The prevalence of autism spectrum disorders: impact of diagnostic instrument and non-response bias

A large part of the variability in rates of autism spectrum disorders (ASD) across studies is non-aetiologic, and can be explained by differences in diagnostic criteria, case-finding method, and other issues of study design. To investigate the effects on ASD prevalence of two methodological issues; non-response bias and case ascertainment. We compared the findings of using a semi-structured parent interview versus in-depth clinical assessment, including an ASD specific interview. We further explored whether including information on non-responders affected the ASD prevalence estimate. A total population of 7- to 9-year olds (N = 9,430) was screened for ASD with the autism spectrum screening questionnaire (ASSQ) in the Bergen Child Study (BCS). Children scoring above the 98th percentile on parent and/or teacher ASSQ were invited to participate in the second and subsequently in the third phase of the BCS where they were assessed for ASD using the Development and Well-Being Assessment (DAWBA), and the Diagnostic Interview for Social and Communication disorders (DISCO), respectively. Clinical assessment using DISCO confirmed all DAWBA ASD cases, but also diagnosed additional cases. DISCO-generated minimum prevalence for ASD was 0.21%, whereas estimated prevalence was 0.72%, increasing to 0.87% when adjusting for non-responders. The DAWBA estimate for the same population was 0.44%. Large variances in prevalence rates across studies can be explained by methodological differences. Both information about assessment method and non-response are crucial when interpreting prevalence rates of ASD.