The 2009–2010 NS-CSHCN was a random-digit-dial telephone survey conducted by the National Center for Health Statistics. Households were the primary sampling units. In each selected household with children, a parent/guardian was administered a screening instrument to determine if any children had special healthcare needs. Special healthcare needs were defined as meeting one or more criteria in association with a long-term health condition: ongoing need for prescription medications; need for more medical care, mental health, or educational services than usual for most same-aged children; limited abilities in comparison to most same-aged children; need for special therapy, such as physical, occupational, or speech therapy; and/or having an emotional, developmental, or behavioral problem necessitating treatment or counseling [30]. If one child was identified as having special healthcare needs, s/he was selected to be the subject of a detailed parent/guardian interview. If more than one child was identified, one was randomly selected to be the subject of the interview.

The final NS-CSHCN data files include screening data on 371,617 children. Of these, 40,242 CSHCN were selected and detailed interviews were completed. The overall response rate was 25.5 %. Most nonresponse occurred due to no answer or person answering did not identify whether a child lived in the household; this nonresponse was more common for cell-phones than for landlines. Among households known to include a child with special healthcare needs, the interview completion rate was 80.8 % [31].

Of the 40,242 CSHCN with completed interviews, 38,034 were aged 3–17 years. For this analysis, we initially selected all 3025 CSHCN aged 3–17 years with parent-report of ASD. We did not include children aged <3 years because many children with ASDs are not diagnosed earlier [7, 15, 17, 20]. We excluded 296 Asian, Native Hawaiian/Pacific Islander, and multiple-race children because small sample sizes in these subgroups precluded race-specific analyses. Our final sample included 2729 non-Hispanic-white, non-Hispanic-black, and Hispanic children with ASD.

ASD was ascertained using two questions: “Has a doctor or other health care provider ever told you that [CHILD] had autism, Asperger’s disorder, pervasive developmental disorder, or other autism spectrum disorder?” and if YES, “Does [CHILD] currently have autism or an autism spectrum disorder?” Children whose parents answered affirmatively to both questions were classified as having ASD. Because 23 % of children reported to have a past ASD diagnosis were not reported to have ASD currently, this ASD definition might be conservative.

Parents of children classified as having “current” ASD were further asked to describe their child’s autism or ASD as either mild, moderate, or severe and were asked their child’s age at first autism/ASD diagnosis. No further instructions were given in regards to assigning the severity level. We initially analyzed diagnosis age as a continuous variable and an ordinal variable: 3–4, 5–8, 9–12 and ≥13 years. Because of sample size constraints, detailed analyses related to profile and predictors of diagnosis age used a dichotomized outcome measure: later diagnosis age (≥5 years, Yes/No). This cut-point was chosen based on sample size and recent US data indicating mean ASD diagnosis age was 4.4 years [17].

Parents’ perceptions of ASD severity were dichotomized as severe versus mild/moderate. Mild and moderate responses were combined because an empiric assessment indicated they were similarly associated with diagnosis age. We also assessed a second indicator of severity, parent-report of intellectual disability (ID) diagnosis (analogous question series as ASD).

Our outcome of interest was diagnosis age and the main risk/predictive factor of interest was race/ethnicity. Four racial/ethnic and primary household language groups were created: non-Hispanic-white, any language (NHW); non-Hispanic-black, any language (NHB); Hispanic-any-race, English (Hispanic-English); and Hispanic-any-race, other language (Hispanic-Other).

Other demographic characteristics included child’s age at survey, child’s sex, highest education of parent/guardian, household income relative to the federal poverty level, total number of children in household, type of health insurance, and family structure. Additionally, we assessed several co-occurring health conditions previously reported to be more common in children with ASDs: difficulty with repeated chronic physical pain (including headaches), asthma, epilepsy/other seizure disorder, allergies (any type), food allergies, and Attention-Deficit/Hyperactivity Disorder (ADD/ADHD) [32]. We did not examine several other co-occurring conditions (e.g. diabetes) due to sample size constraints.

