SEARCH

SEARCH BY CITATION

Keywords:

  • autism;
  • pervasive developmental disability;
  • DSM;
  • prevalence;
  • sample size methodology

Abstract

  1. Top of page
  2. Abstract
  3. NOSOLOGY AND DSM
  4. CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM
  5. SUMMARY AND CONCLUSIONS
  6. REFERENCES

Since its initial description by Kanner in 1943, the criteria by which a diagnosis of autism or autism-like disorders was made—and their alleged etiologies portrayed—have undergone manifold changes, from a psychiatric disorder engendered by “refridgerator” parents to a neurodevelopmental disability produced in the main by genetic abnormalities. In addition, the behavioral characterization of autism has also entered the public consciousness and professional domains increasingly in the past 30 years, the effects of which we are continually coming to terms. A diagnosis of autism that once seemed quite unusual is now considered almost epidemic. Increasing numbers of individuals diagnosed with autism and related pervasive developmental disabilities will, in turn, affect the calculated prevalence of the disorder. In this essay, I attempt to account for the increasing prevalence of autism and autism-related disorders by examining its changing criteria, the individuals and instruments used to make the diagnosis, the reliability and validity of same, and the sample sizes and other aspects of the methodology needed to make an accurate estimate of its prevalence. © 2012 Wiley Periodicals, Inc.

‘When I use a word,’ Humpty Dumpty said in rather a scornful tone, ‘it means just what I choose it to mean—neither more nor less.’

Lewis Carroll

The pitcher stepped onto the mound, leaned forward to read the signs flashed by the catcher, and curled his fingers around the baseball. The batter readied his stance at the plate, awaiting the pitch. The ball was thrown. “Strike,” said the catcher. “Ball,” said the batter. They both looked at the umpire, who said, “It ain't nothin' ‘til I call it’.”

NOSOLOGY AND DSM

  1. Top of page
  2. Abstract
  3. NOSOLOGY AND DSM
  4. CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM
  5. SUMMARY AND CONCLUSIONS
  6. REFERENCES

Kanner's Autism and Childhood Schizophrenia

In his landmark paper published in 1943, entitled “Autistic disturbances of affective content,” Leo Kanner presented medical histories for 11 physically normal-looking children: eight boys and three girls. Despite their individual differences, they manifested a specific set of behaviors whose features formed a distinctive “syndrome,” the most salient of which was an inability to relate to others as any typically developing child of that age. He did not think their behaviors were like those of schizophrenic children, a withdrawal from an already existing social bond. Rather, it was an “extreme autistic aloneness,” that began at birth. Early signs included the child's inability to present an anticipatory posture in preparation to being picked up. Eight of the 11 had acquired expressive speech and language, but their language skills were not used to convey meaning in the social sense. There was a marked absence in initiating conversation. They related better to objects than people. Their speech was described as “parrot-like.” Personal pronouns were repeated rather than corrected grammatically for first- or third-person usage. Some of these children could be described as intellectually disabled (ID); however, most had good cognitive skills, and several had excellent rote memory. They repeatedly performed monotonous rituals. They were terrified of loud sounds or noises. This combination of features, while suggestive of schizophrenia, differed from what Kanner had observed previously in childhood schizophrenia.

Six years later, Kanner [1949] elaborated his account of the syndrome, referring to it as “early infantile autism.” He described it as such a well-defined condition that it could be identified within the first 2 years after birth. Unlike his earlier version, he stated that its features were the essential characteristics of childhood schizophrenia and indistinguishable from it, so much so that it should be regarded as the earliest appearance of childhood schizophrenia.

A PubMed search from the date of Kanner's seminal paper until the year in which the Diagnostic and Statistical Manual of Mental Disorders (DSM-I) was first published in 1952, seeking the terms “autism,” “autistic,” or “childhood schizophrenia,” resulted in no known manuscripts published with any of these terms in the title or abstract. When DSM-I appeared in 1952, under the heading “Schizophrenic reaction, childhood type,” there appeared the following sentence: “Psychotic reactions in children, manifesting primarily autism, will be classified here.”

Following the publication DSM-I in 1952 until its first revision in 1968 (DSM-II), a PubMed search found 127 articles published with the terms “autism,” “autistic,” or “childhood schizophrenia” in the title or abstract, beginning with a article by Peplau [1964] and referring to “infantile autism.” DSM-II does not refer to autism in particular; but, under disorder “295.8 Schizophrenia, childhood type” appears the following description:

This category is for cases in which schizophrenic symptoms appear before puberty. The condition may be manifested by autistic, atypical and withdrawn behavior; failure to develop identity separate from the mother's; and general unevenness, gross immaturity and inadequacy of development. These developmental defects may result in mental retardation, which should also be diagnosed.

This account has little in common with Kanner's 1943 description but is more akin to the one he published in 1949, the latter of which became the one used to make a diagnosis. Of the 127 papers cited in PubMed, 41 (30%) referred to the disorder as childhood schizophrenia. DSM-II description notwithstanding, early attempts to establish a diagnosis of autism capitalized on Kanner's original descriptions of the syndrome [1943], Rimland's E-2 Eorm [Rimland, 1964], the British Working Party's Checklist or Creak's Nine Points [1964], the Behavior Rating Instrument for Autistic Children (BRIAC) [Ruttenberg et al., 1966], or Rendle-Short and Clancy's Checklist [1968]. A first approximation of the prevalence of autism was undertaken by Wing et al. [1967], who surveyed the population of 8–10 year olds in Middlesex, England. Positive cases were referred to as “autism condition of early childhood” which included children whose age of onset was between 2 and 5 years in addition to Kanner's appearance at birth criterion. Diagnoses were based on Kanner's two essential criteria, lack of responsiveness and insistence of sameness. Results found 32 such children—23 boys and nine girls—in a population of 78,000, giving rise to a prevalence of 4.5 per 10,000. According to Wing et al. [1967], the incidence could have been higher, owing to the fact that milder cases may have been missed, and severe cases ending in early death were not known.

Autism and the Diagnostic and Statistical Manuals—III to IV-TR

From 1968 until 1980, the year in which DSM-III was published, a major shift in perspective of the disorder occurred. A PubMed search noted a marked increase to 755 papers were published with the terms “autism,” “autistic,” or “childhood schizophrenia,” in the title or abstract, 98 of which (13%) referred to childhood schizophrenia.

From 1968 until 1980, the year in which DSM-III was published, a major shift in perspective of the disorder occurred. A PubMed search noted a marked increase to 755 papers were published with the terms “autism,” “autistic,” or “childhood schizophrenia,” in the title or abstract, 98 of which (13%) referred to childhood schizophrenia.

While the number of papers published had increased, the proportion referring to the disorder as childhood schizophrenia had decreased sharply. On the other hand, there were 19 papers that used autism or autistic in conjunction with the terms “neurological” or “neurology.” An early paper by Gubbay et al. [1970] found more than three-quarters of the 25 children diagnosed with autism presented with abnormal EEGs.

During the interval between the publications of DSM-II and DSM-III, and in an attempt to assess autism more accurately, several other checklists were developed. One of the first to utilize psychometric techniques to evaluate the reliability and validity of their survey instrument was developed by Krug et al. [1979]—the Autistic Behavior Checklist—which combined items from earlier devices such as BRIAC and Creak. Items were assigned to one of five factors found to be significant in the diagnosis: Sensory Processing; Relating; Body and Object Use; Language; Social and Self-Help. These items, along with their weighted scores, could discriminate children with autism from those who were deaf or severely intellectually or emotionally disabled. Instead of Kanner's two factors—lack of responsiveness and insistence on sameness—three criteria, “severe impairments of social interaction, language abnormalities, and repetitive stereotyped behaviors” [Wing and Gould, 1979] emerged as the “triad of impairment” essential for a diagnosis of autism [Wing, 1981a].

