The microsegmentation of the autism spectrum: research project

5 Intellectual Ability And Disability Across The Spectrum

5.1 A brief summary of the centrality of intellectual ability as an outcome predictor for individuals with autism spectrum disorders was provided in Chapter 3. This chapter develops this theme and the wider question of the distribution of intellectual ability and disability in ASD. It also sets out the results of the research carried out for this study in terms of a systematic review and meta-analysis of this area.

5.2 IQ is the most robust predictor of outcome and level of service needs in ASD, especially in terms of whether or not an individual has an intellectual disability. This has been demonstrated over a considerable period in a large number of outcome, economic and other studies (Beadle-Brown et al., 2000, 2006; Billstedt et al., 2005; Fein et al., 2013; Gillberg & Steffenburg, 1987; Howlin, 2004; Järbrink & Knapp, 2001; Knapp et al., 2009; Lockyer & Rutter, 1970; Lotter, 1974). Three broad groupings may serve as a useful guide in terms of intellectual status. First, there are those with IQ below 50, that is, those with intellectual disability at moderate or more severe level; second, there are those with IQ in the range 50-70, that is, those with levels compatible with mild intellectual disability; third, there are those with IQ 70+, that is, those without an intellectual disability. The last group includes individuals who have received diagnoses both of Asperger’s Syndrome and of childhood autism. The term ‘high functioning autism’ requires caution as it has been used variously to mean (a) individuals with IQ in the average range or above (see, for example, Kumar, 2013) or more broadly, and more commonly, (b) all who do not have an intellectual disability (see, for example, Lake, Perry, & Lunsky, 2014).

5.3 It has been consistently demonstrated that the poorest outcomes are for those with IQ below 50. Very few such individuals achieve good functioning in adulthood, whether in terms of social competence, being in any form of employment or having any meaningful degree of independent living (Billstedt et al., 2005; Gillberg & Steffenburg, 1987; Lockyer & Rutter, 1970; Lotter, 1974).

5.4 In relation to those in the IQ range 50 to 70, outcomes are also on the whole poor, but with progressive change at this higher level. Howlin et al. (2004) followed up 68 individuals with autism and IQ above 50, from mean age seven to mean age 29 years. Outcome measures included standardised cognitive, language and attainment tests and assessment of social, communication and behavioural problems. Although a minority had achieved relatively high levels of independence, most remained very dependent on their families or other support services. Few lived alone, had close friends, or permanent employment. Communication generally was impaired, and reading and spelling abilities were poor. Stereotyped behaviours or interests frequently persisted into adulthood. Ten individuals had developed epilepsy. Overall, 12% were rated as having a very good outcome, while the majority had a poor (46%) or very poor (12%) outcome.

5.5 The pattern changes again for those without intellectual disability, in the IQ range 70+. In the study by Howlin et al. (2004), of 44 individuals with IQ 70+ for whom data were available, 16% had outcomes rated as very good, 16% good, 20% fair and 45% poor or very poor. Overall, the mean Verbal (V)/Performance (P) IQ levels for the whole of the sample in relation to outcomes were: good/very good, V95/P99; fair, V 85/P77; poor/very poor, V65/P38.

5.6 A number of recent studies have focussed on a sector of the population diagnosed with ASD who later ‘lose their diagnosis, or who otherwise have such favourable outcomes that they are no longer autism service users. Fein et al. (2013) reported on 34 individuals with optimal outcome, defined as ‘losing all symptoms of ASD in addition to the diagnosis, and functioning within the non-autistic range of social interaction and communication’. All were high functioning, with mean IQ in the high average range, and none with IQ below 80. While they had milder early social impairments than a matched high functioning autism group who did not have optimal outcome, their early profiles for communication and repetitive behaviours were similar.

