DSM-5 and autism
As with DSM-IV, the task of revision for DSM-5 was taken seriously by the overall work group and members of the committee on neurodevelopmental disorders. Several key executive decisions adopted for all of DSM-5 have had serious implications; these are the decision to eliminate subthreshold categories and the high reliance on diagnostic instruments as a source of criteria and as proof of validity [10]. Issues more specific to autism include the nature of the revision and decisions about how best to outline and organize the proposed criteria and evaluate them.
Some aspects of the DSM-5 approach to autism appear to be well reasoned and quite valuable; for example, the move to a better description of the class of disorder (autism spectrum disorder to replace PDD) and the use of dimensions in combination with categorical approaches. Some of the more practical problems, however, probably arise in the context of the revision process, and the final product has been the subject of much debate [11, 12].
Early on, the decision was made to drop the multiple disorders included in DSM-IV in favor of a single autism spectrum term. A second disorder, social (pragmatic) communication disorder, was added, although its relation to the autism spectrum disorders (ASD) remains unclear (ASD must be ruled out in the diagnostic criteria, but prevalence estimates of ASD have included social (pragmatic) communication disorder). Also unclear is the application and use of this diagnosis in practice and the types of services to which an individual might be entitled. In combination with the use of better specifiers of dimension it was hoped that clinical needs would be well served. Given its single-gene etiology, a decision was also made to remove Rett’s disorder from DSM, although an individual with this genetic condition meeting diagnostic criteria for ASD would still receive an ASD diagnosis probably with a specifier. This is a complicated precedent given the many strong leads it has provided in uncovering the genetics of autism. While a case clearly could have been made for refining the Asperger’s label, the work group elected to eliminate it as a category along with childhood disintegrative disorder. In some respects both moves are controversial, particularly given the inconsistency with which the Asperger’s diagnosis has been utilized (itself a problem but potentially one obscuring a potentially important clinical distinction). Based on a factor analysis of a large body of data from diagnostic instruments a decision was made to collapse social and communication features into a single category and then to have a second category more consistent with Kanner’s ‘insistence on sameness’/restricted interests package with the addition of a sensory sensitivity symptom, which had poor specificity in the DSM-IV field trial [7, 13].
Factor analysis methods have their uses and limitations – depending at the most basic level on which data are included in the analysis and how the analysis is constrained. For the DSM-IV field trial either a two-factor, three-factor or five-factor solution could be derived; others performing similar analyses have noted the complexity of these approaches [14]. The final decision for DSM-IV and ICD-10 to keep the traditional three categories (dating back to Rutter’s 1978 definition [15]) was made partly for reasons of historical continuity and, strategically, having three categories of criteria gave many different combinations of criteria that would yield an autism diagnosis (well over 2,000).
There is no question that social and communicative features are very closely related, but the problem in combining them into one category results in many fewer potential criteria combinations. Another factor contributing to the decreased combinations of symptoms is the move back to a monothetic approach for the social communication domain, where instead of two out of four criteria and one out of four criteria required in the DSM-IV, three out of three criteria are required in the DSM-5. A polythetic approach was retained for the repetitive and restrictive behavioral domain, although the number of symptoms needing to be met was increased from one out of four (zero of four potentially in PDD-NOS) to two out of four. We are not advocating for one approach over the other; they each have advantages and disadvantages and are often used together for diagnostic criteria [16]. Rather, we are illustrating the impact that this decision is likely to have on the composition of the autism spectrum, which might more closely resemble the more classic autism described by Kanner [4] than the broader autism spectrum that might be captured with polythetic criteria. An additional consequence of requiring all three social criteria could be the delayed diagnosis (and consequentially the delay of intervening) for children whose symptoms do not fully manifest until social demands increase.
Data from a large series of well-characterized cases were used to produce the draft DSM-5 criteria for the single new autism spectrum disorder. Two research diagnostic instruments (one a parent report measure and the other observational assessment) were used [17]. The authors rightly noted that this was not a field trial, and their results suggested that when both the parent interview and child assessment were conducted, sensitivity/specificity were maximized; however, in the absence of both, specificity fell. Their data suggested that no more than about 10% of cases would lose their diagnosis. Other data on reliability were also provided from a field trial focused only on this issue [18], although the overall approach to the DSM-5 field trial has also been criticized [19, 20]. Other data using large datasets have also provided some support to the approach undertaken [21]. Given the data available and the major effort undertaken, then, what are the potential problems?
