A Preliminary Epidemiologic Study of Social (Pragmatic) Communication Disorder in the Context of Developmental Language Disorder

Abstract

Background:

There is extremely limited population-based research on Social (Pragmatic) Communication Disorder (SCD). Population-based samples have the potential to better characterize the SCD phenotype by mitigating confounds and biases that are typical of convenience and clinical samples.

Aims:

The aims of this preliminary epidemiologic study were to advance our understanding of the SCD phenotype relative to developmental language disorder (DLD), obtain an estimate of prevalence, identify risk factors and lay the groundwork for future population level research of SCD.

Methods & Procedures:

We analyzed existing data from the EpiSLI Database to examine social communication skills in 393 8^th grade (13–14 years) children with and without a history of DLD. The primary measure used to evaluate SCD was the Children’s Communication Checklist (CCC-2). Two case definitions of SCD reflecting DSM-5 criteria were examined. Both definitions involved significant pragmatic impairment, employing a commonly adopted clinical cut-point of 1.5 SD. In one case pragmatic deficits could occur along with structural language deficits and in the other case (established using principal component analysis) pragmatic and social skills were disproportionately lower than structural language abilities.

Outcomes & Results:

When using the first case definition, SCD was much more common in children with a history of DLD than without DLD and history of language disorder at kindergarten was a significant risk factor for SCD in adolescence. However, it is important to note that SCD could be found in children with no prior deficits in other aspects of language. When the second definition was employed, SCD was equally distributed across children with and without a history of DLD. Male sex was a significant risk factor using this case definition of SCD. The estimated prevalence of SCD ranged from 7% (SE=1.5%) to 11% (SE=1.7%), acknowledging that prevalence depends on the cut-point selected to determine communication disorder.

Conclusions & Implications:

These findings contribute to our understanding of the association between SCD and DLD by recognizing varying profiles of pragmatic and social communication difficulties, which in turn may help refine our diagnostic categories. Preliminary prevalence estimates of SCD can serve as an initial guidepost for identification and planning for intervention services for this condition.

Introduction

Much of the research on child language and communication disorders has employed convenience or clinical samples which can be fraught with bias and confounds, including skewed socio-economic background, co-morbid conditions, and differences in help-seeking behaviors or access to services (Tomblin 2010). Increased use of population-based samples is essential to establish representative characteristics of developmental language and communication disorders to advance our understanding of these conditions, as well as inform clinical practice and public policy decisions (Law et al. 2013). One condition about which there is limited population-based research is Social (Pragmatic) Communication Disorder (SCD) (DSM-5; American Psychiatric Association 2013). SCD entails persistent problems in both verbal and nonverbal communication that lead to functional limitations in communication abilities, social skills, academic performance and/or employment. Symptoms of SCD include problems with pragmatic skills such as topic maintenance or use of social reciprocity; social interaction skills such as code switching or adjusting speech styles are also negatively impacted, as are social cognition abilities and construction of inferences (Norbury 2014, Topal et al. 2018). It is assumed that SCD begins early in development but that it cannot be diagnosed until 4–5 years of age (Swineford et al. 2014), with mild cases perhaps not apparent until adolescence (Topal et al. 2018). According to the DSM-5, SCD can co-occur with other language and communication disorders but cannot be attributed to low abilities in vocabulary and grammar. Further, a diagnosis of SCD cannot be made in combination with a diagnosis of autism spectrum disorder (ASD), intellectual disability (ID) or global developmental delay, or be better characterized by other mental disorders (particularly, social anxiety disorder or attention deficit hyperactivity disorder, ADHD).

There is a good deal of interest among researchers in clarifying the points of distinction among SCD, developmental language disorder (DLD), and ASD (Gibson et al. 2013, Taylor and Whitehouse 2016). Debate exists as to whether SCD represents the condition previously identified in the literature as semantic-pragmatic deficit (Rapin and Allen 1983) and subsequently as pragmatic language impairment (Bishop 2000), or whether SCD was previously identified as mild ASD (Ellis Weismer 2013, Gibson et al. 2013, Norbury 2014). Several researchers have questioned the validity and usefulness of the SCD diagnostic category. Flax et al. (2019) suggest that the value of an SCD diagnosis is unclear and that this condition might best be viewed in terms of the broad autism phenotype (i.e., subclinical ASD). Ash et al. (2017) conclude that their findings indicate that SCD encompasses a group of symptoms that are difficult to differentiate from other language disorders and socioemotional problems.

Among the scant epidemiologic research to date focused on SCD are two studies examining its association with ASD and one investigating SCD relative to other language disorders. Kim and colleagues (2014) conducted a re-analysis of their previously published population-based data for school-age children in South Korea (N=55,266, Kim et al. 2011). They randomly selected 60 cases with full assessments to compare DSM-5 diagnoses with the original DSM-IV-TR criteria. DSM-IV-TR and DSM-5 diagnoses diverged in 22 cases and 17 of those were then determined to meet criteria for SCD (using clinical expert consensus). Based on these findings, Kim et al. (2014) reported an estimated prevalence for SCD of 0.49% with a 95% confidence interval of 0.21–0.77%.

