Journal of Autism and Developmental Disorders

, Volume 39, Issue 4, pp 557–571

Dimensional Structure of the Autism Phenotype: Relations Between Early Development and Current Presentation

Authors

    • Department of Child and Adolescent PsychiatryPhilipps-University Marburg
  • Mardjan Ghahreman
    • Department of Child and Adolescent PsychiatryPhilipps-University Marburg
    • Center for Social Psychiatry Kurhessen
  • Judith Smidt
    • Department of Child and Adolescent PsychiatryPhilipps-University Marburg
  • Helmut Remschmidt
    • Department of Child and Adolescent PsychiatryPhilipps-University Marburg
Original Paper

DOI: 10.1007/s10803-008-0656-5

Cite this article as:
Kamp-Becker, I., Ghahreman, M., Smidt, J. et al. J Autism Dev Disord (2009) 39: 557. doi:10.1007/s10803-008-0656-5

Abstract

The dimensional structure of higher functioning autism phenotype was investigated by factor analysis. The goal of this study was to identify the degree to which early symptoms of autism (measured using the ADI-R) could be predictive of the current symptoms of autism as identified using the ADOS, the adaptive behavior scales, IQ scores and theory of mind scores. Participants included 140 subjects with Full Scale IQ > 70 (104 with autism spectrum diagnosis, 36 with non autism diagnosis, age range 6–24). For the early development as well as for the current presentation a multi-factor solution was found. In line with other studies we found that the social interaction and communication domains are closely related to one factor namely: Social communication. An additional factor implies anxious and compulsive behavior which is associated with current social communication functioning.

Keywords

Autism spectrum disordersDimensional structure of autism phenotypeHigh-functioning autismAsperger syndromeAdaptive behavior

Introduction

Autism spectrum disorders (ASD) are complex neurodevelopmental disorders characterized by impairments in three domains: Social interaction, communication and repetitive, stereotyped behavior. The question arises whether the variability and complexity of the phenotype/clinical representation is consistent with the conceptualization found in DSM-IV and ICD-10: Is the underlying symptom structure of ASD similar to the three behavior domains of DSM and ICD triad? There is evidence that suggests there is only one underlying dimension (Constantino et al. 2004) or possibly three as was, for example, postulated in a study of Robertson et al. (1999) (see Table 1). However, there is controversy about whether variability in clinical expression represents variation in a single or in a number of different underlying dimensions, so that data is needed to develop a dimensional structure of the autism phenotype on the basis of experiential methods.
Table 1

Current studies of the dimensional structure of autism spectrum disorders: whole symptomatology

Author (year)

Sample

Diagnosis

Age (mean/SD or range)

IQ (mean/SD; range)

Methods: statistical procedure and instruments

Results: factors (accounted variance)

Tanguay et al. (1998)

63

Autism, Asperger, PDD-NOS

80 month/33 month

14 subjects: no data available

49 subjects: 79/27; 29–143

Factor analysis:

25 ADI-R items

Affective reciprocity

Joint attention

Theory of mind

(50.5% of variance)

Robertson et al. (1999)

51

Autism, Asperger, PDD-NOS

88 month/–/36 months to 191 months

6 < 70

Factor analysis:

30 ADOS items of communication and social interaction, correlations to the ADI-R factors

Affective reciprocity

Joint attention

Theory of mind

(72% of variance)

Bölte and Poustka (2001)

262

Autism, atypical autism

5–49 years (14.4/8.4)

For 139 subjects: 77.1/27.3; 27–136

For 99 subjects: ~20–85

Factor analysis:

ADI-R all algorithm items

Social communication I

Speech

Social communication II

(46.1% of variance)

Szatmari et al. (2002)

62 low functioning

67 high functioning

Autism

Low functioning: 11.6/5.8

High functioning: 5.5/0.9

 

Factor analysis:

3 domains of ADI-R

VABS scores for communication, activities of daily living, socialization

Autistic symptoms

Level of functioning

69.8% of variance

Tadevosyan-Leyfer et al. (2003)

292

Autism

Range: 2–47

 <30–89

Factor analysis:

98 ADI-R items

Spoken language

Social intent

Compulsions

Developmental

Milestones

Savant skills

Sensory aversions

Constantino et al. (2004)

226

80 PDD (autism, AS, PDD-NOS)

47 ADHD

6 unspecific developmental disorder

35 other Non-PDD

Range: 4–18

 

Factor analysis and cluster analysis

63 items from ADI-R (N = 44)

SRS (maternal report N = 168)

SRS (teacher report N = 61)

Singular, continuously distributed underlying factor

27–40% of variance

Van Lang et al. (2006); Boomsma et al. (2008)

255; 263

130 Autism, AS, PDD-NOS

125 non-autism; 110 autism, 153 PDD-NOS/Asperger

Range: 4–20; 4–24

55/23 20–129; 35–152

Factor analysis: ADI-R algorithm domain items

Impaired social communication

Impaired play skills

Stereotyped language/behavior

Georgiades et al. (2007)

209

165 autism

21 PDD-NOS

23 Asperger

109 month/65 month

28–482 month

26 subjects: no data available

183 subjects:

66/27; 23–132

Factor analysis and correlations with independent variables: ADI-R algorithm domain items, IQ, VABS

Social-communication

Inflexible language and behavior

Repetitive sensory and motor behavior (50% of variance)

Gotham et al. (2007, 2008)

1,139;

1,259

912 autism, 439 non-autism, 279 developmental delays; 970 autism, 98 nonautism ASD, 214 non-ASD developmental delays

14–82 month; 18 month to 16 years

Nonverbal IQ mean ~77; ~79

Factor analysis: ADOS items

Social affect

Restricted, repetitive behaviors

ADI-R Autism Diagnostic Interview-Revised, ADOS Autism Diagnostic Observation Schedule, PDD Pervasive Developmental Disorder, PDD-NOS Pervasive Developmental Disorder not Otherwise Specified, SRS Social Responsiveness Scales, VABS Vineland Adaptive Behavior Scales, ASD Autism Spectrum Disorder

An empirically developed dimensional approach that defines the spectrum on multiple dimensions may offer several advantages. It may, for example, result in more correspondence between the results of genetic research and the phenotype of autistic disorders, provided the pathology can be summarized by empirical and valid behavior dimensions (Volkmar et al. 2004; van Lang et al. 2006; Hus et al. 2007).

