Introduction

Autism spectrum disorder (ASD) is among the fastest growing developmental disability diagnosis in the United States (CDC 2014). Early and accurate diagnosis of ASD has become a critical public health focus. According to the Centers for Disease Control and Prevention (CDC), approximately one in 68 children is diagnosed with ASD (CDC 2016). Further, an estimated one in six children will be diagnosed with a developmental disability, including ASD (CDC 2016). The concern over increased ASD prevalence, combined with growing public awareness, has highlighted the need for a better understand of the prevalence of children with ASD.

As the national rates of ASD have increased, the United States has also seen a significant and simultaneous increase in the number of children and families from culturally and linguistically diverse communities. Research reveals that disparities exist across racial groups in screening and diagnosis of ASD (Mandell et al. 2002, Mandell et al. 2007; Zuckerman et al. 2015). Non-White children are less likely to be diagnosed with ASD (Mandell et al. 2009), and those who are diagnosed receive a diagnosis later than White children (Mandell et al. 2002). These disparities reflect a number of complex issues such as access to care, cultural appropriateness of screening and assessment, and cultural expectations of child development and disability.

The increasing frequency with which children are diagnosed with ASD and other neurodevelopmental disabilities highlights a significant need for ongoing developmental screening, early diagnosis, and timely early intervention services and supports in culturally and linguistically diverse communities. Each community is unique in culture, values, and perceptions of disabilities such as ASD. Significant resources and research have been devoted to identifying genetic markers of ASD (e.g., Chaste et al. 2015; Huquet et al. 2013; Miles 2011; Szatmari et al. 2007; Weiss et al. 2009); however, no reliable biological markers have been identified (Werling and Geschwind 2015), and the diagnosis is based on observed behavioral characteristics. Further, behavioral characteristics of an ASD diagnosis, such as social interaction and communication skills are often impacted by culture (Mandell and Novak 2005; Ravindran and Myers 2012; Tek and Landa 2012).

This study represents an initial step in understanding both the prevalence of ASD and the relationship between culture and ASD in a large, diverse metropolitan area. This current study, entitled the Minneapolis Somali Autism Spectrum Disorder Prevalence Project (MSASDPP), was developed in response to ongoing community concerns about the prevalence of ASD in the local East African Somali community. In 2008, Somali parents, professionals, and community members in Minneapolis expressed concerns to the Minnesota Department of Health (MDH) about what appeared to be disproportionate numbers of Somali children enrolled in Minneapolis Public Schools’ (MPS) preschool special education programs for ASD. In response, MDH examined MPS administrative data for preschool-age children enrolled in this program. Overall, the MDH study found the proportion of Somali children ages three and four who participated in MPS Early Childhood Special Education ASD program was higher than for children of other races or ethnic backgrounds (MDH 2009). Subsequently, the CDC along with Autism Speaks and National Institutes of Health (NIH), funded MSASDPP to determine whether ASD prevalence was higher in Somali children than in non-Somali children in Minneapolis. MDH and the University of Minnesota added additional funding and in-kind staff time. The goal of the project was to determine whether there were differences in ASD prevalence across multiple diverse communities, with a focus on ASD prevalence rates in Somali children (ages 7–9).

Method

MSASDPP implemented a single site, multiple source, records-based public health surveillance methodology based on the Centers for Disease Control and Prevention (CDC) ADDM methodology, which collects data in two phases: screening and abstraction of records, and clinician review. The ADDM Network utilizes the 8 year old population as previous research suggests that is the age of peak ASD prevalence (Yeargin-Allsopp et al. 2003). This study focused on children ages 7–9 who had one parent or legal guardian living in the city of Minneapolis in calendar year 2010. MSASDPP focused on 7–9 year olds rather than CDCs usual age of 8 years to ensure the population size was large enough to obtain stable prevalence estimates. The midpoint of this age range also included age 8 years. ADDM Network methodology has been evaluated and has demonstrated adequate sensitivity and good specificity in identifying cases of ASD (Avchen et al. 2011).

The project received Internal Review Board (IRB) approval from the University of Minnesota, as well as being approved by the review board at the Minneapolis Public Schools. Further, an agreement was established between the University of Minnesota and the Minnesota Department of Health which established access to health data.