Within each race/ethnicity/language group, ASD prevalence was estimated as the number of CSHCN who currently had ASD, divided by the total number of 3–17-year-old children who were included in the initial household screening sample. Thus, the denominator included both CSHCN and children without special healthcare needs. Because children first classified as having special healthcare needs were the subjects of the detailed interviews, information on a child’s ASD diagnosis was only ascertained on these selected children. Although prevalence estimates were weighted to be representative of the US population of non-institutionalized children aged 3–17 years, some ASD cases might have been missed if a parent did not first indicate a special healthcare need. However, independent assessments of data from other nationally representative surveys indicate that 94 % of children whose parents respond affirmatively to the ASD questions are classified as CSHCN [33]. In addition to race/ethnicity/language group, we calculated ASD prevalence within strata based on timing (age) of ASD diagnosis and severity.

We assessed mean and percentage distribution of ASD diagnosis age within strata defined based on both race/ethnicity/language and child’s age at survey. The latter stratification was needed to control for birth-cohort-related factors or differential time frames in which a child could have received a diagnosis.

The remainder of analyses were limited to children aged 5–17 years because younger children, by definition, did not have a later ASD diagnosis. We assessed ASD severity and sociodemographic factors as potential predictors of later diagnoses, and the profile of children with earlier versus later diagnoses in terms of their co-occurring health conditions.

In various analyses, we used t tests and Chi square tests to compare group differences in means and percentage distributions, respectively. We calculated unadjusted odds ratios (ORs) with 95 % confidence intervals (CIs) in which the odds of later diagnoses within each race/ethnicity/language group were compared to those for NHWs. We similarly calculated stratum-specific ORs. We tested for significant interactions between race/ethnicity/language and each severity indicator using likelihood ratio tests.

Because of sample size constraints in some subgroups, we calculated relative standard errors (RSE, standard error/prevalence estimate ×100) for each estimate and indicate those that have a moderately high variability (30–50 %) or high variability (>50–60 %). We do not present estimates where RSE is >60 %.

Adjusted odds ratios (aORs) for associations between later ASD diagnoses and sociodemographic variables were computed using multivariable logistic regression. We constructed two models: the first model included race/ethnicity/language and the aforementioned other sociodemographic factors as independent variables; the second model additionally included ASD severity and an interaction term to examine the potential effect modification between race/ethnicity/language and ASD severity level. We additionally evaluated the first model stratified on both ASD severity and co-occurring ID.

All estimates were weighted to obtain nationally representative estimates [31]. We used multiply-imputed data for child’s race/ethnicity, primary household language, highest education of parents/guardians, and household income [33]. Analyses were conducted using SAS-callable SUDAAN (Research Triangle Institute, Research Triangle Park, NC) to account for the complex sampling design.

Human subjects review was not required for this study since this was a secondary analysis of de-identified datasets.

In this nationally representative study, the estimated prevalence of CSHCN with diagnosed ASD was 15.3 per 1000 NHW children aged 3–17 years (Fig. 1). In comparison, Hispanic-English children had a similar estimate, while NHB and Hispanic-Other children had significantly lower ASD estimates.Fig. 1

Estimated prevalence of CSHCN aged 3–17 years with ASD by race/ethnicity/language, United States, 2009–2010. Racial/ethnic and primary household language groups were categorized as: non-Hispanic-white, any language (NHW); non-Hispanic-black, any language (NHB); Hispanic-any race, English (Hispanic-English); and Hispanic-any race, other language (Hispanic-Other). ^aPrevalence estimate is significantly different from NHW children

Within each race/ethnicity/language group except Hispanic-Other, diagnosis age increased with each successive increase in child’s age at survey (Table 2). Among 3–4-year-olds, mean ASD diagnosis age was comparable across race/ethnicity/language groups (range 2.3–2.5). Among 5–8-year-olds, there was only slight variation in diagnosis age, although mean age for Hispanic-English children (2.9) was significantly lower than that for NHWs (3.5). Among children aged 9–12 and ≥13 years, mean diagnosis age was comparable for NHW, NHB, and Hispanic-English children (overall means 5.0 and 6.3 for 9–12 and ≥13 years, respectively), while Hispanic-Other children had significantly lower mean diagnosis ages (2.9 and 3.7, respectively).Table 2

Mean age at first diagnosis within each race/ethnicity/language and age at survey group of children aged 3–17 years with ASD, United States, 2009–2010

