The Classification of Autism, Asperger's Syndrome, and Pervasive Developmental Disorder Peter Szatmari, MDt

Objective: The current classification ofthe pervasive developmental disorders (PDDs) as conceptualized in both the DSM-IV and lCD-lOis deeply unsatisfying to many parents, pont-line clinicians, and academic researchers. Is the diagnostic validity ofthe various disorders simply lacking empirical dataforfull substantiation, or does the overall conceptualization ofthe category have morefundamental problems, not rejlecting the "true" nature ofthe phenomena? This paper argues the latter hypothesis. I review the historical development ofthe classification ofPDD, summarize recent empirical data on issues ofreliability and validity, and suggest a new approach to classification and understanding. (Can J Psychiatry 2000;45:731-738)

Key Words: autism, Asperger 's syndrome, pervasive developmental disorder, DSM classification utism is a severe neurodevelopmental disorder characterized by a triad of impairments in reciprocal social interaction, in verbal, and in nonverbal communication; it has a pattern ofrepetitive stereotypie activities, behaviours, and interests (1). In 1943, Leo Kanner published a clinical description of 11 children with so-called "infantile autism" (2). These children were characterized by extreme aloneness, poor communication skills, and a resistance to change. Kanner, with remarkable clinical skills, demarcated this pattern of symptoms among all the behaviours shown by children with nonspecific emotional, behavioural, and developmental problems. The term "autism" was borrowed from Eugene Bleuler's description of schizophrenia to characterize the "withdrawal from reality" seen in both conditions. This link with schizophrenia led to the theoretical position that "infantile autism" was in fact a very early form of that disorder. Accordingly, in North America, the term "infantile autism" was supplanted by terms such as "childhood schizophrenia" and "childhood psychosis." Important research by Kolvin (3) and Rutter (4), however, pointed out that children with earlyonset psychosis in childhood were quite different from those with a later onset with respect to symptom patterns, family history, outcome, and neurological history. At around the same time, Lorna Wing completed her epidemiologic work on autism in Camberwell and delineated more carefully a subgroup of children with autism and autistic-like conditions (5). She clearly demonstrated a link, not between autism and

schizophrenia, but rather, between autism and mental retardation. Wing also clearly formulated the notion of a triad ofimpairments-in socialization, in social communication, and in social play. This triad, then, translated into the concept of "pervasive developmental disorders" (PDDs) characterized by a similar (but somewhat different) triad. The term PDD became enshrined in the official classification system of the American Psychiatric Association (APA) in 1980. PDD was considered to be a generic label comprising several different conditions including autism, childhood-onset PDD, residual autism, and atypical autism (6). The term referred to the idea that the impairments in socialization, communication, and play "pervade" all aspects ofa child's life and arise from a developmental disability. However, the term PDD was unfamiliar at that time and puzzled many clinicians and parents. Moreover, considerable controversy surrounded the nonautistic forms of PDD that DSM-III specified.

A

The DSM-III was revised in 1987. The diagnostic criteria for autism were broadened to capture wider variation in expression, and the nonautistic forms of PDD were collapsed into the single category called "PDD not otherwise specified" (PDDNOS) (7). This led to a remarkable (and largely unanticipated) increase in the number ofchildren receiving a diagnosis of autism and PDDNOS. For example, in Wales prevalence rates of autism nearly tripled (8). Such a large increase in diagnosed cases is unlikely due to an actual increase in prevalence but clearly reflects changing diagnostic practices. Nonautistic PDD Subtypes

Manuscript received and accepted August 2000. I Professor, Department of Psychiatry and Behavioural Neursosciences, McMaster University, Hamilton, Ontario. Address/or correspondence: Dr P Szatmari, Hamilton Health Sciences Corporation, Chedoke Campus, Patterson Bldg, Room 207, Hamilton, ON L8N 3Z5 e-mail: [email protected]

In 1944, Hans Asperger, a Viennese pediatrician, described a group ofchildren, also with difficulties in reciprocal social interaction, who demonstrated fluent language but poor conversation skills and who developed unusual and intense interests (9). He too borrowed the term "autism" from Bleuler 731