5.7 In summary, measured intellectual ability has primacy as a determinant of outcome in autism. The scope for overall outcomes to improve from very poor through to very good increases from the more severe levels of intellectual disability, through mild intellectual disability, to the levels of normal functioning seen in those diagnosed with Asperger’s Syndrome or high functioning autism. In addition, those with higher IQ show the greatest increases in skills over time (Beadle-Brown et al., 2006).

5.8 This has major implications for economic impact and the level of service provision required. In a UK study, Knapp et al. (2009) calculated that the lifetime economic cost for someone with autism and intellectual disability was approximately half as much again as for someone without intellectual disability. The costs were calculated at £0.80 million and £1.23 million respectively. A subsequent study by Buescher et al. (2014) revised these costs to £0.92 million and £1.23 million respectively, while comparable lifetime figures for the US were $1.4 million with intellectual disability and $2.4 million without intellectual disability. The availability of accurate figures for the proportion of individuals with autism who have an additional intellectual disability is therefore crucial in its relation to economic impact and to planning of service provision, and even a small variation in the figures used would have very significant economic impact.

5.9 However, a perusal of the literature in this field indicates the almost imponderable difficulty of establishing figures with any degree of confidence. For example, in an economic study by Järbrink and Knapp (2001) it was assumed that 75% of people with autism have intellectual disability, whereas in the further study by Knapp et al. (2009) the figure used was 55%, this change resulting largely from a broader view of autism. For the Buescher et al. (2014) study the estimate was placed at anywhere between 40% and 60%.

5.10 The high levels of variability in the estimates of intellectual disability within the autism spectrum may be attributed to two main factors. The first of these is the time at which the study was conducted. For example, Billstedt et al. (2005), in their population-based, 13-22 year follow-up study of 120 people diagnosed with ASD in the 1970s and 1980s, reported that 82% had ID. This reflected the comparatively limited availability of diagnostic facilities for ASD at that time, which have expanded considerably through the years. When facilities are scarce smaller numbers are diagnosed, and these are likely to be the more severe cases among whom intellectual disability will be more prevalent.

5.11 In addition, the definition of what constitutes autism has expanded significantly from the early view of classical autism as a rare condition to the emergence of the much broader concept that became the autism spectrum (Wing & Gould, 1979). Thereafter, the death of Hans Asperger in 1980 and the renewed interest in his work with Wing’s (1981) clinical account of ‘Asperger’s Syndrome’, followed by Frith’s (1991) translation of his work into English, led to the inclusion of Asperger’s in the diagnostic classification systems from the early 1990s onwards (American Psychiatric Association, 1994; World Health Organization, 1992). Most of the sample studied by Billstedt et al. (2005) had a diagnosis of autistic disorder, and of the remainder who were described as having ‘atypical autism’, almost all were re-diagnosed later as having autistic disorder.

5.12 Early studies of Asperger’s Syndrome indicated a prevalence exceeding that of childhood autism itself (Ehlers & Gillberg, 1993), and since by definition these were individuals who did not have any clinically significant general delay in cognitive development it was clear that their inclusion must have the effect of reducing estimated proportions of intellectual disability in ASD. While we have not found it possible to establish separate prevalence figures for Asperger’s Syndrome owing to a lack of studies of sufficient methodological rigour, the figure of 36/10,000 cited in the Ehlers and Gillberg (1993) study is likely to be an underestimate. It was Gillberg’s view in relation to that study that the figure cited could have been doubled by taking a less strict diagnostic threshold and by including the cases which the authors listed as ‘suspected’ or ‘possible’ Asperger’s Syndrome as well as those described as ‘definite’ (C. Gillberg, personal communication, 7 November, 2014).

5.13 The second factor accounting for the variability in estimates of intellectual disability in ASD is aspects of study methodology. Significant issues include the methods used to determine intellectual status and sample size and representativeness. The age of the sample is of importance, since younger children are likely to include higher proportions with intellectual disability. It is the more severe cases who are diagnosed youngest, with much later average age of diagnosis for the more able children who do not show language or cognitive delay (Howlin & Asgharian, 1999).

