Discussion
Overall, the associations between maternal health conditions and offspring autism appeared broadly consistent across maternal racial and ethnic groups. However, for a number of maternal conditions, including obesity, other maternal disease complicating pregnancy, asthma, and depressive episode, these associations differed across maternal race and ethnicity groups. While these differences were not always in a consistent direction, the point estimates tended to be higher in White mothers compared with those in mothers reporting other races and ethnicities, including Black, Hispanic, and Asian and Pacific Islander.
The reasons for these differences are likely complex and heterogeneous across different maternal conditions, warranting further exploration of these associations to identify specific mechanisms. Potential mechanisms underlying the observed differences include the manner in which these maternal conditions are diagnosed, with overdiagnosis or underdiagnosis potentially leading to differential symptom severity in those classified as exposed. For example, a clinical label of obesity is more likely to be given to White vs Black individuals at the same BMI and comorbidity level,18 with underdiagnosis of obesity in ethnic minority individuals potentially influencing its weaker observed association with autism in our study. Similar mechanisms could explain lower point estimates associated with depression and bipolar disorder (nominally significant) in children of Black and Hispanic mothers.19,20
Additionally, differential causes of certain conditions could also explain the differences in the associations between these maternal conditions and autism. For example, a diagnosis of asthma is more likely to be underlain by environmental factors like air pollution, poorer housing stock, and inadequate ventilation and maternal occupational exposures in Black and Hispanic families compared with White families21—potentially introducing differences in the confounder structure across racial and ethnic groups.
Finally, the relevant exposure may be not the diagnosis itself, but how it is managed.22 Differences across racial and ethnic groups in access to care, insurance continuity, clinical follow-up, and treatment patterns may affect both disease control and the observed associations in our study. These influences may help explain the heterogeneous associations we observed by reflecting both the potential teratogenic effects of some pregnancy medications and the protective effects of appropriate management of pregnancy complications. This, together with the mechanisms we described previously could render the same observable exposure nonequivalent across diverse racial and ethnic groups.1,2
The racial and ethnic differences observed in some associations between maternal health and child autism may, beyond the heterogeneity of factors correlated with maternal diagnoses, reflect disparities in autism diagnosis itself. Historically, autism has been underidentified in racial and ethnic minority children: children of Black mothers were less likely to receive a diagnosis even when scoring above the SRS threshold,23 and when minority children were identified, diagnosis tended to occur later, reflecting barriers to assessment services, fewer specialist referrals, evaluation delays, and practitioner bias.1,24 Consistent with later and more severe-at-detection identification, minority children were also more likely to have co-occurring diagnoses such as intellectual disability, with milder presentations going unrecognized.1 More recently, this identification gap has narrowed, with Black and Hispanic children diagnosed at rates approaching or exceeding those of White children,25 a pattern also observed in the current study. Importantly, this narrowing reflects improved case ascertainment, but not necessarily equitable access to screening. Because racial and ethnic minority children in our sample were more likely to be diagnosed with autism, we could not identify a clear mechanism by which differential autism diagnostic rates alone would explain the pattern of associations observed.
Strengths and Limitations
The strengths of the study include the use of a large, racially and ethnically diverse cohort, detailed EHR data available in KPNC records, robust statistical modeling, and an approach that acknowledges the heterogeneity of both the outcome and exposure. Nonetheless, this study has some limitations. While censoring was accounted for with the use of the Cox model, limited follow-up duration did not enable us to capture autism diagnoses made at later ages, which could be clinically and causally distinct.26 The stratification of the analysis rendered the analyses in certain racial and ethnic groups underpowered, resulting in nonsignificant associations despite the overall (sample-wide) significant associations. Similarly, the lack of significant interaction effects between maternal diagnosis and race and ethnicity should not be interpreted as a lack of heterogeneity of the effects explored in our study. Instead, group-specific estimates should be leveraged for joint meta-analytic efforts, enabling a fuller understanding of how maternal health interacts with broader sociodemographic and clinical factors and how these effects impact epidemiological research in autism.11
In the future, similar designs should be implemented across other settings, enabling the derivation of joint, more precise estimates. Similarly, race and ethnicity represents only one of the dimensions underlying potential heterogeneity of the observed associations, with others related to the family’s socioeconomic status or migration history, as well as the child’s clinical factors, including familial history or presence of comorbidities.
CC BY 4.0. "Racial and Ethnic Differences in Maternal Health and Autism" by Ben Messaoud K, Khachadourian V, Anderson M et al., JAMA network open (2026). https://doi.org/10.1001/jamanetworkopen.2026.37501. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Peer-reviewed and open access; full text from PubMed Central (PMC13639761). Headings and layout adapted for reading; figures and tables are in the original.