Journal of Autism and Developmental Disorders

, Volume 38, Issue 4, pp 616–625

Differentiating Autism and Asperger Syndrome on the Basis of Language Delay or Impairment

Authors

  • Terry Bennett
    • Department of Psychiatry and Behavioural NeurosciencesMcMaster University
    • Department of Psychiatry and Behavioural NeurosciencesMcMaster University
    • Offord Centre for Child Studies
  • Susan Bryson
    • Department of Psychiatry, IWK Health CentreDalhousie University
  • Joanne Volden
    • Speech Pathology and AudiologyUniversity of Alberta
  • Lonnie Zwaigenbaum
    • Department of PediatricsUniversity of Alberta
  • Liezanne Vaccarella
    • Department of Psychiatry and Behavioural NeurosciencesMcMaster University
  • Eric Duku
    • Department of Psychiatry and Behavioural NeurosciencesMcMaster University
  • Michael Boyle
    • Department of Psychiatry and Behavioural NeurosciencesMcMaster University
Original Paper

DOI: 10.1007/s10803-007-0428-7

Cite this article as:
Bennett, T., Szatmari, P., Bryson, S. et al. J Autism Dev Disord (2008) 38: 616. doi:10.1007/s10803-007-0428-7

Abstract

Asperger syndrome (AS) is differentiated from high-functioning autism (HFA) largely on a history of “language delay.” This study examined “specific language impairment” as a predictor of outcome. Language skills of 19 children with AS and 45 with HFA were assessed at 4–6 years of age (Time 1) and 2 years later (Time 2). Children’s symptoms and functional outcome scores were assessed every 2 years (Times 3, 4, and 5) until ages 15–17 years old. Regression analysis revealed that specific language impairment at time 2 more often accounted for the greatest variation in outcome scores in adolescence than the standard diagnosis of AS versus HFA based on history of language delay. Diagnostic implications are discussed.

Keywords

AutismAsperger syndromeLanguage impairmentOutcomes

Introduction

Asperger Syndrome (AS; or Asperger disorder) refers to a condition characterized by impairments in social reciprocity, difficulties in communication, and a set of circumscribed interests or preoccupations that is often highly unusual and is usually carried out in social isolation. AS was first introduced into the official classification system in 1994 but was previously described in the literature under various headings including schizoid disorder (Ssucharewa and Wolff 1996), schizotypal personality disorder of childhood (Nagy and Szatmari 1986) and possibly other labels such as nonverbal learning disability (Stein et al. 2004) and atypical development (Mahler and Furer 1972). The clinical presentation has many characteristics in common with autism, particularly high-functioning autism (often defined as an absence of cognitive delay or, more precisely, mental retardation). There has been much debate as to the relationship between AS and autism. ICD-10 and DSM-IV imply that AS is separate from autism although both state quite correctly that the diagnostic validity of the distinction is uncertain. Both texts base the initial differentiation largely on measures of cognitive and language development. In other words, AS is characterized by an “absence of clinically significant cognitive and language delay” (American Psychiatric Association 2000). In essence, this differentiation comes down to an absence of language delay since it is well known that there are many children with autism who are not severely learning disabled or mentally retarded (i.e., with an IQ below 70). Such children are usually described as high functioning autism (HFA). Most often, the criterion of language delay that distinguishes HFA from AS is “the use of spontaneous phrase speech with a verb by 36 months of age.” Most believe however, that AS has a very close relationship with HFA and in fact, is at one end of the spectrum of Autism Spectrum Disorders (ASD), with autism at the other end. Others conclude that there is no difference between the clinical presentations and that it is not useful to employ the diagnostic label at all (Schopler 1996).

A number of studies have compared the clinical presentation of AS with that of HFA. Howlin (2003) and Macintosh and Dissanayake (2004) conducted systematic reviews while Frith (2004) adopted a narrative approach to summarize studies that investigated differences between the diagnostic labels. Howlin identified 26 studies over the past 15 years that directly compared individuals with HFA and AS. Adequate description of diagnostic criteria and IQ measures were specified as inclusion criteria, and studies were separated into those that either did or did not match for full-scale IQ. The articles reviewed were grouped according to areas of study, including general clinical characteristics, obstetric/early history and motor abnormalities, neuropsychological/language profiles and behavioral/psychiatric disturbances. Howlin noted that early group differences in core symptoms and disability appeared to decrease with age. Overall, AS groups appeared to have better developed verbal skills regardless of IQ compared to children with HFA. It was concluded that, there were “no obvious differences in rates of social, emotional and psychiatric problems, current symptomatology, motor clumsiness or neuropsychological profiles between the two groups” (Howlin 2003).

