Autism spectrum disorder [ASD] is one of the fastest growing neurodevelopmental disorders and eligibility categories for special education in the United States (Maenner et al., 2023; USDOE, 2019). Growing rates of autismFootnote 1 concurrent with increasing culturally and linguistically minoritized [CLM] populations in the United States (Frey, 2021) present a need for accurate identification of autism across populations. The identification of autism and provision of services at a young age can encourage better long-term outcomes, and reduce the financial costs of autism services for the family (Koegel et al., 2014).

Recent results of the Centers for Disease Control and Prevention’s [CDC] Autism Developmental Disabilities Monitoring [ADDM] network reveal shifting demographic trends in autism identification suggesting racial/ethnic differences in autism prevalence have decreased over the past few years (Maenner et al., 2020, 2021, 2023). The most recent ADDM network data found for the first time that the percentage of 8-year-old Hispanic, Black, and Asian or Pacific Islander children identified with autism was higher than the percentage in 8-year-old White children (Maenner et al., 2023). Previous studies though have identified longstanding trends that both the likelihood and timing of autism identification have been inequitable across racial/ethnic groups (Baio et al., 2018; Constantino et al., 2020; Luelmo et al., 2022; Maenner et al., 2020; Safran, 2008; Travers et al., 2018), and both may vary given location or setting of services (Liu et al., 2023; Luelmo et al., 2022; Pettygrove et al., 2013). Children from CLM backgrounds are more likely to experience delays in diagnosis, misdiagnosis, and require more healthcare visits to receive a diagnosis when compared to White children (Mandell et al., 2002, 2007, 2009; Tek & Landa, 2012; Zuckerman et al., 2013, 2017).

Though prevalence rates may be leveling out across racial/ethnic groups, a pattern remains where Black children with autism have significantly higher rates of co-occurring intellectual disability (ID) compared to other groups (Maenner et al., 2023). Access to early intervention may be one contributor to this persistent disparity. Historically, children from CLM communities have received both less access and later access to early intervention services and supports (e.g., Part C, Part B 619) during the early childhood years (Bishop-Fitzpatrick & Kind, 2017; Constantino et al., 2020; Hall-Lande et al., 2021; Hewitt et al., 2016; Shattuck et al., 2009)—years we know are critical for growth in response to intervention (Koegel et al., 2014). More recent research conducted by Cruz and Firestone (2022) reveals unique patterns in early intervention for children with autism, including greater likelihood of White and Asian Pacific Islander (API) students being identified with autism in preschool when compared to other racial and ethnic groups.

Despite evidence of racial/ethnic disparities in both likelihood and timing of autism identification, there is limited understanding of the mechanisms by which these disparities occur. Researchers have examined systemic issues, such as service accessibility, and child and evaluation factors that may influence autism identification (Daniels & Mandell, 2014; Mandell et al., 2009; Morrier et al., 2008). CLM populations may experience inequitable service access, demonstrate behavioral differences in presentation—or at least clinician interpretation of their presentation—while clinicians may experience challenges in adequately assessing CLM populations, or demonstrate racial bias in their identification of autism (Banerjee & Luckner, 2014; Begeer et al., 2009; Harris et al., 2014, 2019; Harrison et al., 2017; Soto et al., 2015; Tek & Landa, 2012).

Multiple factors may impact a diagnostic or eligibility outcome individually, but are also not isolated events; one factor affects the other. To be identified with autism, a child must initially access medical or educational services (and have parental consent for evaluation). Once in contact with such services, a comprehensive autism evaluation should include record review; caregiver interviews on developmental history and current concerns; direct assessment of a child’s development, particularly in cognitive, language, and adaptive skills; and direct assessment of characteristics related to autism, including autism-specific measures. From this, a practitioner should synthesize information and provide a ‘best estimate’ diagnosis or determination of educational eligibility (Bradley-Johnson et al., 2008; Huerta & Lord, 2012; Lord et al., 2022; Ozonoff et al., 2005; Risi et al., 2006; Zwaigenbaum et al., 2015). Given the stepwise nature of the identification process, inequitable experiences in any one step may influence subsequent steps in the process, creating an additive effect of factors contributing to disparities.

Educational settings, though, can increase the reach of autism identification services, removing some service accessibility barriers by reducing transportation and service costs (Williams et al., 2015). But beyond simply accessing services, the evaluation process in educational settings is complex, with potential barriers and contributors to disparity. Despite empirically supported practices, evaluation guidelines lack standardization, including the criteria for autism eligibility, who conducts the evaluation, and what assessments are completed (Barton et al., 2016a, 2016b; Division for Early Childhood; DEC, 2014). There is evidence the experiences within the evaluation process differ for children from a CLM background (Barton et al., 2016a, 2016b; Soto et al., 2015). These differences may be reflected in; (a) the components of the evaluation, including their adequacy for the given population, and (b) the evaluator themselves, including the knowledge and specialty of the individual. These factors may impact the likelihood of an autism diagnosis/eligibility (Cuccaro et al., 1996; Harris et al., 2019; Tek & Landa, 2012; Vanegas et al., 2016). Determining all the ways in which an evaluation may be inequitable is complicated by the data used to inform the current knowledge around disproportionality. A first step is to explore existing data that provides some insight into the evaluation experience of certain populations.

