Early studies focusing on the DSM–5 criteria largely found a concerning loss of sensitivity compared with the DSM–IV–TR criteria. Typically this drop of sensitivity was found in the context of good levels of specificity (between 0.94 and 1.0). For example, Mattila et al. (2011) and Gibbs, Aldridge, Chandler, Witzlsperger, and Smith (2012) reported that strict application of the draft DSM–5 criteria resulted in reduced diagnostic sensitivity for children (a loss of 54 % and 23 % respectively) when compared with DSM–IV–TR. Other studies that applied the two sets of criteria to the same individuals similarly found reduced sensitivity in at-risk toddlers (Matson, Kozlowski, Hattier, Horovitz, & Sipes, 2012) and adults with intellectual disability (Matson, Belva, Horovitz, Kozlowski, & Bamburg, 2012). A more recent study explored this further and reported a loss of 23 % sensitivity for DSM–5 compared with DSM–IV–TR in a sample of children aged between 16 months and 18 years, but with some preliminary evidence suggesting that sensitivity may be lower for those children under the age of 30 months (Harstad et al., 2015). Importantly, this study also suggested that children who met criteria for DSM–IV–TR autistic disorder were significantly more likely to meet criteria for DSM–5 ASD than those who met DSM–IV–TR criteria for PDD-NOS or Asperger syndrome.

The finding of comparatively reduced sensitivity of the DSM–5 criteria to PDD-NOS and Asperger syndrome compared with ‘core’ autism is consistent with previous evidence suggesting that the descriptions may be too narrow to capture the full autism spectrum. For example, a study by Mayes, Black, and Tierney (2013) reported excellent sensitivity of the DSM–5 criteria for both high and low functioning autism groups in two samples of children, but poor sensitivity (between 0.20 and 0.28) for children who met DSM–IV–TR criteria for PDD-NOS. In a similar study, Gibbs et al. (2012) reported that the majority of children in their sample who did not meet criteria for DSM–5 ASD had received a diagnosis of DSM–IV–TR PDD-NOS, while McPar tland, Reichow, and Volkmar (2012) reported low levels of sensitivity for children meeting DSM–IV–TR criteria for Asperger syndrome or atypical autism (including PDD-NOS). This was explored further in a study of data collected with both children and adults conducted by Young and Rodi (2013). They reported that none of the individuals who had received a DSM–IV–TR PDD-NOS diagnosis and just 56.1 % of individuals with Asperger syndrome met criteria for DSM–5 ASD, compared with 73.7 % of those with autistic disorder.

As well as varying according to diagnostic subgroup, research has also looked at the sensitivity of the new criteria in individuals with different ability levels. This work has suggested that the sensitivity of the DSM–5 criteria may vary as a function of IQ. In the study previously described by McPartland et al. (2012), only 46 % of those who met criteria for DSM–IV–TR PDD with an IQ above 70 met the DSM–5 criteria for ASD. In another example, Taheri and Perry (2012) found that while only 22.2 % of their sample who had an IQ above 70 met criteria for DSM–5 ASD, 89.7 % of individuals with an IQ below 40 met the criteria. Not all studies investigating IQ, however, have found a significant effect; while Harstad et al. (2015) reported only a trend indicative of lower sensitivity for higher ability individuals, Young and Rodi (2013) found no significant relationship between IQ and meeting criteria for DSM–5 ASD. Moreover, a recent meta-analysis found that while the sensitivity of the DSM–5 criteria may be reduced for DSM–IV–TR autistic disorder and PDD-NOS, it was not significantly reduced for individuals who met the DSM–IV–TR criteria for Asperger syndrome (Kulage, Smaldone, & Cohn, 2014).