Another consideration is the relation between the autistic symptoms and other behavioral features like adaptive behavior, intelligence or age. The relationship between adaptive functioning, autism symptomatology and intelligence remains unclear (Klin et al. 2007; Saulnier and Klin 2007). Some studies show that intelligence and language abilities are more appropriate predictors for lower functioning individuals with autism than for the higher end of the cognitive spectrum present in autism (Nordin and Gillberg 1998; Howlin 2000, 2003; Howlin et al. 2004; Szatmari et al. 2003; Tsatsanis 2003; McGovern and Sigman 2005). Outcome studies of high functioning individuals with autism seem to indicate that higher intellectual potential is not a good predictor of adaptability in adulthood (Howlin 2003). The ability to transfer cognitive potential into real-life skills is often considerably impaired (Bölte and Poustka 2002; Klin et al. 2007; Saulnier and Klin 2007) and the scores for adaptive behavior in daily life are substantially below the verbal-IQs “highlighting the magnitude of adaptive impairment despite cognitive potential” (Klin et al. 2007, p. 748). The few existing studies have revealed that early language and nonverbal skills are important predictors of outcome regarding communication and social domains (Szatmari et al. 2003), a negative correlation regarding age was established, but no relationship to the current symptomatology of autism (Klin et al. 2007; Saulnier and Klin 2007).

An empirically developed and valid dimensional structure of the symptomatology may also give better indicators of an individual’s progress because it has considerable implications for intervention programs. Factors that enhance the acquisition of adaptive skills have to be found to optimize treatment. If we can identify the autistic symptoms that influence the adaptive skills, we would have an implication for treatment options to improve the adaptive skills in daily life. “This information needs to be used as a platform for research to clarify what factors truly can predict real life success. This issue has received much less attention than it should, both scientifically and clinically “(Klin et al. 2007, p. 758).

Recent studies have been carried out in order to clarify the dimensional structure of the autistic spectrum. Table 1 gives an overview of the relevant studies which include the whole symptomatology. Table 2 includes studies concerning the dimensional structure of the domain of restricted, repetitive behavior and interests. The majority of these studies are based on the data from the Autism Diagnostic Interview-Revised (ADI-R) (Lord et al. 1994, 1997). The ADI-R is a structured interview with the caregivers of an autistic child. The questions concern the autism specific pathology of the subject pertaining particularly to early childhood and contain an algorithm for the classification of autism along the lines of DSM and ICD. The data of the ADI-R gives a historical overview of a child’s symptoms in the early development whereas the Autism Diagnostic Observation Schedule (ADOS) (Lord et al. 1989, 2000) provides direct observation of the child’s current symptoms in respect of autistic traits, especially in communication, social interaction, play and imaginative use of materials. These instruments provide data for diagnostic threshold, domain scores, sub domain scores, and specific items.
Table 2

Current studies of the dimensional structure of autism spectrum disorders: restricted, repetitive behaviour and interests domain

Author (year)

Sample

Diagnosis

Age (mean/SD or range)

IQ (mean/SD; range)

Methods: statistical procedure and instruments

Results: factors (accounted variance)

Cuccaro et al. (2003)

207

Autism

109 month/55 month

Range: 3–21

Not mentioned

Factor analysis and correlations: 12 ADI-R items from the restricted, repetitive behaviour and interests domain

VABS sum score

Repetitive sensory motor actions with a negative correlation to index of adaptive functioning

Resistance to change

(60.7% of variance)

Carcani-Rathwell et al. (2006)

692

573 PDD (autism, atypical autism, Asperger, PDD-NOS)

119 mental retardation in the absence of psychiatric diagnosis

1–18

390 normal intelligence

302 mental retardation

Model’s goodness to fit-Testing: Repetitive and stereotyped behaviours item sheet

Repetitive movements and sensory behaviours are generally associated with lower developmental age and less specific to the autistic syndrome

Cognitive rigidity (higher order group) more autism specific

Szatmari et al. (2006)

339

PDD (autism, Asperger, atypical autism or PDD-NOS)

101 month/66 month

65/28

Factor analysis: 11 ADI-R items from the restricted, repetitive behaviour and interests domain Correlations with Vineland and other ADI-R domains

Insistence of Sameness, positively correlated with autistic symptoms in the communication and language domain

Repetitive sensory and motor behaviours, associated with lower level of functioning (33(ever)–36 (current)% of variance)

Richler et al. (2007)

279

165 Autism Spectrum Disorder ASD (117 autism, 48 PDD-NOS), 49 non spectrum developmental disorders DD, 65 typical development TD

<3; ASD: 29 month, DD: 27 month, TD: 20 month

ASD: 49, DD: 67, TD: 113

Factor analysis: ADI-R items from the restricted, repetitive behaviour and interests domain

Repetitive sensorimotor factor

Insistence on sameness

As shown in Tables 1 and 2, the results of the studies differ in quantitative aspects: This range from only one to six underlying factors. The studies also differ in respect to the content of the resulting factors. One reason may be that there are notable methodological differences among these studies. This concerns the sample size ranging from 51 up to 1,259. The variety of the sample in respect of age and IQ levels also differs: From 80 month old children up to the age of 47, from very low to high IQ’s. The sampling method differs also: Some studies only recruited affected subjects with a diagnosis of autism, some included subjects from the whole spectrum and some studies also included children with normal development or individuals with other diverse psychiatric diagnosis.