732

The Canadian Journal of Psychiatry

and suggested that these children had some type of personality disorder; hence, he used the term "autistic psychopathy." There was already, however, a tradition of describing such children as having "schizoid disorder." A report from a Russian child psychiatrist, published in the 1920s (recently reissued by Sula Wolff), described children with very similar clinical characteristics (10). The description formulated by Asperger did not really reach the attention of the Englishspeaking psychiatric world until Lorna Wing's important 1981 paper. She described several adults with "Asperger's syndrome" (AS) who had a previous diagnosis ofautism (11). Wing did not, however, use the term AS as a specific diagnostic category but rather to describe a different clinical presentation of autism in adults-those who were more fluent in their communication skills. With greater understanding and greater precision in the diagnosis of autism in the 1980s and early 1990s, there was a growing realization that many children who had autistic symptoms failed to meet criteria for that diagnosis. There was also enormous dissatisfaction with the term "POONOS," particularly among parents, who did not appreciate being told that their child had a "nonspecific disorder"! They were in fact looking for something specific to explain their child's developmental difficulties and to give them a concrete plan for intervention. In clinical practice, the "NOS" suffix was often dropped, and people started referring to children as having POD alone. This was in direct contrast to other professionals and researchers, who used the term "POD" to refer to a generic group of disorders, of which autism was but one type. But in the minds of many parents, autism and POD became mutually exclusive diagnostic categories. At the same time, several cross-sectional studies were being published showing symptom patterns in AS, atypical autism, and disintegrative disorder different from those seen in autism (12-15). The OSM-IV task force completed a review of this literature (16), and the new OSM-IV again specified several other types of POD, including AS, disintegrative disorder, and Rett syndrome. POONOS was retained and could also be referred to as atypical autism.

An Appraisal of DSM-IV The OSM classification of this group of disorders has been deeply unsatisfying for several reasons. First, among the academic community, the measurement of these nonautistic forms of POD is quite controversial. It is not at all clear that we have reliable diagnostic criteria for POONOS. Indeed, 2 studies have criticized the current criteria (17,18). POONOS is differentiated from autism largely on the basis of symptom count; children with POONOS have fewer symptoms than do those with autism. The diagnosis of POONOS does not require a specific number or type of symptoms. In contrast, AS is differentiated from autism on the basis of level of

Vol 45,No8

functioning; that is, those with AS have an absence of clinically significant language and cognitive delay. By not also specifying symptom differences, the differentiation of AS from high-functioning autism becomes confused. There are other problems in measurement. In one study agreement among expert clinicians on the diagnosis of POONOS was no better than chance (19). In essence, clinicians could not agree on whether a child with POD met criteria for autism or for POONOS. The diagnostic criteria for AS have also been seriously criticized. Several studies have reported that most children who meet OSM-IV criteria for AS also meet the criteria for autism (20-22). Since OSM-IV specifies that such children should preferentially be diagnosed with autism, the OSM-IV criteria for AS do not seem to be widely applicable. However, clinicians report that the number ofchildren receiving a diagnosis of AS has increased dramatically since the OSM-IV's introduction. Clearly, front-line clinicians do not find the OSM-IV criteria for AS helpful and must be applying the criteria in a different way. Moreover, many clinicians and parents continue to see autism and POD as mutually exclusive categories, despite the attempts to delineate these disorders more precisely. Second, considerable controversy exists as to whether AS is a valid diagnostic category and differs from autism on parameters independent of the defining characteristics. For example, 6 years after DSM-IV's publication, there are still no data that AS and autism differ on biological parameters, other measures of etiology, outcome, or response to treatment. Most of the recent literature is concerned with whether AS is the same as or different from autism on cross-sectional correlates, and the results have been far from conclusive. Third, as a result of that data, many professionals feel that the term PDO is not helpful and that referring to a "spectrum of autistic disorders" is more appropriate (23,24). In other words, the PODs should be considered as lying on a spectrum of severity with (perhaps) Kanner's syndrome at one end and AS at the other. This theoretical model posits that there is considerable variation in severity and that the boundaries between these different types of "disorders" are relatively arbitrary. The distinctions between the disorders subsumed under the POD category do not carry implications with respect to etiology, outcome, or response to treatment beyond differences attributable to severity alone.

The Conceptual Foundation for Same Versus Different OSM-I11 represented a major advance from previous classification systems in that it focused exclusively on symptom patterns as opposed to etiology and emphasized the importance of reliability of diagnosis and categorization. The authors of OSM-I11 argued that unless a classification system is firstreliable, the validity of diagnosis is impossible to establish, a

October 2000

The Classification of Autism, Asperger's Syndrome, and Pervasive Developmental Disorder