A preliminary epidemiologic investigation by Ellis Weismer et al. (2020) sought to characterize demographic and clinical characteristics of a sample of likely SCD cases using data (N=1,094) from the Study to Explore Early Development (SEED), a multi-site case-control study funded by the Center for Disease Control and Prevention. SEED is the largest epidemiologic study in the United States (US) comparing children with ASD, other developmental disorders (DD), and a population sample of children with no known developmental delays. This study used selected items from an autism diagnostic measure that reflected DSM-5 criteria for SCD (based on Foley-Nicpon et al. 2017). Findings indicated that likely SCD appears to fall along a continuum (between ASD and DD with no SCD) involving elevated deficits in social communication and restricted/repetitive behavior that fail to reach clinical threshold for ASD.

Ash et al. (2017) attempted to identify SCD in a community-based sample (N=1,060) from the US of young children participating in language screening. Parents of 208 children completed the Children’s Communication Checklist-second edition (CCC-2, Bishop 2006) which was used to evaluate possible SCD. Ash and colleagues used two different pragmatic scores from the CCC-2: the social interaction difference index (SIDI) and a composite of 5 subscales tapping pragmatic abilities. Factor analyses revealed different factor structures for boys and girls but failed to find a unique factor structure for either pragmatic composite score. The results led these investigators to question the extent to which SCD is a distinct diagnostic category. In contrast to the findings by Ash and colleagues (2017), Tomblin et al. (2014) reported a principal component analysis in which they found a unique factor structure for pragmatics as assessed by the SIDI (SIDC, Social Interaction Deviance Composite in the earlier UK version) from the CCC-2; however, Tomblin and colleagues were focused on characterizing individual differences in language disorders rather than attempting to examine SCD.

As is evident from the description of population-based research above, the field is currently lacking a clinically reliable assessment measure (gold standard) to diagnose SCD (Flax et al. 2019); however, researchers have suggested use of certain existing measures of pragmatic abilities (Norbury 2014). A pre-publication version of the US edition of the CCC-2 was used in the present study of SCD. All of the items were identical to the current version, though two of the scales were titled slightly differently. The CCC-2 is one of the only assessment measures which includes items that tap all of the DSM-5 criteria for SCD (Ash et al. 2017, Yuan and Dollaghan 2018) and has been used in experimental investigations of SCD by various researchers (e.g., Adams et al. 2018, Mandy et al. 2017).

The focus of this study was to examine SCD within the context of other language disorders. This study employed a prior population sample of carefully characterized language disorder status to provide a basis for future prospective epidemiological research of SCD. It is important to study SCD relative to language disorder because it is characterized as a communication disorder rather than as part of the autism spectrum according to the DSM-5. In the absence of a clear guidelines for operationally defining SCD, we applied two different case definitions. The goal of this investigation was to determine the percentage of children with and without a history of language disorder who exhibit SCD, establish a preliminary estimated prevalence of SCD, and identify risk factors for this condition.

Methods

This study employed secondary analysis of data drawn from the EpiSLI Database (Tomblin 2010). The present study focused primarily on 8th grade data, collected between December 20, 2001 and July 06, 2003. Although data from the EpiSLI Database are admittedly dated, these epidemiologic data remain the richest and most recent data on developmental language disorder within the population in the US. At the time the data were collected, the most current versions of assessment measures were administered, and updated versions of those same measures (primarily involving newer normative samples) remain in use today. In fact, the primary assessment tool used to identify SCD in the current study, the CCC-2 (Bishop, 2003, US copyright 2006), is the most recent (English language) version of this measure available and continues to be widely used. The language domains evaluated in the US epidemiologic project (receptive/expressive semantics, grammar, and pragmatics/discourse) match contemporary conceptualizations of language development and disorders. For the current study we adapted the original categories of language impairment (‘specific language impairment’ and ‘nonspecific language impairment’) to align with recent proposals regarding the terminology and criteria associated with ‘developmental language disorder’ (DLD; Bishop et al. 2016). In addition to the numerous earlier papers based on this epidemiologic investigation, more recent publications provide evidence of the continued value of this well-validated, population sample of language impairment (Duff et al. 2015, Lancaster and Camarata, 2019, Park et al. 2015). Use of the EpiSLI Dataset to explore questions about SCD was a cost-effective method of providing preliminary epidemiologic information about SCD to stimulate an iterative process of refining our conceptualization of this condition, determining appropriate content for assessment measures designed specifically for SCD, and ascertaining estimates of children who may benefit from clinical services for deficits in social communication.

Because thes de-identified data from the EpiSLI Database were made publicly available for research, no parental consent was needed for the current analyses. Initial approval of this protocol was granted by the Institutional Review Board at the University of Iowa and parents provided written consent, with older children giving assent to participate.