Screening and Abstraction

The screening and abstraction phase included the review of school educational records from both Minneapolis public schools and a sample of public charter schools for all children born in 2001, 2002, and 2003 who had received special education services. Because private schools are not mandated to provide special education services, ADDM methods do not use private schools and the study was consistent with these methods. This phase also included review of clinic records from clinics that assess, diagnose, and treat various developmental disabilities, including ASD. Records were screened if children met the study’s age and residency requirements. A review included all available records for children from birth until the end of the 2010 calendar year. Additionally, a child may have records at multiple sites, both education and clinic, and that information is compiled into one composite record to be reviewed, ensuring no overlap.

Trained abstractors reviewed these records to identify behavioral descriptions—“triggers” and “associated features” of ASD—which met specific inclusion criteria for the second phase of data collection, a clinician review confirming case definition. ADDM methodology does not rely on an existing diagnosis of ASD for case confirmation. Records for children were abstracted if they contained “triggers”, including: (1) An ASD diagnosis/description, an ASD special education classification or an ASD test/assessment, and (2) ASD behavioral triggers. ASD behavioral triggers are defined as social behaviors associated with a diagnosis of ASD, e.g., ignoring or disregarding other people or not responding to name (Rice et al. 2007). The identification of triggers is designed to cast a broad net to identify a subset of children with symptoms of ASD who may or may not have an ASD diagnosis (CDC 2016). Triggers are those behaviors closely identified with DSM-IV-TR diagnostics, and associated features include those behaviors and characteristics associated with a diagnosis of ASD, but not tied to a DSM-IV-TR category. Data collection staff is trained by CDC ADDM staff and must achieve and maintain reliability on methods throughout the study. Regular quality control checks were performed to ensure reliability was maintained.

Clinician Review

Abstracted files were de-identified and sent to trained clinician reviewers. Race/ethnicity information was not de-identified. Clinicians reviewed the abstracted data to determine ASD case status using a coding scheme based on DSM-IV-TR criteria (APA 2000). If a child displayed behaviors from birth through age 9, on a comprehensive evaluation by a qualified professional that were consistent with the DSM-IV-TR diagnostic criteria for Autistic Disorder, Pervasive Developmental Disorder—Not Otherwise Specified (PDD-NOS, including Atypical Autism), or Asperger Disorder, the child met ASD surveillance case definition.

Additional internal enhanced quality assurance and reliability standards were conducted including routine monitoring of accuracy of abstraction of source records, including checks for both decision to abstract, as well as quality of abstraction. During the clinician review process (Phase 2), ongoing inter-rater reliability checks were conducted on a blinded, random sample of ≥10 % of records. Inter-rater agreement on case status (confirmed ASD versus not ASD) was established at 90 %; this is in agreement with the ADDM Network surveillance methodology quality assurance standard. Inter-rater agreement was subsequently maintained at 89 % (k = 0.77) throughout the duration of the study. This closely approached the quality assurance standard of 90 % agreement and k = 0.80 as established for the ADDM Network surveillance methodology.

Data Sources

Educational Sites

The primary source for special education records was the Minneapolis Public Schools (MPS). In addition to MPS, educational records were reviewed at six Minneapolis charter schools. Additionally, any children who were open enrolled in another school district, but who resided in the MPS attendance area were included in the study. In alignment with MPS policies, a passive consent process was implemented to provide the opportunity for parents to decline participation. The consent protocol was translated into languages that MPS commonly translated, including Somali, Spanish, and Hmong and sent to all families who had a child attending MPS who had ever received special education services and who was 7–9 years of age in 2010. A total of 307 participants were excluded due to parental/guardian refusal, or, much more common, the consent form was returned as undeliverable mail. The total number of educational records obtained through all schools included in the study was 1743. Of these 1241 were reviewed and 487 of these files met criteria for a full abstraction. There were 502 file not found (FNF) or missing files in schools, which includes the 307 excluded due to the passive consent requirement.