Issues in use of DSM-5
Despite the name change to autism spectrum disorder, the concept actually proposed is apparently more restricted than the DSM-IV approach. A series of papers using different approaches and different samples suggest that the issue may be more extensive than would otherwise be thought. If the results of these studies are realized, there would probably be large implications for service eligibility and research for individuals currently receiving support for the disorder. It is important to note in moving DSM-5 into what may be more real-world clinical settings that practitioners will not likely have had the extensive training in diagnostic instruments.
The results of most of the relevant independent studies can be succinctly summarized. It is important to note that in some cases studies were conducted using an earlier version of DSM-5 and that diverse methods and samples were used. Mattila and colleagues used a slightly earlier draft of DSM-5 to assess agreement with DSM-IV [22]. In this epidemiological study a very large sample of 8-year-old children were assessed using the Autism Spectrum Screening Questionnaire and 110 were then seen for more in-depth assessment. The investigators noted that DSM-5 was less sensitive than DSM-IV. Comparisons were also made between DSM-IV and DSM-5, showing individuals with higher IQs being less likely to meet the new diagnostic criteria.
Similarly, Worley and Matson compared symptoms of ASD in several hundred children using DSM-IV and DSM-5. Significant differences were noted in terms of core domain scores on socialization/communication between DSM-IV and DSM-5 [23]. In both cases the groups had significantly higher levels of dysfunction than a control group and the number no longer meeting criteria in DSM-5 was noted to be a potential problem for both clinical service provision and research; for example, relative to epidemiological or longitudinal studies.
Frazier and colleagues evaluated proposed DSM-5 criteria in a large sample of siblings (some with ASD and others without ASD) [21]. They noted that in this sample of children (ages 2 to 18) the specificity of DSM-5 was higher than that of DSM-IV while sensitivity was lower and that relaxing the diagnostic threshold might improve the approach.
McPartland and colleagues reanalyzed a large sample of cases selected from the DSM-IV field trial [24]. Sensitivity and specificity were systematically assessed using a symptom checklist approach to cross-walk DSM-IV criteria to DSM-5. The specificity of DSM-5 was high (94.9%) but sensitivity varied dramatically by clinical group (varying from 0.76 in autism to 0.25 in Asperger’s disorder and 0.28 in PDD-NOS) and by cognitive ability (IQ <70 = 0.70; IQ ≥70 = 0.46).
Mattson and colleagues examined alternative approaches to improve DSM-5 [25]. They evaluated two potential modifications for toddlers, with some degrees of overall improvement but with significant numbers of toddlers apparently left without eligibility for services. They noted that while excluded from the diagnostic categories these toddlers continued to exhibit significant impairment.
Gibbs and colleagues compared DSM-IV-TR and DSM-5 diagnosis in a sample of 132 children [26]. Of the 111 who had received a diagnosis of autism or related PDD in DSM-IV-TR, 26 did not meet the criteria in DSM-5; most of those excluded from ASD in DSM-5 would have received a diagnosis of PDD-NOS in DSM-IV-TR.
Taheri and Perry reviewed over 130 cases of children with previous diagnosis of autism or PDD-NOS and found that about 60% met the new DSM-5 criteria (81% of those with autism but less than 20% of those with PDD-NOS) [27]. They also noted a significant relationship to IQ, with more able cases more frequently losing a diagnosis.
Most recently, Wilson compared DSM-IV, ICD-10, and DSM-5 in a sample of 150 adults with ASD who were more cognitively able [28]. The author noted that about 56% of those meeting ICD-10 also met DSM-5 (although nearly 20% of those not meeting criteria for ASD met DSM-5 criteria for social communication disorder). They noted that this might be an important practical problem in terms of access to service and suggested modification either in the diagnostic threshold (reducing the number of criteria required) or in giving greater leniency to uncertain criteria (allowing them to count). While the first of these proposals would be relatively easy to implement, the second poses significant difficulty. As the authors emphasized, exclusion from clinical services is a potentially significant problem.
On the other hand, Mazefsky and colleagues compared DSM-IV and DSM-5 diagnosis using the research instruments on which the new DSM-5 approaches is based [29]. They found that 93% of the nearly 500 high-functioning participants met the criteria for autism in DSM-5 but that this number was lower when only the parent instrument used and lower still if only the individual assessment was available. While reassuring from the point of view of research, if both instruments are available it is worrisome that, in practice, for adults a parental informant may not be available.