2016 Gatlinburg Conference Poster PS-33 Title: Examining the Influence of Race and Gender on Screening for Autism Spectrum Disorder: A Self-Organizing Map Approach Authors: Marlene V. Strege, Reina S. Factor, Luke E. K. Achenie, Angela Scarpa, Diana L. Robins, Scott McCrickard Introduction: Research suggests delays and inaccuracies in autism diagnoses for racial minorities. For example, one study found that Caucasian children received a diagnosis earlier (6.3 years) than African American children (7.9 years; Mandell, et al., 2002). In addition, there may be gender differences in ASD presentation. ASD is predominately diagnosed in males. Researchers have implicated social and emotional differences as potential explanations for the imbalance in male to female diagnoses, which may result in an under-diagnosis of ASD in females (Head, et al., 2014). Following this, it would be beneficial to explore the role gender plays in ASD symptom manifestation and screening. Machine learning (ML) may serve as an optimal data-analytic approach to examine potential differences between race and gender groups. Unsupervised ML techniques utilize algorithms to "learn" from data to make predictions and detect patterns of responses within the data. The self-organizing map (SOM) algorithm is an unsupervised ML analysis that clusters data according to similar responses (Bock, 2004). Methods: The current study examined potential race and gender response differences on the Modified Checklist for Autism in Toddlers, Revised (M-CHAT-R; Robins et al., 2009). The sample consisted of four groups of toddlers (16-30 months): (1) Caucasians (N=11,156), (2) African Americans (N=2,425), (3) males (N=8,246), and (4) females (N=7,437). Toddlers determined to be at risk were referred for a comprehensive ASD diagnostic assessment. SOM (MATLAB SOM Toolbox) was applied to each group's responses separately, resulting in clusters of varied levels of risk for an ASD diagnosis within each group. Item analysis was used to further explore potential race or gender differences in symptom manifestation. Results: Clusters of varying risk levels were identified in each group. Risk level was determined based on the number of elevated ASD traits. Results demonstrated that SOM resulted in similar cluster results for Caucasians and African Americans. In both groups, three distinct risk status clusters (Typical, Low Risk, and High Risk) were revealed. The application of SOM in the male group also resulted in clusters of three levels of risk status; however, SOM failed to provide these risk differentiations for the female group, resulting in only Low Risk and High Risk clusters. In all groups, 100% of the ASD cases (determined by follow-up testing) fell into the High Risk clusters. Item analysis demonstrated overall consistent symptom manifestation across all four groups. Discussion: Similar results for Caucasians and African Americans suggest race may not be a differential factor in M-CHAT-R screening. This may indicate that race does not directly contribute to potential diagnostic differences between races, but may be the result of other factors (e.g., SES, maternal education, or access to services). In contrast, SOM results differed between males and females. Specifically, SOM was able to differentiate risk status groups between males but was not as successful in designating risk status in females. Poorer performance in females may be because females with ASD tend to show symptoms later than males (Lai, et al., 2014) often not showing clear ASD features until school age when presented with more social opportunities (Robinson et al., 2009). References/Citations: • Bock, T. (2004). A new approach for exploring multivariate data: Self-organizing maps. International Journal of Market Research, 46(2), 189-204. • Head, A. M., McGillivray, J. A., & Stokes, M. A. (2014). Gender differences in emotionality and sociability in children with autism spectrum disorders. Molecular Autism, 5(1), 19. • Mandell, D. S., Listerud, J., Levy, S. E., & Pinto-Martin, J. A. (2002). Race differences in the age at diagnosis among Medicaid-eligible children with autism. Journal of the American Academy of Child & Adolescent Psychiatry, 41(12), 1447-1453.