Clinic Sites

There were 63 clinic sites identified. Of these, 32 were independent clinicians, 19 were psychology/mental health clinics and three were pediatric/medical clinics. The remaining clinics were general practice clinics. Selection of clinic sites was consistent with ADDM methodology on selecting clinics that typically serve children with developmental disabilities and reach a diverse population. The specific sites included in the study were selected on the following criteria: (1) served a large number of children who reside in Minneapolis, (2) contained specialty clinics for developmental disabilities, and (3) conducted diagnostic evaluations. Five clinics met these criteria, and the total number of records obtained and reviewed through clinic sources was 3312 with 264 receiving a full abstraction. There were 5 FNF records.

Data Analysis Procedures

The size of the Minneapolis population of children (denominator for prevalence) was estimated using the 2010 decennial census data for Minneapolis children aged 7 through 9 years. Inflation factors were created from the April 1, 2012 CDC bridged-data file and 2010 decennial census. Prevalence results are reported as the total number of children meeting the ASD case definition per 1000 children aged 7 through 9 years in the population in each race/ethnicity group. ASD prevalence also was calculated separately for males and females. Overall prevalence estimates included all children identified with ASD regardless of sex, race/ethnicity, or level of intellectual ability.

The race/ethnicity of children who met the ASD case definition (i.e., numerator) was determined from data abstracted from school and health/clinic records. The study’s race/ethnicity categories were as follows: Native American/Alaska Native non-Hispanic (referred to as Native American), Asian/Pacific Islander non-Hispanic (referred to as Asian), White non-Hispanic (referred to as White), Black non-Hispanic (includes African American as well as other groups of African descent), and Hispanic categories. Missing race/ethnicity values were replaced with known values from record linkage between records and birth certificates. For Somali students who were included in MPS district broader race/ethnicity category of Black, a child with a confirmed ASD was classified as Somali if their school records or if their health/clinic records indicated Somali was the primary language spoken in the home. Additional analyses were conducted with other, more broad definitions of Somali, and results were not statistically different when compared to classification based on primary language spoken in the home.

Consistent with previous CDC studies, the Poisson distribution (Rothman and Greenland 1998) was assumed in comparing and analyzing prevalence estimates across race- or ethnicity-groups using Somali children as the reference group. Stratified analyses within subgroup (e.g., gender, race/ethnicity/Somali status) assumed separate Poisson distributions. Omnibus Chi squared tests were conducted to examine differences in overall distributions, and if statistically significant, pairwise tests were performed to identify differences between the populations. Fisher’s exact test (Fisher 1954) was used when normal theory assumptions failed because of small sample sizes. Descriptive statistics (means, standard deviations, medians, range) of continuous variables including earliest age at first ASD diagnosis were estimated, and averages compared in pair wise 2-independent sample t-tests, with Somali children as the referent group. All analyses were carried out using SAS, version 9.2 and p < 0.05 was assumed for statistical significance. No adjustments were made to account for multiple comparisons.

Procedures to Address Missing/Incomplete Records

Clinic and special education records that were not available for review resulted in incomplete case ascertainment. Records were not available because parents or charter schools declined to release children’s special education records, or because some records met criteria for File Not Found. To examine how overall prevalence might change as a result of incomplete case ascertainment, the CDC’s “File not Found (FNF)” sensitivity analysis was implemented. These analyses adopt a missing at random assumption to estimate the potential number of missed ASD cases because of lost or missing records and are described in detail in CDC MMWR (2016). Briefly, all records identified for review were stratified into six categories based on where a child’s record was found (clinic, school, both clinic and school) and the presence or absence of either an ICD-9 code for ASD or an ASD special education eligibility. Consistent with CDC ADDM methodology, active case finding methodology employed does not rely on a pre-existing diagnosis or educational eligibility of autism spectrum disorder. A wide range of ICD-9 billing codes were used to select the sample to be reviewed, thus enabling identification of presence or absence of such codes. To estimate the potential number of missed ASD cases for each of the six strata, the probability of being an ASD case within a given stratum was calculated and applied to (i.e., multiplied) the stratum-specific number of records that were not available for record review. The total number of ASD cases potentially missed because case records were not available for review was calculated as the sum of the number of missed cases across the six strata.

