Autism Prevalence, Co-occurring Intellectual Disability, and Support Needs Differ for Somali and Hmong Communities in Minnesota

The current project implemented a multiple-source, records-based public health surveillance methodology as part of the CDC’s ADDM Network (Christensen et al., 2016; Rice et al., 2007). The population for this study included 8-year-old children who resided in the defined geographic area in Minnesota during calendar years 2014 and 2016 (award cycle 2015–2018) and attended one of the participating school districts. Case identification of autism involved two phases. Phase 1, screening and abstraction, included all children born in 2006 or 2008 who had at least one parent residing in the defined geographic surveillance area in 2014 or 2016. Record review included educational records for children who had ever received special education services and clinic source health records from clinics where assessment, diagnosis, and treatment of various developmental disabilities (including autism) occurred. Trained abstractors reviewed for behavioral descriptions that reflected Diagnostic and Statistical Manual-5th edition (DSM-5; APA, 2013) symptoms of autism spectrum disorder—these are referred to as “social behavioral triggers.” For example, poor eye contact, no response to name, or lack of interest in peer interaction were considered social behavioral triggers. Information abstracted from records that contained a social behavioral trigger included verbatim developmental histories, descriptions of autism symptoms, descriptions of co-occurring conditions, results of developmental tests, and documentation of any clinical autism diagnosis or special education eligibility statement. All abstracted information was combined into one composite record if multiple health/education records were abstracted for the same child. In Phase 2, clinician review, clinicians who were licensed psychologists with expertise in diagnosis of autism reviewed the composite records to determine autism case status using a coding scheme based on DSM-5 criteria. A child could meet autism surveillance case definition by having an existing DSM-IV or DSM-5 clinical diagnosis on the autism spectrum. In addition, if a child displayed behaviors from birth through age 8 years on a comprehensive evaluation by a qualified professional that were consistent with the DSM-5 diagnostic criteria for autism, the child met autism surveillance case definition.

The current method allowed for identification of autism cases even when a formal diagnosis of autism had not been made. Similarly, clinician reviewers could determine that autism case status was not met, even in the presence of a formal autism diagnosis or eligibility, if insufficient information was present to support autism case status, or they could overturn autism case status even if behavioral criteria were met if there was sufficient information that the behaviors were better explained by another diagnosis. Clinician reviewers also provided ratings reflecting certainty of autism case status, and secondary reviews of records were performed when the primary reviewer’s case status certainty was low. (Of note, ADDM methodology changed starting in 2018 such that clinician review was eliminated, and prevalence estimates were based on formal community identification either clinically or through special education.) Clinicians and abstractors completed training and ongoing reliability checks. Abstractor training consisted of about 12 h of didactic instruction and practice with sample records. Abstractors established reliability on the decision to abstract at 100% and met monthly to consensus-code, maintaining reliability at over 90%. Clinician reviewers completed an initial didactic training and met monthly to consensus-code and discuss sample records. Inter-rater agreement on DSM-5 case status (confirmed autism versus not autism) was established at 90% and subsequently maintained (k = 0.84 in 2014, Baio et al., 2018; k = 0.89 in 2016, Maenner et al., 2020).

Results of autism prevalence comparisons are provided in Tables 1 and 2. Comparing across race/ethnicity, in the combined study years, Somali children had a higher autism prevalence than Hispanic children, while Hmong and Hispanic children had a lower autism prevalence than White children. Where sample sizes of females with autism were adequate, the male: female ratio was significant. The sex ratios for White and non-Somali Black children were similar to the overall ratio of 4.50:1; the ratio was lower for Somali children, at 2.05:1. Sex ratios for Hispanic, non-Hmong Asian, and Hmong groups could not be calculated as fewer than 10 females met autism case status in each group.

IQ data were available for 85% of the total sample, and IQ availability differed by race/ethnicity such that Hispanic and Somali children were less likely to have IQ data in their records (Table 3). Few Hmong children with autism had co-occurring ID (N=6), and few received ratings of low support needs (N=4); thus, subsequent analyses combined Hmong and non-Hmong Asian children. Clinician ratings of support needs were present for the full sample. Co-occurring ID did not differ by sex (p=.24; Figure 1a). Significant differences in co-occurring ID rates were found by race/ethnicity (p<.0001). Where 14% of White children had co-occurring ID, 37% of non-Somali Black, 39% of Hispanic, and 40% of Somali children had co-occurring ID (Figure 1b). Similarly, clinician rating of support needs differed by race/ethnicity (p<.0001) but not by sex (p=.14; Fig. 2a and b).

