Race and Sex Bias in the Autism Diagnostic Observation Schedule (ADOS-2) and Disparities in Autism Diagnoses

Abstract

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental condition that affects an estimated 1 in 44 children in the US, with a male-to-female ratio of approximately 4:1. 1 The prevalence of diagnosed ASD has increased substantially over the past 2 decades, although this increase in prevalence has been greater in certain demographic groups, such as female patients 2 and those from minoritized racial ethnic groups, 3 suggesting the presence of diagnostic disparities by race and sex. 4 Notably, the prototypical behavioral manifestations of ASD (on which existing diagnostic criteria and standardized diagnostic instruments are based) were derived from samples of children who were predominantly White and male, 4 and, thus, systematic biases in the diagnostic tools used to evaluate individuals with suspected ASD could theoretically contribute to observed diagnostic disparities.Kalb et al 5 sought to investigate this issue by quantifying the magnitude and practical impact of race-based and sex-based bias in the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2), a clinician-administered measure that is widely used in both research and clinical practice to establish or confirm a diagnosis of ASD. Leveraging ADOS-2 data from 6269 youth attending specialist diagnostic evaluations at a university-based ASD clinic, the authors 5 quantified the degree to which different versions (ie, modules) of the ADOS-2 systematically underestimated the features of ASD in Black/African American children (vs White children) and female children (vs male children).Although their analysis did reveal significant race-based bias in 8 items and sex-based bias in 5 items, 5 estimated effect size metrics 6 indicated that for all but 2 items (D4, Repetitive Interest [racebased bias], and D2, Hand Mannerisms [sex-based bias]), these effects were small and unlikely to be of practical significance. Moreover, for all ADOS-2 modules tested, the maximum difference in expected ADOS-2 total scores attributable to measurement bias was less than 1 scale point (range, 0.07-0.91 point). 5 Overall, these findings indicate that the degree of race and sex bias present in the ADOS-2 is low and unlikely to contribute to the systematic underdiagnosis of ASD in Black or female children.A major strength of the study by Kalb et al 5 is its innovative use of item response theory (IRT) models to examine the ADOS-2 items and quantify the degree of bias (or differential item functioning [DIF] in IRT terms) between demographic groups. IRT is a modern, large-sample, psychometric method used to develop, evaluate, and score psychological tests, 7 and IRT models provide mathematical descriptions of how certain item responses (eg, the endorsement of a symptom on the ADOS-2 as 0, 1, or 2) relate to unmeasured latent variables (ie, ASD severity in the case of the ADOS-2) that are assumed to underlie all items on the scale. Within an IRT framework, researchers are able to test for DIF between groups by examining whether a given item is differentially relate...

Accuracy of initial diagnostic impressions of autism in toddlers and behaviors that inform these impressions

“…Future research should examine whether and how these child and family characteristics relate to clinicians' impressions. Replication is especially relevant given the racial and ethnic disparities in the time to diagnosis for Black children with ASD (Aylward et al, 2021; Mandell et al, 2009; Williams, 2022). Additionally, future research should examine the behaviors that clinicians use when forming inaccurate diagnostic impressions in larger samples of female versus male children, given the possibly different presentation of ASD in very young boys and girls, which may contribute to delayed diagnoses in female individuals (Carter et al, 2007; Ros‐Demarize et al, 2020).…”