Can Self-Report Questionnaires Screen for Autism in Adults? Comparison with ‘Gold Standard' Diagnostic Assessments

Background: The prevalence of autism spectrum disorders (ASDs) is increasing. This has resulted in a growth in demand on clinical services, with more people referred for diagnostic assessment. Diagnosing ASD is expensive and time consuming, so there is a need for reliable, cost-effective screening instruments. The UK National Institute for Health and Care Excellence (NICE) has recently recommended using the Autism-Spectrum Quotient-10 item (AQ10) self-report questionnaire as a rapid screening tool for ASD in adults. However, while the AQ10 has been shown to discriminate between already-diagnosed ASD populations and healthy control groups, its performance as a predictor of ASD within real-world clinical populations has not been investigated.

Objectives: Our goal was to determine whether the AQ10 is able to predict ASD caseness in adult clinical settings. We also assessed the original, longer version of the AQ questionnaire, the AQ50, which has also been used as a screening tool.

Methods: 730 participants (548 males, 182 females, mean age 31.2) presenting to outpatient specialist services (N=620), inpatient psychiatric units (N=44) and primary care settings (N=66) completed the AQ. Participants were assessed for ASD using so called ‘gold standard’ instruments, the Autism Diagnostic Interview Revised (ADI-R) and the Adult Diagnostic Observation Schedule Generic (ADOS-G). The AQ10 and AQ50 were evaluated as predictors of meeting standard criteria for ASD caseness on i) the ADI-R and ii) the ADOS-G. We first calculated sensitivity, specificity and Youden’s J (a measure of accuracy) of the AQ10 and AQ50 as dichotomous predictors, using recommended cut-offs. Secondly, Receiver Operating Characteristic (ROC) curve analysis was used to determine optimal cut-off scores.

Results: The AQ10 and AQ50 demonstrated poor informative power across all three patient samples and all comparison measures (Youden’s J= 0-0.24) with higher sensitivity (0.33-0.87) than specificity (0.18-0.63). ROC curve analysis revealed that the ability of the questionnaires to predict ADI-R and ADOS-G caseness was in most cases statistically significant (AQ50: p=0.025, p=0.001 respectively; AQ10: p=0.089, p<0.001 respectively), but modest (area under curve, AUC 0.55-0.61, where 0.5 is chance). Furthermore, optimal performance was seen with higher cut-offs than those currently recommended (maximum Youden’s J at AQ50: 36-37 and AQ10: 8).

Conclusions:  For the first time, we have shown that scores on the AQ10 and AQ50 self-report scales predict caseness in an adult real-world clinical population. However, both screeners performed only slightly better than chance. Increasing the cut-off score above the recommended values improves psychometric properties, yet, this still only provides modest performance. Further work is needed to validate the AQ10 and/or AQ50 as clinical screening tools.