Macintosh and Dissanayake (2004) also conducted a systematic review. Specification of IQ and description of diagnostic method used were not required for inclusion. A review of the studies concluded that little insight into the relationship between HFA and AS was possible and the results have been confounded by other possible differences between the groups. Overall, the authors concluded that, “the validity of Asperger’s disorder as a unique syndrome, separate from HFA, has not yet been either conclusively established or refuted” (Macintosh and Dissanayake 2004).

Frith conducted a more narrative review with the objective of exploring the nature of Asperger syndrome (2004). She concluded that the evidence from the behavioral and neurophysiological literature suggests that AS is a variant of autism rather than a separate disorder and that both autism and Asperger were likely heterogeneous disorders. Frith also concluded that the diagnosis of AS is “hugely important” to clinical practice, although the current boundaries of the disorder lack definition (Frith 2004). The reviews by Howlin, Macintosh and Dissanayake and Frith all highlight several gaps in the research literature addressing the distinctiveness (or lack thereof) of the diagnoses of HFA and AS. They commented on the need for well-designed, controlled studies that include classification based on formal diagnostic criteria that are agreed upon across different groups of researchers. The authors all discussed the importance of avoiding circular research questions and underlined the need to compare the groups on variables that are independent of diagnostic criteria. They highlighted the urgency of pursuing longitudinal research that tracks the trajectories of the different pervasive developmental disorder (PDD) groups, as compared to the cross-sectional studies that comprise much of the current research literature (Macintosh and Dissanayake 2004; Howlin 2003). For example, Macintosh and Dissanayake and Frith commented upon the possibility that the advanced language abilities of children with AS may place them on a less severely disabled trajectory as compared to children with autism (Macintosh and Dissanayake 2004; Frith 2004). Finally, all concluded that it is too early to conclude with final certainty whether the disorders are different or the same (Macintosh and Dissanayake 2004).

This conclusion while certainly consistent with the literature does depend however on how the two groups were defined in the first place. The current differentiating feature diagnostically, as outlined by ICD-10 and DSM-IV, consists of an absence of cognitive or language delay in AS. An important potential problem is that this clinical differentiating feature may lead to substantial misclassification. Among older children, this feature is based on retrospective reports and as a result, may be unreliable. It is also well known that there are many younger children who have delayed speech but who later go on to develop speech and have language skills in the normal range (Fenson et al. 1994; Leung and Kao 1999). In other words, some children may have “speech delay” but not “language disorder” or impairment and it may be the latter which is a more useful and valid differentiating feature.

Tager-Flusberg and colleagues have conducted research outlining different language phenotypes within the autism spectrum (Kjelgaard and Tager-Flusberg 2001). They found that one subgroup of children with PDD tested within the normal range of language skills (regardless of IQ) while another group scored greater than 1–2 standard deviations below the mean and demonstrated language competency profiles similar to those of non-PDD children with specific language impairment (Kjelgaard and Tager-Flusberg 2001). Furthermore, it is our clinical experience that many children with high-functioning autism who are not speaking fluently by 3 years of age are able to do so by 6 years of age and come to resemble more and more the children with AS (Szatmari et al. 2000). Indeed, we have hypothesized that the best way to think of these conditions is not as different disorders but as parallel and potentially overlapping developmental pathways; once children with HFA develop fluent speech and are without structural language impairment (SLI), they may jump to the developmental pathway of the children with AS (Szatmari 2000).

The objective of this study is to see which differentiating feature explains the greater amount of the variation in social and communicative skills and in autistic symptoms over time. The usual clinical distinction (based as it is on the presence or absence of language delay) is compared with an alternative differentiation based on the presence or absence of SLI. SLI is defined as deficits in grammar or syntax, rather than the semantic and pragmatic components that would likely be common to both HFA and AS. Other studies in the literature refer to the term “specific language impairment” (which may incorporate disability in one or more different aspects of language production or comprehension); we chose to study the construct of SLI (impairment in the production and comprehension of syntactic elements of language) in order to exclude other linguistic elements such as pragmatics, which is thought to be part of social cognitive ability. The presence of SLI was assessed at 4–6 years of age and at 6–8 years of age and we hypothesized that its presence at the latter point in time would be the best predictor of outcome over time because it represents a more stable clinical characteristic.