Not satisfied with the inadequate state of classification prior to DSM-III, and in an attempt to determine the prevalence of autism, Wing and Gould [1979] undertook to examine 15,000 children <15 years of age, living in London and diagnosed with at least one these three aberrant behaviors. These researchers found the prevalence of impaired social interaction was 21.2:10,000, of whom 4.9:10,000 exhibited the triad of behaviors associated with typical autism, similar to the earlier results obtained in Middlesex by Wing et al. [1967].

From 1980, when DSM-III was published, until 1987, the year in which DSM-III-R was published, the number of studies found in a PubMed search for the terms “autism,” “autistic,” or “childhood schizophrenia,” increased to 905, of which 65 (7%) referred to autism as childhood schizophrenia. The number and proportion of manuscripts published referring to autism as childhood schizophrenia continued to decline, while the number of papers which included terms such as “neurological” or “neurology” increased to 33. That is to say, during the 20-year period from 1968 to 1987, the emphasis in the description of autism had shifted away from a purely psychiatric disorder to one which could be understood as neurodevelopmental. Indeed, the diagnostic criteria in DSM-III entitled “Infantile Autism” included the triad of impairment set forth by Wing and Gould [1979], but were more specific:

  • (A)
    Onset before 30 months of age.
  • (B)
    Pervasive lack of responsiveness to other people (autism).
  • (C)
    Gross deficits in language development.
  • (D)
    If speech is present, peculiar speech patterns such as immediate or delayed echolalia, metaphorical language, pronominal reversal.
  • (E)
    Bizarre responses to various aspects of the environment, for example, resistance to change, peculiar interest in or attachments to animate or inanimate objects.
  • (F)
    Absence of delusions, hallucinations, loosening of associations, and incoherence as in Schizophrenia.

Controversy over the DSM-III criteria ensued. Volkmar et al. [1986] cited as one of their concerns age of onset. Defined as pervasive development disorders (PDDs), were infantile autism and childhood onset PDD different categorical disorders or different in degrees of severity in the same disorder? They also noted that the term, “pervasive” lack of responsiveness, was not defined for infantile autism. Instead, they preferred the phrase “gross and sustained impairment of social relationships” as employed for childhood onset PDD.

In DSM-III-R [1987], autism was no longer referred to as “Infantile Autism” but as “Autistic Disorder” (AD). The age of onset criterion was increased from 30 to 36 months; and, if noted later than 36 months, to refer to the diagnosis as “childhood onset.” The criterion, absence of delusions, was eliminated. Criteria for the triad of impairment were fleshed out in greater detail. Pervasive lack of responsiveness became Criterion A: “Qualitative impairment in reciprocal social interaction,” with descriptions of five subtypes. Deficits in language development and peculiar speech patterns became Criterion B: “Qualitative impairment in verbal and non-verbal communication,” with descriptions of six subtypes. Bizarre responses to the environment became Criterion C: “Markedly restricted repertoire of activities and interests,” with descriptions of five subtypes. A field trial of DSM-III-R criteria conducted by Spitzer and Siegel [1990] examined data from 11 sites, using various algorithms to optimize diagnostic assessment. The greatest sensitivity (96%) was attained by using four criteria: two from A, one from B, and one from C. Greatest specificity was derived using a total of eight of 16 criteria, including two from A, one from B, and one from C. Thus, for DSM-III-R, at least two subtypes from each triad of impairment must be present, and a total of at least eight of the 16 criteria need be present to make the diagnosis.

In their review of AD and PDD just prior to the release of DSM-III-R, Rutter and Schopler [1987] supported the criteria imparted by DSM-III, but also noted that there were no clear behavioral boundaries separating AD from severe ID or serious language disorders. Moreover, there was no distinct central behavioral feature or biomarker by which to differentiate completely AD from those other disorders. At the other end of the cognitive spectrum, they thought it would be difficult to distinguish high-functioning individuals with AD from those with seemingly normal intelligence, for example, individuals with Asperger syndrome, the consequence of which placed the researcher in another quandary.

According to Wing [1981b], the clinical picture of Asperger syndrome is analogous to that of AD: More common in males than females; speech/language difficulties; lack of facial expression; impaired social interaction; repetitive, stereotyped behaviors; poor motor coordination; superior rote memories. Asperger himself noted that the disorder was permanent feature of behavior, as it is with AD. At the time of Wing's paper, no systematic efforts had been undertaken to determine the prevalence of Asperger syndrome; but, based on the earlier study of the London population by Wing and Gould [1979], a total of six children they observed exhibited features similar to Asperger, produced a prevalence of 1.7:10,000 for the disorder. Surprisingly, Wing [1981b] stated that “Asperger syndrome can be regarded as a form of schizoid personality,” that is, as a psychiatric as opposed to neurodevelopmental disorder. Wing noted further that, while Kanner and Asperger disorders had many behaviors in common, Asperger thought autism was a form of psychosis, while he considered his own syndrome as a personality disorder. But, were they—are they—so different as to classify them as categorically different disorders? Was high-functioning AD different from Asperger syndrome? Despite their similarities, Wing skirted the issue but acknowledged that the triad of impairments common to both syndromes presented the same problems affecting socialization and education. However, if both were elements of the same taxonomic class of PDD, should prevalence estimates for AD and Asperger syndrome be calculated separately?

DSM-IV [1994] and DSM-IV-TR [2000] continued to refer to autism as “Autistic Disorder,” and maintained use of the triad of impairment: qualitative impairment of social interaction; qualitative impairments in communication; and, restricted, repetitive, or stereotyped patterns of behavior. Each of these criteria has four subtypes but the criteria for making the diagnosis were reduced to a total of six or more subtypes, with at least two from impaired social interaction and one each from impaired communication and restricted and/stereotyped behavior. A direct consequence of the reduced number of subtypes to make a diagnosis was to increase the likelihood that such a diagnosis could be made.

The Prevalence of Autism is Linked to Nosology

In the past 40 years, from the time when the frequency of autism in the population was first assessed until the present, a graph of its prevalence—and presumably its incidence, thereby—has increased exponentially (Fig. 1). Many reports in the popular press have suggested that autism has become epidemic. Then again, the diagnostic criteria had been changed five times in the interim. The most dramatic changes occurred between DSM-II and DSM-III, when autism was no longer referred to as a psychiatric disorder—childhood schizophrenia—but as a neurodevelopmental disability. After DSM-III-R was published until the release of DSM-IV-TR, the diagnostic criteria—the triad of impairment—remained essentially the same. What did change were the number of criteria in each of these categories used to make the diagnosis.

thumbnail image

Figure 1. Prevalence studies of autism 1970–2010.

Download figure to PowerPoint

If, instead of envisaging the prevalence of autism as in Figure 1, we examine the prevalence for each interval in which the DSM criteria for autism as a neurodevelopmental disorder for any particular revision was implemented, a very different picture emerges. According to Rutter and Schopler [1992], the criteria employed in DSM-III provided a narrower set of behaviors compared to the broader range in DSM-III-R. Also, according to Rutter and Schopler [1992], most clinicians preferred a broader definition to the criteria than contained in DSM-III. To compare the effect of a broader versus a narrower scope of AD between DSM-III and DSM-III-R, prevalence surveys of autism were selected for the years 1982 to 1992, inclusive, from Table 6-1 from Fombonne et al. [2011]. There were six reported studies in the interval between 1980 and 1987, and nine studies between 1988 and 1992 (Figs. 2 and 3). Given the relative rarity of autism reported at the time, studies in which the sample procured was <10,000 would have had an inordinate effect on rates since one or two cases missed or falsely identified as positive would markedly affect the prevalence in one direction or another. Rutter [2005] noted that, in order to attain a reliable and valid prevalence estimate, one of the factors that must be taken into account is that the sample size, N, must be sufficiently large. Although I will show later why larger minimum sample sizes are necessary, studies in which the sample size was <10,000 persons were eliminated from the present analysis.

thumbnail image

Figure 2. Prevalence studies of autism 1980–1987 under DSM-III.

