Examining shortened versions of the Social Responsiveness Scale for use in autism spectrum disorder prediction and as a quantitative trait measure: Results from a validation study of 3-5 year old children

Abstract

Background: The Social Responsiveness Scale (SRS) is a 65-item measure yielding a continuous score capturing autism-related traits. Scores based on SRS item subsets have been analytically examined but administration of shortened versions has not been evaluated prospectively.

Objective: The goal of this study was to compare psychometric properties of two shortened versions of the SRS to the full 65-item SRS, in young children from both a clinical and general population setting.

Methods: Study participants (aged 3-5 years) were drawn from the AJ Drexel Autism Institute clinic (n = 154) and Kaiser Permanente Northern California (n = 201) and block randomized to receive either the 16-item short SRS, a newly developed computer adaptive testing-SRS, or the published full-length SRS. Total scores across the three SRS administration methods were scaled to facilitate comparisons. Scores were plotted to assess distributional properties, while Receiver Operating Characteristic analysis was used to estimate Area Under the Curve (AUC) and address predictive ability.

Results: Overall, distributional properties of the three administration methods were highly comparable, with shortened measures demonstrating similar ability to capture the range of the distribution and case non-case separation as the full SRS. In addition, AUC values were high (0.91-0.97) and comparable across the administration methods, though there was evidence of difference in predictive ability across measures for females (AUC for full SRS = 0.99 vs. 0.84 for short). Within individual comparisons of short versus full scores (available only for participants at the general population site) suggested underestimation of actual full SRS scores with the CAT-SRS.

Conclusions: Our findings broadly support the construct validity and performance of shortened SRS versions examined here, though the full measure may be needed to more accurately assess traits consistent with ASD diagnosis in females. This work suggests opportunities for collection of ASD-related phenotype in settings where participant burden or feasibility considerations may have otherwise prohibited such measurement.

Keywords: Social Responsiveness Scale; autism spectrum disorder; quantitative traits; validation study.

FIGURE 1

Distributions of Social Responsiveness Scale (SRS) scores by study site. (A) Clinical Site (B) General Population Site (C) By Site and Autism spectrum disorder (ASD) status. Distributional plots using normal density function for full 65‐item raw SRS scores (blue lines), short 16‐item scores (red lines) scaled to full raw scores using equipercentile equating, and CAT‐predicted total raw scores (calculated as summarized in the text) (green lines). (A) Clinical site, including 154 participants referred for autism diagnostic evaluation (with 46, 58, and 50 participants in each of the full, short, and computer adaptive testing (CAT) groups, respectively) and (B) 201 participants from the general population site (with 67 participants in each of the full, short, and CAT groups) (B). Plot (C) shows scores across administration methods separated by child's ASD diagnostic status, with ASD case scores shown as dotted lines and non‐cases as solid lines. Individuals referred for clinical evaluation but not diagnosed with ASD were not included, though the density plot including these individuals was similar.

FIGURE 2

Receiver Operating Characteristic (ROC) curve predicting Autism spectrum disorder (ASD) status by Social Responsiveness Scale (SRS) version in (A) the total study population, (n = 354, including 134 ASD cases) (B) male children only, (n = 180, including 98 ASD cases) and (C) female children, (n = 174, including 36 ASD cases). As for distributional plots, full 65‐item SRS scores are shown in blue, short 16‐item scores (scaled per equipercentile equating) in red, and computer adaptive testing (CAT) scores (predicted raw total scores) in green. One individual from the CAT group is not included due to a missing CAT‐predicted score. No significant differences between AUC values for different SRS versions were observed in the total study population or male comparisons, but within females, there was a significant difference between the full AUC value (0.99) and the short value (0.84; p = 0.02).