Methods

Participants

All children between the ages of 4 and 6 who were either in treatment or coming for assessment at six different centers that serve preschool children with developmental disabilities in Southern Ontario (this was done so that children from non-university centers were also included) were identified and screened. Children who received a clinical diagnosis of ASD but were untestable or achieved a mental age score less than half their chronological age on psychometric testing were dropped from further consideration. The remaining children, including those without psychometric data, entered the study; written informed consent was obtained from the parents of each child. The children’s assent was also obtained on the basis of their willingness to participate in the assessment.

Inclusion criteria for the cohort of high-functioning children with ASD included the following: Children with an ADI based diagnosis of either autism or AS, and who had a Leiter IQ score above 68 or a Stanford Binet Score above 70 (which are above the cut-off for mental retardation) were included in the final cohort of “high functioning” preschool children with ASD. The sampling frame thus stipulated that all children would have an absence of clinically significant cognitive delay. Children without significant language delay were diagnosed with AS. This variable was operationalized using data from the ADI to include spontaneous phrase speech by 36 months and an absence of marked or persistent delayed echolalia, pronoun reversal and neologisms (as coded 2 or 3 on these items from the ADI). Unlike the DSM-IV, we decided that a diagnosis of AS could take precedence over diagnosis of autism. The DSM-IV states that a child with AS who meets criteria for autism will be given the diagnosis for autism (This is not true for the ICD-10). Applying the DSM-IV rule and using the ADI as guide, 15 of 21 children with a diagnosis of AS met ADI criteria for HFA. To identify enough children with AS and to reflect common clinical practice, it was necessary to reverse the hierarchy rule so that if a child met criteria for both AS and high functioning autism, he/she was given a diagnosis of AS. To summarize, the key features differentiating the groups were that the children with autism spoke after 36 months and had evidence of marked deviance in language development as defined above, whereas those with AS did not.

During the enrollment phase, 164 children 4–6 years of age were screened. Of these, 80 children were excluded because either they did not have ASD, their behavior was too low-functioning to obtain an IQ estimate, or previous psychometric testing revealed that they were functioning below the mental age criterion. The remaining 84 children with ASD underwent the full psychometric battery. A further 16 were excluded because their Leiter or Stanford-Binet IQs were below the IQ cutoff for mental retardation. Thus, the size of the study group at enrollment was 68 preschool children with ASD, deemed “high-functioning” because their IQs were above 68 on the Leiter or 70 on the Stanford-Binet scales.

The children in the study were given a psychometric assessment battery, which included a test of language abilities using The Test of Language Development-2 (TOLD-2 Newcomer and Hammill 1988). The families were contacted ∼2 years later (time 2 of data collection) and the assessment battery was repeated. Times 3, 4, and 5 of the study were conducted ∼4, 6, and 8 years, respectively, after the date of inception. During these data collection phases all children were seen for a follow-up assessment of adaptive behavior functioning (using the Vineland Adaptive Behavior Survey Edition) and autism symptomatology (using the Autism Behavior Checklist). A different research assistant than the one who did the original assessment conducted each follow-up assessment to ensure blindness.

Measures

Autism Diagnostic Interview (ADI)

The ADI is a semi-structured interview designed to make a diagnosis of autism and other PDD subtypes according to both the DSM-IIIR and ICD-10 criteria. The interview has good reliability (intra-class correlation for multiple raters ranges between 0.94 and 0.97) (Lord et al. 1994).

Arthur Adaptation of the Leiter Performance Scales

This is a standardized measure of nonverbal problem solving (mean = 100, SD = 15). The Leiter is widely used with ASD and other language impaired children. While the cognitive abilities of the children assessed in this study are relatively comparable, significant variability in verbal abilities exist that can influence comprehension of instructions. The Leiter is especially appropriate to the population under study because it does not require verbal instructions for administration and correlated highly with WISC-R IQ (Levine 1986).