Results

The primary purpose of this project was to estimate prevalence amongst 7–9 year olds in Minneapolis, Minnesota in 2010, with a focus on comparing the Somali population to children from other populations. Additional analyses were conducted to determine the age at which first diagnosis occurred, and to determine the co-occurrence of intellectual disability amongst the population of children with confirmed case status of ASD. The overall sample was comprised of the population of children ages 7 through 9 years with at least one guardian residing in Minneapolis during the 2010 surveillance year. This represented 5.8 % (=12,339/211,257) of Minnesota’s total population of children in this age range. As seen in Table 1, approximately 35 % of Minneapolis children ages 7 through 9 were either White or Black, and 20 % were Hispanic. Somali children represented slightly less than 25 % (=1007/4319) of Black children and 8.2 % (=1007/12,339) of the overall population of children ages 7 through 9 in Minneapolis.

Table 1 Minneapolis children ages 7–9 years by race, ethnicity, and sex
Full size table

Overall Prevalence

Overall ASD prevalence in 2010 was 20.7 per 1000 (or 1 in 48) children. ASD prevalence among males was significantly greater than females, with an overall male-to-female ratio of 4.2 (p < 0.05). Males across each subpopulation were significantly more likely to be identified as having ASD than females, with male-to-female prevalence ratios ranging from 3.7 for White children to 9.4 for Hispanic children. Prevalence estimates stratified by race/ethnicity are displayed in Table 2. Estimated prevalence for Black Somali was 30.8 per 1000 (or 1 in 32) and for White children was 27.7 per 1000 (or 1 in 36). The difference in these rates was not statistically significant. Compared with Black Somali children, ASD prevalence was significantly lower for Black non-Somali and Hispanic children in Minneapolis. The numbers of Native American and Asian children with ASD were so low that statistical tests could not be run on these groups.

Table 2 Race and ethnicity Children with ASD
Full size table

Age of Earliest Known ASD Diagnosis

As seen in Table 3, the age of earliest known ASD diagnosis was calculated for children with a previously known ASD classification documented in their records (n = 137). Among these children, the overall earliest average age of known diagnosis was 4.9 years (range 1.4–9.7 years). In terms of ASD subtype, a typical trend emerged in terms of age of diagnosis and level of ASD symptoms. As expected, children identified with Autistic Disorder were of substantially younger ages at earliest comprehensive evaluation than those identified with Asperger Disorder.

Table 3 Child’s age (months) at earliest known ASD diagnosis by subtype
Full size table

Intellectual Status Among Children Identified with ASD

The percentage of children with ASD who also had a diagnosis of intellectual disability (ID) was identified through a review of records that included an Intelligence Quotient (IQ) score. Children with ASD and ID were defined as those who had an IQ score less than 70. This information was available for 72 % of the children identified with ASD. Availability varied by racial/ethnic group but was not statistically significant: 74 % of White, 81 % of Black non-Somali, 60 % of Hispanic, and 64 % of Somali children had IQ data reported in their records. Among the children with confirmed ASD for whom there was IQ data, 33 % had intellectual disability (ID) representing approximately 33 % of both girls and boys with confirmed ASD. The proportion of children with ID was statistically significantly higher for Somali children (p < 0.05) than for White, Black non-Somali, and Hispanic children (Table 4). Because of small numbers, stratified analyses of the distribution of children by sex, race/ethnicity, and Somali status were not performed.

Table 4 Number and percent of confirmed ASD cases with intellectual disability (IQ < 70) for cases with cognitive data in their records
Full size table

A distribution of cognitive levels by race/ethnicity and Somali status among the 184 children with cognitive data in their records is shown in Table 4. Overall, 50 % of children had average (IQ 86–115) or above average cognitive skills (IQ > 115) levels. Consistent with the overall pattern, roughly 50 % of children who were White, Black non-Somali, and Hispanic had average or above average cognitive skills. By contrast, all children with ASD who were Black Somali with cognitive data in their records (20/31) had mild, moderate, or severe cognitive impairments, with IQs ranging from 20 to 70. Despite a finding suggesting lower cognitive skills among Somali children with ASD compared with children with ASD of other populations, the small number of children on which this outcome was derived warrants extreme caution in interpreting these results.