In attempts to accurately evaluate the prevalence of autism and understand potential disparities, the CDC established the ADDM network to monitor prevalence in the United States through medical and educational record review. ADDM collects data from multiple sources in the community where children are served, including clinics and educational settings. The inclusion of special education data is critical as a significant number of children with autism are identified and served primarily in school-based special education settings (McDonald et al, 2019; Sturm et al., 2021; Wei et al., 2014). As ADDM prevalence estimates through 2016 were obtained through comprehensive record review and did not rely on existing identification, they provide a glimpse into a broader issue; we may be missing children or inaccurately classifying children who have autism and could greatly benefit from services.

Though race/ethnicity are known to play a role, other factors related to autism (e.g., co-occurring intellectual disability, age at evaluation) likely influence the timing and likelihood of an existing identification (Baio et al., 2018). In previous ADDM studies, likelihood of community-identified autism (having a clinical diagnosis or educational eligibility) varied depending on a variety of factors, including race/ethnicity, presence of an intellectual disability, as well as the source (medical or educational) of the abstracted evaluation record used to determine autism case status (Baio et al., 2018; Pettygrove et al., 2013). Children with records from education only sources (no medical records found) were identified later than those with medical records, and were less likely to be identified at all (Esler et al., 2022; Pettygrove et al., 2013). These data may indicate schools may be missing, inaccurately characterizing, or are the sole provider of services for some children with autism, as delays in identification may result from a child being school-aged before accessing services (Wiggins et al., 2006).

Because all U.S. children have the right to a public education, public school systems may provide a more accessible avenue of autism identification, reaching more CLM populations than health systems that rely upon health insurance and face other barriers to service access (Pettygrove et al., 2013). However, evidence suggests autism may be under-identified in educational settings—as school autism rates continue to lag behind epidemiological estimates—and this may be exacerbated for students who are CLM (Nowell et al., 2015; Pettygrove et al., 2013; Safran, 2008). These data bring into question if schools are missing children who could benefit from special education services (Sullivan, 2013). The lack of special education eligibility, or an inappropriate eligibility category for a student with autism, could mean a lack of service delivery or receiving services misaligned with their needs. Though differences in autism prevalence rates across medical and educational systems are not expected to be commensurate given varying identification criteria (i.e., an educational identification of autism must consist of educational impact, which may not occur for some children with a medical diagnosis of autism), disproportionality in special education remains a poorly understood concern (Sullivan, 2013; Sullivan & Bal, 2013; Thomas, 2016). Educational policy emphasizes the role of schools in the identification of disabilities and provision of services (IDEA, 2004), highlighting this as a critical context for understanding the factors that impact accurate and timely identification for CLM populations.

Though contributors to disparities arise throughout the identification process—from initially accessing services to the evaluation measures used—little is known regarding their relative influence and how one factor may impact another. Service accessibility, child characteristics, and evaluation characteristics may be consistent factors impacting identification (Cuccaro et al., 2007; Morrier et al., 2008; Nowell et al., 2015), but the relative contribution of each is complicated by its correlation with other factors, aggregate data, and bivariate analyses that may over simplify the relation between factors (Sullivan & Artiles, 2011; Sullivan & Bal, 2013). To understand the complexity of disproportionality, it is necessary to move beyond a simplistic framing of the issue, with over or under representation explained by race alone (Sullivan & Bal, 2013). Rather, researchers need to take steps to identify the interacting contributors and clustering of factors that may influence the likelihood of identification. Thus, to further explicate disproportionality, the purpose of the current investigation was to characterize students who may be underrepresented in autism eligibility. Specifically, the aim was to identify the clustering of child and evaluation factors documented in service records that may increase the likelihood of lacking autism eligibility, when potentially warranted, with a focused examination of race/ethnicity as a factor that may exacerbate the effect of other variables. To address this aim, the following research questions were examined:

  1. 1.

    Do child and evaluation characteristics documented in evaluation records predict the absence of autism eligibility?

  2. 2.

    Does race/ethnicity significantly predict the absence of autism eligibility when accounting for child and evaluation characteristics?

Methods

Autism Case Ascertainment

The study population included 8-year-old children, who had health or educational records reviewed for developmental evaluations from sites within the ADDM network in the 2014 surveillance year. ADDM personnel review medical or educational records for behavioral descriptions (‘triggers’) of autism characteristics that initiate an abstraction. Once a record is abstracted, all available evaluation information from birth to the current surveillance year for each child are combined into one aggregate record, de-identified, and then reviewed by trained clinicians to determine autism case status. Clinicians documented symptoms and applied a specific algorithm to determine whether cases met autism case criteria. Specifically, clinicians could determine a child met case status based on: (a) having an autism diagnosis and behavioral criteria documented to support the diagnosis (using DSM IV or DSM-5 criteria) or (b) based on meeting behavioral criteria alone. Clinicians could also overrule cases that met behavioral criteria in their record, whether or not they had a formal identification of autism, if the clinician felt another condition could better account for the child’s symptoms (Maenner et al., 2020).