well-known psychometric rule. As part of psychiatry's attempt to approach the scientific validity of other medical disciplines, diagnostic criteria for various psychiatric disorders were made operational such that they could be measured and criteria and diagnostic rules could be clearly specified. This would facilitate agreement on whether a child had a particular diagnosis. The different diagnostic categories were thought of as distinct categories, much like other medical "diagnoses," such as tuberculosis, pneumonia, or inflammatory bowel disease. If reliable diagnostic categories could be specified, each of these would come to be "valid." A specific etiology would be associated with each disorder, which would predict a certain outcome and determine a particular treatment plan. Decisions about medication, for example, were largely based on whether a child met diagnostic criteria for depression, anxiety disorder, or psychosis. The success of psychopharmacologic interventions in clinical trials following DSM-III supported the theoretical underpinning that diagnostic categories represented discrete biological entities. Once this was accepted, the validity ofsome ofthese diagnostic categories was questioned: were they the same or different from each other? Initial studies were very successful in demonstrating that emotional disorders were distinct from disruptive behaviour disorders, that early-onset psychosis was distinct from late-onset psychosis (3), and that autism was distinct from other emotional and behavioural disorders and learning disabilities. The next step was to determine whether 2 conditions within the same major category of disorder might be the same or different. So, for example, regarding disruptive behaviour disorders, an enormous literature appeared on whether attention-deficit hyperactivity disorder (ADHD) was the same as, or different from, conduct disorder (25). The internalizing disorders were also heavily debated: Were anxiety disorders the same or different from mood disorders? Were different types of anxiety disorder different from each other (26)? In fact, the differences among subcategories of disorder are much more difficult to determine than are the differences between the major categories. First, the evidence used to substantiate arguments about same or different is, unfortunately, largely based on etiological studies that are cross-sectional in design. No firm answer about validity can ever be supplied through the use of crosssectional studies. It is virtually impossible to disentangle correlate, risk factor, and cause in such a design. One can never be sure that the potential etiological differences identified in this way are markers ofa simple but trivial association or represent a true risk factor or cause. Our ability to find causes of psychiatric disorders is very poor indeed, and even the hope that genetic studies might provide answers has become fainter in recent years. For example, recent family studies of ADHD demonstrate that mood disorders are more common among first-degree relatives ofADHD probands than those of

733

control subjects (27). This seems to fly in the face of the one classification distinction that everyone agrees is valid-the distinction between internalizing and externalizing disorders. Second, cross-sectional designs often lead to tautological conclusions. For example: 2 disorders are distinguished on the basis of symptoms, and a study reports that other associated symptoms or behaviours differentiate the groups. It is always possible to argue that those differences are due to the way the groups were initially defined. Several studies have demonstrated that children with AS have language skills better than those of high-functioning children with autism (21,22). Does this finding confirm distinctiveness, or does it, in fact, flow from how the groups were already defined? One of the differentiating characteristics specified in DSM-IV for AS is an absence of "clinically significant language delay." If that accounts for differences in language skills, then the finding is tautological. In fact, this argument can be extended to any symptom pattern or behavioural measure distinguishing autism and AS that is determined on the basis of crosssectional designs alone. Virtually any behavioural difference can be accounted for by language. Third, the evidence with respect to whether disorders are the same or different depends on finding qualitative differences between groups. Is it ever possible to find qualitative differences between diagnostic groups? Cross-sectional designs usually demonstrate only quantitative differences. IfAS were diagnosed only in males and autism cases were a mixture of males and females, then the difference would be truly qualitative. But this rarely, if ever, occurs in psychiatry. So when does a quantitative difference become qualitative-90% versus 10%? 80% versus 20%? How does one decide what magnitude of quantitative difference between groups is large enough to constitute a true qualitative difference? If these rules are not established beforehand, the differences between groups can always be argued to represent simple quantitative differences with respect to severity or some other dimension. The same data can be argued both ways-the groups are the same or they are quite different.

The Dimensional Approach: High- and Low-Functioning Autism or PDD An alternative to classifying on the basis of same or different is to eschew the search for qualitative differences between groups and to adopt a dimensional model. It is possible to view autism (or PDD) as a categorical distinction, in the broad sense, from other emotional and behavioural disorders, with the various types ofPDD differing along a single underlying dimension of severity. This is, in fact, the theoretical model underlying the notion ofautistic spectrum disorders. In this context, severity is usually measured using some cognitive measure, such as IQ or verbal ability, and the distinctions between PDD subtypes is largely arbitrary. For example,