Some of the different results of these studies may be explained by another methodological argument: The selection of the item pool for analysis. This may be influenced in some studies by sample size: To reduce the number of variables for the factor analysis only the algorithm items of the ADI-R were chosen. But this is circular argumentation: The algorithm items of the ADI-R were selected due to fact that they most “closely depicted the specific abnormalities described in the clinical descriptions and diagnostic guidelines from the DSM-IV and ICD-10” (Lord et al. 1994). Another limitation is given by the fact that for the development of the algorithm of ADI-R “diagnostically borderline cases had been excluded” and “subjects with PDD-NOS were grouped with other non-autistic subjects” (Lord et al. 1997). So questions arise whether the results are valid for the whole spectrum especially for the high functioning subgroup. Most of the presented results were obtained by factor analysis, but one difficulty with factor analysis is that the characteristics of the sample will profoundly influence the results. Hence it is important to determine whether the factor structure is stable across various subgroups of autism.

There are also some noticeable similarities in these studies. These similarities include in particular the differentiation (as in ICD 10 and DSM-IV) of the symptom categories “communication” and “social interaction”: this differentiation was definitely not found in four studies. There is empirical evidence that these two symptoms categories can not be separated as supposed in ICD 10 and DSM-IV. In the domain of restricted, repetitive behavior there is support that “resistance to change” or “cognitive rigidity” differs from “repetitive sensory motor actions”.

On the one hand the objective of this study is to examine a comprehensive, empirically developed dimensional structure of the autism phenotype by investigating a specific sample of ASD and comparing these results with the existing results of other studies. On the other hand we intended to examine the coherence to other behavioral features like adaptive behavior, age and intelligence (IQ). The clinical relevance of the developed dimensional structure will be tested by examining the connections to other factors (adaptive functioning, theory of mind abilities, age and IQ). The main focus is on the predictive power of early development factors to current functioning. In this way we hope to understand what the emerged factors mean to us. Most previous studies have done this by examining the whole spectrum of autistic disorders and by choosing only the algorithm items of the diagnostic instruments. In our study all items are included in the analysis because it is not yet clear which items are specific to the high end of the autistic spectrum.

Methods

Participants

Our sample included altogether 140 subjects, six of whom were female. The age ranged from 6 to 24 years and the Full Scale IQ from 70 to 139. 52 of them were diagnosed as having Asperger syndrome, 44 as high-functioning autism and 8 as atypical autism, 36 had a non-autism diagnosis. It is important to stress the fact that all subjects came to our clinic with the suspicion of autism. This means that even the subjects without a diagnosis of autism showed considerable “autism-like” behavior. This is reflected by the fact that the non-autism diagnoses are principally differential diagnoses of autism like ADHD (n = 18), emotional disorder of childhood (n = 6), receptive speech disorder (n = 3), schizoid personality disorder (n = 3), other personality disorder (n = 2), delay of development (n = 2) and learning disability (n = 2). Our study therefore examines the “higher functioning end” of the autistic spectrum and its related diagnoses.

There are no statistical differences between the group’s autism and non-autism regarding age and full scale IQ. Descriptive statistics of our sample are presented in Table 3.
Table 3

Sample descriptive statistics

Characteristics

No.

%

Mean

SD

Sex

    Male

134

95.7

  

    Female

6

4.3

  

Best estimate diagnosis

    Asperger

52

37.1

  

        Age

  

11.85

4.40

        Verbal-IQ

  

117.16

16.85

        Performance IQ

  

96.41

17.18

    HFA

44

31.4

  

        Age

  

12.83

5.08

        Verbal-IQ

  

97.61

20.56

        Performance IQ

  

86.02

17.16

    Atypical autism

8

5.7

  

        Age

  

15.10

3.67

        Verbal-IQ

  

93.12

18.17

        Performance IQ

  

87.00

17.26

    Non-autism

35

25.0

  

        Age

  

12.05

4.29

        Verbal-IQ

  

108.39

19.14

        Performance IQ

  

99.47

17.59

Age at testing time

140

 

12.40

4.59

IQ

    Verbal-IQ

140

 

107.32

20.54

    Performance-IQ

140

 

93.37

18.03

    Full-IQ

140

 

100.90

18.31

HFA  high-functioning autism, SD Standard deviation

The data presented in this paper originates from a research project on high-functioning autism and Asperger syndrome which included a comprehensive diagnostic and neuropsychological testing as well as genetic components. All of the participating families gave their written informed consent before participating in the study. The study was approved by the Ethical Committee of the Medical Faculty of our university.

Materials

The participating subjects were diagnosed by experienced clinicians according to the standard Criteria of ICD-10 (World Health Organization 1993) and DSM-IV (American Psychiatric Association 1994) and underwent extensive psychiatric examination. The expression of autistic symptoms was assessed by the German version of the Autism Diagnostic Observation Scale (ADOS-G; Lord et al. 2000; Rühl et al. 2004) and a semi-structured autism specific parent interview (ADI-R; Lord et al. 1997; Bölte et al. 2006); this in-depth assessment was conducted by trained examiners (I. K.-B., M. G).