Sample

The study sample included 2313 children from the 2014 study year population, who met ADDM case criteria for autism and had an evaluation record from an educational setting. The final analytic sample included 2167 and eliminated incomplete cases (those missing values for variables used as predictors in analyses), records from three states (NJ, MO, and MD, given education policy or access to educational records) and included racial/ethnic groups with adequate sample sizes. Cases with missing data were not proportionately different by race/ethnicity. Of complete cases, 29.2% had a documented intellectual disability. Males comprised 81.5% of the sample, 65.1% had an existing clinical diagnosis of autism, and 71.7% had autism special education eligibility documented. The racial/ethnic distribution included 47.6% White, 5.4% Asian/Pacific Islander, 30.3% Black, and 16.7% Hispanic. IQ data was not available for 9.8% of the sample, and rate of missing IQ data did not differ by race/ethnicity.

Measures

From the composite abstracted records, ADDM clinicians reviewed and documented descriptive characteristics (child characteristics, birth information) and data sources (educational or health). Information on evaluation components was also collected; specifically, information on IQ tests, adaptive measures, and autism measures was documented. All measures were those obtained through ADDM study data and were as documented by ADDM study staff. Some variables were not obtained through ADDM personnel, and thus other information available in records served as proxies, as described in Table 1.

Table 1 Descriptions and values of child and evaluation factors
Full size table

ADDM Specific Measures

ADDM clinicians documented a few measures specific to ADDM studies. Autism discriminators are behavioral characteristics documented in a child’s record that are diagnostic of autism (i.e., reliably discriminate autism from other developmental disorders). Cases meeting DSM-IV-TR (APA, 2000) criteria for pervasive developmental disorder-not otherwise specified [PDD-NOS], required the presence of an autism discriminator to meet autism case status. Associated features are behavioral characteristics documented in the child’s record that are not diagnostic of, but often co-occur with, autism. Autism diagnostic status documented represents the diagnostic summary statement resulting from an evaluation and was assigned in a hierarchical fashion; a clinical autism spectrum diagnosis was at the top of the hierarchy, and no mention of autism, “Autism diagnosis not reported,” was at the bottom. Level of impairment was a rating on a 5-point scale representing the clinician reviewer’s interpretation of the child’s level of impairment associated with autism.

Data Analyses

Analyses took place in stages. Literature regarding variable selection in logistic regression informed analysis steps (Agresti, 2002; Bursac et al., 2008). To identify potential predictors of autism eligibility, bivariate analyses were used to evaluate differences in child and evaluation factors between the two eligibility groups. Factors demonstrating significant differences between the eligibility groups (p < .05) in initial analyses and variables of known importance (Agresti, 2002; Bursac et al., 2008) were included in regression models. Variable significance, effect size relative to other variables, and theoretical importance given support in the literature determined the order in which they were stepped into the model. For all chi-squares, Cramer’s V was calculated (Kim, 2017), where 0.10 represented a small effect size, 0.30 a medium effect size, and 0.50 a large effect size. As degrees of freedom increase, the cut points distinguishing effect size decrease. For example, with 5 degrees of freedom, 0.04, 0.13, and 0.22 represent small, medium, and large effect sizes, respectively (Kim, 2017). For all Mann–Whitney U analyses, r was used as a measure of effect size, with cut points falling at 0.10, 0.30, and 0.50 for small, medium, and large effect sizes, respectively (Fritz et al., 2012). Epsilon squared was used for all Kruskal–Wallis analyses, with 0.02, 0.13–0.13, 0.14 + representing small, medium, and large effect sizes, respectively.

To better understand the clustering of factors that may influence lacking autism eligibility despite documented characteristics, hierarchical logistic regression [HLR] was used to estimate how child and evaluation factors may relate to the absence of autism eligibility. The proposed model resulting from bivariate analyses stepped in child factors first, evaluation factors second, and then considered the typical order of events in obtaining a diagnosis or eligibility of autism (e.g., behaviors documented, evaluations conducted, clinical diagnosis given) (Huerta & Lord, 2012). Record site, percentile rank of median household income [PR MHI], and sex served as control variables in all models. Control variables were retained through all regression models regardless of non-significant results in bivariate or regression analyses. To better understand the impact of evaluation factors specifically, post hoc analyses examined specific evaluation components and their relation to the absence of autism eligibility. For all regression models, odds ratios with 95% confidence intervals [CI], and the residual deviance accounted for under each predictor were used to quantify the magnitude of the association between predictors and the absence of autism eligibility. To evaluate model fit and identify the clustering of factors best predicting the absence of autism eligibility, a few indices were used to offer converging evidence, including McFadden’s Pseudo R Squared, Akaike Information Criterion [AIC], and a Likelihood Ratio Test [LRT] to compare nested models.

