Ten Diagnoses Were Associated With Autism in at Least One Subgroup
When researchers analyzed racial and ethnic groups separately, 10 of the 25 maternal diagnoses were statistically associated with autism in at least one group. In each case, the statistically significant subgroup was the White group.
Those diagnoses included overweight or obesity, depressive episode, recurrent major depressive disorder, other anxiety disorders, asthma, bipolar disorder, obsessive-compulsive disorder, diabetes, maternal care for other fetal problems, and other maternal disease complicating pregnancy.
This does not establish that those diagnoses are associated with autism only among White mothers.
Subgroup analyses contain fewer participants and events than analyses of the full cohort, which reduces statistical power. A statistically significant association in one group and a nonsignificant association in another does not automatically mean the groups differ.
That is why the formal interaction analyses are more informative for assessing heterogeneity. After correction for multiple testing, four diagnoses showed statistically significant interaction with maternal race and ethnicity (Ben Messaoud et al., 2026).
How It Compares With Earlier Research
The new findings build on observational research linking several maternal diagnoses with offspring autism. Earlier studies of asthma, depression, and diabetes are particularly informative because those diagnoses also appeared in the new analysis.
Maternal Asthma Has Previously Been Associated With Autism
A large Swedish population and family-based study included 1,579,263 children born from 1992 through 2007. Researchers identified 22,894 autism cases and compared them with 228,940 population controls as well as relatives with different degrees of genetic relatedness.
Maternal asthma was associated with higher odds of autism in offspring, with an odds ratio of 1.43. Paternal asthma showed a smaller association, with an odds ratio of 1.17 (Gong et al., 2019).
Researchers also used comparisons among relatives to investigate whether shared familial factors could explain the association. Estimates for maternal asthma remained elevated in several family-based comparisons, although confidence intervals became wider.
Prenatal exposure to asthma medications among children whose mothers had asthma was not associated with autism in that study.
That distinction matters for the 2026 research. Asthma was one of the four diagnoses whose association with autism showed evidence of variation across racial and ethnic groups, but the study did not establish that asthma caused autism or determine whether disease severity, environmental triggers, treatment, health care access, or other factors contributed to the subgroup patterns.
Maternal Depression Has Also Shown an Association
A UK study using the Clinical Practice Research Datalink examined 194,494 mother-child pairs and identified 2,154 children with autism.
In the study’s reported comparisons, the relative risk was 1.72 for children whose mothers had depression treated with antidepressants and 1.50 for children whose mothers had untreated depression. The detailed adjusted estimates were approximately 1.70 and 1.49, respectively (Hagberg et al., 2018).
Antidepressant use for indications other than depression was not associated with an elevated autism risk, with a reported relative risk of 0.73 and a 95% confidence interval of 0.41 to 1.29.
The researchers also conducted a sibling analysis and analyses related to depression severity. Their interpretation was more consistent with maternal depression and its severity contributing to the observed association than with antidepressant treatment itself.
Even so, this does not establish depression as a cause of autism. Depression can be associated with genetic liability, other psychiatric conditions, smoking, socioeconomic circumstances, stress, medication use, health care utilization, and other factors.
The 2026 study asks a different question. Rather than focusing only on whether depression is associated with autism overall, it found statistical evidence that the magnitude of the association for a recorded depressive episode varied across racial and ethnic groups.
Differences in recognition and diagnosis could matter. If depression is diagnosed differently across populations, the same diagnostic label may capture different distributions of severity, treatment, or health care contact.
Diabetes Research Shows Why Timing Matters
An earlier Kaiser Permanente Southern California cohort demonstrates why a broad diagnostic category can conceal important differences in timing.
The study followed 322,323 singleton children and examined maternal preexisting type 2 diabetes and gestational diabetes according to when gestational diabetes was diagnosed.
After adjustment for maternal age, parity, education, household income, race and ethnicity, comorbidity, and child sex, preexisting type 2 diabetes was not significantly associated with autism, with an adjusted hazard ratio of 1.21 and a 95% confidence interval of 0.97 to 1.52 (Xiang et al., 2015).
Gestational diabetes diagnosed at or before 26 weeks was associated with an adjusted hazard ratio of 1.42, with a 95% confidence interval of 1.15 to 1.74.
Gestational diabetes diagnosed after 26 weeks was not significantly associated with autism, with an adjusted hazard ratio of 0.98 and a 95% confidence interval of 0.84 to 1.15. Antidiabetic medication exposure was also not independently associated with autism.
These results illustrate why a broad label such as diabetes may conceal meaningful differences in timing, severity, metabolic exposure, and treatment.
In the 2026 study, diabetes was statistically associated with autism in the White subgroup but was not among the four diagnoses showing a significant interaction with maternal race and ethnicity after correction for multiple testing.
Together, the asthma, depression, and diabetes studies provide context rather than proof of causality. Associations between maternal diagnoses and offspring autism have appeared in different populations, but their magnitude can depend on how exposures are defined, when they occur, treatment, familial factors, measured confounders, and how autism is identified.
None of these studies demonstrates that a mother’s asthma, depression, diabetes, obesity, or racial identity directly causes autism.
Why the Same Recorded Diagnosis May Not Represent the Same Exposure
Electronic health records allow researchers to study hundreds of thousands of children, but diagnostic codes are imperfect representations of clinical and environmental circumstances.
Two patients with recorded asthma, for example, could differ substantially in disease severity, air pollution exposure, housing conditions, occupational exposures, medication use, access to specialists, and symptom control.
The same problem applies to depression, obesity, and diabetes.