EpiSLI Database and Population Sample

The initial EpiSLI Database sample resulted from a contract (PI, JB Tomblin) from the National Institute on Deafness and Other Communicative Disorders, National Institutes of Health, to establish explicit criteria for the diagnosis of specific language impairment (SLI) and to estimate the prevalence of SLI at school entry (kindergarten). That study was comprised of a stratified cluster sample of 7,218 kindergarten children residing in the Midwestern states of Iowa and Illinois in the United States (Tomblin 2010). The sample was stratified by residential setting such that three population centers each contributed an urban sample and surrounding suburban and rural samples. The EpiSLI sample was comparable to national census data at that time with respect to residential strata, sex, and race/ethnicity with the exception of a lower proportion of Hispanics and Asians. This difference reflected the need to exclude bilingual children due the known differences between development of language in monolingual and bilingual speakers. All kindergarten children at a sampled school participated in the screening which focused exclusively on language ability. The kindergarten testing was completed from Feb 24, 1993 to Nov 10, 1995. At the diagnostic testing phase exclusionary criteria were implemented such that children were not asked to participate further if they had intellectual disability, autism, neurological problems, blindness, hearing impairment, or came from a home where English was not the primary language spoken. Other than for children meeting these exclusionary criteria, the diagnostic phase of the study included all children who failed the screening and a random sample (33%) of those who passed the screening test (see Appendix: Flowchart). A total of 3,877 children were invited to participate in the diagnostic phase (1,933 screening failures/1,944 screening passes); parental consent was obtained for 2,084 (54%) of those invited but 75 were reported to speak more than one language leaving 2,009 monolingual children. Complete diagnostic data were obtained from 1,929 children. The diagnostic battery included assessment of hearing, language, speech, cognitive abilities, pre-reading skills, and gross motor skills. Examiners conducting the diagnostic testing were blind to screening results.

Diagnoses in the EpiSLI sample were based on language performance in addition to hearing and nonverbal cognitive assessment. The EpiSLI criteria, described by Tomblin et al. (1996), resulted in classification into one of four diagnostic categories: 1) specific language impairment (SLI) – failed language assessment but passed hearing and cognitive testing; 2) typically developing (TD) – passed all assessments; 3) nonspecific language impairment (NLI) – passed hearing but failed language and nonverbal cognitive testing; and 4) cognitive failure (CF) – passed hearing and language assessment but failed cognitive testing. The breakdown was 216 of 1,929 identified as SLI, 1,287 TD, 209 identified as NLI, and 222 were classified as CF; the estimated prevalence of SLI was reported to be 7.4% based on a language composite score of at least −1.25 SD on two of five domains in the presence of average nonverbal IQ (standard score > 87, Tomblin 2010, Tomblin et al. 1997).

A clinical research center was established (PI JB Tomblin) to follow the kindergarten EpiSLI sample from elementary school into high school (2nd grade through 10th grade, see Tomblin and Nippold 2014). The longitudinal study consisted of numerous experimental tasks, standardized tests, and questionnaires, including measures of language, reading, phonological awareness, intellectual abilities, memory, processing speed, social skills, and academic abilities (Tomblin and Nippold 2014).

History of Language Disorder

Examination of the occurrence of SCD in the presence/absence of language disorder was made based on initial kindergarten diagnosis of language disorder using EpiSLI criteria (Tomblin et al. 1996). Rather than retaining the four original EpiSLI diagnostic categories, a dichotomous indicator of language abilities was used in this study (DLD and No DLD). The rationale for combining the SLI and nonspecific language impairment (NLI) categories was threefold: Current DSM-5 criteria for language disorder do not include discrepancy criteria between language and nonverbal cognition. A recent population-based study of language disorder at school entry in the UK found no significant differences in developmental outcomes for children with average or low-average nonverbal IQ (Norbury et al. 2016). Further, cluster analysis of the EpiSLI kindergarten data revealed no evidence of separate clusters for the SLI and NLI diagnoses (Lancaster and Camarata 2019). Therefore, ‘DLD’ (Bishop et al. 2016) was used to reflect language impairment with a broader range of nonverbal cognition (>70), and ‘No DLD’ designated typically developing and low cognitive groups with normal range language abilities.

