Developmental Profiles of Young Children Referred for Concern for Autism Spectrum Disorder: DBPNet Study

Children ages 18 months-5 years, 11 months referred to academic medical center clinics for a concern for possible ASD were consecutively enrolled between May 2019 and February 2020. This study was approved by the Institutional Review Boards of the Children’s Hospital of Philadelphia (Developmental-Behavioral Pediatrics Research Network (DBPNet) coordination center and the 8 participating sites 7 DBPNet and 1 Austrian site each of which provides subspecialty assessment for children with ASD. Participants were excluded if they had a prior multidisciplinary team diagnosis of ASD, or the child was not English speaking (not German speaking at the European site). Clinicians were board eligible or certified in DBP (Developmental-Behavioral Pediatrics) and/or NDD (Neurodevelopmental Disabilities) or physician specialists trained to diagnose and treat ASD at the European site. Using usual clinical protocols specific to each site, a clinical assessment was completed that included a detailed history and demographics (age, sex, race/ethnicity, caregiver education level, and insurance type), physical examination, behavioral observations, an Autism Diagnostice Observation Schedule (ADOS-2) and site-specific selection of other measures. These evaluations included the some of the following measures: (1) cognitive (Bayley Scales of Infant Development (Albers & Grieve, 2007), Mullen Scales of Early Learning (Mullen, 1995), Differential Ability Scales (Elliott et al., 2018), Stanford–Binet Intelligence Scales (Roid & Pomplun, 2012), Wechsler Preschool and Primary Scale of Intelligence (Wechsler, 2002), or Wechsler Intelligence Scale for Children (Wechsler, 2003)); (2) language (Preschool Language Scales (Zimmerman et al., 2011), Receptive–Expressive Emergent Language Test (Bzoch et al., 2003), Expressive One-Word Picture Vocabulary Test (Michalec & Henninger, 2011), Receptive One-Word Picture Vocabulary Test (Martin & Brownell, 2010), or Clinical Evaluation of Language Fundamentals (Wiig et al., 2013), and (3) adaptive (Vineland Adaptive Behavior Scales (Sparrow et al., 2016) or Adaptive Behavior Assessment System (Harrison & Oakland, 2015)) domains. Since this study was embedded in naturalistic clinical assessments, the types and the number of tests administered varied by site. To compare results across disparate scales with each domain, each of these measures were collapsed into categories based on standard score ranges: average to above average (standard scores > 84 to maximum), borderline (standard scores > 69 to 84), mild impairment (standard scores > 54 to 69), moderate impairment (standard scores > 39 to 54), and severe/profound impairment (standard scores ≤ 39 to minimum) based on tool normative data. DBP clinicians made a Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) (DSM-5) criteria based diagnosis (ASD-yes or ASD-no) based on all the available data listed above and the results of an Autism Diagnostic Observation Schedule, 2nd Edition (ADOS-2) (Lord et al., 2006, 2012). Included in the analysis are all the children who had at least one data point for a developmental domain. Each analysis used fully available data using pairwise deletion in that a subject is not deleted from an analysis if they have a missing value in one variable because they may have a valid value in another variable (in multivariate models).

Descriptive statistics characterize participant demographics. Developmental profile frequencies were compared between groups in the domains of cognitive, language, and adaptive function. Different instruments were used across sites. In order to establish homogeneity of domain functioning across sites and enable meaningful comparisons between ASD and no ASD groups, quantitative rating scores were converted into five categorical ratings ranging from "0" reflecting severe/profound impairment to "4" reflecting average to above average function. The categorical rating scale was adopted in which participants were rated as follows: 0 = severe/profound impairment, 1 = moderate impairment, 2 = mild impairment, 3 = borderline, and 4 = average to above average. The Mplus (https:www.​statmodel.​com) and Stata (https://​www.​stata.​com) packages were utilized in the present report. The level of significance was set to 5% for a two-tailed test.

Since specific assessments were administered based on clinical practice patterns at each participating site, data on each developmental domain were not available for all participants. To ensure that missing data across assessments did not bias the diagnosis status, we conducted two analyses. First, we conducted a cross tabulation between the group of children that had complete assessments (for all three domains of functioning) and contrasted them with the group that had missing data on any one of the assessments. Next, we contrasted 2X5 tables for diagnosis (2 levels) and level of functioning in cognitive, language and adaptive domains (5 levels) for the two levels of a third variable, defining missingness. In other words, we tested for the homogeneity of findings in the 2X5 tables for children with complete data versus those who had missing values in any one of the assessments.

Among the 349 children enrolled, there were 250 with ASD and 99 no ASD. Overall, these groups showed no differences in race, ethnicity, or gender (Table 1). Participating children were from diverse backgrounds, with 30% of the no ASD and 42% of the ASD groups being none White. Of the entire sample, 19% were of Hispanic ethnicity. A male predominance was seen in both groups as is typical among children with developmental concerns. Those without ASD were more likely to be publicly insured, [63% vs 49%; p = 0.021] and older [42 vs 39 months; p = 0.042].

As shown in Table 2, there are significant differences in language and adaptive skills between the two group (Language chi-square 21.258 p < 0.001; Adaptive chi-square 19.386 p < 0.001). There was no difference in cognitive skill (Cognitive chi-square 6.282 p = 0.1790).

Further investigation of between group differences focused on a series of category curve tests. Developmental profile frequencies differed between groups (Fig. 1). These tests contrasted the frequency of responses between groups using omnibus chi-square tests followed by pairwise z-tests. The latter was corrected using a Bonferroni correction. As shown in Table 2, there were no significant differences between groups in cognition, although in the no ASD group there were higher occurrence of “average and above average” responders compared to the ASD group (No ASD 56.1% vs ASD 38.7%. In language, however, the no ASD group had a significantly higher presence of “average and above average” responders and the ASD group a higher prevalence of “moderate impairment” compared to the no ASD group. Finally, there was a significantly higher prevalence of “moderate impairment” and “mild impairment” in the ASD group in adaptive functioning, with a higher prevalence of “borderline” and “average and above average” functioning in the no ASD group.

Differences between developmental domains were compared within the ASD and no ASD groups to assess for patterns. As shown in Table 3, there were no differences between any 2 areas of function in the no ASD group. In the ASD group, cognitive function was significantly above both language function and adaptive function; however, language function was comparatively lower than adaptive function but not significantly.

Overall, there were complete developmental domains assessment results for 28.4% of the sample. To ensure that the entire sample represented consistent profiles, missing data analysis was conducted. First, results pointed to no significant differences between children who had complete versus missing assessments with regard to their ASD diagnosis [(Chi-square (1) = 3.482, p > 0.05], thus, the pattern of results remained the same regardless of missingness. Finally, there were 15 statistical tests evaluating the homogeneity of the two samples (full data vs. missing) with two significant effects pointing against the conclusion of homogeneous samples. These two significant effects represented 13.3% of the total number of comparisons, slightly above the alpha level. Given the large number of tests conducted, we adjusted the p-value using the Benjamini–Hochberg correction using a false discovery rate equal to the level of significance a = 5%. Results indicated that the two significant effects were no longer significant after applying the correction for family-wise error. Thus, we conclude that there were no systematic effects of missingness in that samples with full and missing cases were largely equivalent regarding the substantive findings of the study for each of the three developmental function domains.