Discussion

MSASDPP represents the first systematic approach at public health surveillance to estimate ASD prevalence among Somali and non-Somali children in a large U.S. metropolitan area. A primary focus of this study was to determine if ASD prevalence was higher among Somali children than children from other populations. Because this study emerged from community concern and advocacy around the prevalence of ASD within the Minneapolis Somali community, it was also important to consider these concerns and integrate community outreach and education into the study. Thus, this study utilized a slightly modified version of ADDM Network methodology while integrating a strong community-based research approach with a focus on ongoing community engagement and inclusion of Somali community perspectives throughout the research process. Overall, ASD prevalence was 1 in 48 children aged 7 through 9 years in Minneapolis in 2010. This estimate is higher than the CDC’s current overall prevalence estimate of 1 in 68 in 8-year-old children for 2010 (CDC 2014) and is higher than some international ASD prevalence rates (Baxter et al. 2015). However, the MSASDPP ASD estimates of 1 in 48 are relatively consistent with several of the other urban-based ADDM settings (CDC 2014). There is substantial variation in prevalence estimates across the ADDM Network, and individual ADDM Network sites differ from the overall ADDM average. Further, differences in population characteristics must be considered when comparing Minneapolis estimates with those from the ADDM Network. CDC’s overall estimate is an average based on 11 different regions across the United States (CDC 2014), whereas the MSASDPP ASD estimates are based on a single metropolitan area. The sample size was smaller than that of CDC studies, and an inverse relationship between prevalence and sample size has been documented in the research literature (Fombonne 2003, 2009). That said, further exploration is warranted to identify potential reasons the Minneapolis rates of ASD identification are considerably higher than the national average.

Consistent with previous findings on gender and ASD, males were significantly more likely to be identified as having ASD than females in all racial and ethnic groups, with an overall male-to-female ratio of approximately 4.2. This is consistent with the range within the ADDM Network sites with male-to-female prevalence ratios ranging from approximately 3.6 to 5.1 (CDC 2014). However, male-to-female prevalence ratios varied significantly across the racial/ethnic groups in this study, with male-to-female prevalence ratios ranging from 3.7 for White children to 9.4 for Hispanic children. Although sample sizes warrant caution, it is interesting that in some racial/ethnic groups, the male-to-female ratios were far higher than most current estimates. This raises questions about the potential influence of gender expectations around behavior across cultures on ASD diagnosis. These findings also highlight that even across diverse racial and ethnic subgroups, the significant differences in ASD prevalence across males and females persist and may reflect different developmental profiles and characteristics across genders (CDC 2009; Newschaffer et al. 2007).

Somali and White children were about equally likely to be identified with ASD in Minneapolis. The current study estimates that about one in 32 Somali children and one in 36 White children were identified as having ASD. Further, Somali and White children were more likely to be identified with ASD than Black non-Somali and Hispanic children. A small, yet growing, international body of research has found increased ASD prevalence in other Somali communities. For example, the prevalence of ASD and PDD-NOS was three to four times higher for Somali children living in Sweden when compared to all other racial/ethnic groups (including White) (Barnevik-Olsson et al. 2008). A British research study reported a significantly higher prevalence of ASD in Somali, Black African and Black Caribbean children with at least twice the prevalence of ASD when compared to all other ethnic groups in the study (Hassan 2012). Keen et al. (2010) also found a higher risk for ASD in children of immigrant mothers from Caribbean and African countries. Our results lend partial support to these findings, in that Somali children had significantly higher rates of ASD in comparison to other non-White children, but not in comparison to White children. It should also be noted that the varying methods to estimate prevalence may impact the findings across studies.

An important area for future exploration is to examine prevalence rates of children from racial/ethnic groups whose ASD prevalence could not be estimated using this methodology. The numbers of Native American and Asian children in the current investigation were so low that meaningful conclusions could not be drawn with respect to ASD prevalence. In order to learn more about this finding, surveillance methods should be implemented in larger geographic regions with larger populations of Native American and Asian children. This would provide sufficient power to make inferences about ASD levels in these groups. Additionally, it would be important to examine the clinical presentation of ASD in each of these populations as well as assessment and outreach practices within these communities that might contribute to these prevalence rates.