Download figure to PowerPoint

thumbnail image

Figure 3. Prevalence studies of autism 1988–1994 under DSM-III-R.

Download figure to PowerPoint

In one study during this period, Matsuishi et al. [1987] found a prevalence rate almost five times that of the other five studies. I compared that study from Japan with another published in the same year, albeit from a US population but using the same DSM-III criteria [Burd et al., 1987]. I constructed a 99% confidence interval around the difference between the two proportions and found no overlap in the two prevalence rates. Therefore, the Matsuishi et al. study was designated an outlier and removed from subsequent analysis. Another study by Wignyosumarto et al. [1992] in which the population sampled was N = 5,120 was also removed from analysis because of its small size. The prevalence rates for the remaining studies using DSM-III and DSM-III-R were then compared. I found the median prevalence of autism under DSM-III to be about 3.26:10,000. A regression line computed from the prevalence rates was relatively flat between 1980 and 1987 (Fig. 2) and comparable to the earlier studies by Wing et al. [1967] and Wing and Gould [1979]. When the broader criteria were implemented under DSM-III-R, the median prevalence almost tripled, to 9.5:10,000. Nonetheless, the computed regression equation also remained relatively flat in the interval between 1988 and 1992. In other words, the broader criteria in DSM-III-R produced a quantum leap in the prevalence rate, but the increased rate remained stable over time.

In the years between 1993 and 2000, the criteria for AD were modified again under DSM-IV. The total number of criteria used to make the diagnosis was reduced from 8 to 6. During this period, Fombonne et al. [2011] report 10 studies of prevalence for autism. Of these, three assessed populations with fewer than 10,000 inhabitants [Honda et al., 1996 (N = 8,537); Arvidsson et al., 1997 (N = 1,941); Kadesjö et al., 1999 (N = 826)]. In these studies, each of which reported using ICD-10 criteria, the prevalence rates ranged between 47.1:10,000 and 50:10,000, six times that which had been noted previously under the DSM-III-R criteria. Although ICD-10 criteria, not DSM-IV criteria, were used to identify autism in these studies, both employ the three key elements of impairment in reciprocal social interaction, communication, and restricted, repetitive patterns of behavior. In fact, only two studies published in 2000 reported using DSM-IV criteria. Figure 4 presents the prevalence rates of all 10 studies. When all 10 studies are included in the analysis, the resulting regression line remains flat, but the median prevalence rate, 14.7:10,000, has jumped again by more than 50% over the prevalence from previous DSM-III-R set of criteria, primarily as a result of these three small studies.

thumbnail image

Figure 4. Prevalence studies of autism 1995–2000 under DSM-IV.

Download figure to PowerPoint

The most interesting pattern results when prevalence rates obtained during the period in which only DSM-IV and/or ICD-10 criteria are used to make a diagnosis of autism. Volkmar et al. [1992] found ICD-10 had the best combination of sensitivity and specificity regarding diagnostic accuracy of autism compared to DSM-III and DSM-III-R. Since ICD-10 and DSM-IV criteria are essentially the same, it was of interest to see how prevalence rates fare when comparable criteria are employed. Results are shown in Figure 5 in which 24 studies from 1996 to 2010 were amassed. Regression analysis again presents an essentially zero slope during the period in which median prevalence rate is about 21.34:10,000, a 50% jump in the median prevalence compared to the period 1995–2000. According to data from the most recent interval 1995–2010, the prevalence rate for autism has remained constant over past 15 years; and, as a result of broadening and operationalizing the criteria, among other factors, the prevalence more or less quadrupled over the rate found in the early studies by Wing et al. [1967] and Wing and Gould [1979].

According to data from the most recent interval 1995–2010, the prevalence rate for autism has remained constant over past 15 years; and, as a result of broadening and operationalizing the criteria, among other factors, the prevalence more or less quadrupled over the rate found in the early studies by Wing et al. [1967] and Wing and Gould [1979].

thumbnail image

Figure 5. Prevalence studies of autism 1996–2010 under DSM-IV-TR or ICD-10.

Download figure to PowerPoint

CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM

  1. Top of page
  2. Abstract
  3. NOSOLOGY AND DSM
  4. CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM
  5. SUMMARY AND CONCLUSIONS
  6. REFERENCES

Clearly, there are serious methodological issues regarding estimates of the prevalence of autism, as Fombonne [2009] and others have noted. In addition to the noted changes in diagnostic criteria, Wing and Potter [2002] identified several possible causes. Among others, they mention:

  • (1)
    Different study methodologies.
  • (2)
    Increased awareness among professionals, parents, caregivers, and the general public.
  • (3)
    Pervasive developmental disabilities may be associated with other DSM disorders.
  • (4)
    Development of specialized services for PDD.
  • (5)
    Identification of possible etiologies, for example, genetic mutations.

In order to make valid and reliable estimates, Rutter [2005] listed five factors that must be taken into account:

  • (1)
    A proper, systematic screening of the population.
  • (2)
    The sample size, N, must be sufficiently large (to which I would add: and representative of the population from which it is drawn).
  • (3)
    A well-defined population at risk for AD and PDD.
  • (4)
    Concentrate on a cohort for which diagnoses of AD or PDD will be reliable and valid.
  • (5)
    Diagnoses made by trained professionals using standardized assessment tools.

Systematic Screening Using Standardized Assessment Tools

Screening protocols of individuals for autism can utilize one or more of several tools available. One can classify individuals as AD by using trained professionals and/or clinicians, the gold standard for autism; or, one can administer one of several instruments standardized for diagnosing individuals for AD or PDD, for example, the CHAT, M-CHAT, ADI-R, CARS, or ADOS-G, depending on the age of the individual tested. Given the variability of behavior in 2–3 year olds, the CHAT and M-CHAT should probably not be implemented for prevalence estimates, nor should such a young cohort be included [Fisch, in press]. That said, the age group between 5 and 20 years would likely provide the optimal cohort for assessment.

Screening instruments most frequently invoked for prevalence estimates are the ADI-R and ADOS-G, which are considered the standards for research in autism [Mazefsky and Oswald, 2006].

Screening instruments most frequently invoked for prevalence estimates are the ADI-R and ADOS-G, which are considered the standards for research in autism [Mazefsky and Oswald, 2006].

The ADI-R was introduced by Lord et al. [1994] as a replacement for an earlier version, the ADI, the scores for which are based on a semi-structured interview. Items for the ADI-R were generated by following clinical descriptions in DSM-IV and ICD-10. These researchers reported on the reliability and validity for each of the items [Lord et al., 1994]. Initial reliability was based on 20 children, 10 of whom were previously diagnosed as autistic, and 10 who were non-autistic children with ID, all between 3 and 5 years of age. Chance-corrected weighted kappa coefficients for items ranged from 0.64 to 0.89. A second study in which 25 autistic and 25 non-autistic children with ID were recruited found 24/25 children with AD and 23/25 with ID correctly diagnosed by the ADI-R algorithm. As a result, Lord et al. [1994] concluded that the ADI-R was both a reliable and valid instrument for assessing autism. A much larger follow-up multicenter study which collected assessments from eight sites found very high sensitivity and specificity of the ADI-R's ability to differentiate AD from non-AD individuals [Lord et al., 1997].

