Sex Bias in a Gold-Standard Autism Assessment: An Analysis of 19,569 Children and Adults

Authors

Abstract

Importance. Autistic females are consistently underidentified relative to males. Whether item-level sex bias in the Autism Diagnostic Observation Schedule (ADOS), the gold standard observational assessment of autism spectrum disorder, contributes to this disparity across the full developmental and language range remains unknown. Objectives. To examine differential item functioning (DIF) by sex in ADOS algorithm items across all modules. Design. Cross-sectional study. Setting. Item-level ADOS data were drawn from the NIMH Data Archive, Simons Simplex Collection, Simons Foundation Powering Autism Research for Knowledge, and Autism Inpatient Collection. Participants. Individuals with item-level ADOS or ADOS-2 data who were administered the age-appropriate module. Exposure. Sex (female vs. male) Main Outcome and Measures. Sex-based DIF was examined separately within each of 8 module and language-level combinations, covering 112 algorithm items total. For each combination, a graded response item response theory model was fit to the item set, and DIF was evaluated using likelihood ratio test approach. Results. A total of 19,569 individuals (14,807 male [75.7%]; 4,760 female [24.3%]) aged 12 months to 75 years were included. Of 112 items analyzed, eleven unique items showed significant sex DIF after correcting for type I error. No significant DIF was detected in the Toddler Module, Module 1 Low Language, or Module 2 after correction. Significant DIF was detected in Module 1 High Language (3 items) and Module 3 (1 item), with Module 4 showing the greatest number of significant items (7 of 14). Overall, adult females tended to be underscored on social communication items and overscored on items concerning restricted and repetitive behaviors and interests. All effect sizes were small except one (stereotyped language, Module 1 High Language). Conclusions and Relevance. These findings suggest that the ADOS is largely robust to sex-based measurement bias for younger individuals and those with less language, but significant bias exists for verbally fluent adolescents and adults. The domain-specific pattern of bias is consistent with female camouflaging of social difficulties and sex differences in the presentation of restricted, repetitive behaviors and interests. Clinician training and future instrument development should account for these differences to improve diagnostic equity across sex.

No cited bibliography is currently integrated for this reference. This does not necessarily mean that no bibliography exists.