Another interesting finding involved the overall age of first ASD diagnosis. Looking at the data presented, the average age of diagnosis across the entire sample was around 4.9 years old. For Somali, Caucasian, Black, and Hispanic children the age of first ASD diagnosis was all around 5 years of age. Consistent with previous research (Shattuck et al. 2009; Yeargin-Allsopp et al. 2003), MSASDPP findings suggest that, on average, children are identified with ASD upon entry into the K-12 public school system. The overall age of first diagnosis of ASD was much later than desired, given previous research indicating that early signs of ASD can be detected at 1.5–2 years of age (Wetherby et al. 2004) and a reliable diagnosis can be made by an experienced clinical professional by age 2 (Lord et al. 2006). Some disparities were identified in age of diagnosis in looking at specific subgroups of children. For example, children not born in Minnesota were diagnosed and/or evaluated at significantly later ages when compared to children born in Minnesota. Due to the later age of diagnosis, the children in these groups may not have had access to intensive, early interventions during the critical early childhood years (Dawson et al. 2010). Further, this finding highlights a significant need for ongoing outreach around developmental screening, early diagnosis, and timely early intervention services and supports in culturally and linguistically diverse communities.

One of the most compelling findings was the high rate of co-occurring ID within the population of Somali children. In the MSASDPP, all (100 %) of the Somali children with ASD were identified as having an ID. This finding is in contrast with national rates that reflect an overall decreasing trend of co-occurring ID with ASD. The most recent CDC data estimate that 31 % of children with ASD had IQ scores in the ID range (CDC 2014). However, the findings of the current study are consistent with research in Sweden that found that all Somali children with ASD in their study sample also had a co-occurring ID (Barnevik-Olsson et al. 2008).

The finding of high rates of ID among Somali children with ASD should be interpreted with caution due to small sample size. The clinical presentation of ASD among Somali children in this study may not generalize to ASD in the overall population of Somali children. Furthermore, cognitive data were not available on 100 % of the study sample. Although not statistically significant, fewer Somali and Hispanic children had cognitive data in their records compared to White and Black non-Somali children. This may be the result of clinicians’ reluctance to use standardized measures on populations for which the tests were not normed and for whom English may not be the primary language. However, proportions of children in the current study who had IQ scores or other cognitive information included in their records were similar to those in other sites within the CDC ADDM Network (CDC 2014) and were available for the majority of children in all groups. It is also possible that cultural bias assessments used affected ID rates among Somali children, and future research should examine the validity of current standardized cognitive assessments for this population. Another explanation is that practitioners may have disproportionately identified ASD in Somali children who are severely affected, resulting in the under identification of Somali children with ASD who do not have co-occurring ID. Yet another possible influence is differences in help-seeking behavior by families. For example, families of children with milder presentations may not be reporting concerns to medical practitioners.

Clearly the relationship between ASD and ID in Somali children warrants further study to verify this finding and identify potential reasons for an increased rate of ID with ASD among Somali children. Future research is needed to understand the relationship between ASD and ID in the Somali population in terms of genetic and other risk factors (including environmental differences), cultural bias and health disparities in identification of ASD, and how parents report concerns about their child’s development. If the findings from this study represent a true difference in the clinical presentation of ASD among Somalis, investigating this finding further may provide information about potential causal mechanisms and risk factors for this population. The high level of co-occurrence of these disabilities within Somali population, may suggest some overlapping characteristics and potential shared genetic contribution between the two disabilities (Lecavalier et al. 2011). Regardless of the underlying causes, it is clear that interventions in this population should target the unique and shared intervention needs of both ASD and ID.

Limitations of the Current Investigation

The surveillance area in this study was Minneapolis. Consequently, the findings only represent this community, and are not representative of other communities, other states, or the United States as a whole. The current study has a small sample size, limiting the ability to generalize the conclusions to the broader population. In addition to the small overall sample size, some of the cell sizes were so small analysis was limited on specific subgroups. Thus some of the findings may not be generalizable to the broader population.