According to DSM–IV–TR, a diagnosis of PDD-NOS is given when an individual has either (a) impairments in reciprocal social interaction together with impaired communication or (b) impairments in reciprocal social interaction and the presence of repetitive or restricted interests. However, based on a study of 66 individuals who met criteria for PDD-NOS, Mandy, Charman, Gilmour, and Skuse (2011) found that the majority of people (64 of the 66 cases seen) had impaired social interaction and communication in the absence of repetitive and restricted behaviours and interests (type (a), above). Given that DSM–5 required the presence of RRBs for a diagnosis of DSM–5 ASD, and assuming that the majority of individuals with PDD-NOS do not have these behaviours, as described by Mandy et al., it is perhaps not surprising that so many studies have reported reduced sensitivity of the DSM–5 criteria for PDD-NOS. One suggestion emerging from the DSM–5 field trials was that the apparently reduced prevalence of DSM–5 ASD compared with the combined prevalence of DSM–IV–TR autistic disorder, Asperger syndrome, and PDD-NOS may be accounted for by movement into the newly defined social (pragmatic) communication disorder (SCD) category (Regier et al., 2013). However, this was not universally accepted, and Bishop and Norbury (2002) noted that the majority of children they identified with a pragmatic language impairment³ used stereotyped language, with a minority also reporting unusual sensory interests. Both sensory symptoms and the more repetitive and stereotyped aspects of communication impairments are included within the RRB domain of DSM–5 ASD; Norbury, therefore, suggested that some children with PDD-NOS may continue to receive an ASD rather than SCD diagnosis (Norbury, 2014; Swineford, Thurm, Baird, Wetherby, & Swedo, 2014). Whether individuals with DSM–IV–TR PDD-NOS best meet the DSM–5 criteria for ASD or SCD—or neither—remains to be seen as further research is conducted following the publication of the DSM–5 guidelines. However, the potential implications should these individuals qualify for the SCD diagnosis will be discussed below.

Overall, the findings reviewed so far in this chapter lend support to the view that the DSM–5 descriptions may be too narrow to capture the broad range of subgroups included within DSM–IV–TR. While some researchers may argue that this could in fact reflect the overly inclusive nature of DSM–IV–TR rather than an overly rest rictive approach by DSM–5, several studies have indicated that those individuals missed by DSM–5 had significantly higher autism symptom severity than individuals with non-autism clinical diagnoses and individuals with no clinical diagnoses (Matson, Belva, et al., 2012; Matson, Hattier, & Williams, 2012; Matson, Kozlowski, et al., 2012; Mayes et al., 2013; Worley & Matson, 2012). The studies presented so far are, therefore, consistent with the idea that DSM–5 may underdiagnose individuals with significant clinical need consistent with autism, although some of those who do not meet criteria for ASD may meet criteria for SCD. It is important to note, however, that not all studies have found reduced sensitivity for the DSM–5 criteria, and this literature will be reviewed in the next section.

Following the release of the draft DSM–5 criteria, the earliest research findings suggested that the new criteria may lack sensitivity whilst maintaining good levels of specificity. In the following years, additional large-scale studies were conducted that reported the opposite pattern; that is good levels of sensitivity but poor specificity. The three studies that first reported good levels of sensitivity for DSM–5 ASD mapped items from two well-established clinical tools onto the DSM–5 criteria (Barton, Robins, Jashar, Brennan, & Fein, 2013; Huerta, Bishop, Duncan, Hus, & Lord, 2012; Mazefsky, McPartland, Gastgeb, & Minshew, 2013). The tools that they used were the Autism Diagnostic Interview (ADI–R; Lord, Rutter, & Le Couteur, 1994) and the Autism Diagnostic Observation Schedule (ADOS ; Lord et al., 2000), both of which were developed specifically to help guide diagnosis according to the DSM–IV–TR criteria for autism and PDD. Both the ADI–R and the ADOS include diagnostic algorithms, which can be run to determine whether an individual meets the criteria for autism, and the ADOS includes an additional classification of autism spectrum, which relates to the broader category of PDD. When using information collected with both tools, Mazefsk y et al. (2013) found sensitivity of 0.88 in a sample of 498 children and adults with clinical DSM–IV–TR autism (or PDD) diagnoses. This high level of sensitivity could be further improved (to 0.93) by including additional items measuring repetitive behaviours that were not included in the diagnostic algorithm. Despite these excellent levels of sensitivity, it was not possible to assess the true efficacy of the DSM–5 criteria in this study; this was because the specificity of the criteria could not be explored as the sample did not include individuals with non-PDD diagnoses. In this study, therefore, it was not possible to be certain that the high levels of sensitivity reported did not reflect a tendency for individuals with any form of developmental disability—or indeed typical development—to meet the criteria.

