The Predictive Value of AQ and SRS-a in Adults with Suspected ASD

Background:

ASD questionnaires are often used as screeners in the assessment of adults with suspected ASD. Their diagnostic value depends on their capability to properly assess the likelihood that the disorder is present (sensitivity: true positives) or not (specificity: true negatives).

In clinical practice, the AQ (Autism Quotient, open source) and the SRS-A (Social Responsiveness Scale - Adults, paid source) are most often used as tools for quantitative autism assessment. Their predictive value for diagnostic classification has seldom been compared.

Objectives:

The aim was to study the predictive value of the AQ and the SRS-A for diagnostic classification in a general mental health care population.

Methods:

The adult patients, who were referred for ASD assessment, filled in both an AQ (max score = 50) and a SRS-A (max score = 192) self-report at the beginning of the diagnostic process. An independent researcher scored these questionnaires. The results remained unknown to the diagnostician and the patient until after the ASD diagnosis was officially confirmed or rejected, resulting in an ASD-group and a non-ASD group. In total, there were 92 participants (M = 33.51 years, SD = 12.33), of which 68% received an ASD diagnosis. The ASD and non-ASD group did not differ on important characteristics, such as age (M = 33.68 vs. M = 33.14) and gender ratio (1.3:1 vs.1.4:1). T-tests and ROC-analyses were performed.

Results:

The ASD group had significant higher scores than the non-ASD group on both the AQ (M = 29.17, SD = 7.75; M = 20.97, SD = 8.13 respectively; t = 4.65, df = 90, p < .001, Cohen’s d = 1.03) and SRS-A (M = 70.87, SD = 10.76; M = 63.59, SD = 12.56 respectively; t = 2.86, df = 90, p < .01 Cohen’s d = 0.62). The correlation between both questionnaires was high (r = .80, p < .001).

The ROC-analysis for the AQ yielded an AUC of .78 (p < .001) for ASD vs. non-ASD. A cut-off score of 26 (as recommended for clinical use, but also determined by the best Youden’s Index in this research) had a sensitivity of .76 and a specificity of .72 for ASD. The ROC analysis for the SRS-A yielded an AUC of .69 (p < .01) for ASD vs. non-ASD. A cut-off score of 70 (determined by the best Youden’s Index) had a sensitivity of .63 and a specificity of .72 for ASD.

Conclusions:

Both questionnaires could differentiate between the ASD and the non-ASD group. However, the AQ had a higher effect size (large vs. medium), a better predictive reliability (moderate vs. poor) and a higher sensitivity. Thus, based on the current results, the AQ seems to be superior as a screening tool for general mental health care patients with suspected ASD. However, replication studies are needed before advising which one to use for clinical practice. Also, note that questionnaires are not intended to be diagnostic in itself. If there are clinically significant levels of autistic traits, a comprehensive diagnostic evaluation is warranted.