5.14 In relation to methodological issues, the figure of 55% with ID used by Knapp et al. (2009) was based on the Baird et al. (2006) estimate. Subsequent to that time, Charman et al. (2011) provided an analysis of the same dataset from the specific standpoint of IQ. They state regarding a particular limitation to this sample: ‘the decision to only screen cases with a local clinical diagnosis and/or children with a statement of SEN means that we will not have captured all higher IQ children with an ASD’ (p.625). This is borne out by the fact that the statement of special educational needs is a legal document for children requiring substantial additional support in school in England. Not all children with ASD require such substantial support and therefore those with a statement are likely to represent those with more severe difficulties. In addition, selection was from those who were willing to be followed up, and it is not known whether these were representative in terms of severity. The implication therefore is that the removal of this methodological limitation would result in a lower occurrence of ID in the ASD population.

Method

5.15 The present study reports a systematic review of peer-reviewed published studies of distribution of intellectual disability in ASD using meta-analysis to provide a weighted, pooled estimate to inform statistical modelling and economic analysis. The online journal databases ‘Medline’, ‘PsychInfo’ and ‘PsychArticles’ were searched for English-language, peer-reviewed papers published since 31 December 2002 which investigated, or commented upon, the IQ of individuals with ASD, or the level and presence of intellectual disability amongst this population. The database search returned papers which included ‘child developmental disorders’ or ‘pervasive’ (a term which covered all terms relating to pervasive developmental disorders) as well as any of the following terms in the main body of the article: IQ, intelligence, cognitive disability, cognitive impairment, learning disability, learning difficulty, WAIS, WISC, Stanford-Binet, Vineland, British Picture Vocabulary Scale.

5.16 The initial search (Stage 1) returned 68,651 results (40,315 from Medline and 28,336 from PsychInfo, with no unique articles identified from PsychArticles), validating both the databases searched and the terms used. The majority of the papers identified in Stage 1 (n = 68,613) were removed from further analysis as they did not contain primary data relating to the intelligence levels or intellectual disability status of individuals with ASD (Stage 2). The literature reviews and reference lists of the remaining 38 papers were searched for mentions of previously unidentified studies. No unique, previously unidentified papers, were found as part of this process.

5.17 These 38 papers were scrutinised for relevance and quality using 11-point data extraction forms (a copy of which is shown at Appendix B.2). Quality assessments of the studies were based on the grading of five key factors concerning the level of detail studies had provided about the sample from which the IQ data were collected, the diagnostic criteria used, the tools and professionals involved in diagnosis, sample size and representativeness, the methods used for collecting IQ data and the assessment measures used. The grading criteria used to assess these aspects of a study’s methodology are also shown in Appendix B.1.

5.18 Quality assessments were carried out by two of the authors following training on a random set of seven papers from the 38 reaching Stage 3 of the paper selection process. Following this, these authors then independently coded a further random sample of 6 papers (17% of the total). They agreed on all of the papers that were to be excluded at this stage, and overall there was a 95% level of agreement between the independently coded quality of evidence scores (score range 0 – 20). The final coding of the 5% of cases which were the subject of disagreement was agreed upon by both authors following detailed discussions regarding the papers concerned.

5.19 Following the data extraction stage of analysis, 33 of the papers were removed from the final analysis (Stage 3) as they had (a) based their analysis on samples considered to be unrepresentative or skewed (n = 14); (b) used non-standardised measures of IQ (n = 4); (c) reported only mean IQ scores for an entire ASD sample (problematic in that in this context overall mean scores would be heavily influenced by the proportion of higher and lower functioning ASD cases within each sample; n = 12); (d) failed to provide important methodological details relating to recruitment and the diagnostic process (n = 2); or (e) based their analysis of a sample of less than 30 (n = 1). The remaining 5 papers were included in a meta-analysis.