Later studies of the psychometric properties of the ADI-R were less convincing, however. Some researchers noted differences in the factor structure of the ADI-R as well as its other psychometrics. For example, Lecavalier et al. [2006] found evidence to support the three-factor structure of the ADI-R, but noted that most items loaded onto a single factor. van Lang et al. [2006] found an alternate three-factor model that fit the ADI-R better. In a much larger study in which test and cross-validation subsamples were generated, Frazier et al. [2008] identified a two-factor structure for the ADI-R, with factor loadings onto stereotyped language and restricted, stereotyped behavior on one factor, and impaired socialization and communication on a second factor. That two-factor structure was confirmed by another large sample study by Snow et al. [2009].

One consequence of a two-factor structure for the ADI-R may have found its expression in its ability to discriminate AD from certain groups of non-AD individuals. For example, Mildenberger et al. [2001] used the ADI-R to differentiate a small sample of children with AD from those with receptive language disorder, and found that 10/11 children with AD were diagnosed correctly, while 15/16 children with receptive language disorder were also correctly identified. Another outcome of a possible two-factor structure may be the extent to which ADI-R corresponds (or not) to other instruments' and/or clinicians' diagnoses. When comparing the ADI-R to the CARS, Pilowsky et al. [1998] examined 83 individuals with suspected autism and found chance-adjusted agreement, kappa = 0.36, while Saemundsen et al. [2003] assessed 54 children and found kappa for the most stringent ADI-R criteria = 0.40. When the ADI-R was compared to the ADOS-G, de Bildt et al. [2004] found that for children evaluated between 5 and 8 years of age, kappa = 0.67, but for children older than 8 years, kappa = 0.16. When the ADI-R was compared to clinicians' diagnoses, Mazefsky and Oswald [2006] observed that, for children aged 2–8 years, kappa = 0.46; Cox et al. [1999] used the ADI-R to examine children at 20 and 42 months of age and found kappa = 0.61; and Gray et al. [2008] assessed children and discovered kappa = 0.46. Although some researchers have argued that middling values of kappa are satisfactory, there are no benchmarks for kappa as there are for the Pearson correlation coefficient. Shrout [1998] advised that only kappa ≥ 0.80 should be considered substantial. Therefore, if the ADI-R is used to screen for a diagnosis of autism for prevalence studies, its results should be viewed with some caution, since kappa estimates for the ADI-R do not attain the criterion level suggested by Shrout.

Sample Size

To my knowledge, no published study of autism prevalence has proposed what should constitute a sufficiently large sample. Wing and Potter [2002] assert that smaller sample sizes permit a more precise assessment of individuals with autism, but Fombonne [2005] found a negative correlation between prevalence estimate and sample size in the 34 studies he examined. Neither author recommended a minimum sample size. Curiously, the epidemiological and statistical literature provide limited information for establishing prevalence in case-only cohorts; but, age group, case ascertainment, reliability and validity of diagnostic assessment, and whether the study is prospective or retrospective, are some of the nuisance parameters that would have bearing on variability in the estimates.

One relatively simple approach to estimating minimum sample size makes use of the sensitivity and specificity of the diagnostic instrument, as well as the estimated prevalence of the disorder, as provided by Buderer [1996]. Previous concerns about its use notwithstanding, the instrument most frequently made reference to in prevalence studies has been the ADI-R. Therefore, for the purposes of this analysis, I employed the sensitivity and specificity of the ADI-R to obtain rough estimates for minimum sample size. Sample size estimates were based on two sources: Buderer's Table 3, which provides sample sizes based on sensitivity; and, second, her computation of sample size based on the equation:

  • equation image

where TN, true negative; FP, false positive; P, prevalence estimate.

Expected sensitivity was based on four published reports of sensitivity and specificity for the ADI-R [de Bildt et al., 2004; Mazefsky and Oswald, 2006; Gray et al., 2008; Papanikolaou et al., 2009]. Sensitivity ranged from 0.72 [de Bildt et al., 2004], 0.73 [Gray et al., 2008], 0.75 [Mazefsky and Oswald, 2006] to 0.87 [Papanikolaou et al., 2009]. Sample size estimates based on the equations presented in Buderer [1996] were computed for three sensitivity estimates in her Table 3: 0.75, 0.80, 0.85; and, for two prevalence estimates: P = 0.00095 [the median estimate from 1988 to 1992 (DSM-III-R studies) [Fombonne et al., 2011; Table I]; and, second, P = 0.00213 [the median estimate from studies between 1996 and 2010 (DSM-IV or ICD-10 studies) [Fombonne et al., 2011; Table I]. Results are shown in Figure 6 and indicate that, for the most upbeat estimate of sensitivity [85%], a sample size N = 23,136 would be needed using the median prevalence estimate from studies using DSM-IV or ICD-10 criteria from 1996 to 2010. Sample size based on the more likely obtained sensitivity estimates, 0.75, and using earlier DSM-III-R criteria, would require a sample size N = 72,301.

Table I. Prevalence Estimates Based on True Positive (TP) and False Negative (FN) Calculations for the ADI-R From Three Published Studies
 TP + FNDSM-III-RDSM-IVDSM-IV-TR + ICD-10DSM-IV + ICD-10
Median Prevalence estimate (P)/Study 0.000950.001470.002160.00213
Papanikolaou et al. [2009]6568,42244,21830,09330,517
Gray et al. [2008]120126,31681,63355,55656,339
de Bildt et al. [2004]123129,47483,67456,94557,747
thumbnail image

Figure 6. Sample size computations based on sensitivity, specificity, and estimated prevalence rates.

Download figure to PowerPoint

To compute sample size estimates from the above equation, expected sensitivity was based on the aforementioned published reports of sensitivity and specificity for the ADI-R, from which I enlisted three assessment studies employing the ADI-R with the largest samples of individuals with PDD [de Bildt et al., 2004; Gray et al., 2008; Papanikolaou et al., 2009]. Results are shown in Table I. Median prevalence estimates were based on the studies presented in Table I [Fombonne et al., 2011], for each of four intervals in which different DSM criteria were in effect, as noted in the above paragraph. Resulting sample sizes range from N = 30,093 (using DSM-IV-TR or ICD-10 criteria) to N = 129,474 (using DSM-III-R criteria).

From these results, one can surmise that the very minimum number of individuals enrolled in an at-risk population and needed for a proper prevalence study of autism should be between 25,000 and 30,000. Of the studies listed in Fombonne et al. [2011], Table I, a third of the studies (16/48) should therefore be eliminated from any meta-analysis because of inadequate sample size.

A Well Defined At-Risk Population and the Problem of Ascertainment Bias

Barbaresi et al. [2009] note that, in order to obtain accurate prevalence rates, it is essential that estimates be based on research diagnoses independent of clinical diagnoses, in the same population, and during the same time period. Research diagnoses, however, can be problematic. Laidler [2005] examined the methods used by the US Dept. of Education to identify children with autism, which bases its data on assessment criteria for autism established by each state legislature and school district. These data are then used to determine the prevalence of autism in various communities in the US. Laidler noted that, until 1984, prevalence based on cohort was relatively unchanged. Cohorts born after 1984 exhibited not only higher prevalence rates at the earliest age examined—6 years—but increased at increasing linear rates at later ages. These increases were due, in part, to mandated reporting of PDD after 1992. On the other hand, Laidler [2005] also noted that cutoffs for diagnoses of autism were subjective and had become less stringent among medical practitioners during the ensuing years.

Prevalence studies recruit their participants by one of two means: retrospectively or prospectively. Retrospective studies tend to ascertain their cases from medical, clinic, and/or school records in which a history of the individual, typically a child or adolescent, is described. From these reports, trained professionals are said to make a diagnosis of AD or PDD, using research criteria based on the reports and/or a follow-up clinical evaluation. According to Laidler [2005], these reports may not be accurate because of the criteria each state and school district uses, and prevalence estimates based on them will, therefore, be biased. Of studies in which cases were obtained retrospectively in this manner, the ones of greatest significance and concern are from New Jersey, Georgia, California, and the Centers for Disease Control (CDC).

