Background
Autism Spectrum Disorder (ASD) is characterized by onset during the early developmental period and manifested by deficits in social communication, social interaction and restricted, repetitive behavior [
1]. Emerging evidence suggests that ASD traits are continuously distributed in the general population, where individuals with an ASD diagnosis represent the extreme end [
2,
3]. A diagnostic assessment of ASD, consistent with the prevailing diagnostic systems DSM-5 and ICD-10, includes developmental history as well as assessment of social and communication skills and repetitive and stereotyped behaviors [
1,
4]. It may also include medical history, physical examination, systematic considerations about coexisting conditions to establish differential diagnoses, observations in different environments and taking the child’s needs, strengths, skills, and impairments into account [
5]. In addition, assessment in multidisciplinary teams (e.g. child and adolescent psychiatrists, psychologists and speech and language therapists) is generally recommended [
6].
The absolute majority of children referred for diagnostic assessments have been identified via less specialized instances such as mandatory developmental check-ups, or through parental concerns leading to primary care visits. In addition to observation, the initial visit could encompass broadband and/or narrowband screening instruments. For ASD, a number of instruments are available such as the Modified Checklist for Autism in Toddlers (which includes 23 items, reported sensitivity/specificity: .87/.99) [
7]), the Autism Spectrum Screening Questionnaire (27 items [
8], .62/.90 [
9]), the Social Communication Questionnaire (40 items [
10], .85/.75 [
11]), the Social Responsiveness Scale (65 items, .85/.75 [
12]), and the Autism–Tics, ADHD and other Comorbidities inventory (A–TAC, 17 items, .96/.88 [
13]). These instruments range between 17 and 65 questions and even though they might be considered to be brief, there is a need for shorter and reliable instruments. An initial screening should be broad and tap into the whole child and adolescent psychiatric field in order to allocate the child to the most suitable assessment service where a full clinical investigation is conducted.
Classical test theory has widely been used to determine scale reliability by analyzing the scale as a whole with little regard to the underlying distribution or item specific characteristics. In classical test theory, the reliability of a scale is connected to the correlations between all items that constitute the scale. Cronbach’s coefficient alpha has widely been used as a measure of reliability and the alpha value can be improved by adding items or by enhancing the average inter-item correlation [
14]. Item response theory (IRT), however, offers an opportunity to examine each individual item in a scale and enhance reliability by identifying items with maximum precision. In IRT, an individual’s response to an item is assumed to be explained by an underlying (latent) trait. An IRT analysis offers an assessment of an item’s psychometric quality in relation to its position on the latent trait continuum, and its discrimination abilities on the same continuum. Thus IRT has an advantage in the ability to guide scale reduction with retained reliability and measurement precision [
15,
16].
To the best of the author’s knowledge only one article has employed IRT on a screening instrument for ASD. In this article, the Social Responsiveness Scale was reduced from 65 to 16 items while maintaining high reliability [
17]. However, the sample consisted of 21,426 individuals where >50% had a registered ASD diagnosis and the ‘controls’ were mainly siblings without a diagnosis of ASD but, most likely, with an elevated ASD symptomatology. Taken together this primarily generates information about the sensitivity but very little about the specificity, rendering the usage of the instrument limited as a screener in primary care where the entire continuum of autistic traits is represented.
The A–TAC has been part of the nationwide Child and Adolescent Twin study in Sweden (CATSS) for over a decade, and has been validated using classical test theory [
13,
18‐
22]. The purpose of this paper is to advance our understanding of the autism domain in the A–TAC, by taking the growing needs for a short and time-effective ASD screening instrument with high precision and minimal response burden into consideration. More specifically the aims of this study were to: (a) determine the psychometric properties of each item in the ASD domain in A-TAC using IRT, (b) construct a short form that could be used as a screening instrument in the general population, and (c) validate the proposed short form and determine cut-off values.
Discussion
The primary aim of this study was to conduct an IRT analysis in order to construct a valid and reliable short form of the ASD domain in A–TAC. Our main finding is that four questions can satisfactorily discriminate among subjects in the far end of the autism trait continuum, especially in children below the age of nine. The short form reported excellent previous validity and the results were in agreement with a previous validation of the full ASD domain (17 items) [
22].