Figure 5.1 Flowchart for IQ paper selection process

5.20 Methodological details of the five studies included in the final meta-analysis are shown in Table 5.1. Of these five studies, two provided IQ data only in relation to those with a diagnosis of childhood autism (Honda, Shimizu, and Nitto, 2005 and Oliveira et al. 2007), two provided data relating to individuals across the spectrum (Ellefsen, Kampmann, Billstedt, Gillberg, & Gillberg, 2007 and Keen & Ward, 2004) and one provided IQ data relating to a PDD population (Chakrabarti & Fombonne, 2005). Figure 5.1 presents a PRISMA flow chart summarising the paper selection process.

5.21 The five studies were carried out in four different countries (the United Kingdom, the Faroe Islands, Japan and Portugal), and the size of the samples that IQ data were collected from ranged between 41 and 138 (m = 98.2, SD = 36.55). All of the studies collected IQ data from children and young adults within the age range up to 17, although the data provided by three of the studies (Chakrabarti & Fombonne, 2005, Honda et al., 2005 and Oliveira et al., 2007) related only to children under the age of nine.

5.22 All five studies collected their data as part of a larger investigation into the prevalence of ASD. To confirm the diagnoses identified as part of these prevalence investigations, two studies used DSM-IV criteria (Chakrabarti & Fombonne, 2005; Oliveria et al., 2007) while three used ICD-10 criteria (Ellefsen et al., 2007; Honda et al., 2005; Keen & Ward, 2004). However, there were some differences in the methods used to assess and obtain levels of IQ as shown in Table 5.1 below.

5.23 Random effects meta-analyses using the non-parametric method of moments (Borenstein, Hedges, Higgins, & Rothstein, 2009) were carried out on weighted logit-transformed event rates of intellectual disability ( ID).

Table 5.1 Summary of the samples assessed by the five studies and the measures of IQ used

Study	N	Diagnoses included in sample			Age range (years)	IQ measures used
		Autism	Asperger’s	Other ASD
Chakrabarti & Fombonne (2005)	57	21	11	25	4 – 6	WPPSI, Merrill-Palmer Scale & Griffiths Mental Development Scale
Ellefsen et al. (2007)	41	12	20	9	8 – 17	WISC-R & DISCO*
Honda et al. (2005)	95	95	-	-	0 – 5	Stanford-Binet (Japanese version)
Keen & Ward (2004)	138	138			5 – 18	BAS, WISC-III & WPPSI
Oliveira et al. (2007)	120	120	-	-	6 – 9	Griffiths Mental Development Scale & WISC-III

*Used for estimating IQ when other tests could not be completed.

Results

5.24 Table 5.2 below provides details of the distribution of IQ scores in these five studies. The number of individuals in these studies with IQ scores of (a) less than or equal to 49; (b) 50-69; and (c) greater than or equal to 70 are shown. The 95% CIs for the sub-total and grand total means are also reported. As the table reveals, some 55% of the individuals in these studies had IQ scores <70.

Table 5.2 Distribution of IQ scores across the five studies included in the final meta-analysis

Study	N	Distribution of IQs by Study
		IQ < 49	IQ 50 - 69	IQ > 70
Chakrabarti & Fombonne (2005)	57	8	9	40
Ellefsen et al. (2007)	41	11	2	28
Honda et al. (2005)	95	46	25	24
Keen & Ward (2004)	138	28	19	91
Sub-Total	331	93 (28.10%) (95% CI 23.26%-32.94%)	55 (16.62%) (95% CI 12.61%-20.63%)	183 (55.29%) (95% CI 49.93%-60.65%)
Oliveira et al. (2007)	120	100		20
Grand Total	451	248 (54.99%) (95% CI 50.40%-59.58%)		203 (45.01%) (95% CI 40.42%-49.60%)

5.25 Figure 5.2 below shows the forest plots (point estimates of the proportion of individuals in each study with ID and associated 95% confidence intervals) which show the level of variability in the estimate for each study.