The scoring instructions of the ADI-R cover two points of time: “current” (within 3 month before the interview) and “at age of 4–5 or ever” (behaviors had not developed by age 4–5 years or throughout the individual’s life). In the present study only the judgments concerning the “at age 4–5 or ever” were included in the analysis, so the focus is on early development.

The ADOS-G consists of four modules, appropriate for children and adults of differing language levels, and which range from nonverbal to verbally fluent. For the purposes of this study only Module 3 and 4 for verbally fluent children and adolescents were used. Furthermore the inter rater reliability between pairs of raters was investigated using 17 videotaped ADOS-G assessments: The kappa values ranged from 0.42 to 1.0. The mean of the kappa values was 0.75 and complete agreement was established for autism/non-autism distinction.

The Vineland Adaptive Behavior Scales (VABS) (Sparrow et al. 1984) is an interview-based tool assessing adaptive behavior in three domains: communication (receptive, expressive, and written), daily living skills (personal, domestic, and community), and socialization (interpersonal relations, play and leisure time, and coping skills). The VABS is sensitive to severity of impairments in ASD (Carter et al. 1998; Klin et al. 2007; Saulnier and Klin 2007).

For intellectual testing the German version of the Wechsler Intelligence Scales was chosen. These are a series of standardized tests used to evaluate cognitive abilities and intellectual abilities in children (WISC-III; Tewes et al. 1999) and adults (WAIS-R; Tewes 1993). The theory of mind abilities are evaluated by a battery of tests which highlights relevant aspects of theory of mind and associated social perception: face recognition (Benton et al. 1983); facial emotion recognition (www.candit.com: Facial Emotion Matching); and social attributions (social attribution test, Klin and Jones 2006). The number of correct answers in all three tests was summarized to one value.

Estimated Diagnoses

For each patient, DSM-IV/ICD-10 psychiatric diagnosis had been established by at least two expert clinicians among the authors. Subjects with an autistic diagnosis fulfilled the cut-off criteria of the ADOS-G and ADI-R for autism or ASD.

Procedure

To examine the dimensional structure of the autism phenotype the data from the ADI-R and the ADOS-G was selected for analysis. We used all relevant items of the ADI-R and the ADOS-G which display the pathology of our high functioning sample regardless of their inclusion in the diagnostic algorithm or not. The criteria for judging items as relevant was that the item measured the autistic symptomatology itself (not comorbidity or general behavior) and was judged as irrelevant if more than 90% of the sample coded with a zero (i.e. that the behavior shows no evidence of abnormality). A test of sampling adequacy establishing whether the partial correlations among variables are too small was obtained (Kaiser–Meyer–Olkin test). The Bartlett’s test of sphericity (establishing whether the correlation matrix is an identity matrix, which would indicate that the factor model is inappropriate) was also undertaken. Principal components analyses were used. These are a statistical technique that transforms an original set of variables into a smaller set of uncorrelated variables or factors that represent most of the information of the original variables (Morrison 1990). The analysis was done with the FACTOR procedure from the software packages SPSS for Windows (Version 14.0). For all principal-components analyses, VARIMAX rotation (Kaiser 1958) was used to achieve a simpler structure while maintaining independence between the rotated factors. Use of PROMAX rotation yielded results similar to those reported below. To evaluate the extent to which the obtained factors were correlated with other variables, we calculated scores based on each of the derived factors for every participant. Optimally weighted factor scores were automatically generated by SPSS so that the mean factor score for the entire sample was zero and the standard deviation was one. We used Pearson correlation coefficients (level of significance ≤ 0.01) to examine the association between the empirically derived factors, IQ and age.

To establish which variables predict the current pathology and the current adaptive behavior abilities best, a linear regression analysis was calculated. The emerged factors of the ADOS-G were used as criteria variables and in another analysis the Vineland data. The emerged ADI-R-Factors—which present the pathology of the early development—the verbal IQ, performance IQ, the theory of mind-skills variable and age were used as predictor variables.

Results

Factor Analysis ADI-R: Early Development

The Items “loss of verbal abilities”, “comprehension of speech”, “general level of speech”, “actual communicative speech”, and “midline hand movements” were excluded because of irrelevance for a majority (>90%) the sample (see above). Altogether 47 items of the ADI-R were considered in the factor analysis.

The Kaiser–Meyer–Olkin measure of sampling adequacy indicates a good adequacy of the sample (0.837), the Bartlett’s test of sphericity was 0.000, indicating the data was suitable for factor analysis. A cutoff point for factor loading of 0.365 was chosen to minimize cross-loadings. 13 eigenvalues were >2.0, six had an eigenvalue >3.0. The three-, four-, and five-factor solutions were examined. In the three-factor solution the amount of explained variance was only 38%, in the four-factor solution it was 42%. The five-factor solution could explain 46% of variance, but one factor included only two items and there were 15 cross-loadings. The selection of the number of factors to extract was based on examination of the first five eigenvalues, the scree plot, and conceptual interpretability of the emerged factors. As a result, the four-factor solution was selected. Table 4 presents the factor loadings on the four factors.
Table 4

Factor loadings on social-communication, anxiety and compulsions, stereotyped behavior and inadequate behavior

ADI-R domain

 

Factor loadings

Factor 1: social communication (SOCOM early)

Com

Imitation of activities/play

0.722

   

Com

Imaginative play

0.694

   

Com

Conventional/instrumental gestures

0.667

   