To understand the effect of race/ethnicity on the absence of autism eligibility, initially, race was added as a predictor in the last iteration of the HLR model. Next, given documented racial/ethnic disparities in factors examined, it was of interest to see if similar racial/ethnic disparities existed in predictors found to be significant in the HLR. Lastly, among children who met ADDM autism case criteria without documented autism eligibility, the association between racial/ethnic group membership and autism diagnostic status documented in the record was examined. It was of interest to assess if CLM populations were less likely to be considered for autism (e.g., Autism Eligibility Rule Out documented) or have concerns of autism reported (e.g., Report of Autism Concerns documented) more frequently than White children, based on level of concern reported. For all analyses, significance was set at p < .05. Data analyses were performed using R, version 1.2.5042.

Results

Bivariate analyses identified group differences in child and evaluation factors between the educational autism eligibility and the no autism eligibility group. When examining demographic factors, eligibility groups differed by race/ethnicity, largely attributable to a higher percentage of Hispanic children lacking autism eligibility (14.9% of autism eligible group and 21.2% of the non-eligible group; p = .006). Eligibility groups did not differ by PR MHI (W = 474,558, p = .866), or sex, X2(1) = 0.051, p = .822. When examining child factors, the presence of a non-autism clinical diagnosis, X2(1) = 40.532, p < .001, the presence of a clinical autism diagnosis, X2(1) = 96.044, p < .001, presence of an intellectual disability, X2(2) = 21.19, p < .001, level of impairment (W = 636,493, p < .001), total number of autism discriminators (W = 612,808, p < .001), total number of associated features (W = 502,568, p = .049), and age at first evaluation (W = 429,250, p < .001) were all significantly related to eligibility group membership (Table 2). Proportionally, significantly fewer children had a clinical autism diagnosis and an intellectual disability in the no autism eligibility group, while significantly more children in this group had a non-autism clinical diagnosis documented. The no autism eligibility group also had a significantly lower level of impairment documented, significantly fewer autism discriminators and associated features documented, and received their first evaluation at a later age. Effect sizes were medium for a clinical autism diagnosis, and small to medium for the presence of intellectual disability and child race/ethnicity.

Table 2 Eligibility group differences in child factors
Full size table

Table 3 presents differences by eligibility group in evaluation components completed, the evaluator, and the site from which the record was obtained. All three variables; record site, X2(7) = 62.053, p < .001; evaluation comprehensiveness, X2(3) = 125.03, p < .001, and evaluator, X2(4) = 36.251, p < .001, were significantly related to eligibility group membership, with a large effect size for evaluation comprehensiveness. Differences in site of evaluation record by eligibility group were largely attributed to group differences observed for North Carolina, Tennessee, Colorado, and Georgia, based on the standardized residuals. Differences in evaluation comprehensiveness were largely attributable to group differences in having three documented evaluation components. Evaluator level differed between eligibility groups with significantly more children in the no autism eligibility group with evaluations completed by an ‘educator—NOS’ and significantly fewer by an ‘unknown’ evaluator.

Table 3 Eligibility group differences in evaluation factors
Full size table

To better understand how child and evaluation factors differing between eligibility groups in bivariate analyses may predict the absence of educational autism eligibility, a HLR was conducted. Table 4 presents results with odds ratios [OR] and significance levels. The initial model examined all control variables (site, PR MHI, sex) as predictors. Of note, in model 6, the only evaluator levels to significantly predict the outcome were those unknown or unspecified in evaluation records. As such, this outcome provided limited interpretation, and so, was dropped from future models. Of all models, model 8 proved to have the best fit, given the relatively lower AIC value, higher pseudo-R squared value (compared to previous models), and LRT results indicating significant improvement over the previous model. This model indicated the combination of fewer autism discriminators (OR 0.77; 95% CI 0.71, 0.83), fewer evaluation components (OR 0.32; 95% CI 0.26, 0.40), no intellectual disability (OR 1.46; 95% CI 1.13, 1.88), no clinical autism diagnosis (OR 1.88; 95% CI 1.49, 2.38), the presence of a non-autism clinical diagnosis (OR 2.30; 95% CI 1.80, 2.98), and more associated features (OR 1.06; 95% CI 1.02, 1.10) best predicted the absence of autism eligibility. When examining specific evaluation components, post hoc analyses revealed the presence of an autism specific measure, especially those conducted in an educational setting, and the presence of an adaptive test significantly increased a child’s likelihood of autism eligibility (OR 0.13; 95% CI 0.10, 0.16; p < .001; OR 0.18; 95% CI 0.14, 0.22; p < .001; OR 0.72; 95% CI 0.55, 0.93; p = .012, respectively) (Online Table 5).

Table 4 Factors predicting the absence of autism eligibility with associated odds ratios and measures of model fit
Full size table

To understand the influence of race/ethnicity on the absence of autism eligibility, race/ethnicity was added as a predictor in model 9 to see the unique contribution, accounting for previous factors. Though in initial bivariate analyses, race/ethnicity was significantly related to the outcome with a small effect size, it did not significantly predict the outcome in the HLR. In model 9, after controlling for all previously added and significant predictors, race/ethnicity did not significantly predict autism eligibility for Asian/Pacific Islander, Black, or Hispanic groups (OR 0.95; 95% CI 0.57, 1.52; p = .822; OR 1.02; 95% CI 0.77, 1.35; p = .865, and OR 1.36; 95% CI 1.00, 1.85; p = .052, respectively).