Figure 5.2 Summary of random effects meta-analysis of ID event rates from 5 final studies

Figure 5.3 Funnel plot of standard error by point estimate of ID event rates from a random effects model showing 95% confidence intervals.

Logit event rate

5.26 The results revealed an overall pooled ID event rate of 0.522 (95% CI 0.280-0.754), based upon a weighted logit analysis. However, high levels of heterogeneity were observed (Q=90.87, df=4, I ²=95.60, τ ²=1.32), as revealed by a funnel plot of logit event rate of ID by standard error shown in Figure 5.3.

5.27 A moderator analysis was carried out to investigate the heterogeneity. A comparison of the two studies with samples more representative of lower functioning individuals with autism (Honda et al., 2005; Oliveira et al., 2007) with the remaining three studies which analysed samples more representative of the autism spectrum as a whole was carried out (see also Table 5.2 for further details of the distribution of IQ scores in these studies). The findings revealed a significant difference (Q =88.14, df = 1, p = .0001) between a pooled mean ID event rate of 79.2% (95% CI 73.2 – 84.2) for the studies focussing on those with autism and a pooled mean ID event rate of 32.7% (95% CI 27.0 – 38.9) for those studies focussing on the whole autism spectrum.

5.28 On the basis of the above, the following figures are noted for the distribution of IQ scores for the three studies focussing on the whole autism spectrum: (a) less than or equal to 49 (moderate to severe ID), n = 47 (19.9%); (b) 50-69 (mild ID), n = 30 (12.7%); and (c) greater than or equal to 70 (no ID), n = 159 (67.4%). It was not possible to calculate comparable figures in relation to those with autism alone as there were only two studies and the larger of these did not provide a breakdown of scores of those with ID.

Discussion

5.29 The accurate assessment of intellectual ability is of considerable importance in relation to autism spectrum disorders. In diagnostic terms it has been necessary since the publication of the current classification systems in the early 1990s in order to determine specific criteria for Asperger’s Syndrome, which requires no clinically significant delay in cognitive function. Since the publication of DSM 5 (American Psychiatric Association, 2013), together with the anticipated publication of ICD-11, and the resultant abandonment of the separate diagnostic categories of childhood autism, Asperger’s Syndrome and atypical autism, it is more rather than less important. The new category of autism spectrum disorder requires an axial classification involving intellectual function in every case, being defined in terms of the dimensions of presence or absence of an intellectual disorder and of functional language, together with whether there has been loss of previously acquired skills.

5.30 The studies included in this meta-analysis met rigorous standards in regard to diagnostic criteria, diagnostic procedures, sample size, statistical analysis and all other relevant aspects of methodology. There were some limitations to the investigation, however. The final sample size of papers was small, which impacted on the statistical power and generalisability of the moderator analyses. There may have been additional relevant papers which were not included in the databases searched. In addition, the paper selection process focussed only on English language papers, and it is possible that there were additional relevant papers published in other languages or in non-peer reviewed sources. Further, our final analysis only included studies from four countries (the UK, the Faroe Islands, Japan and Portugal). However, as noted in Chapter 4, para. 4.41, there was no evidence to suggest that there were regional variations in the prevalence estimates associated with ASD.

5.31 Any attempt to consider the distribution of intellectual ability and disability within the autism spectrum raises a number of issues. First, there is the issue of IQ measurement and its meaning. Many different measures are used across studies of intelligence. The five final studies included in this analysis used one or more of four main assessment approaches in deriving IQs – tests in the Stanford tradition (Stanford-Binet, Merrill-Palmer), the Wechsler scales, the British Ability Scales and the Griffiths Mental Development Scale. Each of these approaches the measurement of ability in ways that reflect theoretical and practical differences.