Com

Pointing to express interest

0.630

   

Inter

Shared pleasure

0.629

   

Inter

Direct gaze

0.627

   

Inter

Group play with peers

0.626

   

Inter

Showing and directing attention

0.619

   

Inter

Quality of social overtures

0.600

   

Inter

Social smile

0.595

   

Inter

Offers comfort

0.594

   

Inter

Interests in children

0.567

   

Com

Reciprocal conversation

0.523

   

Inter

Imitative social play

0.516

0.376

  

Com

Imaginative play with peers

0.511

   

Inter

Responses to approaches of other children

0.497

   

Com

Vocal expressions

0.474

0.415

  

Com

Social vocalization/chat

0.472

   

Inter

Offering to share

0.455

   

Inter

Appropriateness of social responses

0.410

   

Inter

Initiation of appropriate activities

0.394

   

Factor 2: anxiety and compulsions (anx)

Stereo

Resistance to change in environment

 

0.700

  

Stereo

Compulsions

 

0.627

  

Stereo

Resistance to change in daily routines

 

0.615

  

Stereo

Unusual negative reaction to sensory stimuli

 

0.608

  

Com

Verbal rituals

 

0.607

  

Stereo

Sensitivity to noise

 

0.535

  

Com

Neologism

 

0.460

  

Stereo

Circumscribed interests

 

0.454

−0.376

 

Stereo

Repetitive use of objects or interest in parts of objects

 

0.454

  

Stereo

Unusual attachment to objects

 

0.379

  

Factor 3: stereotyped behavior (stereo)

Com

Use of other’s body to communicate

  

0.599

 

Com

Nodding

  

0.548

 

Com

Pronominal reversal

  

0.544

 

Com

Head shaking

  

0.523

 

Com

Stereotyped utterances and delayed echolalia

 

0.439

0.453

 

Stereo

Mannerism

  

0.451

 

Stereo

Complex mannerism

  

0.380

 

Stereo

Unusual sensory interests

  

0.374

 

Factor 4: inadequate behavior (inad)

Inter

Social disinhibition

   

0.615

Com

Inappropriate questions or statements

 

0.374

 

0.587

Com

Intonation

   

0.560

Inter

Range of facial expressions

0.559

  

0.510

Inter

Inappropriate facial expression

0.447

  

0.497

Com

Attention to voice

   

0.391

Com = Communication domain; Inter = social interaction domain; Stereo = restricted repetitive behaviors and interests domain, factors with a loading >0.365 are listed. Items in bold text are algorithm items

The first factor is named: Social communication (“SOCOM early”), it explains 17.72% of variance. 21 out of 47 items with high loadings are included in this factor and are mainly items of the algorithm of the ADI-R. This factor combines items from the original social interaction (13 items) and communication domain (8 items) of the ADI-R. It contains play and imitativebehavior, nonverbal behavior as well as reciprocal behavior. There is no correlation to Full Scale IQ. The second factor is named: anxiety and compulsions (anx), 10 items with high loadings are included and it explains 10.74% of variance. It is dominated by anxiousness and compulsive behavior but also circumscribed interests (which in some cases have a compulsive character). Interestingly some other aspects of verbal behavior are included: verbal rituals and neologism. No correlation to IQ was found. The third factor (eight items, 7.54% of variance explained) is characterized by stereotyped behavior (stereo) in aspects of nonverbal as well as verbal behavior. The circumscribed interests load negatively on this factor. There is a significant negative correlation to full scale IQ (−0.30, Pearson coefficient, p < 0.01). The fourth factor (explaining 6.38% of variance) is best described by inadequate behaviors (inad) (six items).

The comparison of the autism and the non-autism group shows that especially the first two factors differ considerably (p-value = 0.00 for factor “anxiety and compulsions”, p = 0.01 for “SOCOM early”).

Factor Analysis ADOS-G: Current Presentation

For this analysis data from module 3 and 4 were taken together (“emphatic or emotional gestures” was omitted, because it is not included in module 3; “language production and linked nonverbal communication” was also omitted because it depends on the scoring of 3 other items). Three items (“tantrums, aggressions”, “immediate echolalia” and “self-injurious behavior”) were not relevant to a majority (>90%) of the sample. The remaining 24 items were included in the factor analysis. The Kaiser–Meyer–Olkin measure of sampling adequacy points out a good adequacy of the sample (0.87), the Bartlett’s test of sphericity was 0.000. As for the factor analysis for the ADI-R a cutoff point for factor loading of 0.365 was chosen to minimize cross-loadings. 6 eigenvalues were >1.0. The three-, four-, and five-factor solutions were examined. In the three-factor solution the amount of variance explained was 47%, in the four-factor solution it was 52% and the five-factor solution explains 57% of variance. The selection of the number of factors to extract was based on examination of the first five eigenvalues, the screen plot, and conceptual interpretability of the emerged factors. The five-factor solution was selected which is presented in Table 5.
Table 5

Factor loadings of the ADOS-G

ADOS Domain

 

Factor loadings

1

2

3

4

5

Factor 1: social communication (SOCOM current)

Inter

Shared enjoyment

0.776

    

Inter

Overall quality of rapport

0.755

    

Com

Reporting of events

0.714

    

Inter

Quality of social overtures

0.708

    

Inter

Empathy/comments on others emotions

0.702

    

Com

Conversation

0.690

    

Com

Offers information

0.688

    

Inter

Amount of reciprocal social communication

0.676

    

Inter

Quality of social responses

0.656

    

Com

Asks for information

0.629

    

Inter

Insight

0.569

 

0.434

  