Huerta et al. (2012) conducted a large-scale study of three samples of children, which included a total of 4,453 children with DSM–IV–TR PDD clinical diagnoses as well as 690 with non-PDD diagnoses. When analyses were based on parent-report (ADI–R) data only and using the rule that an individual would need impairment on one item in all three of the social communication subdomains and in at least two of the four RRB subdomains, sensitivity of the DSM–5 criteria was 0.91 for the sample as a whole. Sensitivity was generally lower for non-autism PDD and Asperger syndrome when explored in the three samples independently (varying between 0.76 and 0.94). The inclusion of data collected using the ADOS largely resulted in improved sensitivity. Specificity, however, was unacceptably low, both for the sample as a whole (0.53) and in the two samples where these data were available (0.49 and 0.63). In a similar study of toddlers (Barton et al., 2013), sensitivity of the combined ADI–R and ADOS data was 0.84 when applying the same principles reported by Huerta et al. (2012), but specificity was again unacceptably low (0.55). These two studies, therefore, reflect a tendency to over-diagnose ASD.

One way in which the diagnostic performance of DSM–5 could be adjusted would be to change the rules governing both the pattern of symptoms needed for a diagnosis (the symptom profile) and how many symptoms are needed. Both Huerta et al. (2012) and Barton et al. (2013) explored whether adjusting these rules could improve the diagnostic performance of the DSM–5 criteria relative to DSM–IV–TR. Huerta et al. investigated the effect of increasing the number of items an individual would need to score on in each subdomain of the DSM–5 criteria from one to two. This adjustment would be expected to improve specificity, but could at the same time reduce sensitivity, and indeed this was what was found; sensitivity dropped from 0.91 to 0.88 and although specificity was improved, this improvement was only marginal (from 0.53 to 0.66). In a second adjustment focusing more on the symptom profile, Huerta et al. explored the effect of decreasing the total number of subdomains in which an individual needed impairment, so that an individual would need a minimum of two symptoms in either (a) all three social communication subdomains and at least one or more of the RRB subdomains or (b) at least two of the three social communication subdomains and at least two of the four RRB subdomains. This adjustment increased sensitivity to 0.99 but further reduced specificity to 0.42.

Barton et al. (2013) conducted more detailed investigation of the impact of varying the DSM–5 rules, exploring different combinations of thresholds for the individual subdomains (the number of symptoms) and also the number of subdomains required (the symptom profile). The solution that they found to achieve the best combination of sensitivity (0.93) and specificity (0.74) required toddlers to score on at least one item in one of the RRB subdomains and above a statistically defined threshold in at least two of the social communication subdomains. The most frequent adjustment to the DSM–5 rules that has been explored, however, has been the requirement that individuals need exhibit impairment in just two rather than all three of the social communication subdomains (symptom profile). This adjustment has typically been reported to improve sensitivity (Frazier et al., 2012; Huerta et al., 2012; Matson, Hattier, et al., 2012; Mayes et al., 2013; Wilson et al., 2013) with only a minimal loss of specificity (Matson, Hattier, et al., 2012; Mayes et al., 2013).

As outlined earlier in this chapter, effective diagnostic criteria should have good levels of both sensitivity and specificity. Although alterations to the DSM–5 rules should certainly be considered if research supports the need to do so, the next section will review evidence suggesting that it may be possible to achieve good levels of sensitivity and specificity using the DSM–5 rules as they currently stand.

