The Autism and Developmental Disabilities Monitoring (ADDM) Network is an active surveillance system funded by the Centers for Disease Control and Prevention (CDC) to track the prevalence and characteristics of children with ASD in the United States. The findings from ADDM-reported data have yielded valuable insights on the changing prevalence and characteristics of ASD. However, research shows that early diagnosis and intervention of ASD improves long-term outcomes, and the American Academy of Pediatrics recommends ASD screening for all children between the ages of 18 and 24 months. Even with early screening, most children with ASD are not diagnosed until age 4. To better understand ASD in young children, the Early ADDM Network was established in 2010 to track ASD prevalence and characteristics among children aged 4 years old. The Early ADDM Network is a subset of the ADDM network and covered 7 of 13 ADDM sites, including Arizona, Colorado, Missouri, New Jersey, North Carolina, Utah, and Wisconsin, for at least one of the surveillance years included in this report (2010, 2012, and 2014).

Case determination for both the ADDM and the Early ADDM is conducted in a two-phase process using data collected from multiple sources, including children’s health and school records. Children met the case definition of ASD if their behaviors were consistent with the DSM-IV-TR criteria for autistic disorder, pervasive developmental disorder-not otherwise specified (PDD-NOS, including atypical autism), or Asperger disorder. Researchers also conducted an analysis using DSM-5 criteria, which were published in 2013, as compared to older DSM-IV-TR criteria.

Overall, ASD prevalence among 4-year-old children was 13.4 per 1,000 in 2010, 15.3 in 2012, and 17.0 in 2014—ranging from 8.1 (Missouri in 2012) to 28.4 (New Jersey in 2014). Prevalence was higher among boys at all sites, with boy-to-girl ratios ranging from 2.6:1 (Arizona and Wisconsin in 2010) to 5.2:1 (Colorado in 2014). Prevalence among white children ranged from 7.7 per 1,000 (Missouri in 2014) to 29.3 (New Jersey in 2014), among black children from 3.8 (Missouri in 2010) to 24.7 (New Jersey in 2014), and among Hispanic children from 9.1 (Arizona in 2010) to 28.2 (New Jersey in 2014). Overall, few differences in ASD prevalence were found by race/ethnicity among children aged 4 years, and those that were identified occurred in 2010 but not in later years.

For two of the sites (Arizona and New Jersey), scores on intellectual ability tests were available for all surveillance years. These data revealed that the percentage of 4-year-old children with ASD who had co-occurring intellectual disabilities remained stable over time; 47.0%, 43.6%, and 46.0% in 2010, 2012, and 2014, respectively. The proportion of children with ASD who had co-occurring intellectual disabilities was significantly higher among 4-year-olds than among 8-year-olds for all sites and surveillance years, with the exception of Arizona in 2010. The percentage of children who had received their first comprehensive developmental evaluation by 36 months of age ranged from 48.8% (Missouri in 2012) to 88.9% (Wisconsin in 2014). The median age for first ASD diagnosis ranged from 28 months (North Carolina in 2014) to 39 months (Missouri and Wisconsin in 2012). Interestingly, the overall prevalence estimate using DMS-IV-TR diagnostic criteria was about 20% higher than the overall prevalence estimate using DSM-5 criteria. The authors hypothesize that this may be due to the fact that the DSM-5 surveillance case definition requires documentation of the more extensive behavioral criteria for ASD in the DSM-5 compared with the DSM-IV-TR or an ASD diagnosis by a community provider, which 4-year-old children may not yet have had the opportunity to receive. Moving forward from the 2016 surveillance year, all Network sites have used the DSM-5 criteria.

This study provides valuable information about the prevalence of ASD in younger children. Prevalence varied widely across sites, in part due to differences in the availability of records. Prevalence among 4-year-old children was 60-80% higher in sites that reviewed both health care and education records than those that reviewed only health care records. This suggests that the public education system plays a critical role in early identification of children with ASD. Overall, ASD prevalence was found to be higher for 8-year-old children than 4-year-old children, which reflects the challenges in diagnosing younger children and the variability in how ASD symptoms can manifest as children age. Improving parental awareness of early ASD signs and symptoms, as well as encouraging public school systems to identify early social and behavioral challenges, may help identify and treat children with ASD at a younger age, thus contributing to better outcomes in the future.

To produce research results that are valid and generalizable, researchers must ensure that their study participants are representative of the entire population of interest. They need to consider accessibility of the target population, similarity among participants, and characteristics of participants that might confound results. The exclusion of a subset of a population is called selection bias. Individuals with intellectual disability (ID) are vulnerable to exclusion from ASD research studies, which can challenge the validity and generalizability of results to the full range of autistic individuals. Recent estimates suggest that 50-55% of individuals with ASD have co-occurring ID. Exclusion of these individuals from research could result in interventions that are ineffective in addressing their needs, as well as other negative consequences. It is therefore important to explore the presence of selection bias in ASD research.

The researchers in this study conducted a cross-sectional review and meta-analysis of published ASD research studies to determine the number of studies that do not include individuals with ID. They also sought to determine the quality of reporting potential selection bias in these publications. The researchers reviewed all studies published in the following ASD-specific journals in the year of 2016: Molecular Autism, Autism Research, Journal of Autism and Developmental Disorders, and Autism: International Journal of Research and Practice. Only empirical research studies were included for analysis.

The researchers identified the population of interest in each study. Studies were included that stated that their population of interest was the entire autism spectrum and that their findings were applicable to this. Studies were excluded that stated that their population of interest was specific sub-groups of ASD in terms of high or low ‘cognitive functioning.’ Studies of children below the age of 2 years were also excluded. To identify potential selection bias in each study, the researchers recorded the proportion of the ASD study population that did not have ID, as defined by an IQ above 70. They also assessed the quality of reporting selection bias by identifying the number of studies that did not report any data on the participants’ intellectual ability or did not acknowledge the potential that the generalizability of the results could be limited due to the exclusion of individuals with ID.

The resulting meta-analysis included 301 studies that claimed to include a population across the entire autism spectrum, which reflected data from 7,215,166 participants, including 100,245 participants with ASD. Out of the 301 studies, only 55% provided data on ID status and 25% specifically excluded individuals with ASD and co-occurring ID. Of the 165 studies that reported ID data, 82% showed selection bias against individuals with ID. Over half of these studies did not mention the potential lack of generalizability of their results.

Only 17% of the 301 studies reported the proportion of ASD participants who were either verbal or non-verbal. The researchers estimated that in total, 94% of the ASD study participants did not have ID and 6% did have ID. The researchers also estimated that only 2% of ASD study participants were non- or minimally verbal.

Next, the researchers reviewed publications that cited the studies that did not include ASD participants with ID. It was found that 91% of these citations erroneously referred to the original research findings as applicable or generalizable to all individuals with ASD.

This study provides evidence that ASD research tends to underrepresent individuals with ID, despite reporting results that suggest generalizability to the entire ASD population. This selection bias may occur in part because it is more challenging to recruit individuals with ID. It is nonetheless important to overcome barriers to including individuals with ID in research in order to further our understanding of all subtypes of ASD. The authors recommend the development of inclusive strategies for participants with ID and suggest that future studies that do exclude participants with ID be required to provide justification for the exclusion and report results with greater transparency.