Com

Gestures

0.517

    

Imagin

Imagination

0.488

   

−0.384

Factor 2: non/verbal behavior

Inter

Unusual eye contact

 

0.676

   

Com

Speech abnormalities

 

0.659

   

Inter

Facial expressions

0.485

0.562

   

Com

Stereotyped phrases

 

0.529

0.509

  

Factor 3: hyperactivity

Other

Anxiety

  

−0.724

  

Other

Over activity

  

0.587

0.488

 

Factor 4: stereotyped behavior

Stereo

Sensory interests

   

0.816

 

Stereo

Mannerism

   

0.660

 

Factor 5: interests and compulsions

Stereo

Excessive interests

    

0.605

Stereo

Compulsions or rituals

    

0.587

Com

Overall level of language

    

−0.411

Com = Communication domain; Inter = social interaction domain; Stereo = restricted, repetitive behaviors and interests domain; Imagin = Imagination

Other = other abnormal behaviors; 1 = factor SOCOM; 2 = factor non/verbal behavior; 3 = factor hyperactivity; 4 = factor stereotyped behavior; 5 = interest and compulsions, factors with a loading >0.365 are listed. Items in bold text are algorithm items

Analog to the results of the ADI-R, the first factor includes 13 items of the communication and social interaction domain, mainly algorithm items. So the first factor is also named “Social communication (SOCOM current)”, which explains the main part (26%) of variance. Similar to the results of the ADI-R play behavior is also included (item “imagination”). No correlations to IQ and age were found.

The second factor is mainly influenced by eye contact and speech abnormalities and is named: “non/verbal behavior”. This factor explains 10% of variance. There was no significant correlation to IQ. The third factor explains 8% of variance and is named “hyperactivity”. This factor is mainly influenced by the item “hyperactivity” while “anxiety” loads negatively meaning that anxious behavior correlates negatively with hyperactivity. Participants with high hyperactivity show low anxiety. This factor reveals a significant correlation to age (−0.46 Pearson coefficient, p < 0.01) demonstrating that the older the subjects are the lower hyperactivity is. The fourth factor is named “stereotyped behavior” (7% of variance) which correlates significantly with the performance IQ (−0.26 Pearson coefficient, p = 0.01). The last factor “interest and compulsions” explains 5% of variance.

The first two factors (“SOCOM current” and “non/verbal”) differ enormously between the autism and the non-autism group (p = 0.000 for both, controlled for IQ). Table 6 displays the correlations between the ADOS and ADI factors.
Table 6

Correlations (Pearson coefficients) between the ADOS and ADI factors

  

Early development

ADI SOCOM

Anxiety and compulsions

Stereotyped behavior

Inadequate behavior

Current presentation

ADOS SOCOM

0.137

0.259**

0.053

0.030

Non/verbal behavior

0.153

0.016

0.229**

0.152

Hyperactivity

0.128

0.161

0.112

0.234**

Stereotyped behavior

0.069

0.081

0.143

0.234**

Interests and compulsions

−0.181*

0.031

−0.085

0.122

ADI SOCOM Factor “Social Communication” of the ADI-R, ADOS SOCOM Factor “Social Communication of the ADOS

* Correlation is significant at p < 0.05 level (2-tailed)

** Correlation is significant at p < 0.01 level (2-tailed)

Vineland Adaptive Behavior Scales

Our sample had a mean full IQ of 100.90 (with a standard deviation of 18.31) and a mean sum score of “adaptive behavior composite” of 63.24 (SD = 24.60) for the whole sample. However all scores indicate a great deal of scatter, especially the communication domain scores. The adaptive skills are more than three standard deviations below the cognitive abilities, which make abundantly clear that there is a high magnitude of social disability despite good cognitive abilities.

Regression Analysis: Relations Between Early Development and Current Presentation

We conducted linear multiple regression analysis models to explore the predictive power of the received factors of the ADOS-G for the current symptomatology and the three Vinland domains for the autism group. For this the factors of the ADOS-G analysis were taken consecutively as dependent variables and the independent variables included the ADI-R factors for the symptomatology of the early development and the variables age, IQ and the sum score of the theory of mind abilities. They were entered as a block. The same procedure was undertaken for the Vineland scores as dependent variables. Results are displayed in Table 7.
Table 7

Results of linear regression analysis

Criteria variables: current presentation

Predictor variables

Adjusted R2

Variable

Beta

p

ADOS SOCOM actual

0.133

Anxiety and compulsions

0.315

0.002

Age

0.334

0.005

Theory of mind

−0.314

0.012

ADOS: hyperactivity

0.186

Age

−0.279

0.014

Communication skills

0.218

Verbal-IQ

0.357

0.002

Daily living skills

0.232

SOCOM early

−0.305

0.002

Anxiety and compulsions

−0.242

0.012

Socialization skills

0.230

Age

−0.563

0.000

Predictor variables: derived factor of the analysis of ADI-R (symptomatology of the early development), age, IQ and the sum score of the theory of mind abilities. Criteria variables: derived factors of the analysis of ADOS (current symptomatology) and Vineland adaptive behaviors scores

SOCOM Social Communication

One result of the first regression analysis confirms that the regression model for the sub-sample of the autism subjects (F = 2.748, p = 0.010) attained significant explanation, but accounted only for a small amount of explained variance (R2 = 0.209, adjusted R2 = 133) The amount of autistic symptoms in the current presentation is best predicted by the amount of anxiety and compulsions in the early development, in combination with age and theory of mind abilities. Of interest is the fact that age and theory of mind abilities are of relevance for the actual autistic symptoms only in combination with the factor “anxiety and compulsions”. There was no significant correlation between this factor and age alone. The factor SOCOM of the ADOS is higher the more symptoms of “anxiety and compulsions” are shown in early childhood, the older the child and the lower theory of mind abilities are. The factor “anxiety and compulsion” also predicts the noticeable problems in nonverbal and verbal behavior (unusual eye contact, speech abnormalities, reduced facial expression and stereotyped phrases) but this relation is not of satisfying significance (p = 0.025), while hyperactivity is best predicted by age.