Overall, 53% of the sample had a documented clinical diagnosis of autism spectrum disorder in their records, and this did not differ by sex or race/ethnicity (Table 4). Median age of diagnosis did not differ by race/ethnicity or sex. Median age of first evaluation did differ by race/ethnicity; White children had a median age of evaluation of 48 months, while the median age was under 42 months for Hispanic and Somali children.

This study represented the largest to date on autism prevalence among children from Somali communities in the U.S. and the first study of autism prevalence in Hmong children. Findings supported higher autism prevalence for Somali children compared to Hispanic and Hmong and non-Hmong Asian children, with a trend toward higher prevalence relative to White and non-Somali Black children. Although the Somali prevalence rate was statistically higher, prevalence ratios with other race/ethnicities ranged from 1.32 to 2.11 and were lower than those found in previous studies of children of Somali immigrant mothers in Sweden (Barnevik-Olsson et al., 2008, 2010), as well as those found when examining attendance within Minneapolis Public Schools’ preschool ASD programs (MDH, 2009). The latter finding is interesting, as the city of Minneapolis and Minneapolis Public Schools were included in the surveillance area in the current study. However, a direct comparison of the results of the current study with the Minneapolis preschool ASD program study may not be appropriate for three reasons: (a) the Minneapolis preschool study only examined educational records, while the current study examined medical and educational records to identify autism cases; (b) the current study drew from a broader geographic area; and (c) school attendance in Minnesota is not mandatory until age 7 years; thus, the current study of 8-year-old children is likely to have more comprehensive ascertainment than the Minneapolis preschool study.

Findings also revealed a significantly lower autism prevalence for Hmong children compared to White, non-Somali Black, and Somali children. Hmong prevalence did not differ from that of non-Hmong Asian children, and prevalence for non-Hmong Asian children was also lower than that of White, non-Somali Black, and Somali children. These findings supported community concerns of potential under-identification of autism within Hmong and other Asian communities in Minnesota (MDH, 2009).

Consistent with previous findings on sex and autism, males were significantly more likely to be identified with autism than females, with an overall male-to-female ratio of 4.5. However, male-to-female prevalence ratios were markedly different for Somali children compared to other groups for whom sex ratios were calculable, with male-to-female prevalence ratio of 2.05 compared to 4.26 for White children and 4.02 for non-Somali Black children. Although sample sizes warrant caution, a potential influence may be that historically, females with autism have shown higher rates of co-occurring ID, potentially contributing to a higher rate of co-occurring ID in our Somali sample. However, we did not find a higher rate of co-occurring ID in females compared to males in our overall sample. Another possible implication is the potential influence of gender expectations around behavior across cultures on autism diagnosis. These findings also highlight that even across diverse racial and ethnic groups, the significant differences in autism prevalence for males and females persist (Lyall et al., 2017; Werling & Geschwind, 2013).

The finding from the Minneapolis study that 100% of Somali children with autism had co-occurring ID was not replicated in this study. Somali children had a rate of ID (40%) similar to ID rates for non-Somali Black and Hispanic children in Minnesota but higher than ID rates for White and Asian children (Hmong and non-Hmong combined), which were 14 and 24%, respectively. Given that the overall ID rates for children with autism across the national ADDM network were 31% in 2014 and 33% in 2016, an overall rate of 40% is not unexpectedly high and was slightly lower than national rates of co-occurring ID for Black children (44% in 2014 and 47% in 2016) (Baio et al., 2018; Maenner et al., 2020). It is important to note that the geographic area expanded in 2014 and 2016 beyond Minneapolis to include parts of Hennepin and Ramsey counties, including Saint Paul, which is the second largest and most diverse school district in Minnesota. Differences in school district policies and practices with regard to IQ testing, particularly of children from culturally and linguistically diverse backgrounds, may have impacted findings. Our study also found differences in clinician ratings of support needs by race/ethnicity, with a higher proportion of White children receiving ratings of the lowest support needs than other groups—twice that of Somali children and 2.5 times that of Asian children. This fits with previous findings from the ADDM network (Wiggins et al., 2020; Young et al., 2024). Interestingly, Asian children had the highest proportion (39%) rated with high support needs but a relatively low rate of co-occurring ID (24%). Co-occurring ID can contribute to but is not equivalent to having high support needs, and this finding may suggest a higher level of autism-related symptoms compared to other groups.