Assessment of Social Communication Disorder

The CCC-2 parent-report questionnaire consists of 10 separate scales, focusing on structural language and discourse skills (4), pragmatic communication (4), and behaviors typically indicative of ASD (2). The CCC-2 yields two summary measures, the General Communication Composite (GCC) used to identify clinically significant communication disorders and the Social Interaction Difference Index (SIDI) which provides an index of social pragmatic abilities compared to structural language skills. A validation study (Norbury et al. 2004) provided overall support for the CCC-2 but indicated that it did not reliably distinguish between subtypes of language disorder. Tomblin et al. (2014) examined patterns of spoken language abilities in the sample comprising the EpiSLI Database. A principal component analysis for 8^th grade data revealed a four-factor solution in which the fourth component (Eigen value of 1) was formed soley by the SIDI. The GCC from the CCC-2 loaded with other standardized language tests and language sample measures to form the first component (Eigen value of 4.5) reflecting receptive and expressive word and sentence use. It should be noted that SIDI also loaded quite strongly on the first component. Given the reported limitations of the SIDI measure and use of alternate indices derived from the CCC-2 to confirm pragmatic weaknesses by other researchers (e.g., Adams et al. 2012, Ash et al. 2017, Geurts and Embrechts 2010, Helland 2014) we decided to attempt to empirically establish patterns of performance on the CCC-2 using principal component analysis.

Exclusionary/Validation Measures and Covariates

Performance intelligence quotient (PIQ) in 8^th grade was measured using the Block Design and Picture Completion subtests of the Wechsler Intelligence Scale for Children – Third Edition (WISC–III, Wechsler 1991). Eighth grade data from the Attention Problems scale of the Child Behavior Checklist (CBCL, Achenbach 1992) were used to identify clinical ADHD, defined as scoring above the 97^th percentile. Principal component analysis results from the CCC-2 were validated against 8^th grade scores from the Social Problems scale of the CBCL (Achenbach, 1992) and the Social Skills Rating Survey (Gresham and Elliott 1990). Other variables from the dataset used in this study included sex, race (white, non-white), residential stratum (rural, suburban, urban) and kindergarten EpiSLI language composite scores.

Weighting and Adjustments for Participation Bias

The EpiSLI sample intentionally over-represented language disorder; the initial weighting procedures used to estimate prevalence of language difficulties in the population at kindergarten are described in Tomblin et al. (1997). The assessment of SCD for the current study was based on 8^th grade EpiSLI data (N=527) and used the EpiSLI weights for the 8^th grade sample. However, only a subset of 8^th graders (N=393) who were assessed had CCC-2 questionnaire data used to identify SCD. In order to correct for potential bias due to non-participation in that particular measure we examined independent variables that predicted the probability of being in the 8^th grade sample with versus without CCC-2 data. Logistic regression analyses revealed that a combination of kindergarten nonverbal cognition (block design subtest of the Wechsler Preschool and Primary Scale of Intelligence-Revised, Wechsler 1989), race, and sex, along with screening pass/fail status and kindergarten language diagnosis, significantly predicted the probability of inclusion in the samples with or without CCC-2 data. The original 8^th grade weights were multiplied by the estimated inverse conditional probability of CCC-2 participation rescaled to add up to 527 and 393. After applying the adjusted weights, the samples with and without CCC-2 data did not differ significantly in terms of parent education, sex, race, kindergarten assessment of performance IQ, and z-scores from the kindergarten EpiSLI language composite as summarized in table 1.

Table 1.

Clinical and demographic characteristics (weighted mean estimates and frequencies) at Kindergarten for children with the Children’s Communication Checklist (CCC-2) data available (N=393) and children with no CCC-2 data (N=134) at 8^th grade.

Sample Characteristics at Kindergarten	CCC-2 (N=393)	No CCC-2 (N=134)	Comparison of CCC-2 and No CCC-2
	Mean (Std. Error)	Mean (Std. Error)	Estimate	Std. Error	t value	Pr (>|t|)
Performance IQ	100.46 (.69)	99.18 (1.62)	1.278	1.761	0 .726	0.468
Block Design	9.62 (.17)	9.11 (.39)	0.518	0.424	1.220	0.222
Picture Completion	10.56 (.16)	10.57 (.40)	−0.007	0.430	−0.017	0.987
Language Compositeb	0.02 (.06)	− 0.19 (.11)	0.212	0.128	1.659	0.098
Parent Education (years)	14.02 (.14)	13.57 (.28)	0.450	0.316	1.424	0.155
Sex
Male	56%	56%	NA
Female	44%	44%
Race/Ethnicity		(missing =1)
White	89%	92%	X21 = 1.455e-05
Non-White	11%	8%	p = .998
Native American	0%	0%
African American	8%	5%
Asian	0%	0%
Other	3%	3%
Hispanic	0%	4%

^aPerformance IQ standard score on the Wechsler Preschool and Primary Scale of Intelligence – Revised (Weschler, 1989)

^bEpiSLI kindergarten language composite z-score per Tomblin et al. (1997)