For the adaptive behavior the regression model for the sub-sample of the autism subjects (F ≥ 4.132, p = 0.000) shows significant explanation and the amount of explained variance is greater (R2 = 0.287–0.301, adjusted R2 = 0.218–0.230). The abilities of the autistic subject in everyday life in the domain of communication are best predicted by the verbal-IQ, in the domain of daily living skills they are best predicted by the autistic symptoms of early development (factor “SOCOM early”) and by the factor “anxiety and compulsion”. Socialization is best predicted by the age of the subject, the relation to the autistic symptoms of early development (factor “SOCOM early”) is once again not sufficient (p = 0.020).

Discussion

The results of our analyses suggest that the Asperger/HFA phenotype is structured by 4 respectively 5 dimensions which differ to the conceptualization of DSM-IV or ICD-10. The results of the factor analyses account for a sufficient amount of variance (42% for ADI-R, 57% for ADOS-G). In other studies (Table 1) mostly a three factor solution was chosen but furthermore with this differentiated factor structure some important relations to other variables could be identified.

Dimensional Structure of the Symptoms of Early Development

The ADI-R Factor “SOCOM early” of our study is predominantly identical to the factor SOCOM of the Canadian study (Georgiades et al. 2007) with the exception of only four items (“use of other’s body to communicate”, “nodding”, “head shaking”, and “inappropriate facial expression”) which are included in other factors in our study and one additional subdomain (“total, relative failure to initiate or sustain conversational interchange”). The emerged factor “SOCOM early” is also very similar to the factor: “impaired social communication skills” of the Dutch study of van Lang et al. (2006): Just the items “use of other’s body to communicate”, “inappropriate facial expression”, “nodding” and “head shaking” are not included, but the items of the Dutch study “impaired make believe and play skills” are included in our factor “SOCOM early”. In comparison with the German study (Bölte and Poustka 2001) our factor “SOCOM early” is also similar but not identical to their factor “social communication” as it also includes some items of the factor “speech” of the study of Bölte and Poustka (2001).

The factor “SOCOM early” combines the “reciprocal social interaction” and “communication” domain of the ADI-R: This result seems to be stable in several studies and is replicated in our study for a sample of high-functioning autism. Furthermore, we could demonstrate that this factor is of relevance for the adaptive behavior in daily life, especially for daily living skills. In contrast to the Canadian study (Georgiades et al. 2007) we found in our high functioning sample no correlation to IQ, neither verbal nor performance IQ.

Consistent with other studies (“Compulsions”, “resistance to change”, “Cognitive rigidity”, “Insistence of Sameness”, “insistence on sameness”, see Table 2) we found a second factor named “anxiety and compulsions”. In our study this factor also included items from the domain of communication (“verbal rituals”, “neologism”) demonstrating that compulsive behavior also influences verbal behavior. This is in accordance with the result of the study of Szatmari et al. (2006) which found a positive correlation between the factor “insistence on sameness” and autistic symptoms in the communication and language domain. It could be hypothesized that these behaviors result from anxiety and compulsions but on the other hand the diversity of behavioral aspects limits the interpretation of the factor. As a new result we could demonstrate that this factor is of high relevance for further development. It best predicts the main factor of autistic symptomatology in the current presentation. It differs highly between autism and non autism groups. This result is similar to the study of Carcani-Rathwell et al. (2006) who found that this “higher order” factor is more autism specific. Another important result is that despite its demonstrated relevance only two out of ten items of this factor are included in the algorithm of the ADI-R. This could be an indication that further validation studies in support of the ADI-R algorithm are necessary for high-functioning autism subjects.

The third factor also combines aspects of stereotypes and communication in the general aspect of stereotyped behavior. Here, the influence of intelligence becomes evident, as the circumscribed interests correlate with IQ itself, they load negatively on the factor stereotyped behavior. There is a considerable relation to the cognitive abilities and fewer to the autistic specific symptomatology as represented in other factors. This is also reflected by the fact that the autism group doesn’t differ from the non autism group in aspects of stereotyped behavior. In the existent studies such a factor was not found: Indeed the factors “stereotyped language and behavior” (van Lang et al. 2006) and “inflexible language and behavior” (Georgiades et al. 2007) seem to be similar, but in both studies these factors imply circumscribed interests and compulsive behavior. This is possibly explained by the influence of the cognitive abilities which are relatively high in our sample. Another interesting fact is the result that gestures are not included in this factor but in the main factor both in the analysis of ADI and ADOS.

In summary the amount of variance explained in our study as well as in all other studies in this area is only moderate. One could argue that the usefulness of the 4 factor solution for the ADI-R analysis is disputable: it potentially encompasses a variety of behavior that does not represent any single entity. But by demonstrating that the derived factors are of relevance for the actual symptoms and adaptive behavior, the usefulness becomes evident.

