The practical impact of inequitable item responding on the diagnostic accuracy of screening tools for identifying Black and Hispanic children with autism. - Health disparities for under-represented racial and ethnic groups with autism have persisted for decades. For example, children who are Black and Hispanic with autism are identified at later ages than White children despite there being no biologically plausible reason for delays in diagnosis. This suggests that other systemic barriers likely contribute to lower prevalence rates and delayed identification. There is a growing awareness that developmental screening instruments could potentially contribute to these disparities. That is, many front-line screeners perform poorly when the child is Black or Hispanic (e.g., Dahl, et al, in press; Moody, et al, 2018; Rosenberg, et al, 2019). While previous research has found that these instruments are not as accurate for autism identification in Black and Hispanic children, there is no research available determining whether items within these screeners contribute to the poorer identification, or if differential item functioning (DIF) impacts the overall performance of these screeners. This project will use factor analytic and item response theory methods to identify specific items that differentially perform across Black, Hispanic, and White children on the two most widely used screening tools for children 3 to 5 years old (Social Communication Questionnaire and Social Responsiveness Scales). We will then use a recently published method to quantify the impact that DIF has on overall instrument performance. To do this we will 1) determine if the SRS and SCQ factor structures are consistent across groups; 2) use Item Response Theory (IRT) to measure test level and item level effect sizes; 3) use a new method by Gonzalez and Pelham (2021) to determine if items that have Differential Item Functioning (DIF) impact the overall performance of the instruments; and 4) conduct a sensitivity analysis to determine if race, ethnicity, socioeconomic status or other demographic variables impact the outcomes. This will be a secondary analysis of data from the Study to Explore Early Development (SEED), the largest case-control study of autism risk factors in the world. SEED has an extremely well characterized and diverse sample. Data come from six states (California, Colorado, Georgia, Maryland, North Carolina, and Pennsylvania) and includes over 7,271 children. This research will allow for more effective developmental screening and reduce early identification health disparities for Black and Hispanic children. Findings from this study will reduce the public health burden of later identified autism in Black and Hispanic populations by leading to improved screening algorithms that improve the early identification process resulting early intervention for young children.