In New Jersey, Bertrand et al. [2001] evaluated all children, 3–10 years old, whose parents resided in Brick Township. In addition to medical, clinic, and school records, parents also self-referred their children to the researchers. The census count for children 3–10 years—the denominator—was increased by a 25% inflation factor, producing an estimate of N = 8896. Of the 75 children identified with autism, 53 (71%) were assessed clinically. The remainder were evaluated by record review. Of these 75, 60 were confirmed as PDD, 36 who met the full criteria for AD, producing an extremely high prevalence of 67:10,000. While the clinical follow-up procedure likely produced an accurate assessment of many of the childrens' status, evaluation of younger children aged 3 years may not be correct for the reasons stated here earlier. Also, it is not known how many of the children were provided by parents' self-report, which may have biased the outcome as well. In addition, based on the aforementioned estimates for minimum sample size, the cohort selected was probably too small to gauge a proper prevalence estimate, especially as the number is a weighted estimate and not an actual count. Finally, and most importantly, Brick Township was selected because of the high rate of autism that had been noticed anecdotally previously. As the authors state, the prevalence may have been different if adjacent townships had been included in the study. All these factors probably produced the unusually high and likely biased estimate of prevalence.

The California study of autism prevalence may be among the most controversial. Croen et al. [2002] examined a cohort of children ages 5–12 years, born between 1987 and 1994, who had been diagnosed previously by a qualified professional as having any neurological disorder related to ID, including autism, and eligible to receive services from the Department of Developmental Services (DDS). The DDS contains files for all such referrals. Other children identified as autistic were obtained from a Special Education database. Diagnoses of autism for this cohort would then have been based on DSM-III, DSM-III-R, or DSM-IV criteria, depending on the birth year. However, the recorded diagnosis of autism was not confirmed by a follow-up clinical assessment. As expected, prevalence for each cohort year increased from 4:10,000 in 1987–1988 to 16:10,000 in 1993–1994. These researchers also noted a decrease in diagnosis of ID correlated with an increase in the diagnosis of autism. Blaxill et al. [2003] challenged these findings, stating that Croen et al. had made several methodological errors. In particular, they noted that the inverse correlation between autism and ID prevalence was an artifact of Croen et al.'s statistical analysis. In response, Croen and Grether [2003] re-analyzed their data and found that, in fact, the inverse relationship between autism and ID prevalence was spurious.

However, the California experience has not been typical of other states in the US where diagnostic substitution appears to occur [Shattuck et al., 2007]. What was not contested in the California study was the observed increase in autism diagnoses among children who were identified with disorders other than ID, a different form of diagnostic substitution. Given that reporting PDD was mandated as of 1992, and the services that would be available for children who would be given a diagnosis of autism, it would not be unexpected that a number of children would have been given such a diagnosis, when, in fact that may not have been appropriate. There is much anecdotal evidence to suggest that parents often pressured pediatricians and other professionals to make a diagnosis of autism in order to obtain the specialized services, for example, speech/language therapy, that would be provided as a result. Indeed, in her follow-up commentary of the California study, Eagle [2004] noted that an increasing number of children with ADHD, OCD, or LD, among other psychiatric disorders, were now being diagnosed as high-functioning AD or Asperger syndrome.

At about the same time, and in a similar fashion, Yeargin-Allsopp et al. [2003] examined the prevalence of autism in Atlanta, Georgia. They examined the records of 289,456 children 3–10 years old residing in Atlanta in 1996. Many of their cases were referred from the Atlanta Developmental Disabilities Surveillance Program, but public schools were their primary source of case identification. Cases were scored and confirmed by an expert in autism using DSM-IV criteria. To check reliability, a random sample of abstracted evaluations was scored by another reviewer. Although percent agreement between reviewers was high (96%), chance-corrected agreement was middling (kappa = 0.46). Of the children enrolled, a diagnosis of autism was given to 987, producing a prevalence rate of 34:10,000. Also noted were significant increases and decreases in the prevalence of autism for each age–year cohort, but which were not explained by the researchers. Yeargin-Allsopp has stated elsewhere [Lewis and Dictenberg, 2010] that an increase in prevalence occurred in conjunction with the implementation of broader criteria, which we have shown earlier to be the case. However, that would not explain the parabolic increase-then-decrease in prevalence in the Atlanta study. Regrettably, as in the California study, cases were not confirmed by a clinical assessment. That is, these researchers relied upon the written record as the primary source for the diagnosis. Curiously, although data for other developmental disabilities were also available, those prevalence rates were not shown. As Eagle [2004] noted in the California study, other psychiatric diagnoses may have decreased in conjunction with the increased diagnoses of autism. Shattuck et al. [2007] reported that, from 1976 to 1992, the number of children with ID decreased by 41% while the number of children diagnosed as LD increased 198%. Shattuck also observed that, despite US federal mandates, the prevalence of autism varied widely from state to state, which suggests further that factors apart from bona fide diagnoses were involved.

Other reports of prevalence by the CDC [Surveillance Summaries, 2007] regarding the prevalence of autism and related disorders based on retrospective review of medical, clinical, and/or school records underscore similar difficulties in determining the proper ascertainment of valid cases of autism. Results from the population-based incidence study in Olmstead County, Minnesota [Barbaresi et al., 2009] also provide evidence of the problems associated with retrospective record review. Barbaresi et al. examined medical records from the Mayo clinic from 1994 to 1998 of children diagnosed with any developmental, psychiatric or neurological disorder. Using DSM-IV criteria, they identified 124 who fulfilled the conditions for research-based autism. Using the same Medical Diagnostic Index, they identified individuals who had actually received a clinician's diagnosis of AD or PDD. These researchers found that, while both the incidence of research-identified and clinically diagnosed autism increased as DSM criteria changed, the clinician's incidence of autism was always much lower than the research-identified rate.

A Cohort for Which Diagnoses of AD or PDD Will Be Reliable And Valid

An accurate assessment of AD and its prevalence not only depend on the criteria employed in DSM and ICD, but also on the reliability and validity of the assessment device and the behavioral characteristics of the individual assessed. As noted earlier, diagnoses at or before the age of 2–3 years can be problematic. At that age, variability in behavior of the developing infant can be great. Ventola et al. [2007] examined children between 16 and 32 months of age with suspected developmental delays and diagnosed 150/192 (78%) with AD or pervasive developmental disability (PDD). For those who returned for a follow-up evaluation, 38/46 (83%) continued to meet the criteria for AD or PDD. Of 18-month-old infants diagnosed as AD or PDD, Baird et al. [2000] found that the M-CHAT detected 33/94 (35%), with a false positive rate of 72%. Chlebowski et al. [2010] found that the CARS correctly identified 181/243 (74%) 2-year olds diagnosed as AD or PDD.

Chakrabarti and Fombonne [2001] surveyed 15,500 children residing in Staffordshire, England in June, 1998. Children ranged in age from 2½ to 6½ years. Unfortunately, there was no breakdown of the sample by age. Children were assessed in four phases, the fourth of which involved a clinical assessment by a multidisciplinary team. Of those surveyed, 26 were diagnosed with autism, for a prevalence of 16.8:10,000. Altogether, 97 were diagnosed as PDD, which produced an unusually high prevalence of 62.6:10,000. As prevalence surveys go, this one was one of the more rigorous with regard to clinician confirmation. However, as noted earlier, the sample size is small comparatively; and, lamentably, the number of children evaluated between 2½ and 3½ years of age was not furnished. As noted above, a positive diagnosis for children in this age range can be problematic. How independent raters were presented cases to confirm a positive diagnosis was not elaborated either. Were negative cases presented as well?