As race/ethnicity added no significant improvement in model fit for the HLR, it was of interest to determine if other significant predictors of autism eligibility differed by racial/ethnic group. Online Table 6 presents results from analyses examining racial/ethnic group differences in child factors. PR MHI was included given demonstrated disparities among racial/ethnic groups, and its potential approximation for service accessibility, a large contributor to disparities in autism identification for CLM populations (Liptak et al., 2008; Nowell et al., 2015; Palmer et al., 2005, 2010). There were significant differences by racial/ethnic group for PR MHI, H(3) = 212.41, p < .001, number of associated features, H(3) = 25.028, p < .001, level of impairment, H(3) = 30.032, p < .001, presence of a clinical autism diagnosis, X2(3) = 29.447, p < .001, and presence of an intellectual disability, X2(6) = 56.98, p < .001. The presence of a non-autism clinical diagnosis did not significantly differ by racial/ethnic group. Given this factor’s relative importance in the HLR, it was of interest to determine if the types of non-autism clinical diagnoses differed by racial/ethnic group. Post hoc analyses revealed the type of non-autism clinical diagnosis was not equivalent across racial/ethnic groups. White children were more likely to have a diagnosis of ADHD, where all CLM groups were more likely to have a diagnosis of a developmental delay—cognitive.

Online Table 7 presents results from analyses examining differences by racial/ethnic groups in evaluation factors, including evaluation comprehensiveness and the presence of specific components. There was not a significant difference by racial/ethnic group for evaluation comprehensiveness, H(3) = 7.52, p = .057, but the presence of an autism specific measure, completed in a clinical or educational setting, was significantly different among racial/ethnic group. Examining the standardized residuals revealed differences in the presence of an autism measure anywhere in a child’s record were largely driven by the Hispanic group.

To further understand the potential influence of race/ethnicity on the diagnostic/eligibility outcome of an evaluation, autism diagnostic status documented in the record was examined for those children without autism eligibility. For children with multiple evaluations on record, the highest diagnostic status documented was compared across racial/ethnic group. Results demonstrated significant differences by racial/ethnic group in the status documented, H(3) = 11.336, p = .010. Post hoc analyses revealed disparities were attributed to significant differences between the White and Black groups, and the White and Asian/Pacific Islander groups. When examining specific diagnostic statuses, there was a higher percentage of White children with the highest status (Diagnosis of autistic disorder) documented compared to all other racial/ethnic groups, and higher percentages of the lowest status (No report of concerns) documented for Asian/Pacific Islander, Hispanic, and Black groups than White children. Notably, across all racial/ethnic groups, the diagnostic status of Autism Eligibility Rule Out, indicating autism was tested for and ruled out, was rarely documented, with only 2.4% of the no autism eligibility sample with this diagnostic statement.

Discussion

The current study expanded on prior research by examining characteristics previously identified in the literature and evaluating their individual and additive predictive ability of educational autism eligibility. Factors predicting the absence of autism eligibility included fewer autism discriminators documented, a less comprehensive evaluation, no intellectual disability present, no clinical autism diagnosis present, the presence of a non-autism clinical diagnosis, and more associated features documented. Of these factors, having fewer autism discriminators, less comprehensive evaluations, absence of a clinical autism diagnosis, and presence of non-autism diagnoses were relatively more important in predicting absence of autism eligibility. Though race was not significantly predictive when accounting for previous factors in the HLR, post hoc analyses demonstrated disparities exist within those factors most likely to predict autism eligibility. Percentile rank of median household income, number of associated features, presence of an intellectual disability, presence of a clinical autism diagnosis, and presence of an autism specific measure all demonstrated significant differences by racial/ethnic group. Thus, factors that predict the absence of autism eligibility may be underlying determinants of disproportionality in autism.

Racial/ethnic group differences in predictors may indicate children have similar behavioral concerns (Chaidez et al., 2012; Cuccaro et al., 2007), but how these concerns are interpreted and further evaluated may differ given a child’s race/ethnicity (Begeer et al., 2009; Liptak et al., 2008; Ravindran & Myers, 2012). When examining racial/ethnic disparities in diagnoses received prior to a diagnosis of autism, Mandell and colleagues (2007) suggested disparities seen could result from practitioner bias regarding the occurrence of autism in some populations. In the current study, though all children had documented behavioral deficits consistent with autism, among children without educational autism eligibility, children from CLM populations had proportionally more diagnostic statuses of “no report of autism concerns;” in other words, they were less likely to have any mention of an autism concern in their records. Practitioner bias could have influenced their likelihood to consider and assess for autism in certain racial/ethnic groups. Indeed, the evaluation experiences of children who are CLM may differ from the beginning of the evaluation process. As race/ethnicity added no unique contribution in predicting autism eligibility, but multiple predictors of educational autism eligibility differed for CLM groups, it is necessary to explore why some characteristics may be more predictive of autism eligibility than others and interpret how these characteristics may contribute to disproportionality.