The applications of the short form can be multi-faceted, even though a short form cannot be used as a diagnostic instrument. Its strengths lie in quickly identifying those individuals that would benefit from a further assessment of ASD in order to facilitate early identification of symptoms, and by extension, early diagnosis and intervention. ASD affects approximately 1% of the population [
31] and early interventions can give developmental gains, such as enhanced cognitive function and adaptive behavior [
32]. Today, it is still unclear if general screening of ASD should be incorporated in primary care [
33], however, some organizations recommended general screening for toddlers [
34,
35]. In this instance a check-up will most likely cover several areas, such as food intake, communication, language development, sleeping habits, motor skills and somatic examination. Therefore, it is important that relevant general screening instruments for ASD are time-effective and brief as well as validated in general population settings. Sweden, as well as several other countries, provides regular check-ups for children in primary care settings. Robins [
36] reported that out of 21 children diagnosed with ASD only four were identified by health-care providers as suitable for further psychiatric assessment. Taken together this suggests that screening instruments may indeed aid the identification of ASD in health care settings even in the presence of trained professionals conducting developmental surveillance. The ASD domain short form reported excellent validity before the age of 9 or 12 and can therefore be used when concerns are raised during a regular check-up, primary care visits or in elementary school settings.
The predictive validity was fair while the sensitivity values were rather low (.519 and .248) for the proposed cut-off scores. This indicates that the short form may not be optimal as a screening instrument for older age-groups in clinical settings. However, Arvidsson et al. [
37] reported a substantial decrease in autism symptom score in A-TAC for individuals who were diagnosed with ASD at ages 7–12, but with a less explicit decrease for the language module. Future research could include an examination of age-specific cut-off values and consider if language deficits should be further highlighted during assessment of ASD in older children.
In research settings brief and time-effective instrument with minimal response burden can be a valuable resource in large-scale epidemiological studies. Primarily, where the goal is to determine prevalence figures and when there is no need for an assessment that encompasses the broader phenotypic variance. Furthermore, in low-income countries where societal resources may be limited, brief and easy to access instruments that are free of charge are needed. The A–TAC is an open access instrument that can be downloaded in Swedish or English from the Gillberg Neuropsychiatry Centre website,
http://gnc.gu.se [
38], and it is also included as an appendix in Larson et al. [
13].
The primary strength in the present study is that it consists of a large population-based sample, and its linkage to the NPR, which contains best-estimate clinical diagnoses. However, the result in this study should be considered in light of some limitations. First, the scores from A-TAC were retrieved in connection with the 9th or 12th birthday and the short form should be used with awareness of possible differences between age-groups. However, the questions in A-TAC are asked in a “whole-life” frame and the respondents are asked to consider if a specific problem has been pronounced compared to peers and the questions are modelled around the DSM-IV definition of Autistic Disorder (299.00) [
27], which are general descriptions that vary greatly depending on the developmental level and chronological age. Secondly, the validation of the short form was conducted in a sample of respondents who had completed all 96 A-TAC items, thus item order bias cannot be ruled out. However, the Cubo et al. [
19] validation only consisted of the ASD-domain and came to very similar conclusions as previous articles as to why the effect of order is most likely not a major problem. A future validation of the short form should nevertheless be completed in an independent sample. Thirdly, differences in response pattern in possible sub-groups, such as boys and girls, were not examined, on the other hand the reported male-female sex ratio were at par with previous publications [
39]. Finally, the sample was based on twins and it has been argued that twins may have an increased risk for ASD [
40,
41]. However, this assumption has not been confirmed in large-scale epidemiological studies [
42‐
44] or within the CATSS [
31].
Conclusions
The ASD domain in A–TAC has the ability to discriminate among subjects in the far end of the autism trait continuum. The proposed short form, with 4 out of the original 17 items, showed excellent previous validity while the predictive validity was fair. The ASD domain short form can be a valuable instrument as a screener in primary care settings in order to identify individuals in need of further assessment and in epidemiological studies.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.