Stanford-Binet Intelligence Scale, Fourth Edition

The Stanford-Binet measures overall cognitive development as four different cognitive domains-verbal reasoning, quantitative reasoning, abstract/visual reasoning and short-term memory skills (mean = 100, SD = 15) (Thorndike et al. 1985).

The Test of Language Development-2 (TOLD-2)

The Grammatic Completion, and Grammatic Understanding subtests of the TOLD-2 were used to measure grammatical comprehension (GC) and usage. Standard scores (mean = 10, SD = 3) were calculated for each child. If the child was mute or unable to obtain a basal score on the TOLD he/she was given an imputed score of one below the lowest score of a child who was able to complete the test. This was done to minimize the number of missing values, and to provide a reasonable valid, if conservative, estimate of abilities. Twenty-two (22) children with autism and one child with AS were given an imputed score on grammatical understanding (GU) and comprehension (Newcomer and Hammill 1988).

Vineland Adaptive Behavior Scales (VABS)

The Vineland is a semi-structured interview designed to assess personal and social sufficiency of individuals from birth to adulthood. Specifically, it measures an individual’s strengths and weakness in socialization, communication, and activities of daily living in children 0 to 18–11 years of age (mean = 100, SD = 15). The Vineland does not require an assessor to directly administer tasks to individuals, but instead requires an interview with a respondent familiar with the individual’s behavior (Sparrow et al. 1984).

Autism Behavior Checklist (ABC)

The ABC (Krug et al. 1980) is a parent completed, questionnaire that is a convenient, reliable (95% agreement for inter-rater reliability) (Krug et al. 1980) and valid (sensitivity = 80%; specificity = 70%; Volkmar et al. 1988) measure of autistic symptoms. The Autism Behavior Checklist has 57 questions divided into five categories: (1) sensory, (2) relating, (3) body and object use, (4) language, and (5) social and self-help.

Analysis

Outcome measures selected for use in our sample included the communication, socialization and daily living skills domains from the Vineland and the total number of autism symptoms as estimated by the total score on the ABC. Outcomes were measured at Times 3, 4, and 5. Raw scores were used from the Vineland in order to increase sensitivity to developmental growth. To determine which best predicts a child’s outcome, the presence or absence of clinical diagnosis, SLI at time 1 or SLI at time 2, were used to classify the children. For the purposes of the present paper, SLI is defined as 1.5 standard deviations below the standard mean score of 10 when the scores from both the GC domain and GU domain of the TOLD-2 (SLI = GC + GU/2 < 5.5) were averaged. See Table1  for sample characteristics of each of the three different ways the ASD children were classified.

Agreement in group membership between all three different methods of grouping the children—by clinical diagnosis of AS or HFA, by assessment of presence or absence of SLI at time 1 (age 4–6), and by assessment of SLI at time 2 (age 6–8)—was determined using the statistical package PC-Agree (Cook 1987). Pairwise agreement between the three different methods of classification using cross-tabulation was also estimated (Table 2). Separate regression analyses were then run for the each method of classifying the children to determine which method accounted for more of the variation in the outcome scores at times 3, 4, and 5. All analyses were done using SPSS Version 13.0.1 (SPSS Inc 2004).
Table 1

Sample characteristics by classification; mean and standard deviation (SD)

Group classification

Clinical diagnosis

SLI time 1

SLI time 2

High- functioning autism

Asperger syndrome

SLI

No SLI

SLI

No SLI

Number

N = 45

N = 19

N = 42

N = 20

N = 41

N = 21

Age Time 1 in months (SD)

64.96 (11.94)

69.21 (8.71)

65.29 (11.99)

67.25 (9.35)

65.07 (11.38)

67.76 (11.13)

Mean TOLD grammar score at Time 1 (SD)*

3.81 (2.04)

7.18 (3.07)

3.26 (1.35)

8.18 (2.19)

3.34 (1.78)

7.4 (2.73)

Leiter IQ time 1 (SD)

87.24 (18.04)

99.68 (16.38)

87.19 (17.49)

98.20 (18.99)

86.12 (16.79)

101.33 (18.03)

Gender

Males = 42

Males = 15

Males = 37

Males = 18

Males = 36

Males = 19

Females = 3

Females = 4

Females = 5

Females = 2

Females = 5

Females = 2

*Combined average of grammatical completion and grammatical understanding subscores of the test of language development (TOLD)