5.32 The Stanford-Binet scales represent a continuation of the original intelligence tests which developed from the work of Alfred Binet in the early 1900s. From a wide range of subtests they produce an overall IQ based on ‘mental age’. The Stanford-Binet (Japanese version) was used in one of the five studies (Honda et al., 2005). Merrill played a central role in the early revisions of the Stanford-Binet, for which the Merrill-Palmer was designed as a largely non-verbal substitute, with additional discriminatory facility for younger children and for those with intellectual disability. It includes verbal items where appropriate and is suitable for use up to age six. It too was used in one of the five final studies (Chakrabarti & Fombonne, 2005).

5.33 The Wechsler tests in their various revisions, including the Wechsler Preschool and Primary Scale of Intelligence ( WPPSI) for younger children, are the most widely used intelligence tests in the world (Camara, Nathan, and Puente, 2000) and have gained wide recognition as representing the ‘gold standard’ for this purpose (Hunt, 2011). They were designed as a more refined alternative to the Stanford-Binet scales, and in addition to providing a Full Scale IQ they gave separate scales for a Verbal IQ and a Performance IQ. They were later developed further to provide a cognitive profile with separate scores for Verbal Comprehension, Perceptual Reasoning, Working Memory and Processing Speed. Wechsler tests were used in four of the final five studies (Chakrabarti & Fombonne, 2005; Ellefsen et al., 2007; Keen & Ward, 2004; Oliveira et al., 2007).

5.34 The British Ability Scales aimed to provide more sophistication and theoretical rigour than tests in the Stanford-Binet and Wechsler traditions. They yielded individually interpretable subtests, divided into ‘core’, ‘diagnostic’ and ‘achievement’ domains. Only those tests that measured the most complex cognitive processes, the core scales, were considered to provide the best estimates of g or general intelligence, and they were used to produce a General Cognitive Ability ( GCA) score. This was taken to be a purer measure of g than the composites of other batteries which included all cognitive tests in their final figure irrespective of their g loading (Elliott, 1997). It was felt that in the Wechsler scales, for example, weaknesses in diagnostic tests such as those for working memory, which is frequently impaired in autism, could artificially depress overall scores. The British Ability Scales were used in one of the five final tests (Keen & Ward, 2004).

5.35 The Griffiths Mental Development Scale was designed to provide an overall developmental level for children from birth, the original version covering only from birth to two years, but subsequently extended to eight years. Its items overlap considerably with subtests on general intelligence tests and are divided into six scales: locomotor, personal/social, language, eye-hand coordination, performance and practical reasoning. Its General Quotient ( GQ) has been found at age three to be a good predictor of IQ at age five (Bowen et al., 1996). It was used in two of the final five tests. Chakrabarti and Fombonne (2005) used it in only six cases in order to derive IQ scores where these were not otherwise available, and Oliveira et al. (2007) used it for children with lower cognitive ability.

5.36 In addition to the different approaches outlined to assessing intelligence using wide-ranging intelligence tests, there is the further question of the extent to which such tests provide accurate measures of actual ability in ASD. The greater the range of functions measured the more likely it is that these will reflect functions which are known to be compromised in autism. The example of working memory has already been noted, but in addition the Wechsler scales assess processing speed, social comprehension and other areas in which people with autism are likely to have lower scores. From the late 1930s Raven designed a much more clearly-defined testing format using two tests designed to measure the key aspects of g – eductive ability (the ability to forge new insights, to discern meaning in confusion, to perceive, to identify relationships, in short, the ability to generate new, largely non-verbal concepts which make it possible to think clearly) and reproductive ability (the ability to recall, and use, a culture’s store of explicit, verbalised concepts). These tests covered the whole child and adult age range from five years and upwards: the Crichton or Mill Hill Vocabulary Scales, and the Coloured or Standard Progressive Matrices (Raven, 1966; Raven, Raven, & Court, 1998).

