An Item Response Theory Evaluation of the Autism Diagnostic Interview-Revised (ADI-R)

Background: Early and accurate diagnosis of autism spectrum disorder (ASD) is essential for access to interventions and subsequent improvement in children’s outcomes (Itzchak & Zachor, 2011; Stone & Yoder, 2001). Yet, the Autism Diagnostic Interview- Revised (ADI-R; Rutter, Couteur & Lord, 2003), one component of the gold standard assessment, has been shown to have reduced diagnostic agreement with clinical impressions across contexts (de Bildt et al., 2015; Zander, Sturm, and Bolte, 2015). There is additional evidence that parental factors such as level of ASD concern can may be related to the sensitivity of the ADI-R (Havdahl et al., 2017).

Objectives: This study aims to determine if the age at which parents were first concerned about their child’s development influences their report of their child’s ASD symptoms on the ADI-R.

Methods: A sample of 3557 individuals with ASD (Mage= 101.15(46.25), 79% Caucasian), was culled from two large national data repositories (NDAR and Simons Simplex Collection). A categorical variable representing parental concern was created: early (first concern before 12 months of age), average (first concern between 12 and 36 months) and late (first concern after 36 months). Differences between the three time of concern groups on ADOS severity scores and age were tested using a one-way analysis of variance (ANOVA). Factor structure of the ADI-R was evaluated using exploratory factor analysis with half the sample, followed by confirmatory analyses to cross-validate findings with the second half. To compare groups on the latent variables, multiple group models were tested including non-verbal IQ as a covariate to identify items that showed statistical redundancy and confirm the overall similarity of the groups across latent traits. Differential item functioning using item response theory was then applied to examine ADI-R item bias between the three timing of concern groups following procedures of Hansen et al (2014).

Results: There were minimal group differences on ADOS severity and age between concern groups (h² = .004; h² = .02 respectively). A two-factor structure representing social communication (SC) and restricted and repetitive behavior (RRB) domains was found to best fit the data (RMSEA= .056, CFI= .958, TLI= .955).

Six items demonstrated substantial statistical redundancy (i.e., local dependence) with other items and were excluded from evaluation of factor mean differences. These items were adjacent to each other on the test form (e.g. items 30 and 31). The latent variable means for the SC and RRB factors were higher in the early concern group compared to the late concern groups (.69 and 1.22 SD higher, respectively). The same pattern was observed when NVIQ was included as covariate (.37 and 1.05 SD higher), despite no group differences on ADOS scores. Eleven SC domain items and three RRB domain items were biased (i.e., DIF) based on parental concern group.

Conclusions: This study demonstrates that the timing of initial parent concern about their child’s development influences parents’ item-level responses on the ADI-R. This suggests that excluding items with item bias by age of concern may strengthen the independent diagnostic accuracy and precision of the ADI-R.