Chi-square, Rao & Scott adjustment

Data Analysis

Analyses were completed using the survey (Lumley 2020) routine in R (version 4.0) or SAS. We examined two case definitions of SCD. The first definition identified cases displaying significant pragmatic impairments with or without structural language disorder, using five scales from the CCC-2 – four pragmatic scales plus the Coherence scale (Ash et al. 2017). We refer to the first case definition as ‘Concomitant SCD’ because pragmatic/social communication deficits could occur along with DLD. Prevalence of Concomitant SCD was estimated as the weighted proportion of children, with/without a history of DLD, who had a mean composite pragmatic score ≤ 1.5 SD below the mean according to test norms and did not meet exclusionary criteria (low cognition, ADHD). This particular cut-point was selected based on a recent epidemiologic study in the UK that used 1.5 SD below the mean to define language disorder (Norbury et al. 2016). For our second case definition, referred to as ‘Discrepant SCD,’ we analyzed whether there was a profile of performance on the CCC-2 indicative of SCD. We applied principal component analysis (using adjusted weights) to the correlation matrix and extracted the first two components. A second principal component (PC2) emerged which captured cases in which pragmatic and social skills were disproportionately worse than structural language abilities. Prevalence of discrepant SCD was estimated as the weighted proportion of children who scored ≥ 1.5 SD beyond the mean on PC2, minus those meeting exclusionary criteria. We examined scores on other assessment measures to validate use of PC2 scores and to assess exclusionary criteria for SCD. Weighted logistic regression analyses were used to establish predictors of being in the sample with SCD for both case definitions.

Results

We first established overall performance of our sample on the primary measure of interest, the CCC-2. We used standardized scoring procedures for the CCC-2 to determine summary measures and subscale scores for children with and without a history of DLD. The assessment protocol on which EpiSLI diagnoses were made did not include the CCC-2; therefore, scores on this measure provided an independent index of language and communication abilities. Table 2 summarizes CCC-2 weighted scores broken down by language diagnosis in kindergarten. Independent weighted t-tests (two-tailed) were used to examine differences between groups (DLD vs. No DLD). Children with a history of DLD in kindergarten scored significantly lower than children without DLD in terms of the General Communication Composite (p <.0001), as well as significantly lower on each subscale (see statistics reported in table 2). The GCC score for the group with a history of DLD was >1.25 SD below the mean according to test norms. Six of the CCC-2 subscale scores for the DLD history group (Speech, Semantic, Coherence, Initiation, Context, Nonverbal Communication) fell more than 1 SD below the mean. On the other hand, the Social Interaction Difference Index (SIDI) scores did not differ significantly between the No DLD and DLD history groups (p=.40),

Table 2.

Weighted scores for the Children’s Communication Checklist, Second Edition (CCC-2) at 8^th grade for children with a history of Developmental Language Disorder (DLD) at kindergarten and children with a history of No DLD (normal language) at kindergarten.

CCC-2 Summary Variables	DLD History	No DLD History	Comparison of DLD and No DLD
	Mean (Std. Error)	Mean (Std. Error)	Estimate	Std. Error	t value	Pr (>|t|)
GCCa	80.21 (1.94)	95.43 (1.08)	−15.22	2.22	−6.86	<0.0001***
SIDIb	−0.036 (0.85)	−0.43 (0.50)	0.40	0.99	0.38	0.40
CCC-2 Subscales	DLD History	No DLD History	Comparison of DLD and No DLD
	Mean (Std. Error)	Mean (Std. Error)	Estimate	Std. Error	t value	Pr (>|t|)
Speech	6.85 (0.37)	8.95 (0.16)	−2.10	0.40	−5.24	<0.0001***
Syntax	7.24 (0.34)	9.17 (0.17)	−1.93	0.38	−5,07	<0.0001***
Semantics	6.91 (0.39)	9.09 (0.20)	−2.17	0.44	−4.89	<0.0001***
Coherence	6.81 (0.37)	9.13 (0.21)	−2.32	0.43	−5.40	<0.0001***
Initiation	6.60 (0.35)	9.17 (0.23)	−2.57	0.42	−6.14	<0.0001***
Scripted Language	7.27 (0.34)	9.35 (0.20)	−2.07	0.39	−5.29	<0.0001***
Context	5.72 (0.36)	8.94 (0.22)	−3.22	0.42	−7.69	<0.0001***
Nonverbal Communication	6.75 (0.39)	9.21 (0.22)	−2.46	0.45	−5.50	<0.0001***
Social Relations	7.06 (0.32)	9.20 (0.20)	−2.14	0.38	−5.65	<0.0001***
Interests	7.32 (0.36)	8.36 (0.22)	−1.03	0.42	−2.45	0.0147*

^aGCC = General Communication Composite (identifies clinically significant communication disorder)

^bSIDI = Social Interaction Difference Index (provides index of social pragmatic abilities compared to structural language)

^*p<.05;

^***p<.001

Identification of SCD: Two Case Definitions

Using the criteria described above for Concomitant SCD, we identified 79/393 children who met these criteria. After we established cases of possible SCD, we reviewed additional assessment data at 8^th grade to consider potential exclusionary criteria for SCD, including PIQ in 8^th grade < 70 (−2 SD) and clinical ADHD. After removing children who met these exclusionary criteria (table 3), there were 52/393 children identified with Concomitant SCD; this resulted in a prevalence estimate of 0.11 (11%), with a standard error (SE) of 0.0167 (1.7%). Figure 1 represents the breakdown of Concomitant SCD according to history of language disorder. Concomitant SCD in 8^th grade was estimated to occur in 30% of the children with DLD history and 9% of children without DLD history.

