Commentary: Advancing measurement of ASD severity and social competence: a reply to Constantino and Frazier (2013)

The Social Responsiveness Scale (SRS) is currently being used in clinical and genetic studies of autism as both a screener and as a quantitative measure of autistic traits. Our article (Hus, Bishop, Gotham, Huerta & Lord, 2013) assessed the influence of nonspecific factors on SRS scores to aid researchers in their interpretations of these scores. In their commentary, Constantino and Frazier (2013) argue that the strong influence of behavior problems on the SRS represents the overlap between neuropsychiatric syndromes, and that behavioral symptoms reflected in SRS scores ‘might actually be caused by the autistic syndrome’ (p. 1). They cite evidence for substantial overlap in genetic associations between ASD, ADHD, and other child psychiatric disorders. Our concern, however, is that if scores on a measure do not distinguish between general behavior problems and autism symptoms, questions about overlap cannot even begin to be answered. Moreover, what was not noted in the commentary was the fact that we found equally large effects of behavior problems and smaller effects of social competence on SRS scores in typical siblings, who did not have ASD. It is widely recognized that children with ASD have varying levels of general behavior problems, and that many have comorbid conditions, such as ADHD, language delay and intellectual disability (Lundstr€ om et al., 2011; Simonoff et al., 2008). However, for researchers seeking to identify causal or riskrelated genetic influences, behavioral measures not confounded by these other behaviors are needed to draw conclusions that a particular finding is contributing to specific risk for ASD. Similarly, for researchers who want to evaluate associations between particular regions of the brain and core autism symptoms or social competence, evidence that measures such as the SRS are strongly influenced by general behavior problems would seem to present a problem in identifying the specificity of that region. Thus, our goal in Hus et al. (2013) was to determine the influences of non-specific factors known to affect scores on other ASD measures (Charman et al., 2007) to provide more informed interpretations of SRS scores. Given the SRS’s widespread use in genetic and neurobiological research, we hoped that we could increase the degree to which the SRS measured social competence and ASD symptoms, and thereby extend its utility in drawing associations between behavioral phenotypes and underlying biology. In a previous study, we found this approach was useful in improving the validity of scores on the autism diagnostic observation scale (ADOS). After the introduction of the ADOS, we learned that expressive language level and age (e.g., de Bildt et al., 2004) strongly influenced raw ADOS totals. Through expressive language and age-based algorithms (Gotham, Risi, Pickles & Lord, 2007) and the introduction of severity scores calibrated by the same dimensions, we were able to improve the degree to which the ADOS domain scores represent autism severity within the context of a clinical observation. The calibrated severity scores in the ADOS (Gotham, Pickles & Lord, 2009; Hus, Gotham & Lord, 2012) allow researchers to expand the boundaries of constructs of social-communication deficits and repetitive behaviors and quantify difficulties across an interval scale, rather than simply providing categorical cut-offs – contributing to the need for dimensional measures highlighted by Constantino and Frazier (2013). We approached the current analysis of the SRS from the same point of view. The SRS is particularly valuable because it provides a range of scores even within a typical population. We wondered if we could better understand what child factors could be controlled to make the SRS a more specific measure of social competence. Recently, Duku et al. (2012) took on a similar task with the SRS, resulting in selection of a subset of 30 items intended to measure social impairments in preschool children, but which still correlated with the CBCL internalizing and externalizing scales, r = 0.65–0.68. We started by looking for how the SRS related to the social domain of the Vineland Adaptive Behavior Scales, a parent report measure widely used as a measure of social competence (Gillespie-Lynch et al., 2012; Klin et al., 2007). We chose social competence because the SRS is commonly referred to as a quantitative measure of social reciprocity that