5.37 Dawson, Souliéres, Gernsbacher and Mottron (2007) assessed a sample of 38 autistic children and 13 autistic adults using Raven’s Progressive Matrices and found that they scored significantly higher than on Wechsler scales. These differences were not found in non-autistic controls. They concluded that the intelligence of individuals with autism is being underestimated. At the same time it is appropriate to note that a ‘pure’ test of non-verbal reasoning such as the Matrices may not indicate how an individual with autism will actually function at a practical level, because the weaknesses which are identified on wide-ranging intelligence tests like the Wechsler scales highlight functional abilities that are necessary for day-to-day performance.

5.38 Despite the differences in approaches to the measurement of ability in the final studies included in this analysis, they all have in common the fact that they serve as wide-ranging tests which measure recognised cognitive skills and developmental levels and which are able to yield a composite score designed to reflect general ability. Nevertheless, the fact that there is no single measure of intelligence used in all studies of autism and intellectual ability is a factor which needs to be taken into account in assessing results.

5.39 Second, there is the issue of age, for which two relevant factors must be considered. The first relates to the reliability of IQ measures at different periods. It has long been established that tests taken at about age six show a high correlation with results obtained in later years. In an early study, Jones and Bayley (1941) found that IQ at age six had a .77 correlation with IQ at age 18, rising to .89 for the period age 12 to age 18. Nevertheless, the average change in the latter period still amounted to seven IQ points. Comparable findings were reported in a later study by Moffitt, Caspi, Harkness and Silva (1993). However, correlations become decreasingly lower with lower age of testing (Flensborg-Madsen & Mortensen, 2015). The second factor relates to the age of the sample in terms of how representative it is of the autism spectrum. As it is the most severe cases who are diagnosed youngest, with those who have no cognitive or linguistic delay being diagnosed later, very young samples are likely to include children with lower levels of ability.

5.40 Third, there is the issue of seeking to establish a more differentiated breakdown of IQ within the broad category of intellectual disability. Such a breakdown has practical utility given the differential outcomes based on whether there is moderate to severe ID (scores less than 50), mild ID (scores from 50-69) or no ID (scores at or above 70) (Billstedt et al., 2005; Gillberg & Steffenburg, 1987; Howlin et al., 2004; Lockyer & Rutter, 1970; Lotter, 1974). This must be done with caution owing to the small number of studies available and the relatively low sample size. Our best current estimate across the whole autism spectrum are the figures we have cited of 19.9% moderate to severe ID, 12.7% mild ID and 67.4% with no intellectual disability.

5.41 There are significant difficulties surrounding any attempt to establish precise figures for the spread of intellectual ability and disability in ASD. While a large number of studies contain information relevant to intellectual functioning, few have reliable data in terms of how such functioning has been assessed, how data have been gathered or how representative the samples are of individuals on the autism spectrum. Studies are also marked on the whole by small sample size, and many focus on special populations such as those admitted to hospitals.

5.42 In conclusion, the estimate of the percentage of individuals with ASD and a co-occurring diagnosis of intellectual disability of 32.7% (95% CI 27.0 – 38.9) best takes into account the representativeness of the sampling across the autism spectrum of the included studies in this review.

5.43 Although this figure is significantly lower than figures previously reported, we would propose that it is intuitively accurate in terms of the known clinical parameters of ASD. The earliest studies reported the highest level of ID in the ASD population as they were based on more severe cases and on a narrow definition of autism. The expansion of diagnostic resources facilitated the inclusion of less severe and higher-functioning cases, and the inclusion of Asperger’s Syndrome significantly changed the ASD landscape in terms of proportions with ID. The acknowledgement by Charman et al. (2011) that the figure of 55% with ID proposed by Baird et al. (2006) would not have taken account of all higher-functioning cases provides further confirmation that the true figure was likely to be a lower one.

5.44 Finally, by way of caveat, it should be noted that the figures yielded by this study are based upon data from a small number of individuals and sources and much of the research identified through our systematic review reported IQ only as part of a small-scale study. Given that IQ is a strong predictor of outcomes for individuals with ASD, and the implications for planning and service provision, there is a need for further large-scale research studies of the co-occurrence of ID and ASD.