Given the problem associated with diagnosis at an early age, most studies of autism prevalence evaluate older cohorts of children and/or adolescents. However, as children age, changes in behavior also occur. Symptom frequency, especially repetitive and stereotyped behaviors decrease with increasing age [Starr et al., 2003; Shattuck et al., 2007]. Fisch [in press] noted that most studies that re-examine children with AD or PDD find a significant shift in diagnoses away from AD to PDD. If the prevalence of AD is evaluated among older children, some may not be included in the numerator but will be expressed as an increase in the prevalence of PDD.

A recent study by Baron-Cohen et al. [2009] illustrates many of the concerns raised here. These researchers attempted to assess children 5–9 years of age in the county of Cambridgeshire, England. Letters of participation were sent to schools that were part of the Special Educational Needs register. Questionnaires were then sent to schools that responded positively, as well as parents of children in those schools. Confirmation of diagnosis was made by a health professional, based on the questionnaires returned. Fewer than half of the schools responded, providing a register of 8,824 children. Results obtained found 83 children with so-called Autism Spectrum Condition, which produced a prevalence of autism of 11:10,000; and, of PDD, 94:10,000. In his commentary, Fombonne [2010] observed that there were several problems associated with the study. First, the response rate from schools was low, as was the parent participation rate. Apparently, there were instances of children either younger or older than the specified age range were included in the analysis, affecting both the number of positive cases, and the total N sampled. Third, case confirmation-lacked consistency; and, finally, the use of weighting procedures to estimate the total number of children in that age range may have been imprecise. I would add that the small size and lack of clinician confirmation makes the prevalence estimate less than useful.

Regarding study design, few investigations of prevalence offer little in the way of methodological support for the reliability of the procedure, as Fombonne [2005] observed. Even detailed multistage studies, such as those performed by the CDC, show no indication of the reliability of the estimates other than to provide 95% confidence intervals for their respective binomial distributions, that latter of which may be inappropriate as well. There is no known biomarker for autism, therefore, classification depends entirely on the individual performing the assessment, by whatever means employed. Consequently, a diagnosis of autism is subjective [Laidler, 2005], and always has been. Moreover, as Rutter [2005] and others noted, as the boundaries defining autism and autism-like disorders expand with each change in official criteria, the margins between autism disorder and atypical autism, and between atypical autism and ordinary eccentricity, have become increasingly fuzzy.

Moreover, as Rutter [2005] and others noted, as the boundaries defining autism and autism-like disorders expand with each change in official criteria, the margins between autism disorder and atypical autism, and between atypical autism and ordinary eccentricity, have become increasingly fuzzy.

So, when parental reports are used to determine prevalence, responses to questionnaires or live telephone interviews should be interpreted with great caution.

Several such parental-report studies have been performed [Gurney et al., 2006; Kogan et al., 2008, 2009]. Using the National Survey of Children's Health (NSCH), Gurney et al. examined the records based on 85,272 parental interviews conducted between January, 2003 and July, 2004. Among the questions asked in the interview was, “Has a doctor or health professional ever told you that your child has autism?” Based solely on the parents' responses, 483 said “yes,” from which Gurney et al. calculated that the prevalence rate of children ages 3–17 years was 53:10,000. In collaborative effort with the CDC, Kogan et al. [2009] conducted a random-digit dialed interview of households in the US in which children also resided. The screening question used by the interviewer was “[Has] a doctor or other health care provider ever told you that your child had … autism [or other PDD]?” Of the interviews completed, the study had enrolled 78,037 children, of whom 453 were “diagnosed” as autistic or autistic-like, producing a prevalence rate of 58:10,000. Whatever the response, there was no confirmation by medical, clinic, or school record; nor was there a follow-up assessment by a trained clinician. Contrary to Fombonne's 2009] assertion that prevalence estimates must be viewed as under-estimates, studies from survey research show that disclosure rates on sensitive items such as drug or alcohol use is significantly higher when live telephone interviews are employed than if information was taken from anonymous internet surveys [Hines et al., 2010]. Insofar as accuracy is concerned, Lee et al., [2010] found only moderate concordance between telephone interview and face-to-face diagnosis of major depression (sensitivity was 77%; specificity was 75%).

Not all is bleak. Retrospective studies can show great rigor. Kielinen et al. [2000] estimated the prevalence—actually the incidence—of autism in two Finnish provinces. The at-risk population born 1979–1984 (3–18 year olds) was 152,732. Cases were found through a central registry of hospitals and institutions. Although children were initially diagnosed using one of several sets of earlier diagnostic criteria, all children were reviewed using ICD-10 and DSM-IV criteria. Each child diagnosed as autistic or PDD had the diagnosis confirmed by a psychiatrist or neurologist. The number of cases of autism was 212, which produced an incidence of 13.9:10,000, with a larger-than-expected 50% of children with IQ scores in the normal range.

Methodologically rigorous prospective studies offer a more optimal procedure by which to estimate the prevalence of autism and/or pervasive developmental disabilities. One such is a study by Baker [2002], although it too has its drawbacks. She examined children residing within the Australian Capital Territory (ACT) who were referred to their Child and Adolescent Mental Health Service (CAMHS). Referrals were made for 2 years: 1989 and 1997. Assessments for the first cohort were made by the author; for the second cohort, other psychologists were enlisted. Both cohorts were assessed with standardized instruments. DSM-III-R criteria were used in 1989 to make a diagnosis of AD or PDD; DSM-IV criteria were used in 1997. Of those referred, nine children from the 1989 cohort, and 27 from the 1997 cohort were diagnosed as AD or PDD. The total population of 0–19 year olds in 1989 was 92,088, producing a prevalence of 9.8:10,000. In 1997, the total population of 0–19 year olds was 91,673, producing a prevalence of 29.5:10,000, triple that of the 1989 prevalence estimate. Baker attributes the threefold increase in prevalence to increased awareness, but the change in criteria was also likely a factor. Unfortunately, the largest proportion of positive diagnoses of AD or PDD was made among the 0- to 3-year-old group—67% in 1989 and 48% in 1997—which, as noted previously can prove problematic in terms of accuracy and may over-estimate the prevalence of autism in this population.

Having noted the problems of case definition and detection associated with prevalence studies, Magnússan and Saemundsen [2001] prospectively examined two birth cohorts in Iceland. The first cohort was composed of individuals born between 1974 and 1983; the second were born between 1984 and 1993. The first cohort was assessed primarily by ICD-9 criteria; the second, by ICD-10 criteria. Cases were detected initially at elementary schools and referred for diagnosis to one of two centers specializing in assessment of psychiatric disorders. The population of the first cohort was 42,403; for the second cohort, 43,153. The prevalence of autism or atypical autism in the first cohort was 4.2:10,000; the prevalence of autism or atypical autism in the second cohort was 13.2:10,000.

SUMMARY AND CONCLUSIONS

  1. Top of page
  2. Abstract
  3. NOSOLOGY AND DSM
  4. CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM
  5. SUMMARY AND CONCLUSIONS
  6. REFERENCES

The description and criteria by which autism and autism-related disorders are defined have changed markedly since Kanner's initial publication in 1943, as has its suspected etiology. In the absence of a gold standard biomarker, however, the diagnosis of any disorder based on behavioral evaluation will be influenced by the apparent manifestation of criteria used to define the disorder. Inevitably, this will impinge upon any epidemiologist's ability to determine accurately the prevalence and/or incidence of the disorder. Notwithstanding the difficulties regarding definition, assessing the prevalence of any behavioral disorder is fraught with methodological issues: Study design (retrospective or prospective); population sampled and recruited; sample size; cohort age range; public and professional awareness of the disorder; validity, reliability, and stability of criteria used for assessment and case detection; reliability and validity of the screening instrument and/or clinician; the probability model used to construct confidence intervals; all are factors affecting the obtained prevalence of the disorder. As I have attempted to show, prevalence rates for autism have increased markedly in the past half century or so, not so much as a developing epidemic, but as a result of the changing parameters associated with the factors just noted.