Higher ID rates and support needs for Black (and Hispanic) children has been a national trend that needs further exploration. A potential contributor for Black children could be later age of identification, which has been documented in the ADDM network as well as other sources (Constantino et al., 2020; Bishop-Fitzpatrick & Kind, 2017; Mandell et al., 2009; Shattuck et al., 2009; Hall-Lande et al., 2021; Wiggins et al., 2020; Zuckerman et al., 2013). In 2016 in the ADDM Network, Black children with autism plus ID were diagnosed 6 months later than their White counterparts and were less likely to have been evaluated by age 36 months (Maenner et al., 2020). Access to early intervention has been shown to positively impact cognitive development and may reduce the likelihood of persistent ID in children with autism (Dawson et al., 2010, 2012; Reichow, 2012), which could also reduce support needs over time. Another important factor to consider is that the ADDM network defines ID based on a score below 70 on the most recent IQ test. There is a long history of concerns of racial/ethnic bias in IQ testing that remains today (Aston & Brown, 2021), and including information about adaptive skills was emphasized in the diagnosis of ID in DSM-5 in part due to concerns about racial/ethnic bias in IQ testing (Boat et al., 2015). We recognize that a DSM-5-TR diagnosis of intellectual disability requires deficits in both cognitive and adaptive functioning; however, the availability of adaptive behavior measures was lacking among the clinic and educational records reviewed. The use of IQ scores less than or equal to 70 was considered a reasonable proxy for ID but may not align with a true diagnosis of ID in all cases.

The racial disparity in co-occurring ID rates and level of support needs in autism also could be driven by under-identification or misidentification of autism without ID or with mild support needs. In previous analyses of ADDM data, Black and Hispanic children, despite meeting behavioral criteria for autism, were less likely to have an autism clinical diagnosis (Wiggins et al., 2020) or special education eligibility (Young et al., 2024) compared to White children and also more likely to have a non-autism diagnosis or special education eligibility than White children. Based on results of a national survey, Jo et al. (2015) found racial/ethnic disparities in prevalence and age of diagnosis of autism were largely attributed to “mild/moderate” cases, suggesting there may be underrepresentation, and possibly under-identification, of culturally and linguistically minoritized groups with autism with mild/moderate support needs.

Findings of the current study suggested earlier evaluation of Somali and Hispanic children compared to White children, but this did not translate to earlier clinical diagnosis of autism. It is important to note that the ADDM definition of an evaluation encompassed any developmental evaluation, not just those specific to autism. Children with autism with ID are typically referred earlier than those without ID; thus, higher rates of ID for Somali and Hispanic children may be driving earlier referral for an evaluation but perhaps not one specific to autism. Indeed, previous research using ADDM data found that even though they had documented behavioral characteristics consistent with autism, Black, Hispanic, and Asian children were more likely to have no mention of an autism concern in their records compared to White children (Young et al., 2024).

A variety of modifiable systemic factors may influence accurate early identification of autism in culturally and linguistically diverse children, including systemic inequities, provider knowledge of autism and comfort discussing developmental concerns with families, provider dismissal of parent concerns, confusion with the diagnostic process, and a lack of properly translated and culturally competent screening tools (Aylward et al., 2021; Fenikilé et al., 2015; McNally et al., 2020; Parish et al., 2012; Vanegas, 2021; Zuckerman et al., 2014). Barriers at the family level also can play a role, including language barriers, lack of autism knowledge, and different cultural perceptions of disability including stigma around disability and mental health (Donohue et al., 2019; MDH, 2014; Zuckerman et al., 2014, 2017). Although culturally responsive autism screening and diagnostic practices have received increased attention (Harris et al., 2014; Soto et al., 2015), more research is needed on specific provider and parent education and awareness models to reduce stigma and increase equitable access to services.

The current study found a higher autism prevalence for Somali children compared to Hispanic and Asian children (both Hmong and non-Hmong), with a trend for a higher prevalence compared to White and non-Somali Black children. In contrast, we found a lower autism prevalence for Hmong children compared to White and Black children (both Somali and non-Somali). The study also found these findings generally matched with community concerns and impressions that prompted this research. Results highlight the importance of reducing disparities in the identification of children with autism characteristics so that equitable access to appropriate interventions can be promoted across communities.