734

The Canadian Journal of Psychiatry

lower-functioning individuals with autism tend to have lower IQ and more severe autistic symptoms, whereas those at the other end of the spectrum (AS or high-functioning autism) tend to have fewer autistic symptoms and to be brighter. In fact, the utility or "validity" of classifying individuals as high- and low-functioning has quite a bit of evidence. Many studies have reported that lower-functioning individuals have more symptoms than those who are higher-functioning (28,29). They also differ on etiological parameters (30), on natural history and outcome (32), and on response to treatment. A massive study undertaken by Rapin and others (33) of a large number of preschool children clearly demonstrates the existence of 2 "taxa," or subgroups, of children with autism: higher-functioning and lower-functioning. Ironically, the clinical distinction that has the most evidence of validity-that is, high versus low functioning-is not captured by either of the 2 official classification systems (DSM or ICD). There are, however, 3 problems with this purely dimensional approach. First, we do not really know how to define higher and lower functioning. What measure should be used as a dimensional phenotype, and what threshold or cut-off should be used to classify children as higher and lower functioning? Most commonly, IQ is used, but should we use full-scale or verbal (or nonverbal) IQ? Perhaps other dimensional measures should be used instead. After all, IQ is an average ofseveral cognitive skills, and perhaps other, more specific cognitive functioning measures such as executive function or language comprehension would be better tools. In addition, IQ or other psychometric measures may not reflect functioning in the real world. An IQ test assesses functioning under highly structured situations. People with autism and PDD usually function much more poorly in the real world, simply because it is unstructured. Volkmar (31,34) and Carter (35) have argued that the Vineland Adaptive Behavior Scales (VABS), which measure functioning and developmental competence in everyday life, are much more accurate indicators of level of functioning and are very useful metrics for diagnosis. Second, there are no studies demonstrating the stability over time ofthis classification approach. Tests ofIQ in very young children tend to involve visual, motor, and nonverbal measures of matching and discrimination (such as object assembly and block design). These represent particular strengths for people with autism or PDD. However, as they age and mature, more conceptually complex tests of nonverbal ability are used in IQ tests, which may require more problem-solving and greater cognitive flexibility. Children with autism or PDD tend to do more poorly on these because they depend on executive-function skills, a weakness in PDD adolescents and adults (36). For example, in a longitudinal study of preschoolers with PDD (37), nonverbal IQ decreased steadily

Vol 45, NoS

over time, not because there was a loss of cognitive skills but rather because the nature of the IQ tests for older children changes and taxes the individual's executive-function skills to a greater extent. Therefore, a child may be highfunctioning as a preschooler but then fall below the threshold at a later age and be considered lower-functioning (37). Other studies have also found some instability in the classification ofcognitive functioning over time within individual children, although crude correlations over time are quite high (38). Third, this approach assumes that there is a single underlying dimension along which all individuals with PDD differ. It is a common assumption that symptom severity correlates highly with level offunctioning. However, this correlation may vary depending on the age and developmental level of the child. This is most clearly seen with respect to symptoms in the domain of communication. For example, lower-functioning or nonverbal children will not be able to display autistic symptoms such as echolalia, pronoun reversal, neologisms, and difficulties in conversation, and so they will have fewer symptoms on the communication domain than those who are higher-functioning. This inverse relation may also apply to repetitive activities. Resistance to change and rituals tend to be uncommon in very young children or those with severe developmental delay (39,40), perhaps because a certain level of cognitive sophistication is required to notice changes or to be able to plan and anticipate the performance of a ritual. Severity may represent 2 independent dimensions-one measuring symptom severity and another, level of functioning. One way to resolve this issue is to do a factor analysis, a statistical technique that determines whether a simpler measurement model underlies several observable and measurable constructs. A factor analysis could determine whether symptoms and level of functioning represent a single underlying dimension or 2 separate constructs. My colleagues and I have completed a factor analysis of 129 children with autism and AS (41). These children came from 2 separate samples, onea family study of largely lower-functioning children with autism (42) and the other a longitudinal study ofpreschoolers with high-functioning autism and AS (21). Factor analysis quite clearly demonstrated 2 dimensions: measures ofautistic symptoms in reciprocal social interaction, in communication, and in repetitive activities constituted one domain, and the VABS (that is, level offunctioning) represented a second, orthogonal, and independent domain. Combined, these 2 factors accounted for 70% of the phenotypic variation among children with high-functioning autism, low-functioning autism, and AS. The finding of2 orthogonal dimensions underlying phenotype variation is contrary to the usual assumption of a single dimension of severity. Which dimension should we use to classify children as higher or lower: symptom severity or level of functioning? The finding also challenges the notion of a single autistic spectrum. Clearly, a classification system is needed that can take account of both

October 2000

The Classification of Autism, Asperger's Syndrome, and Pervasive Developmental Disorder

dimensions. Using either the notion ofPDD as articulated in the DSM-IV or the idea of an autistic spectrum is inadequate. A third way is needed.

