The Autism Diagnostic Interview- Revised (ADI-R) is a standardised semi-structured interview that provides a framework for the developmental history needed when considering diagnosis of an Autism Spectrum Disorder (Lord et al, 1994; Rutter et al, 2003). The ADI-R has also been widely used as a measure of the autism phenotype in research studies and as a criterion in validity studies of other measures. The published version provides a diagnostic algorithm for ICD-10 childhood autism only, with no algorithm for the broader diagnosis of ASD. In 2013 DSM-5 will be published with new diagnostic criteria for ASD.
Objectives:
This paper presents the findings from an exploratory factor analysis, proposes algorithms for ASD and considers the impact of the new DSM 5 criteria on current clinical and research practice.
Methods:
A dataset was collated from nine clinical academic and research centres (8 UK; one US). The subjects were aged from 18 months to 19 years at time of assessment. The combined dataset includes children referred for diagnostic evaluation for ASD, or recruited to genetic or intervention studies, children attending speech and language specialist education, children with conduct disorder and a school based general population study. Exploratory Factor Analysis was conducted using MPLUS, to investigate the underlying factor structure of the ADI-R and the results informed the selection of items to create an algorithm for the diagnosis of ASD. Further analyses were undertaken to investigate the performance of the new algorithms with respect to the child’s original research descriptor, age, gender, language and cognitive ability.
Results:
Complete data on 873 participants were used to investigate the underlying factor structure of the ADI-R. Additional samples of 270 individuals with autism and 92 typically developing children were available for external validation. The factors were rotated with a PROMAX transformation in order to maximise the contrast in factor loadings. Models with 2 factor solutions with Root Mean Square Error (RMSEA) of 0.052 for non-verbal and 0.050 for verbal cases were the best fit. This 2–factor solution was used to design the ASD algorithms. Threshold cut-off scores were identified using Receiver Operating Characteristic (ROC) curves to maximise sensitivity and specificity. Algorithm cut-offs achieved good discrimination for children with a research diagnosis of ASD.
Conclusions:
A 2- factor structure, with domains labelled “Social Communication” and “Stereotyped Speech, Rigidity and Repetitive Behaviour”, was identified using ADI-R items and new algorithms were successfully developed with high sensitivity and specificity for autism and ASD. The implications of these findings will be discussed in relation to the DSM5 criteria for ASD and the wider impact on diagnosis for children and young people from preschool to older adolescence.
This work was funded by the Health Foundation with additional support from Northumberland Tyne and Wear NHS Foundation Trust flexibility and sustainability funding .
See more of: Clinical Phenotype
See more of: Symptoms, Diagnosis & Phenotype