Table 2

Agreement between different methods of classification

 

Presence/absence of SLI at T1

Presence/absence of SLI at T2

Clinical diagnosis

K = 0.51 (SE = 0.12)

K = 0.37 (SE =  0.13)

p < 0.0001

p < 0.04

Presence/absence of SLI at T1

K = 0.51 (SE =  0.12)

 

p < 0.01

Results

At time 1, 64 children aged 4–6 were included in the sample of high-functioning PDD; 19 children were diagnosed as having AS and 45 were diagnosed with HFA (see Table 1).

Of the original sample of 64 children, two children were lost to follow-up at time 2 (due to a family move) and one of these was not included in this study although he did return for inclusion in the larger longitudinal study at time 4. Of the individuals assessed at times 1 and 2, all but one were assessed at follow-up at time 3. Three were unavailable at time 4 (all either declined follow-up at that time or had moved out of the region). At time 5, two of the children assessed at time 4 did not return for follow-up due to refusal to stay in study and moving out of the area (see Fig. 1).

Group Membership at Time 1 and 2

Agreement in group membership across the classifications was moderate (K = 0.46, SE (K) = 0.07). Cross-tabulation of the group as classified by clinical diagnosis (T1) and by SLI (T1) revealed a kappa coefficient of 0.51 (SE = −0.12, p < 0.001). Of the subjects with autism, 83.7% were also classified as SLI, compared to 31.5% of subjects with AS. Eighty-six (85.7%) percent of children classified as SLI were also diagnosed as HFA. By Time2, the kappa coefficient for agreement between clinical diagnosis (T1) and SLI (T2) had fallen to 0.37 (SE = 0.13, p < 0.04). The agreement between SLI (T1) and SLI category at T2 was also examined, revealing a kappa coefficient of 0.51 (SE = 0.12, p < 0.01). Of the 42 children classified as SLI at T1, eight (19%) were no longer language-impaired at T2. Conversely, of 18 children who were determined by testing to be free of SLI at T1, five (27.8%) fell within the range of impairment on structural language at T2 (see Table 2).

In summary, there were only moderate levels of agreement between the three-ways of classifying the children emphasizing the lack of agreement between early language delay and concurrent or later SLI.

Outcomes at Time 3

All three predictor variables, clinical diagnosis and SLI at time 1 and 2 accounted for a significant amount of the variation observed in the socialization subscore on the Vineland at time 3. Clinical diagnosis accounted for a greater amount of the variance, though the difference was marginal (see Table 3).
Table 3

Time 3 outcomes

 

B

SE

R2

Vineland social

Clinical diagnosis

−19.01

5.48

0.15**

SLI group time 1

−18.19

5.50

0.14**

SLI group time 2

−8.75

3.65

0.09*

Vineland communication

Clinical diagnosis

−20.69

6.49

0.13**

SLI group time 1

−23.66

6.33

0.18**

SLI group time 2

−24.04

6.20

0.19**

Vineland daily living skills

Clinical diagnosis

−24.66

5.96

0.21**

SLI group time 1

−21.36

6.14

0.16**

SLI group time 2

−20.86

6.09

0.15**

Autism behavior checklist

Clinical Diagnosis

24.81

6.67

0.17**

SLI group Time 1

13.09

7.20

0.04

SLI group Time 2

20.49

6.82

0.12**

SLI = structural language impairment, B = beta co-efficient, SE = standard error, R2 = amount of variance explained

**p ≤ 0.01

*p ≤ 0.05

Variance in daily living skills on the Vineland was significantly related to all predictors: Clinical diagnosis (R2 = 0.21, p < 0.001), SLI at time 1 (R2 = 0.16, p < 0.01) and SLI at time 2 (R2 = 0.15, p < 0.001).

All independent variables predicted a significant amount of variation in the communication subscale of VABS: Clinical diagnosis contributed to 12.9% of the variance (p < 0.02), SLI at T1 contributed 17.8% (p < 0.001), and SLI at T2 made the largest contribution of 19% (p < 0.02) of the total variance on the communication score at age 10–12. Clinical diagnosis accounted for the greatest amount of variance in the total ABC score at age 10–12, at 16.9% (p < 0.001). SLI at T1 was not significantly predictive (R2 = 0.04, p = 0.07), unlike SLI group at age T2 (R2 = 0.12 p < 0.004).