Table 3.

Number of children with potential social communication disorder (SCD) who were excluded for low (<70) performance intelligence scores (PIQ) and clinical range scores for attention deficit hyperactivity disorder (ADHD). Children with No SCD who met exclusionary criteria (in brackets) were included in the sample.

Exclusionary Criteria	Concomitant SCD (N=393)		Discrepant SCD (N=390)
	SCD	No SCD	SCD	No SCD
PIQ	12	[12]	0	[24]
ADHD	15	[ 5]	6	[14]

Figure 1.

Percentage of ‘Concomitant SCD’ at 8^th grade based on history of developmental language disorder (DLD) or No DLD at kindergarten. Children were identified based on performance on the Children’s Communication Checklist, 2^nd Edition (CCC-2) pragmatic scales (Initiation, Nonverbal Communication, Scripted Language, and Context) plus the Coherence scale. Concomitant SCD was defined as an average weighted composite pragmatic score (across 5 scales) that was at least 1.5 SD below the mean according to test norms. Percentages reflect exclusion of cases with low performance IQ (<70) or ADHD.

Our second case definition, referred to as ‘Discrepant SCD,’ was derived from principal component analysis of scores from all scales on the CCC-2. The component loadings (reflecting correlations between each scale and the respective component) are shown in table 4. In principal components analysis, the components are always ordered with respect to statistical importance. Consequently, we attended to the statistical strength of the component (as reflected by its eigenvalue) as well as its theoretical interpretability, in deciding which components to retain. While the second component was relatively weak, as evidenced by an eigenvalue below 1, we nevertheless retained it due to its interpretability and perceived value. Subsequent components were even weaker statistically and lacked perceived value and so were not retained. The first principal component (PC1) reflected a general communication ability component having high positive correlations with all scales. The second principal component (PC2) shows positive correlations with the structural language scales (particularly Speech and Syntax) and negative correlations with pragmatic and ASD symptom scales (particularly Initiation and Interests). As a result, this component distinguishes children with profiles reflecting high structural and low pragmatics/ASD symptom scores (on the positive end) from children with profiles reflecting low structural and high pragmatics/ASD symptom scores (on the negative end).

Table 4.

Component matrix for the principal component analysis of scores for 8^th grade children (N=390) on ten scales of the Children’s Communication Checklist – second edition (CCC-2), broken down by the first principal component (PC1) and second component (PC2).

CCC-2 Scale	Component 1 (PC1)	Component 2 (PC2)
Speech	.88	.31
Syntax	.89	.30
Semantics	.92	.16
Coherence	.93	.03
Initiation	.90	−.21
Scripted Language	.93	.02
Context	.93	−.02
Nonverbal Communication	.91	−.16
Social Relations	.93	−.13
Interests	.86	−.31

The eigenvalue and percent variance for PC1 were 8.269 and 82.693, respectively. For PC2 the eigenvalue was 0.406 and the percent variance was 4.058, indicating that the first component was quite strong, accounting for approximately 83% of the variance in the 10 scales while the second component was considerably weaker, accounting for just over 4% of the variance. There were 3 children for which 1 or more CCC-2 scale scores was missing so principal components were computed on a total of 390 cases. PC2 scores ranged from −3.416 to 3.517, with a mean of 0.000, SE of 0.051, and SD of 0.999.

Using this case definition, there were 29/390 cases of possible SCD. We sought internal and external validation of our use of the second principal component to identify SCD. First, we examined a score derived from the CCC-2, the SIDI, which provides an index of pragmatic/social skills compared to structural language abilities. The possible SCD group scored significantly lower than the group without SCD (M= −15.85, SE=1.04 and M=.88, SE=.38, respectively; Estimate = −16.73, SE = 1.10, t = −15.16, p<.0001). According to the CCC-2 manual (Bishop 2006), SIDI scores within the range of −10 to 10 are considered typical and include 90% of the normative sample, whereas scores of −11 or less and those of 11 or greater are more characteristic of clinical groups The group with possible SCD exhibited a significantly higher (worse) mean t-score at 8^th grade on the Social Problems scale of the Child Behavior Checklist (CBCL, Achenbach 1992)than the group without SCD (M= 63.28, SE=1.52 and M=55.58, SE=.39, respectively; Estimate = 7.70, SE = 1.43, t = 5.39, p<.001). Additionally, the possible SCD group displayed significantly lower positive social skills on the Social Skills Rating System, Parent Form Secondary Level (SSRS, Gresham and Elliott 1990) compared to the group without SCD (M=90.79, SE=2.45 and M=99.10, SE=.82, respectively; Estimate = 8.31, SE=3.15, t=2.64, p<.01).