Two studies to date have reported good levels of sensitivity and specificity of the DSM–5 criteria without adjustment to the rules. Using questionnaire data collected from a large registry of siblings where at least one child in the family ha s an autism diagnosis, Frazier et al. (2012) reported that the sensitivity of the DSM–5 criteria relative to clinical judgement was 0.81 with specificity of 0.97. Although this sensitivity value is commonly accepted as good, the authors noted that adjusting the DSM–5 rules as described above (i.e. impairment in two rather than all three social communication subdomains) further improved sensitivity to 0.93 with only a minimal decrease in specificity (0.95 rather than 0.97). The improved level of sensitivity was attributed to the identification of more individuals with Asperger syndrome. One potential limitation of this study, however, was that the comparison group included siblings of children with autism who had typical development as well as those with non-autism clinical diagnoses. As such, this sample was not a typical clinical comparison group, and this may have somewhat inflated the reported specificity in comparison with other studies.

Using a diagnostic instrument called the Diagnostic Interview for Social and Communication Disorders (DISCO ; Leekam, Libby, Wing, Gould, & Taylor, 2002; Wing, Leekam, Libby, Gould, & Larcombe, 2002), researchers and clinicians developed an algorithm based on the draft DSM–5 criteria that had good levels of sensitivity (0.85) and specificity (0.89) for autism in comparison with an entirely clinical control group, which included children with either language impairment or intellectual disability (Kent, Carrington et al., 2013). When typically developing children were also included in the comparison sample, sensitivity and specificity was 0.85 and 0.95 respectively, which is comparable to the figures reported by Frazier et al. (2012). Kent, Carrington et al. (2013) also explored the effect of relaxing the DSM–5 rules so that an individual needed impairment in two of the three social communication subdomains. As in other studies, improved sensitivity was found (0.96 compared with 0.85); however, this improvement was not statistically significant. Moreover, specificity was decreased (0.69 compared with 0.89 when only clinical controls were included), although this was again not significant. Finally, the sensitivity of the algorithm did not vary as a function of age or ability level in a sample of 200 children (n = 112; 68 higher ability (HFA); 44 lower ability (LFA)), adolescents (n = 33; 19 HFA; 14 LFA), and adults (n = 45; 33 HFA; 12 LFA). Although the results from this study support the DSM–5 criteria for ASD, it is important to note that the analyses were conducted on relatively small, well-defined research samples in which the majority of individuals in the autism group had diagnoses of childhood autism. A clear test of their validity will be to investigate their accuracy when used in standard clinical care pathways.

In summary, research focusing on the impact of revisions made in DSM–5 has raised concerns that the new criteria may be overly restrictive, with a lack of sensitivity particularly for those who met criteria for the non-autism PDD subgroups within DSM–IV–TR. The majority of studies have provided evidence supporting this concern, and there has been some discussion as to whether the difference between the two sets of criteria may be due to overly inclusive descriptions in DSM–IV–TR rather than overly restrictive descriptions in DSM–5. Evidence of higher symptom severity in those missed by DSM–5, however, may suggest the apparent loss of sensitivity of DSM–5 should not be disregarded, regardless of whether this loss reflects ‘over-diagnosis’ by DSM–IV–TR. There are studies, however, that have reported good levels of sensitivity. Although these studies typically reported poor specificity and, therefore, still indicated less than optimal performance of the DSM–5 criteria, there is some preliminary evidence that it may be possible to achieve good levels of both sensitivity and specificity.

The variability in the research findings to date is not reassuring at a time when families and clinicians are looking for resolution of the concerns regarding the diagnostic criteria. One explanation for the different findings in these studies may be that the potential to fully investigate the accuracy of the DSM–5 criteria may be limited by the diagnostic tools that are used to gather information about individuals within the sample. This possibility will be explored in the next section.