As I have attempted to show, prevalence rates for autism have increased markedly in the past half century or so, not so much as a developing epidemic, but as a result of the changing parameters associated with the factors just noted.

Some researchers have suggested that the prevalence of autism will likely range from 30 to 60:10,000 [e.g., Rutter, 2005]. However, based on recent population-based studies such as those in Finland, Australia, and Iceland, I suspect the rate will be much lower, perhaps more nearly half or a third that range of rates. That is not to say that the occurrence of autism or autistic-like behaviors in the population should be minimized. To the contrary, however, over-estimating the rate will not necessarily translate into programs needed by these individuals to address the issues and difficulties they confront.

REFERENCES

  1. Top of page
  2. Abstract
  3. NOSOLOGY AND DSM
  4. CRITERIA FOR ESTABLISHING VALID AND RELIABLE PREVALENCE OF AUTISM
  5. SUMMARY AND CONCLUSIONS
  6. REFERENCES
  • Arvidsson T, Danielsson B, Forsberg P, Gillberg C, Johansson M, Kjellgren G. 1997. Autism in 3–6 year old children in a suburb of Goteborg, Sweden. Autism 1: 163173.
  • Baird G, Charman T, Baron-Cohen S, Cox A, Swettenham J, Wheelwright S, Drew A. 2000. A screening instrument for autism at 18 months of age: A 6-year follow-up study. J Am Acad Chil Adolesc Psych 39: 694702.
  • Baker HC. 2002. A comparison study of autism spectrum disorder referrals 1997 and 1989. J Autism Dev Disord 32: 121125.
  • Barbaresi WJ, Colligan RC, Weaver AL, Katusic SK. 2009. The incidence of clinically diagnosed versus research-identified autism in Olmsted County, Minnesota, 1976–1997: Results from a retrospective, population-based study. J Autism Dev Disord 39: 46.
  • Baron-Cohen S, Scott FJ, Allison C, Williams J, Bolton P, Matthews FE, Brayne C. 2009. Prevalence of autism-spectrum conditions: UK school-based population study. Br J Psychiatry 194: 500509.
  • Bertrand J, Mars A, Boyle C, Bove F, Yeargin-Allsopp M, Decoufle P. 2001. Prevalence of autism in a United States population: The Brick Township, New Jersey, investigation. Pediatrics 108: 11551161.
  • Blaxill MF, Baskin DS, Spitzer WO. 2003. Commentary: Blaxill, Baskin, and Spitzer on Croen et al., 2002, the changing prevalence of autism in California. J Autism Dev Disord 33: 223226.
  • Buderer NM. 1996. Statistical methodology: I. Incorporating the prevalence of the disease into the sample size calculation for sensitivity and specificity. Acad Emerg Med 3: 895900.
  • Burd L, Fisher W, Kerbeshian J. 1987. A prevalence study of pervasive developmental disorders in North Dakota. J Am Acad Child Adolesc Psychiatr 26: 700703.
  • Chakrabarti S, Fombonne E. 2001. Pervasive developmental disorders I n preschool children. JAMA 285: 30933099.
  • Chlebowski C, Green JA, Barton ML, Fein D. 2010. Using the child autism rating scale to diagnose autism spectrum disorders. J Autism Dev Disord 40: 78799.
  • Cox A, Charman T, Baron-Cohen S, Drew A, Klein K, Baird G, Swettenham J, Wheelwright S. 1999. Autism spectrum disorders at 20 and 42 months of age: Stability of Clinical and ADI-R diagnosis. J Child Psychol Psychiatr 40: 719732.
  • Creak M. 1964. Infantile autism. Englewood Cliffs, New Jersey: Prentice-Hall Inc. pp 221236.
  • Croen LA, Grether JK. 2003. Response: A response to Blaxill, Baskin, and Spitzer on Croen et al. (2002) “The changing prevalence of autism in California.J Autism Dev Disord 33: 227229.
  • Croen LA, Grether JK, Hoogstrate J, Selvin S. 2002a. The changing prevalence of autism in California. J Autism Dev Disord 32: 207215.
  • de Bildt A, Sytema S, Ketelaars C, Kraijer D, Mulder E, Volkmar F, Minderaa R. 2004. Interrelationship between Autism Diagnostic Observation Schedule—Generic (ADOS-G), Autism Diagnostic Interview—Revised (ADI-R), and the Daignostic and Statistical Manual of Mental Disorders (DSM-IV-TR) classication of children and adolescents with mental retardation. J Autism Dev Disord 34: 129137.
  • Eagle RS. 2004. Commentary: Further commentary on the debate regarding increase in autism in California. J Autism Dev Disord 34: 8788.
  • Fisch GS. Autism and Epistemology III: Child development, behavioral stability, and reliability of measurement. Am J Med Genet (in press).
  • Fombonne E. 2005. Epidemiology of autistic disorder and other pervasive developmental disorders. J Clin Psychiary 66: 38.
  • Fombonne E. 2009. Epidemiology of pervasive developmental disorders. Pediatr Res 65: 591598.
  • Fombonne E. 2010. Estimated prevalence of autism spectrum conditions in Cambridgeshire is over 1%: Commentary. Evid Based Ment Health 13: 32.
  • Fombonne E, Quirke S, Hagen A. 2011. Epidemiology of pervasive developmental disorders. In: Amarel DG, Dawson G, Geschwind DH, editors. Autism spectrum disorders. Oxford: Oxford University Press. pp 90111.
  • Frazier TW, Youngstrom EA, Kubu CS, Sinclair L, Rezai A. 2008. Exploratory and confirmatory factor analysis of the Autism Diagnostic Interview—Revised. J Autism Dev Disord 38: 474480.
  • Gray KM, Tonge BJ, Sweeney DJ. 2008. Using the Autism Diagnostic Interview—Revised and the Autism Diagnostic Observation Schedule with young children with developmental delay: Evaluating diagnostic validity. J Autism Dev Disord 38: 657667.
  • Gubbay SS, Lobascher M, Kingerlee P. 1970. A neurological appraisal of autistic children: Results of a Western Australian survey. Dev Med Child Neurol 12: 422429.
  • Gurney JG, McPheeters ML, Davis MM. 2006. Parental report of health conditions and health care use among children with and without autism: National Survey of Children's Health. Arch Pediatr Adolesc Med 160: 825830.
  • Hines DA, Douglas EM, Mahmood S. 2010. The effects of survey administration on disclosure rates to sensitive items among men: A comparison of an internet panel sample with a RDD telephone sample. Comput Human Behav 26: 13271335.
  • Honda H, Shimizu Y, Misumi K, Niimi M, Ohashi Y. 1996. Cumulative incidence and prevalence of childhood autism in children in Japan. Br J Psychiatry 1692: 228235.
  • Kadesjö B, Gillberg C, Hagberg B. 1999. Brief report: Autism and Asperger syndrome in seven-year-old children: A total population study. J Autism Dev Disord 29: 327331.
  • Kanner L. 1943. Autistic disturbances of affective contact. Nervous Child 2: 21750.
  • Kanner L. 1949. Problems of nosology and psychodynamics of early infantile autism. Am J Orthopsychiatry 19: 416426.
  • Kielinen M, Linna SL, Moilanen I. 2000. Autism in Northern Finland. Eur Child Adolesc Psychiatry 9: 162167.
  • Kogan MD, Strickland BB, Blumberg SJ, Singh GK, Perrin JM, van Dyck PC. 2008. A national profile of the health care experiences and family impact of autism spectrum disorder among children in the United States, 2005–2006. Pediatrics 122: e1149e1158.