After establishing possible cases of Discrepant SCD using PC2 scores, we examined exclusionary criteria for SCD, namely, low PIQ and clinical levels of ADHD. Once cases meeting exclusionary criteria were removed (table 1), we found a ratio of 23/390, resulting in an estimated prevalence of 0.07 (7%) for Discrepant SCD with a SE of 0.0150 (1.5%). Figure 2 illustrates the breakdown of Discrepant SCD based on kindergarten history of DLD. Discrepant SCD was estimated to be 7% in children with and without DLD history.

Figure 2.

Percentage of ‘Discrepant SCD’ at 8^th grade based on history of developmental language disorder (DLD) or No DLD at kindergarten. Discrepant SCD was identified based on a principal component analysis of weighted scores from the 10 scales comprising the Children’s Communication Checklist, 2^nd Edition (CCC-2). This analysis (second component) revealed cases in which pragmatic and social skills were disproportionately worse than structural language abilities; children who scored above 1.5 SD beyond the mean were defined as having discrepant SCD. Percentages reflect exclusion of cases with low performance IQ (<70) or ADHD.

Separate weighted logistic regression analyses were conducted to identify Concomitant or Discrepant SCD outcome using the following demographic and clinical predictors from kindergarten: sex, race, residential stratum, language diagnosis (DLD, No DLD), and performance IQ. For Concomitant SCD, regression results revealed that kindergarten language diagnosis was the only significant predictor; Estimate = 0.218, SE=0.057, t=3.801, p<.001. Fifty-eight percent (30/52) of the children with Concomitant SCD had a history of DLD whereas 24% (71/297) of children without Concomitant SCD had a DLD history. Regression findings for the Discrepant SCD definition indicated that male sex significantly predicted discrepant SCD status: Estimate = −0.075, SE= 0.032, t value = −2.359, p<.05. The estimated percentage of males is 82% (SE=8.96) for Discrepant SCD and 54% (SE=2.98) for No SCD.

Discussion

Relation of SCD to DLD

The two case definitions of SCD that we employed yielded somewhat different patterns. Although Concomitant SCD was much more common in children with a history of DLD than without DLD, results showed that SCD could be found in children with no prior deficits in other aspects of language. Discrepant SCD, defined as cases in which pragmatic and social skills were disproportionately lower than structural language abilities, were equally distributed across children with and without a history of DLD. Although SCD can co-occur with DLD (DSM-5), SCD should not be a direct result of impaired vocabulary or grammar. Given the difficulty of establishing the true nature of this relationship we elected to consider two variants of SCD – one in which structural language deficits were free to vary along with pragmatic impairment and another in which pragmatic and social impairment was the dominant feature and structural language abilities were stronger (and therefore, unlikely to be the cause of the SCD).

Preliminary SCD Prevalence and Risk Factors

Our preliminary prevalence estimates suggest that we might expect approximately 7 to 11% of 8^th grade children in the population to present with SCD based on a 1.5 SD cut-point. Note that the EpiSLI Database sample excluded children with hearing impairment, intellectual disability, ASD, and bilingualism. In calculating prevalence estimates we further took the conservative approach of excluding cases that exhibited low PIQ scores and scores in the clinical range of ADHD at 8^th grade. When we identified SCD more inclusively (Concomitant SCD), allowing pragmatic deficits to occur with/without DLD, history of language disorder at 5 years of age was a significant risk factor for SCD during adolescence. Specifically, the percentage of children with a history of DLD who had SCD was three times greater than the percentage of children without a history of DLD with SCD. When SCD was identified such that pragmatic/social deficits predominated relative to structural language (Discrepant SCD), cases of SCD were distributed equally across children with/without a history of DLD but male sex was a significant risk factor. It is noteworthy that the Discrepant SCD sex ratio was similar to the 4 to 1 ratio of males-to-females observed in ASD (Maenner et al. 2020).

To our knowledge, the only previous prevalence estimate for SCD was reported by Kim et al. (2014) to be 0.5% (CI 0.2–0.8%) based on a prior epidemiologic study of school-age children in South Korea. This figure is an order of magnitude smaller than the prevalence estimates derived from the current study. The two investigations used population samples that had been obtained for different reasons (studying language disorder vs. autism) and used very different assessment measures and clinical definitions to identify SCD. The primary goal of the Kim et al. (2014) study was to assess the convergence of autism/ASD diagnoses using DSM-IV compared to DSM-5 criteria. Only in the small number of cases in which the diagnoses diverged was SCD considered as a possible alternative diagnosis. In the present study, our preliminary prevalence estimate of SCD (7–11%) is comparable to the reported prevalence of 7% for language disorder of unknown origin (Norbury et al. 2016; Tomblin et al. 1997). It is important to note that when using a cut-point around 1.5 SD for any clinical condition, that the prevalence of that condition will necessarily be in the range of 7% if the population (after exclusions) is normally distributed.