Outcomes at Time 4

Assessment of variables at time 4 revealed that SLI at Time 2 accounted for a greater percentage of variation in all outcome scores compared to the other classification methods. For example, on the socialization subscale on the Vineland, clinical diagnosis contributed to 8.2% of variance (p < 0.02) while SLI at T1 was not significantly predictive. SLI at age T2 however accounted for 10.0% of the total variance in his/her socialization scores at time 4 (p < 0.013) (see Table 4).
Table 4

Time 4 outcomes

 

B

SE

R2

Vineland social

Clinical diagnosis

−14.67

6.16

0.08*

SLI group time 1

−10.79

5.94

0.04

SLI group time 2

−14.89

5.80

0.10*

Vineland communication

Clinical diagnosis

−18.03

7.12

0.09*

SLI group time 1

−17.80

6.92

0.10*

SLI group time 2

−24.04

6.37

0.21**

Vineland daily living skills

Clinical diagnosis

−12.0

7.60

0.03

SLI group time 1

−15.46

6.83

0.08*

SLI group time 2

−23.34

6.64

0.19**

Autism behavior checklist

Clinical diagnosis

19.51

6.96

0.11**

SLI group time 1

18.17

6.79

0.11**

SLI group time 2

21.91

6.66

0.16**

SLI = structural language impairment, B = beta co-efficient, SE = standard error, R2 = amount of variance explained

**p ≤ 0.01

* p ≤ 0.05

For the daily living measure from the Vineland, clinical diagnosis was not a significant predictor (R2 = 0.03, p < 0.12), SLI at T1 accounted for 7.6% of the variation in the score (p < 0.001) and SLI at T2 contributed 8.5% (p < 0.019). All predictors were significantly related to outcomes on the communication subscale at age T4: clinical diagnosis accounted for 9.4% of the variation (p < 0.01), SLI at T1 contributed 10.1% (p < 0.01) and SLI at T2 accounted for twice as much of the variance (R2 = 0.21; p < 0.001). Variation in the total score of the ABC at T4 was significantly predicted by all three classifications: But once again SLI at T2 accounted for most of the variation, 15.9% (p < 0.02).

Outcomes at Time 5

A similar pattern was seen for outcomes at time 5: SLI at T2 accounted for most of the variation on three of the four outcome variables. The exception involved the socialization score on the Vineland scale at time 5: clinical diagnosis was a somewhat more robust predictor, accounting for 9.5% of the variance (p < 0.01) as compared to 8.2% predicted by SLI at T2 (p < 0.082). SLI at the earlier T1 was not associated with outcome on socialization. Clinical diagnosis accounted for 8.5% (p < 0.019) of the variation in the daily living measures of the Vineland at T5, whereas the SLI classification at age T1 was not significant (R2 = 0.05, p < 0.07). Time 2 SLI, however, was the most robust predictor of later daily living scores, and for the communication subscale of the Vineland. Clinical diagnosis accounted for 12.5% of the variation in the total ABC score (p < 0.01), while the presence or absence of SLI at T1 again failed to predict any outcome significantly. Once again the SLI categories at the later time T2 accounted for more of the variation than either of the other two ways of classifying children (see Table 5)
Table 5

Time 5 outcomes

 

B

SE

R2

Vineland social

Clinical diagnosis

−17.35

6.77

0.10*

SLI group time 1

−12.98

7.00

0.05

SLI group time 2

−15.52

6.61

0.08*

Vineland communication

Clinical diagnosis

−18.21

7.59

0.08*

SLI group time 1

−18.75

7.61

0.09*

SLI group time 2

−22.84

7.11

0.15**

Vineland daily living skills

Clinical diagnosis

−21.91

9.03

0.09*

SLI group time 1

−16.63

8.94

0.05

SLI group time 2

−22.89

8.73

0.10*

Autism behavior checklist

Clinical diagnosis

23.31

8.56

-0.13**

SLI group time 1

13.13

8.46

0.03

SLI group time 2

27.42

7.82

0.21**

SLI = structural language impairment, B = beta co-efficient, SE = standard error, R2 = amount of variance explained