  • Kogan MD, Blumberg SJ, Schieve LA, Boyle CA, Perrin JM, Ghandour RM, Singh GK, Strickland BB, Trevathan E, van Dyck PC. 2009. Prevalence of parent-reported diagnosis of autism spectrum disorder among children in the US, 2007. Pediatrics 124: 13951403.
  • Krug D, Arigk J, Almond P. 1979. Autism Screening Instrument for Educational Planning (ASIEP). Autism Behavior Checklist. Portland: Autism Screening Instrument for Educational Co.
  • Laidler JR. 2005. US Department of Education data on “autism” are not reliable for tracking autism prevalence. Pediatrics 116: e120e124.
  • Lecavalier L, Aman MG, Scahill L, McDougle CJ, McCracken JT, Vitiello B, Tierney E, Arnold LE, Ghuman JK, Loftin RL, Cronin P, Koenig K, Posey DJ, Martin A, Hollway J, Lee LS, Kau AS. 2006. Validity of the autism diagnostic interview-revised. Am J Ment Retard 111: 199215.
  • Lee S, Tsang A, Mak A, Lee A, Lau L, Ng KL. 2010. Concordance between telephone survey classification and face-to-face interview diagnosis of one-year major depressive episode in Hong Kong. J Affect Disord 126: 155160.
  • Lewis MJ, Dictenberg JB. 2010. Genes, brain, and behavior: Development gone awry in autism? Ann NY Academy Sci 1205: E21E36.
  • Lord C, Rutter M, Le Coutour A. 1994. Autism Diagnostic Interview—Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J Autism Dev Disord 24: 659685.
  • Lord C, Pickles A, McLennan J, Rutter M, Bregman J, Folstein S, Fombonne E, Leboyer M, Minshew N. 1997. Diagnosing autism: Analysis of data from the Autism Diagnostic Interview. J Autism Dev Disord 27: 501517.
  • Magnússan P, Saemundsen E. 2001. Prevalence of autism in Iceland. J Autism Dev Disord 31: 153163.
  • Matsuishi T, Shiotsuki Y, Yoshimura K, Shoji H, Imuta F, Yamashita F. 1987. High prevalence of infantile autism in Kurume City, Japan. J Child Neurol 2: 268271.
  • Mazefsky CA, Oswald DP. 2006. The discriminative ability and diagnostic utility of the ADOS-G, ADI-R, and GARS for children in a clinical setting. Autism 10: 533549.
  • Mildenberger K, Sitter S, Noterdaeme M, Amorosa H. 2001. The use of the ADI-R as a diagnostic tool in the differential diagnosis of children with infantile autism and children with receptive language disorder. Eur Child Adolesc Psychiatry 10: 248255.
  • Papanikolaou K, Paliokosta E, Houliaras G, Vgenopoulou S, Giouroukou E, Pehlivanidis A, Tomaras V, Tsiantis I. 2009. Using the Autism Diagnostic Interview—Revised and the Autism Diagnostic Observation Schedule—Generic for the diagnosis of autism spectrum disorders in a Greek sample with a wide range of intellectual disabilities. J Autism Dev Disord 39: 414420.
  • Peplau LA. 1964. Infantile autism. Perspect Psychiatr Care 5: 112122.
  • Pilowsky T, Yirmiya N, Shulman C, Dover R. 1998. The Autism Diagnostic Interview—Revised and the Child Autism Rating Scale: Differences between diagnostic systems and comparison between genders. J Autism Dev Disord 28: 143151.
  • Rendle-Short J, Clangy HG. 1968. Infantile autism. Med J Aust 1: 921922.
  • Rimland B. 1964. Infantile autism. Englewood Cliffs, New Jersey: Prentice-Hall Inc. pp 221236.
  • Ruttenberg BA, Dratmann ML, Fraknoi J, Wemar C. 1966. An instrument for evaluating autistic children. J Am Acad Child Psychiatry 5: 453478.
  • Rutter M. 2005. Incidence of autism spectrum disorders: Changes over time and their meaning. Acta Paediatrica 94: 215.
  • Rutter M, Schopler E. 1987. Autism and pervasive developmental disorders: Concepts and diagnostic issues. J Autism Dev Disord 17: 159186.
  • Rutter M, Schopler E. 1992. Classification of pervasive developmental disorders: Some concepts and practical considerations. J Autism Dev Disord 22: 459482.
  • Saemundsen E, Magnússan P, Smári J, Sigurdardóttir S. 2003. Autism Diagnostic Interview—Revised and the Child Autism Rating Scale: Convergency and discrepancy in diagnosing autism. J Autism Dev Disord 33: 319328.
  • Shattuck PT, Mailick Seltzer M, Greenberg JS, Orsmond GI, Bolt D, Kring S, Lounds J, Lord C. 2007. Change in autism symptoms and maladaptive behaviors in adolescents and adults with an autism spectrum disorder. J Autism Dev Disord 37: 17351747.
  • Shrout PE. 1998. Measurement reliability and agreement in psychiatry. Stat Methods Med Res 7: 301317.
  • Snow AV, Lecavalier L, Houts C. 2009. The structure of the Autism Diagnostic Interview—Revised: Diagnostic and phenotypic implications. J Child Psychol Psychiatry 50: 732742.
  • Spitzer RL, Siegel B. 1990. The DSM-III-R field trial of pervasive developmental disorders. J Am Acad Child Adolesc Psychiatry 29: 855862.
  • Starr E, Szatmari P, Bryson S, Zwaigenbaum L. 2003. Stability and change among high-functioning children with pervasive developmental disorders: A 2-year outcome study. J Autsim Dev Disord 33: 1522.
  • Surveillance Summaries. 2007. Prevalence of Autism Spectrum Disorders—Autism and developmental disabilities monitoring network, Six Sites, United States, 2000; 14 Sites, United States, 2002. Morbidity and Mortality Weekly Report, pp 1–40.
  • van Lang ND, Boomsma A, Sytema S, de Bildt AA, Kraijer DW, Ketelaars C, Minderaa RB. 2006. Structural equation analysis of a hypothesised symptom model in the autism spectrum. J Child Psychol Psychiatry 47: 3744.
  • Ventola P, Kleinman J, Pandey J, Wilson L, Esser E, Boorstein H, Dumont-Mathieu T, Marshia G, Barton M, Hodgson S, Green J, Volkmar F, Chawarska K, Babitz T, Robins D, Fein D. 2007. Differentiating between autism spectrum disorders and other developmental disabilities in children who failed a screening instrument for ASD. J Autism Dev Disord 37: 425436.
  • Volkmar FR, Cohen DJ, Paul R. 1986. An evaluation of DSM-III criteria for infantile autism. J Am Acad Child Psychiatry 25: 190197.
  • Volkmar FR, Cicchetti DV, Bregman J, Cohen DJ. 1992. Three diagnostic systems for autism: DSM-III, DSM-III-R, and ICD-10. J Autism Dev Disord 22: 483492.
  • Wignyosumarto S, Mukhlas M, Shirataki S. 1992. Epidemiological and clinical study of autistic children in Yogyakarta, Indonesia. Kobe J Med Sci 38: 119.
  • Wing L. 1981a. Language, social, and cognitive impairments in autism and severe mental retardation. J Autism Dev Disord 11: 3144.
  • Wing L. 1981b. Asperger's syndrome: A clinical account. Psychol Med 11: 115129.
  • Wing L, Gould J. 1979. Severe impairments of social interaction and associated abnormalities in children: Epidemiology and classification. J Autism Dev Disord 9: 1129.
  • Wing W, Potter D. 2002. The epidemiology of autism spectrum disorders: Is the prevalence rising? Ment Retard Dev Disabil 8: 151161.
  • Wing JK, O'Connor N, Lotter V. 1967. Autistic conditions in early childhood: A survey in Middlesex. Br Med J 3: 389392.
  • Yeargin-Allsopp M, Rice C, Karapurkar T, Doernberg N, Boyle C, Murphy C. 2003. Prevalence of autism in a US metropolitan area. JAMA 289: 4955.