Issues of Resources: Access

In the current study, the presence of a clinical autism diagnosis was an important predictor of autism eligibility in regression analyses, after accounting for predictors that may represent level of impairment. Given this, the presence of a clinical diagnosis may better serve as a proxy for family resources or service access. Results of the current study support previous research citing challenges in accessing medical services for CLM populations (Nowell et al., 2015; Zuckerman et al., 2017), as all CLM groups were significantly less likely to have a documented clinical diagnosis of autism. It is possible that caregivers of children with autism who have the resources, knowledge, and access to obtain a medical diagnosis, may also have the resources to advocate for their child in an educational setting. These results may highlight an educational setting as a critical context for autism identification for CLM populations. As some children will only ever access evaluations within the education setting (Esler et al., 2022; Pettygrove et al., 2013), it is important that educational evaluations are able to accurately identify autism, even in the absence of a clinical diagnosis.

Issues of Impairment: Child Level

Results from the current and previous studies may demonstrate practitioners can identify autism when observing overt cases, or higher levels of impairment, but struggle identifying less apparent presentations. In the current study, more documented autism discriminators (i.e., behaviors unique to and highly representative of autism) increased the likelihood of autism eligibility. Among factors examined for racial/ethnic group differences, presence of an intellectual disability demonstrated one of the largest effect sizes, with children with an intellectual disability more likely to receive autism eligibility. While in the current study, rates of co-occurring intellectual disability were higher among CLM children compared to White children, cases of autism without a co-occurring intellectual disability in CLM populations may have been unidentified. 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 CLM groups with mild/moderate autism (Jo et al., 2015). In the current study, both the autism eligibility group and all CLM groups had significantly higher clinician-rated levels of impairment, but CLM populations were not more likely to be identified with autism. As ADDM study data relies on existing evaluations, it is possible children from CLM populations with lower levels of impairment or without an intellectual disability are yet to be evaluated.

Issues of Misidentification: Evaluation Level

Best practice guidelines recommend performing comprehensive evaluations, using autism specific, intellectual, and adaptive assessments (DEC, 2014; Huerta & Lord, 2012). Results from the current investigation suggest use of all three assessments could increase the accuracy of autism identification. In the current study, children without autism eligibility were significantly less likely to have an autism measure documented. Previous research has often attributed the lack of identification to not directly assessing for autism or utilizing a standardized autism measure, especially in CLM populations (El-Ghoroury & Krackow, 2012; Mandell et al., 2007). Though a comprehensive evaluation is critical to identifying autism, results from the current investigation suggest identification of a co-occurring disorder from a comprehensive evaluation may actually decrease the chances of receiving autism eligibility. The presence of a non-autism clinical diagnosis appeared to be the strongest predictor of lacking autism eligibility in regression analyses. In addition, the presence of more associated features (i.e., behaviors often seen in autism but also in many other diagnoses) decreased the likelihood of autism eligibility. Despite the high prevalence of co-occurring disorders with autism (Lord et al., 2018), practitioners could either assess for certain disabilities and fail to further assess for co-occurring disabilities or misinterpret the features of autism as evidence of another disorder.

Though evaluation comprehensiveness overall did not differ between racial/ethnic groups, when examining the specific components, there were significant differences in the presence of an autism assessment for the Hispanic group. This aspect of the evaluation may have predicted the absence of autism eligibility for CLM populations. Consistent with previous research, practitioners may have assumptions about the likelihood of autism in certain racial/ethnic groups or in varying SES levels (Begeer et al., 2009; Cuccaro et al., 1996), and these assumptions may have contributed to different approaches related to evaluation of children’s skills. With assumptions regarding the occurrence of autism and disparities in the misdiagnoses given (Mandell et al., 2007), these could indicate practitioners are less likely to assess for autism in CLM populations. In the current study, practitioners may have been less likely to use measures that would facilitate evaluating for the possibility of autism in certain racial/ethnic groups.

Though the presence of a non-autism clinical diagnosis did not differ among racial/ethnic groups, when examining the most commonly documented non-autism clinical diagnoses, the type of diagnosis was not equivalent across racial/ethnic groups. Despite differences between the most commonly occurring non-autism clinical diagnoses between the current and previous studies (Mandell et al., 2007), results suggest practitioners may still be assigning alternative diagnoses to children who meet criteria for autism, these differ given a child’s race/ethnicity, and may influence the educational services they receive.

Improving Autism Identification

Though results of the current study do not point to one avenue to address concerns of disproportionality, they do confirm previous literature identifying both child characteristics (Maenner et al., 2013) and the evaluation process (Harris et al., 2014) as potential contributors to disparities in autism identification. The current study extended this work by identifying the unique contribution and additive effect of these characteristics in predicting disparities in autism identification. Though the potential accessibility of school services may reduce some disparities in autism eligibility by race/ethnicity, there are other factors that may help to explain why some children are not found to have educational autism eligibility when possibly warranted. These factors may be unduly influential for Hispanic populations. To address disproportionality in special education, results from the current study suggest these factors and the utilization of empirically supported practices in autism identification warrant further attention.