Whether a condition is recognized and entered into a medical record can also depend partly on health care access and diagnostic practices. Treatment after diagnosis may differ as well.
The authors discussed several possible explanations for the patterns they observed, including diagnostic practices, environmental exposures, access to and continuity of medical care, treatment patterns, and differences in autism ascertainment.
The study did not directly demonstrate that any of these factors caused the differences in association estimates. They remain possible explanations requiring further investigation (Ben Messaoud et al., 2026).
What the Results May Mean
The findings should not be reduced to the claim that maternal race changes the biological effects of obesity, asthma, or depression on a developing child.
The study did not establish that.
A more defensible interpretation is that epidemiological associations between a recorded maternal diagnosis and later autism may not always have the same magnitude across populations.
Race and ethnicity are intertwined with social, environmental, economic, and health care circumstances. Differences in epidemiological estimates can reflect structural inequalities, environmental exposures, access to care, diagnostic practices, treatment patterns, and other measured or unmeasured factors rather than inherent biological differences.
The findings also do not provide evidence that changing treatment for a maternal condition would prevent autism.
Patients should not stop prescribed asthma, diabetes, antidepressant, or other treatment because of these observational findings. Medication and treatment decisions during pregnancy should be discussed with the treating clinician, who can consider the risks of both treatment and leaving the underlying condition inadequately controlled.
Limitations
The observational design is the central limitation. The study can identify associations but cannot establish that recorded maternal diagnoses caused autism or that maternal race or ethnicity caused differences in those associations.
The exposures were based on diagnoses recorded in electronic health records. Those records may not fully capture disease onset, severity, duration, treatment, medication exposure, or control.
The exposure windows also require careful interpretation. Nonchronic diagnoses could have been recorded within 12 months before delivery, while chronic diagnoses could have been recorded up to 48 months before delivery. A recorded diagnosis therefore does not necessarily mean that the condition was newly diagnosed or continuously active during pregnancy.
Residual confounding is another concern. Genetic factors, environmental exposures, socioeconomic circumstances, family history, migration history, treatment differences, and other variables could contribute to the observed relationships.
Insurance type and health care utilization provide only partial measures of socioeconomic circumstances and health care contact. They cannot fully capture income, education, neighborhood conditions, structural barriers, or quality of care.
Follow-up was relatively short. Median follow-up was 3.7 years, and the median age at autism diagnosis was 3.0 years. Children whose autism was identified later in childhood may therefore have been missed.
That limitation could be particularly important if age at autism diagnosis differs across populations.
Stratification also reduced statistical power in some racial and ethnic groups. Nonsignificant subgroup findings should therefore not be interpreted as proof that an association is absent.
Finally, the cohort came from Kaiser Permanente Northern California. Although the health system serves a large and diverse population, people at the extremes of the income distribution are underrepresented. The findings may not generalize completely to all U.S. populations or health systems (Ben Messaoud et al., 2026).
Final Thoughts
In this cohort of more than 224,000 children, associations between recorded maternal diagnoses and offspring autism were generally similar across racial and ethnic groups. Four of 25 diagnoses showed evidence that the magnitude of those associations varied: overweight or obesity, depressive episode, asthma, and other maternal disease complicating pregnancy.
Earlier research provides useful context rather than a simple causal explanation. Large studies have associated maternal asthma, depression, and early-diagnosed gestational diabetes with offspring autism, but they also demonstrate why treatment, timing, familial factors, disease severity, exposure definitions, and confounding matter.
The 2026 study adds another reason for caution. The same diagnosis recorded in two populations may not necessarily represent identical clinical, environmental, or health care circumstances.
Future research with longer follow-up and more detailed information about disease severity, treatment, medication exposure, socioeconomic circumstances, environmental factors, family history, and autism ascertainment may help explain why some association estimates differ across populations.
For now, the study supports neither a biological racial explanation nor the conclusion that the recorded maternal diagnoses themselves cause autism.
Funding and Disclosures
The 2026 study was supported by the National Institute of Mental Health through grant MH124817 to Magdalena Janecka and award T32-MH122394 to Vahe Khachadourian.
The authors reported that the funder had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or the decision to submit it for publication.
Lisa A. Croen reported receiving National Institutes of Health grants during the study. No other disclosures were reported (Ben Messaoud et al., 2026).
References
Ben Messaoud, K., Khachadourian, V., Anderson, M., Janecka, M., & Croen, L. A. (2026). Racial and ethnic differences in maternal health and autism. JAMA Network Open, 9(10), e2637501. https://doi.org/10.1001/jamanetworkopen.2026.37501
Gong, T., Lundholm, C., Rejnö, G., Bölte, S., Larsson, H., D’Onofrio, B. M., Lichtenstein, P., & Almqvist, C. (2019). Parental asthma and risk of autism spectrum disorder in offspring: A population and family-based case-control study. Clinical & Experimental Allergy, 49(6), 883-891. https://doi.org/10.1111/cea.13353
Hagberg, K. W., Robijn, A. L., & Jick, S. (2018). Maternal depression and antidepressant use during pregnancy and the risk of autism spectrum disorder in offspring. Clinical Epidemiology, 10, 1599-1612. https://doi.org/10.2147/CLEP.S180618
Xiang, A. H., Wang, X., Martinez, M. P., Walthall, J. C., Curry, E. S., Page, K., Buchanan, T. A., Coleman, K. J., & Getahun, D. (2015). Association of maternal diabetes with autism in offspring. JAMA, 313(14), 1425-1434. https://doi.org/10.1001/jama.2015.2707