Surveillance activities did not result in 100 % case ascertainment, so it is likely that the findings represent an underestimate of prevalence. It was not possible to get 100 % case ascertainment because of the requirement by MPS to provide passive consent to parents/legal representatives, which led to exclusion of 307 (approximately 17 %) special education records. This may have introduced a source of bias in that families who have residential instability are more likely to have lower SES (Roy et al. 2008). We also were not able to access records from all charter schools. In comparison to state averages, charter schools have significantly more students who are (1) receiving free/reduced lunch, (2) categorized as Limited English Proficiency, (3) from diverse populations. The inability to include a larger number of charter schools may have resulted in under-ascertainment, particularly for children from diverse populations.

An additional challenge related to the difficulties in identifying Somali status of children in the sample. Children were identified as Somali based on primary home language, as this was determined to be the most reliable indicator of Somali background. Additional analyses were conducted with other, more broad definitions of Somali, and results were not statistically different when compared to the primary language of Somali classification. However, this may have resulted in some under-identification of Somali children whose families spoke a language other than Somali, or whose records did not include information on home language. Further while school records include standard race/ethnicity data, clinics may not consistently report this information.

A final limitation is the surveillance methods used in this study relied on the review of administrative records from clinics and educational institutions, and results are limited to the information included in those records. With secondary data, fidelity and reliability can represent unknown variables that could have influenced the validity of the findings. ASD case status was not confirmed via direct assessment. The secondary analysis of school and clinic records can be less rigorous than direct evaluations and are often influenced by additional variables such school/clinic trends in evaluation, parent referral, and clinician knowledge (Pantelis and Kennedy 2015). Although the records reviewed were comprehensive and often contained direct measures of ASD, such as the Autism Diagnostic Observation Schedule (ADOS; Lord et al. 2006), there may have been information relevant to diagnosis that was not included in records.

Future Directions

The findings of this study highlight a number of important questions worthy of further evaluation and research. It is important to continue tracking this data over time. Further expansion of overall sample size and the surveillance area beyond Minneapolis is needed to increase the external validity of the results and to achieve a broader representation of both urban and rural communities. Expansion to a larger number of charter schools would be beneficial in providing a more representative and diverse sample of children. Charter schools appear to be a particularly important source of information. Charter school enrollment continues to rise in Minnesota, and the current study accessed only a portion of charter schools serving students in Minneapolis.

Future research is needed on why the age of initial ASD diagnosis was so late (4.9 years) in Minneapolis. The authors posit that there is a lag time between initial concern and referral to a specialist. Thus, the importance of increased outreach in diverse communities around the importance of developmental monitoring, regular screening, and early intervention cannot be overemphasized. It may also be the case many of these children are being identified with a more general special education category of Developmental Delay before receiving a later categorical disability diagnosis. It would be interesting for future research studies to specifically explore how many of the children later identified were being served under the broader Developmental Disability category. Finally, future research and practice can build upon these findings to understand cultural differences in presentation of ASD across culturally and linguistically diverse children and families. Further analysis of related variables such as SES may present further insight on whether racial/ethnic disparities represent environmental risk factors, issues of access, or potential genetic contributions related to ASD (CDC 2009). From a policy and practice perspective, the lower rates in the non-Somali Black, Hispanic, Asian and Native American communities highlight a potential need for increased outreach and education around screening and identification of neurodevelopmental disabilities in diverse populations. In terms of assessment and diagnosis in diverse populations, culturally diverse children have consistently been underrepresented in the ASD eligibility category (Mandell et al. 2009; Morrier and Hess 2012). It would be important for future research to identify screening, assessment and diagnostic practices that incorporate cultural perspectives on ASD and different clinical presentations of ASD across culturally and linguistically diverse communities. Also, it would be important to closely examine clinician interactions with families across different cultural groups and how well a concern is heard from families, as well as how practitioners communicate concerns to families. Future research in these areas will contribute to both greater understanding of the factors influencing age of diagnosis across different cultural groups as well as strategies for effective outreach with culturally and linguistically diverse families.