Improving Developmental Monitoring, Screening, and Early Intervention

A mechanism by which disparities in autism identification occur may trace back to access to early developmental screening and early intervention services. Although there are some promising improvements in early identification, historically, children from CLM backgrounds have been identified with autism later than White children (Constantino et al., 2020; Bishop-Fitzpatrick & Kind, 2017; Mandell et al., 2009; Shattuck et al., 2009; Hall-Lande et al., 2021; Hewitt et al., 2016; Wiggins et al., 2020; Zuckerman et al. 2013). Subsequently, children from CLM groups have been less likely to access to specialized autism early intervention services during the early childhood years. A variety of modifiable factors may influence this delay in access to early screening and early intervention, including systemic inequities, clinician knowledge of autism and comfort discussing developmental concerns with families, language barriers, and a lack of properly translated and culturally competent developmental screening tools (Aylward et al., 2021; Fenikilé et al., 2015; Gordon et al., 2016; McNally et al., 2020; Vanegas, 2021). To address these issues, early childhood special education (ECSE) programs, such as Part C, must continue to emphasize the importance of early developmental screening and partner with health care systems and public health to promote the connections between developmental screening and early referrals to ECSE programs (both Part C and Part B 619).

These findings highlight the practical importance of increasing access to early developmental monitoring and screening resources for young children and families with a particular focus on children from CLM backgrounds. Public health outreach campaigns such as CDC’s “Learn the Signs. Act Early” (http://www.cdc.gov/actearly) promote increased knowledge of the early signs of developmental delays and provide access to high quality, evidence based developmental resources and tools for families to monitor development (Raspa et al., 2015). Several states, such as Minnesota, have utilized Parent Connectors to connect with parents in CLM communities to promote early developmental monitoring and developmental screening to increase awareness and access to early identification and early intervention services such as Part C and Part B 619. This focus on community outreach and providing high quality developmental resources promote increased access to early intervention services directly in community-based settings that serve and support families with young children (e.g., culturally focused family organizations, faith communities, community cultural events, etc.). In turn, this outreach promotes early identification of autism and other developmental delays in CLM communities and has the potential to reduce disparities in autism identification and access.

Improving Practitioner Knowledge

Results of the current investigation emphasize the need to identify and understand differences in the documentation and interpretation of behavioral concerns for different racial/ethnic groups and to improve identification of subtler presentations of autism. The characteristics of autism and the challenges in identifying these in CLM populations have been well documented (Tek & Landa, 2012; Zwaigenbaum et al., 2015). Educators may require further training in identifying the social and behavioral characteristics associated with autism, and how these may be understood or evaluated in CLM populations (Harris et al., 2014; Harrison et al., 2017; Norbury & Sparks, 2013). Though previous research has suggested racial/ethnic groups have similar presentations of core autism symptoms (Chaidez et al., 2012; Cuccaro et al., 2007), if a practitioner has limited knowledge regarding cultural norms, they may inaccurately characterize a child’s behavioral concerns (Ravindran & Myers, 2012). Interpretation of behaviors may subsequently influence the evaluation and eventual diagnosis or educational eligibility (Begeer et al., 2009; Liptak et al., 2008; Tek & Landa, 2012). Beyond the ability to simply identify the core characteristics of autism in CLM populations, practitioners may also lack knowledge and training in evaluation practices and subsequent intervention services for students with autism (Aiello et al., 2017; Harris et al., 2020), creating either barriers or hesitancies in providing this identification.

Improving Practitioner Evaluation Practices

Though there are empirically supported guidelines for the assessment of autism, including in educational settings (DEC, 2014; Esler & Ruble, 2015; Huerta & Lord, 2012), there is evidence these are underutilized and may be contributing to disproportionate rates of autism (Aiello et al., 2017; Allen et al., 2008; Barton et al., 2016a, 2016b). Educators may be consistently under-utilizing best practices, including conducting developmental histories and caregiver interviews, as well as directly assessing adaptive, intellectual, and social or behavioral characteristics of autism (Allen et al., 2008; Barton et al., 2016a, 2016b; Esler et al., 2022; Harris et al., 2019). A recent study using data from the ADDM network found that 52% of special education evaluations for educational autism eligibility included an autism measure (Esler et al., 2022). In the current study, high percentages of children with autism had an autism measure documented somewhere in their record (93% of those with autism eligibility and 64% of those with no autism eligibility), but within educational evaluations, this dropped to 78% for children with autism eligibility and 42% for children with no autism eligibility. There are approaches that, when used in combination, may help to reduce inequities in the evaluation experience. Given predictors of educational autism eligibility identified in the current study, changes in practice may entail (a) encouraging use of a culturally and linguistically responsive caregiver interview, (b) conducting more comprehensive evaluations that integrate cultural and linguistic considerations, and (c) encouraging the use of autism specific measures with an emphasis on culturally responsive data interpretation and collection.