Validity of the Multivariate Phenotype for PDD Obviously, it is not enough to show by factor analysis that 2 orthogonal dimensions underlie phenotypic variation in autism or PDD, or however it is termed. For this to be a useful or valid finding, it is important to determine whether this model also carries implications with respect to etiology, outcome, and response to treatment. There is, in fact, some evidence that the genetic mechanisms responsible for level of functioning might be different from those responsible for variation in PDD symptoms. In a large sample of families having 2 children with some type ofPDD, phenotypic variation was examined within each sibship and compared between sibships (43). This is an important preliminary strategy to identifying genetically homogeneous subgroups. If both phenotypic dimensions tend to be consistent within a sibship, these dimensions might be governed by the same genotype. In our sample of multiplex sibships, PDD subtype did not run true within families; within a particular sibship, the proband might have autism and an affected sibling might have atypical autism or AS. Neither was there any tendency for symptom severity to run true within families. One sib might have many symptoms, another might have very few. In contrast, measures offunctioning (IQ and VABS) did tend to run true within families. Probands who were "higher-functioning" tended to have siblings who were also higher-functioning and vice versa. In other words, level of functioning runs true within families, but symptom severity does not. This suggests that the etiological mechanism underlying level of function may be different from the mechanism underlying symptom severity or PDD subtype. This observation was extended in a study of family history (44). Biological relatives of PDD probands had more autistic-like characteristics (the so-called broader autism phenotype) than those who had adopted a child with autism or were step-parents. The study also investigated the level of risk for the broader autism phenotype in relatives depending on the clinical characteristics of the proband. No significant associations were found between symptom severity or PDD subtype in the proband and risk of the broader autism phenotype in relatives. In contrast, relatives of high-functioning probands (identified either by IQ or the VABS) were at higher risk of the broader autism phenotype than were lowerfunctioning probands. This suggests that the genotype associated with high-functioning PDD probands (regardless of subtype) can also be expressed as the broader autism phenotype in family members. A variable expression of the lowerfunctioning phenotype does not include the broader autism phenotype and provides further evidence that the genetic mechanisms are different for higher and lower functioning.

735

Both these lines of evidence suggest that the neurobiological mechanism underlying variation in symptoms is different from that for level of functioning. The multivariate phenotype for PDO appears to be a valid representation of the underlying variation in PDD.

A Third Way: PDD Subtypes as Different Developmental Trajectories The PDDs are better conceptualized not as different subtypes or as lying on a spectrum but rather as different developmental pathways that are a function of both symptom severity and level of functioning. The evidence for this model comes from a longitudinal follow-up study of preschoolers with highfunctioning autism and AS. Sixty-eight preschool children with PDD have been followed since they were 4-6 years of age. Two follow-up assessments have been completed-one at 6-8 years ofage and another at 10-13 years ofage. The primary research question is whether autism and AS have a different outcome over time. The first analysis was cross-sectional, comparing the groups at 4-6 years of age (21). Children with AS were defined on the basis of having normal IQ, at least 1 impairment in reciprocal social interaction, at least 1 impairment in communication, and at least 1 example of a repetitive stereotypie behaviour. Also, as children, they had to be speaking spontaneously in phrases by 36 months ofage, with no evidence ofdeviant language development as indicated by persistent delayed echolalia, pronoun reversal, or neologisms. The children could meet the Autism Diagnostic Interview (ADI) algorithm diagnosis for autism, but as long as they met the communication-skill threshold outlined above they were considered to have AS. If the DSM-IV criteria had been strictly applied (using the ADI diagnostic algorithm as a template), few children that met our criteria for AS would have met the DSM-IV criteria for the disorder (21). Compared with those with high-functioning autism, at 4-6 years of age those with AS had higher nonverbal IQ; better language skills; fewer autistic symptoms in socialization, communication, and repetitive activities; and better adaptive functioning as assessed by the VABS socialization and communication domains (21). There were, however, no differences between the groups on their sex ratio or visual motor abilities. These initial differences were maintained at 6-8 years of age and at 10-13 years of age (45,46). Moreover, the developmental pathways were parallel; that is, the groups were neither coming together nor drifting apart. In other words, the magnitude ofthe differences between the groups was stable at the 3 points in time 4-6, 6-8, and 10-13 years of age The differences on follow-up could not be explained by potentially confounding differences in nonverbal IQ or in

736

The Canadian Journal of Psychiatry

D

e v e I

o p m

e

"\ a I

c

- - - - - _..._ - - j

i

o m p

e t

/ /

n

We stratified the autistic children into those at time 2 (age 6-8 years) who had attained the same level of language abilities demonstrated by the AS children at time 1 (age 4-6 years). With regard to outcome, the autistic children who developed the same level oflanguage fluency as the AS children at time I now very closely resembled the AS children on measures of socialization and communication (42). In other words, a subgroup of children with autism had "joined" the developmental pathway ofthe children with AS. The cohort ofnonverbal, but still high- functioning, autistic children were being left behind. With development, the differentiation ofPDD children appears to increase and becomes more specific.

/

/

e

e

Vol 45, No8

36 mas

60 mos

ss mos

e

Figure 1. Possible developmental pathways for preschool children with PDD

language abilities. For the most part, the differences were clinically meaningful and amounted to roughly a standard deviation, a substantial effect size.