Dimensional Structure of the Symptoms of Current Presentation

Again we found a highly autism specific factor, which includes items from the social interaction and communication domain of the ADOS-G and also play behavior. Similar to the results for ADI-R there is no correlation to IQ. This factor is very similar to the factor analysis of ADOS data on large independent datasets recently published (Gotham et al. 2007, 2008) named “social affect”. Only the item “unusual eye contact” loads in our study on a separate factor. The second factor shows similarities to the factor “affective reciprocity” of Robertson et al. (1999) which included the items “facial expression” and “speech abnormalities” but the item “eye contact” loaded on another factor (joint attention). In summary we could also confirm some aspects of the few existing studies of the dimensional structure of the ADOS data.

We did not find a correlation between the main social communication factors of ADOS and ADI (see Table 6). This may be caused by the fact that social communication skills change greatly from age 4–5 upwards, even though many of the subjects still have sufficient problems in this domain. Another objection may be that the parent’s recollection of the age 4–5 behaviors is simply inaccurate after a long time period. No other study has investigated the relation between ADOS and ADI factors, further studies are needed to answer these open questions.

Predictive Power of Early ADI Factors to Current ADOS Factors and Adaptive Functioning

The emerged factor “anxiety and compulsion” in early development is associated with current social communication functioning. One could hypothesize that anxiety and compulsive behavior inhibits the development and limits the reduction of autistic symptoms. A young autistic child with low anxiety and compulsive behavior is more able to learn and to develop adequate behaviors. Although we found only modest associations between early ADI and current ADOS scores, the impact of various behavioral aspects pictured in the factor “anxiety and compulsions” is of interest and has to be examined in further studies.

Relation to Age and Intelligence

We could demonstrate that “hyperactivity” shows a high correlation to age. In line with the studies of Saulnier and Klin (2007) and Klin et al. (2007) we couldn’t find this correlation with the specific autistic symptomatology indicating that these symptoms are stable over time. In contrast, the more symptoms of “anxious and compulsive behavior” in early childhood were shown, the older the child and the lower the theory of mind abilities are, the more autistic symptoms are actual present. Unfortunately, autistic symptoms don’t get lost with age. In combination with anxious and compulsive behavior in early childhood and low theory of mind abilities they increase. Limitations are given because our study only correlates age with the autistic symptoms and the age spectrum was only 6–24 so no long term study was undertaken. In summary the IQ doesn’t predict the main autistic symptoms. In addition to other studies (Klin et al. 2007; Saulnier and Klin 2007) these results emphasize that the IQ is not an appropriate predictor of the autistic symptomatology for the higher end of the cognitive spectrum of autism: Intelligence can predict the communicative behavior but not the severity of autistic symptoms.

Relations to Adaptive Behavior

In addition to some other studies we could demonstrate that even a sample of high-functioning autism subjects show a remarkable impairment in adaptive behavior. We could demonstrate that adaptive skills and autistic symptomatology are not independent domains as suggested in other studies: Szatmari et al. (2002) found two independent factors: autistic symptoms and adaptive behavior. Klin et al. (2007) and Saulnier and Klin (2007) found only a weak relation between adaptive behavior and the symptomatology measured by the ADOS. In contrast, we could state that adaptive behavior and autistic symptoms of early development are not independent domains.

In Summary

This data supports the notion that variation in autistic symptoms is not simply a reflection of developmental level (in respect of age or cognitive abilities) but may rather represent variation in some other underlying etiological factors (Georgiades et al. 2007). Our analysis was based on a micro level of symptoms and included many variables. In this way we could demonstrate the relevance of the emerged factors. By demonstrating correlations with the variables adaptive behavior, age, IQ and theory of mind the factors represent construct validity. In line with several other studies we point out that the social interaction and communication domains are so closely related that they emerge as a single factor called: Social communication. The most important conclusion of our study is that the emerged factor “anxiety and compulsion” in early development is associated with current social communication functioning. This finding has never been reported in previous work. The implication of this is that a great effort has to be made to reduce symptoms in this domain by intervention programs in early childhood because these symptoms affect further development. Another important finding is that IQ scores do not correlate with earlier or current social communication functioning. The latter has been reported previously (Gilchrist et al. 2001; Szatmari et al. 2003; Howlin et al. 2004), but the current results are an important confirmation of this observation.

Nevertheless it is necessary to investigate the validity of the proposed factors in another study since our study has a number of limitations. First of all the entire sample size was small considering the undertaken statistical procedures: The ratio of sample size to items analyzed is the main limitation of our findings. The findings presented in this paper must be regarded as a first explorative examination of the dimensional structure of high functioning ASD. Further investigation is needed to confirm our results in a larger sample. In particular, a study focusing on the comparison of a low and a high IQ sample is needed. This could clarify whether the factors established in our study are specific for our high functioning sample. Yet another limitation is the methods used (ADOS-G, ADI-R) which were not constructed to be used as quantitative measures for dimensions or even severity. Also factor analysis is sensitive to sample size and method of ascertainment. Like the authors of many other studies we opted for VARIMAX rotation and treat ADI and ADOS scores as if they are continuous, interval data although they are ordinal. Although a check demonstrated that a PROMAX rotation gave similar results this fact also places limits on the results. An advantage of our study is the homogeneity of the sample in terms of high functioning level (IQ > 70) and the inclusion of various variables for validation. It could be pointed out that the dimensional structure of higher functioning ASD using standard measures (ADOS and ADI) generates some similarities to other studies but also some new findings: By studying the relationship between autistic symptoms, adaptive behavior and other particular characteristics we may sharpen the focus of intervention for particular developmental groups.

Acknowledgments

This research was supported by a research award of the German Max Planck association H. Remschmidt received 1999. We thank all of the families for their participation in our research program.

Copyright information

© Springer Science+Business Media, LLC 2008