Limitations of the Current Investigation

One significant limitation of this study is the use of existing data collected from pre-specified sites that were not randomly obtained. Though the ADDM study provides a comprehensive look at evaluations and allows for comparisons between children with and without existing autism identification, the sample is not chosen to be representative of the population as a whole (Baio et al., 2018). As such, any interpretations regarding specific racial/ethnic groups should be made with caution. Of note, though adequate for logistic regression analyses, the Asian/Pacific Islander sample was also small in comparison to the other groups and may be less representative of this racial/ethnic group as a whole than the other groups. Use of existing data also created limitations in determining autism case status. Though clinician reviewers are highly trained, determinations are limited to the information provided in evaluation records and were not validated through in-person assessment. Practitioners performing the evaluations may have had valid reasons for excluding autism that were not documented in the evaluation record. Further, given the variety of evaluations reviewed and the complexity of the review process, there are many missing and unknown data points. Though missing data cases only made up 6% of the original sample, and these were not proportionately different by race/ethnicity, there is the possibility these cases were significantly different on some other aspect relevant to the current investigation.

A direct measure of individual level SES was not available, and thus the percentile rank of median household income for the Census block group in which a child resides was used as a proxy. Though offering some information regarding SES, this may not represent an individual child’s SES. Given the importance of SES in accessing services and likelihood of receipt of autism identification (Liptak et al., 2008; Nowell et al., 2015; Palmer et al., 2005), this may impact the outcome more than what was reflected in the current analysis.

Lastly, though all children in the sample had an educational evaluation record abstracted, it was unknown if IQ tests, adaptive tests, or behavioral features documented were conducted/documented in an educational or medical setting. Without knowing the source, this provides less information about the educational evaluation process. Only an autism assessment could be traced to the source from which it was documented. Though the source of assessments and behavioral features documented are unknown, schools had identified these children in some capacity, as they had an educational evaluation documented.

Future Directions for Research

Given the large portion of cases included in this study that were not identified with autism eligibility despite meeting ADDM case criteria for autism, there is a need for more exploratory studies such as this to better understand how to address disparities in the provision of autism services for children who would benefit. In the current study, the number of autism discriminators predicted likelihood of lacking autism eligibility, but the number documented per child did not differ by race/ethnicity. With evidence certain diagnostic features both impact identification (Maenner et al., 2013) and may show item level differences among CLM populations on autism measures (Harrison et al., 2017; Kalb et al., 2022), future investigations should examine not only the number, but types of diagnostic features examined by race/ethnicity to assess their predictive ability of autism eligibility. Examining other ways in which documentation and interpretation of behavioral characteristics may contribute to closing the identification gap warrants consideration.

Future research should include further nuanced approaches in evaluating severity and presentation of autism and how it may predict likelihood of identification. With the Individuals with Disabilities Education Improvement Act [IDEIA] policy necessitating the need to demonstrate educational impact of a disability and varying interpretations of this statement (Thomas, 2016), children without significant impairment may be under identified. To inform understanding around which children get identified and why, it may be of use to examine the decision-making process in determining when to evaluate for autism and if that disability “adversely affects educational performance”. In the current sample, the diagnostic status of autism eligibility rule out was rarely documented, suggesting whether or not there was educational impact of a disability, autism was rarely considered.

Lastly, given the importance of a clinical diagnosis in receiving educational autism eligibility, it would be important to further explore the effect this has on likelihood of autism eligibility in a variety of ways. First, though examined as a child factor, a clinical diagnosis may be a better indication of a systemic factor representing access to services. Previous research has identified White families with higher levels of maternal education and a higher SES more easily obtain a clinical diagnosis (Durkin et al., 2017; Nowell et al., 2015). To explore these as predictors of autism eligibility, it would be important for future research to characterize families obtaining both a medical diagnosis and educational eligibility of autism. Second, while previous literature has identified state policies around school-based autism evaluation, including accepting outside evaluations (Barton et al., 2016a, 2016b), it would be of use to also understand informal practices regarding outside evaluations, and how educators may weigh or consider a clinical diagnosis of autism, as well as non-autism diagnoses, when determining educational eligibility.

Conclusion

This study used multivariate analyses to move beyond examining predictors in isolation or a simplistic framing of disproportionality, and instead exploring multiple factors predicting the absence of autism eligibility, their additive effect, and their potential influence and occurrence given a child’s race/ethnicity. This analysis revealed the combination of fewer autism discriminators, absence of an intellectual disability, less comprehensive evaluations, absence of a clinical autism diagnosis, presence of a non-autism clinical diagnosis, and more associated features best predicted the absence of autism eligibility. Race/ethnicity did not uniquely contribute to the absence of autism eligibility beyond those combined factors, but many significant predictors did differ by racial/ethnic group. As such, it is important to evaluate and reduce inequities experienced within the autism identification process for populations who are CLM. Though disparate rates in special education are the observable outcome, how behaviors are documented and subsequently evaluated may be the mechanisms by which disproportionality occurs. With the knowledge that many children with behavioral characteristics consistent with autism may not receive educational services aligned to those needs, it is important to explore the evaluation process in special education to determine where inequities may arise and unknowingly contribute to disparities in autism identification.