There were also differences with respect to early variables that predicted outcome. We developed composite scores of verbal and nonverbal cognitive abilities at age 4-6 years and examined the extent to which these 2 variables, plus the clinical distinction of autism versus AS, predicted 3 different outcomes: the VABS socialization scores, the VABS communication scores, and a composite measure of autistic symptoms from the Autism Behaviour Checklist (47). The important finding was the significance of the interaction between clinical diagnosis and the early language measures (46). Language skills were a much stronger predictor ofsocial and communication skills at outcome among children with autism than among children with AS. Indeed, language measures predicted little of the outcome among the children with AS. This may not be surprising in view ofthe very strong differences in language between the groups, but it does emphasize that not only are outcomes between the groups different but also the predictors of outcome are different.

These results, however, obscure the fact that there was much more heterogeneity in the development of the children with autism than among the children with AS. By definition, children with AS were speaking fluently by 36 months ofage. By definition, children with autism were not. With age, the language skills of the children with autism could potentially improve more dramatically than those of AS children. The latter group may have reached a ceiling on their language skills.

Figure I represents a schematic representation of this model. Up to approximately 36 months ofage, there may exist an undifferentiated group of PDD children. Those who develop fluent language by 36 months ofage (by definition those with AS) progress on a different developmental pathway than do other higher-functioning children with PDD (the autism pathway). Some of these latter children will also develop fluent language, but only at a later age. Once they do, they may proceed to the developmental pathway ofAS children. There isa diminishing probability that children on the autism pathway will develop fluent language as chronological age increases. The end result is that there is a canalization of early development such that different subtypes ofPDD progress along different pathways and the possibility of crossing over diminishes with age. The key variable in determining pathway is the timing of the development offluent language. Some PDD children will develop fluent language early, others may never develop it at all. But the developmental pathway (hence the PDD subtype) that a child follows is critically determined by the timing of that event. The earlier it occurs, the more likely the child has a diagnosis of AS; the later it occurs, the more likely the diagnosis of autism. The different PDD subtypes or, better, pathways, then become markers for this key variable. Is AS different from high-functioning autism? Yes, no, andit depends. There is evidence that they have different symptoms, different level of functioning, different outcome, and different predictors of outcome. But there is also evidence that they are the same with respect to genetic etiology. It also depends on whether the children are assessed prior to 36 months ofage or after that crucial period defining the onset of AS. Whether PDD subtypes are the same or different is not a very helpful question to ask and leads only to circular debates about the meaning and interpretation of cross-sectional differences.

Conclusion The current classification system of PDDs as articulated in the DSM-IV and ICD-IO is flawed because it is based on a

October 2000

The Classification of Autism, Asperger's Syndrome, and Pervasive Developmental Disorder

mistaken medical model-the idea that discrete biological entities exist within PDD. This leads to a preoccupation with searching for cross-sectional differences between PDD subtypes, a strategy which has not been very useful in classification or in clinical practice. Moreover, it is hopelessly confusing to parents and to front-line clinicians. A dimensional approach, as articulated in the notion of autistic spectrum disorders, is better but still problematic, since more than 1 underlying dimension exists. An alternative approach is needed that incorporates both dimensions in a developmental context. PDD subtypes might be conceptualized as representing different developmental pathways that differentiate with chronological age. This does not necessarily mean that the diagnostic criteria themselves need to be changed; however, greater precision in differentiating autism from AS and PDDNOS is definitely required. It is rather the underlying model that must be reconceptualized according to the empirical data available. The benefit ofthis approach is that it would focus attention on interventions that move children from one PDD pathway to another over time-from, say, the autism pathway to the AS pathway. Research could more profitably focus on the genetic, epigenetic, and environmental variables that determine the pathway a child will follow. For example, what variables determine the timing of the onset of fluent and useful speech? Can some of these be modified biomedically or through the current practices of early intervention? To be truly useful, a classification system should point to intervention strategies. Perhaps that can be accomplished most readily by using a developmental context to classify these developmental disorders. Acknowledgement This research was supported by funding from the Ontario Mental Health Foundation and the Medical Research Council of Canada. The author thanks the many colleagues and students who have actively collaborated in the ideas and data presented.

References 1. American Psychiatric Association. Diagnostic and statistical manual ofmental disorders. 4th ed. Washington (DC): American Psychiatric Association; 1994. 2. Kanner L. Autistic disturbances of affective contact. Nervous Child 1943;2:217-50. 3. Kolvin I. Studies in the childhood psychoses. Br J Psychiatry 1971; 118:381-419. 4. Rutter M. Childhood schizophrenia reconsidered. Journal of Autism and Childhood Schizophrenia 1972;2:315-38. 5. Wing L. Gould J. Severe impairments of social interaction and associated abnormalities in children. Epidemiology and classification. J Autism Dev Disord 1979;9: 11-29. 6. American Psychiatric Association. Diagnostic and statistical manual ofmental disorders. 3rd ed. Washington (DC): American Psychiatric Association; 1980. 7. Waterhouse L, Wing L, Spitzer R, Siegel B. Pervasive developmental disorders: from DSM-III to DSM-III-R. J Autism Dev Disord 1992;22:525-49. 8. Webb EVJ, Lubos S, Hervas A, Scourfield J, Fraser WI. The changing prevalence of autistic disorder in a Welsh health district. Dev Med Child Neurol 1997;39: 150-2. 9. Asperger H. Die "autistischen Psychopathen" in Kindesalter. Archiv fur Psychiatrie une Nervenkrankheiten 1944; 117:76-136. 10. Ssucharewa GE, WolffS. The first account of the syndrome Asperger described? (Die schizoiden Psychopathien im Kindesalter) Eur Child Adolesc Psychiatry 1996;5:119-32.