**p ≤ 0.01

*p ≤ 0.05

Discussion

According to DSM-IV criteria, children with autism and children with AS are distinguishable diagnostically by the presence of cognitive impairment and language delay. However, many children have HFA, that is their cognitive scores fall within the range of “average” IQ As reviewed above, the ASD literature remains equivocal overall as to whether and how HFA and AS differ with respect to etiology, neuropsychological and psychiatric profile and prognosis. Developmental and psychiatric diagnoses should convey not only an understanding of how a disorder is distinguished from others with respect to cross-sectional presentation, but also clarify its etiology and prognosis. We hypothesized that the use of the psychometrically measured construct of SLI measured at 6–8 years of age would be a better predictor of later variance in outcome scores with respect to autistic symptoms and functioning as compared to the present DSM-IV differentiation that is based on language delay. We decided that specifically investigating the construct of SLI, i.e., difficulties with the production and comprehension of syntactic or grammatical elements of language, would be more helpful than that of specific language impairment, which may also include difficulties with pragmatic or semantic components.

We found that classifying high-functioning children with ASD according to the presence or absence of SLI created different yet overlapping groups of children compared to classifying them according to the standard DSM-IV diagnosis of HFA or AS. Furthermore, the classification based on the presence or absence of SLI at age 6–8 predicted the greatest amount of variance in many outcomes.

This study also found that, with few exceptions, the classification of children with high-functioning autistic spectrum disorder according to SLI at T2 was a more robust predictor of variance in outcome scores on the Vineland Adaptive Behavior Scale and the Autism Behavior Checklist at times 4 and time 5. Furthermore, we found that assessing a child’s grammatical language abilities at the later time of assessment (T2) was more predictive of later outcome, as compared to the clinical standard of retrospective recall of language milestones achieved at age 2–3 (as is understood by the DSM-IV criteria) or to the assessment of structural language at 4–6 years of age. While the other classification using clinical diagnosis often accounted for more variation in outcome at time 3 there was a clear pattern across times 4 and 5 that the presence or absence of SLI at time 2 consistently accounted for a larger amount of variation than the contributions of the other two methods of grouping the children.

Limitations of this study include the number of comparisons drawn as well as the finding that the differences between the various methods of classification were not of great size across the range of results and could not be tested statistically. However, conclusions may be drawn from the overall pattern of results, which is consistent. Of the 12 outcomes measured across times 3, 4, and 5, the measurement of SLI at time 2 predicted the greatest amount of variation in seven later outcomes as compared to four outcomes for which clinical diagnosis yielded the highest R2-values. It is possible that the presence or absence of SLI at time 2 was more consistently predictive simply because it is temporally more proximal to the outcomes being measured rather than being due to a more latent developmental process, as we would argue. However, Time 2 SLI did not account for a greater amount of variation in outcomes at Time 3, measured only 2 years later. Further exploration of potential causal pathways involving language and other developmental deficits and outcomes will be required to clarify these early findings.

Why would the assessment of language at 6–8 years of age be more predictive of outcome in later adolescence? It may be that, similarly to the variations in rate of vocabulary and grammar acquisition among typical children, language development in autistic-spectrum children is quite variable and that the time at which individual differences “settle out” may be further delayed; grammar acquisition may be too heterogeneous during the first 5 years in ASD children to declare a persistent SLI that would account for poorer outcomes in later years (Tager-Flusberg 2004). This may explain why SLI at time 1 was not found to be as predictive as impairment at T2.