Likewise, among children aged 5–8, 9–12, and ≥13 years, the percentage distributions of the ordinal categories of ASD diagnosis age were fairly similar for NHW, NHB, and Hispanic-English children (Table 3). However, 89 % of Hispanic-Other children aged 9–12 years had an earlier diagnosis compared to 50 % of NHW children, and Hispanic-Other children aged ≥13 years showed a trend (albeit non-significant) toward younger diagnosis age than NHW children (P = 0.10).Table 3

Percentage distribution^a of age at first diagnosis by race/ethnicity/language and age at survey in children aged 3–17 years with ASD, United States, 2009–2010

Overall, NHW children aged 5–17 years had the highest proportion of later diagnoses while Hispanic-Other had the lowest (Table 4). However, these findings varied by ASD severity. Likelihood ratio tests testing for interactions between race/ethnicity/language and both severity indicators, parent-reported severity and co-occurring ID, were significant. Stratified analyses illustrated the magnitude of these effect modifications. Among children whose parents perceived their severity level as mild/moderate, a higher proportion (50.8 %) of NHW children had later diagnoses than NHB (33.5 %) or Hispanic-Other children (18.0 %). However, among children with severe ASD, a lower (albeit non-significant) proportion of NHW children (16.4 %) had later diagnoses than NHB (37.8 %), and Hispanic-English (30.8 %) children; the proportions appeared fairly comparable for NHW and Hispanic-Other (12.0 %) children. A similar pattern was seen when comparing children with or without co-occurring ID. Overall, the associations between race/ethnicity/language and later diagnoses were not substantially impacted by adjustment for sociodemographic factors. However, the difference in odds of later diagnosis for NHB children with severe ASD increased slightly and was statistically significant.Table 4

Association between race/ethnicity/language and later diagnosis of ASD based on ASD severity indicators among CSHCN aged 5–17 years with ASD, United States, 2009–2010^a

In addition to associations with race/ethnicity/language, the findings from the multivariable models also indicated that children living in two-parent stepfamily households, single-mother-households, or households with other structures were more likely to have a later diagnosis than those living with two biological/adoptive parents (aOR 1.83, 95 % CI 1.35–2.48), and that children living in households with incomes <200 % of the federal poverty level were more likely to have a later diagnosis than those from higher income households (aOR 1.44, 95 % CI 1.02–2.03). Neither household members’ education levels nor child’s type of health insurance were significantly associated with later diagnosis (data not shown).

NHW and NHB children with a later ASD diagnosis were more likely to have co-occurring ADD/ADHD than NHW and NHB children with an earlier diagnosis (Table 5). NHB children with a later diagnosis were also more likely to have difficulty with repeated chronic physical pain. While the reasons varied for this condition, 16 % of NHB children with repeated chronic physical pain also reported as having sickle cell disease.Table 5

Differences in co-occurring health conditions in children aged 5–17 years with ASD based on age at diagnosis and race/ethnicity/language, United States, 2009–2010^a,b

Given the findings observed in our analyses of diagnosis age, we further assessed ASD prevalence within subgroups based on diagnosis age and severity (Fig. 2). Among children aged 5–17 years, the prevalence differential between NHW and Hispanic-Other children was much larger for mild/moderate ASD (13.9 vs. 3.0) than severe ASD (2.4 vs. 1.1). Likewise, the prevalence differential between NHW and NHB children was notable for mild/moderate ASD (13.9 vs. 8.4), while prevalence of severe ASD was comparable (2.4 vs. 2.5). The prevalence variation across race/ethnicity/language groups was also greater for those with later versus earlier diagnoses and was particularly prominent for those with later diagnoses of mild/moderate ASD.Fig. 2

Estimated prevalence of ASD by race/ethnicity/language, parent-assessed severity, and age at diagnosis among children aged 5–17 years at survey, United States, 2009–2010. Weighted prevalence estimates are per 1000 non-institutionalized children in the US population. ^aAge at diagnosis is <5 years. ^bAge at diagnosis is ≥5 years. ^c P < 0.05 two-tailed z test comparing the prevalence estimate within each race/ethnicity/language group to the prevalence estimate among NHW children