737

Clinical Implications • The POD subtypes are better conceptualized as representing different developmental pathways. • Level of functioning does not always correlate with symptom severity. • The timing of the onset of fluent language is a key variable influencing outcome.

Limitations • We need a better understanding of the longitudinal course of POD. • Interventions need to be developed to move POD children from one pathway to another.

11. Wing L. Asperger's syndrome: a clinical account. Psychol Med 1981; 11: 115-29. 12. Gillberg C. AS in 23 Swedish children. Dev Med Child NeuroI1989;81:520-31. 13. Szatmari P, Bartolucci G, Bremner R. Asperger's Syndrome and autism: comparisons on early history and outcome. Dev Med Child Neurol 1989;31 :709-20. 14. Volkmar FR. The disintegrative disorders: childhood disintegrative disorder and Rett's disorder. In: Psychoses and pervasive developmental disorders in childhood and adolescence. Washington (DC): American Psychiatric Press Inc and American Academy of Child and Adolescent Psychiatry; 1996. p 223-43. 15. Sparrow SS, Rescorla LA, Provence S, Cordonson SO, Goudreau D, Cicchetti DV. A follow-up of atypical children. J Am Acad Child Adolesc Psychiatry 1986;25: 180-5. 16. Szatmari P. The validity of autistic spectrum disorders: a literature review. J Autism Dev Disord 1992;22:583-600. 17. Buitelaar JK, Van del' Gaag R, Klin A, Volkmar F. Exploring the boundaries ofpervasive developmental disorder not otherwise specified: analyses of data from the DSM-IV Autistic Disorder Field Trial. J Autim Dev Disord 1999;29:33-43. 18. Waterhouse L, Morris R, Allen D, Dunn M, Fein D, Feinstein C, and others. Diagnostic and classification in autism. J Autism Dev Disord 1996;26:59-86. 19. Mahoney WJ, Szatmari P, MacLean JE, Bryson SE, Bartolucci G, Walter SD, and others. Reliability and accuracy of differentiating pervasive developmental disorder subtypes. J Am Acad Child Adolesc Psychiatry 1998;37:278-85. 20. Miller IN, Ozonoff S. Did Asperger's cases have Asperger disorder? A research note. J Child Psychol Psychiatry 1997;38:247-51. 21. Szatmari P, Archer L, Fisman SF, Streiner DL, Wilson FJ. Asperger's syndrome and autism: differences in behavior, cognition and adaptive functioning. J Am Acad Child Adolesc Psychiatry 1995;34:1662-71. 22. Eisenmajer R, Prior M, Leekam S, Wing L, Gould J, Welham M, and others. Comparison of clinical symptoms in autism and Asperger's disorder. J Am Acad Child Adolesc Psychiatry 1996;35: 1523-31. 23. Myhr G. Autism and other pervasive developmental disorders: exploring the dimensional view. Can J Psychiatry 1998;43:589-95. 24. Prior M, Eisenmajer R, Leekam S, Wing L, Gould J, Ong B, and others. Are there subgroups within the autistic spectrum? A cluster analysis of a group of children with autistic spectrum disorders. J Child PsychoI Psychiatry 1998;39:893-902. 25. Hinshew SP. Externalizing behavior problems and academic underachievement in childhood and adolescence. Causal relationships and underlying mechanisms. Psychol Bull 1992;111:127-55. 26. Bernstein GA, Borchardt CM, Perwien AR. Anxiety disorders in children and adolescents: a review of the past 10 years. J Am Acad Child Adolesc Psychiatry 1996;3:1110-9. 27. Faraone SV, Biederman J. Do attention deficit hyperactivity disorder and major depression share familial risk factors? J Autism Dev Disord 2000;30:163-7. 28. Freeman BJ, Ritvo ER, Schroth PC, Ronick I, Guthrie D, Wake L. Behavioral characteristics of high and low IQ autistic children. Am J Psychiatry 1981; 138:25-9. 29. Bartak 1,Rutter M. Differences between mentally retarded and normally intelligent autistic children. Journal of Autism and Childhood Schizophrenia 1976;6: 109-20. 30. Goldberg J, Szatmari P, Nahmias C. Imaging of autism: lessons from the past to guide studies in the future. Can J Psychiatry 1999;44:793-801. 31. VolkmarFR, Sparrow SS, Goudreau D, Cicchetti D, Paul R, Cohen DJ. Social deficits in autism: an operational approach using the Vineland Adaptive Behavior Scales. JAm Acad Child Adolesc Psychiatry 1987;26: 156-61. 32. Lotter B. Factors related to outcome in autistic children. Journal of Autism and Childhood Schizophrenia 1974;4:263-77. 33. Rapin I, editor. Preschool children with inadequate communication. London: MacKeith Press; 1996. 34. Volkmar FR, Carter A, Sparrow SS, Cicchetti DV. Quantifying social development in autism. J Am Acad Child Adolesc Psychiatry 1993;32:627-32.