Clinically, the use of SLI rather than parental or caregiver report of their child’s language milestones confers several advantages as a diagnostic criterion. It eliminates parent recall of past events relating to child health or development, which has been shown to be of variable reliability (Majenemer and Rosenblatt 1994; McCormick and Brooks-Gunn 1999). By using a construct such as SLI measured at or after 6 years of age, valid, reliable and widely available psychometric tests such as the TOLD may be used to address diagnostic accuracy on a case-by-case basis. Furthermore, differentiating between language-impaired and unimpaired groups may address the widespread variability and frequent lack of agreement between experts in how to classify children as AS versus autism. For example while some groups currently use the exact DSM-IV criteria others use a modified version for their research studies. The use of retrospective recall of language milestones as a means of differentiating between AS and HFA has been criticized due to potential bias and also that it is too inclusive a criterion (Klin et al. 2005). Our findings suggest that the commonly used developmental reference points, that the child speaks in words by age 2 and phrases by age 3, is not as helpful (at least as far as predicting outcome is concerned) as waiting until the later age of at least 6 years old to determine whether a child is language impaired or not. Given that one of the important roles of diagnostic labels is to highlight or predict outcome, it may be that the assessment of structural language at a later age is a more useful criterion for further subcategorizing children with ASD. The question as to why abilities in grammar are predictive of later outcomes in function and symptoms is important to our understanding of the types of neuropsychological deficits underlying autism. Furthermore, considerable discussion continues about the nature of grammatical development itself across all children. Within psycholinguistics, schools of thought range from concepts of Universal grammar, wherein syntactic elements of language are considered innate, to the theory that much of the structure of language is more socially embedded as it is learned by children through processing and interacting with the linguistic output of others (Chapman 2000; Toppelberg and Shapiro 2000). Therefore, the key predictor of later outcome in this study may not be grammar but the earlier social ability that it reflects, such as the joint attention necessary to engage with the words and sentences of others.

Structural language may be a marker of some other process within cognitive development. For example, increasing grammatical sophistication relies upon a child’s improving ability to recognize patterns and representations within language, to apply rules and create shortcuts within these analyses such that reception and expression of language is fairly automatic. It may be that a cognitive deficit such as poor executive function may be reflected in this diminished ability to sequence and structure language, and that this underlying cognitive weakness may also be impacting upon adaptive function and autistic symptoms. Studies of the association of executive function with symptoms and outcomes in PDD have resulted in mixed conclusions, according to a 2001 paper by Liss et al. (Liss et al. 2001). They found that the literature generally supports perseveration and cognitive inflexibility as the subtype of executive dysfunction that has been most consistently detected, and that general executive function deficits are not universal in PDD. Furthermore, their study comparing IQ-matched children with ASDs and those with developmental language delay found that performance on executive function tests was associated with adaptive functioning and symptom burden, however these differences disappeared when verbal IQ was covaried. They discussed the possibility that executive function and verbal IQ are interrelated, either with a deficit in one impacting upon performance in the other domain, or that another cognitive process may cause overlapping variance in the two (Liss et al. 2001). This may be reflected in our findings as well.

The TOLD grammar subtests were used to assess the syntax, or structure of language, as independent from the components involving elements of social communication such as understanding meaning and context. The intent was to separate out the children’s ability to grasp the “form” of language from their understanding of semantics and pragmatics. However, grammar development may be aided to some degree by “bootstrapping”—Using other domains of language, e.g., prosodic, semantic or pragmatic elements, to further build upon knowledge and use of grammar (Karmiloff and Karmiloff-Smith 2001; Chapman 2000). Children with ASD would clearly be at a disadvantage with respect to these skills in social communication. Therefore, the predictive power of SLI in these children may reflect the social deficits that would be significant enough to both “derail” syntactic understanding and impact upon later symptoms and function. These ongoing discussions within the domain of psycholinguistics of grammar and the cognitive and developmental psychology of ASD may therefore inform each other. It will be important to further explore the role of structural language and its effect on the trajectories of children, as well as its relationship to other aspects of cognitive function and social impairment.

In conclusion, we argue that an alternative way of differentiating between groups of children with ASDs than the current criterion of cognitive and language delay, or the diagnoses of autism versus AS may be to use the notion of SLI. Classifying the children with ASD according to whether or not they met criteria for more persistent language impairment, identified a somewhat different group of children with ASD and accounted for greater variation in later outcomes than did the typical grouping according to autism or AS. The application of language impairment or disorder as a diagnostic criterion differentiating HFA from AS also avoids recall bias as it is measured with psychometric testing as opposed to recalling past milestones. Further research will be needed to explore possible causal pathways and the relationship between language and other facets of development in ASDs.
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-007-0428-7/MediaObjects/10803_2007_428_Fig1_HTML.gif
Fig. 1

Flow diagram in attrition of sample over time

Acknowledgment

This research project was funded by the Ontario Mental Health Foundation. We would like to acknowledge the children and families who have participated so faithfully in this project over several years.

Copyright information

© Springer Science+Business Media, LLC 2007