738

The Canadian Journal of Psychiatry

35. Carter AS. Volkmar FR. Sparrow SS. Wang JJ, LordC, Dawson G, and others. The Vineland Adaptive Behavior Scales: supplementary norms for individuals with autism. J Autism Dev Disord 1998;28:287-302. 36. Ozonoff'S, Roger SJ, Pennington BF. Asperger's syndrome: evidence ofan empirical distinction from high-functioning autism. J Child Psychol Psychiatry 1991;31: 1107-22. 37. Szatmari P, Streiner DL. The effect of selection criteria on outcome studies of children with pervasive developmental disorders (POD). EurChildAdolesc Psychiatry 1996;5: 179-84. 38. Ventner A, Lord C, Schopler E. A follow-up study of high functioning autistic children. J Child Psychol Psychiatry 1992;33:489-507. 39. Piven J, Jarper J, Palmer P, Arndt S. Course ofbehavioral change in autism: a retrospective study ofhigh-lQ adolescents and adults. J Am Acad Child Adolesc Psychiatry 1996;35:532-9. 40. Starr E, Szatmari P, Bryson SE, Zwaigenbaum L. Stability and change among high functioning children with pervasive developmental disorders: a two-year outcome study. Paper presented at the annual meeting of the American Academy of Child and Adolescent Psychiatry. October, 1999. Chicago.

Vol 45, NoS

41. Szatmari P, Merette C, Thivierge J, Maziade M. Quantifying severity in autism;a factor analysis [unpublished]. 42. Thivierge J, Bedard C, Cote R, Maziade M. Brainstem auditory evoked response and subcortical abnormalities in autism. Am J Psychiatry 1990; 147: 1600-1l 43. MacLean JE, Szatmari P, Jones MB, Bryson SE, Mahoney WJ, Bartolucci G and others. Familial factors influence level of functioning in pervasive developmental disorder. JAm Acad Child Adolesc Psychiatry 1999;38:746-53. 44. Szatmari P, MacLean JE, Jones MB, Bryson SE, Zwaigenbaum L, Bartolucci G, and others. The familial aggregation of the lesser variant in biological and nonbiological relatives of POD probands: a family history study. J Child Psychol Psychiatry 2000;41 :579-86. 45. Szatmari P, Bryson SE, Streiner DL, Archer L, Wilson FJ, Ryerse C. The two-year outcome ofpreschool children with autism and Asperger syndrome. Am J Psychiatry [Forthcoming]. 46. Szatmari P, Bryson SE, Boyle MH, Duku E. Predictors of outcome among preschool children with autism and Asperger Syndrome. Paper presented at the annual meeting of the American Acaademy of Child and Adolescent Psychiatry. October 1999. Chicago. 47. Krug DA, Arik J, Almond P. Behavior checklist for identifying severely handicapping individuals with high levels of autistic behaviour. J Child Psychol Psychiatry 1980;21:221-9.

Resume-e-La classification de l'autisme, du syndrome d'Asperger et des troubles envahissants du developpement (TED) Objectif: La classification courante des troubles envahissants du developpement comme la conceptualisent tant Ie DSM-IV que le Cllvl est extremement insatisfaisante pour de nombreux parents et cliniciens de premiere ligne ainsi que pour une proportion importante de chercheurs universitaires. La question est de savoir si cela est simplement attribuable au manque de donnees empiriques qui corroboreraient plus pleinement la validite du diagnostic ou s 'il y a un probleme plus fondamental dans la conceptualisation generale de la categorie qui ne reflete pas la « vraie » nature du phenomene. L 'objectifde cet article est de defendre I'exactitude de cette derniere hypothese. Pour prouver le bien-fonde de cet argument.j 'examinerai le developpement historique de la classification des TED, resumerai les donnees empiriques recentes sur les questions de fiabilite et de validite, et suggererai une nouvelle approche de la classification et de la comprehension.