Conceptualization of SCD

The present study examined SCD within the context of language disorder. Ash et al. (2017), using a language screening community sample, found no evidence of a unique factor structure in CCC-2 scores reflecting a distinct SCD category (however, see Tomblin et al. 2014). Although principal component analysis of all ten scales on the CCC-2 in the present study revealed a component that reflected one case definition of SCD, it accounted for a small percentage of the variance in CCC-2 performance and would need to be replicated. Prior population-based studies of SCD have also examined this condition in the context of ASD (Ellis Weismer et al. 2020, Kim et al. 2014). That research has shown that children who have some ASD features but do not meet criteria for ASD may qualify for an SCD diagnosis. Clinical and convenience sample studies have largely followed this same pattern of exploring the association between either SCD and DLD (e.g., Adams et al. 2018) or SCD and ASD (e.g., Mandy et al. 2017). More work is needed that takes a broad view of SCD to test the hypothesis that this condition falls somewhere between ASD and DLD.

There has been longstanding research and clinical interest in pragmatic and social communication deficits without an accompanying autism/ASD diagnosis, particularly in the UK (Bishop 2000). Current results confirm the existence of SCD in the absence of intellectual disability, DLD, or ADHD, but could not clearly rule out ASD or the broader autism phenotype. Additional research is warranted to examine the claim that SCD is not a separate category but represents a range of symptoms that cross various diagnoses (Ash et al. 2017, Norbury 2014). Certain overlaps in symptoms have been observed across diagnostic categories. For instance, some research suggests that the absence of restricted and repetitive behaviors in SCD compared to ASD, as specified in the DSM-5, may not be so clear-cut (Ellis Weismer et al. 2020, Flax et al. 2019). There are various challenges and considerations to keep in mind when attempting to make direct comparisons and determine category boundaries among DLD/SCD/ASD. DLD is typically identified on the basis of standardized test scores that fall below some designated level, but no particular pattern of linguistic deficits must be displayed. We applied the same cut-off (−1.5 SD) that has been used in identifying DLD to identify SCD in the current study, using a measure that evaluated all four DSM-5 diagnostic criteria for SCD. ASD is diagnosed by ascertaining that an individual displays a particular profile of deficits corresponding to the eligibility criteria specified in the DSM-5 rather than simply requiring that overall social skills fall a certain number of standard deviations below the group mean. Because of these different approaches to diagnosis of DLD/SCD/ASD, direct comparison of phenotypic characteristics and estimates of prevalence are complicated.

Clinical Implications

We assume that the Concomitant SCD case definition would be most useful clinically. That is, the main concern would be that the child/adolescent was demonstrating social communication deficits that needed to be addressed whether or not structural language (vocabulary/grammar) impairments existed along with the social/pragmatic issues. Based on the current findings, children with DLD at 5 years of age are at increased risk for SCD as adolescents. Specifically, we would estimate that roughly 30% of children with a history of DLD would display substantial enough pragmatic problems that they would benefit from services for SCD. Thus, clinicians should be especially vigilant about assessing social communication skills in children with a current or prior diagnosis of DLD. It would also be important to keep in mind that these results suggest we would expect a small proportion (approximately 9%) of children with no background of difficulties in other aspects of language to display Concomitant SCD. Therefore, broad-based screening would be needed rather than only focusing on children with DLD if these other children are to be identified and served.

The Discrepant SCD case definition provides a conservative estimate of social communication disorder that clearly cannot be attributed to problems with vocabulary or grammar. Although the SIDI score from the CCC-2 was designed to assess weaknesses in social pragmatic abilities compared to structural language skills, several studies have revealed limitations of this index (Ash et al. 2017, Helland 2014) such that clinicians should be cautious about using it to identify SCD. From a clinical standpoint, children with Discrepant SCD may represent those with social communication deficits that do not reach threshold for a diagnosis of ASD, as intimated by the fact that male sex was a significant risk factor for this case definition of SCD. When children present with autistic features but do not meet criteria for ASD, a diagnosis of SCD should be considered.

Attention to social/pragmatic communication deficits, other than in the case of ASD, is relatively limited in the US compared to the UK and children with these types of communication problems often go unserved. To date, the only published prevalence data for SCD (Kim et al. 2014) suggests that this is a very rare condition that would require few clinical resources. The current findings, though preliminary, offer a markedly different perspective which would indicate a need for educational supports and clinical services for a number of children with deficits in social communication to mitigate potential problems with academic achievement, friendships (bullying), and related psychosocial difficulties.

Limitations and Future Directions

This study has several limitations that are important to acknowledge. We analyzed data from an existing dataset and the primary data (8^th grade) had been collected nearly 20 years ago. That said, the EpiSLI sample represents the richest source of data and most contemporary epidemiologic study of DLD in the US. This sample was a valuable and cost-effective starting point for examining the association between SCD and other language disorders. The EpiSLI sample only included children for which English was the primary language spoken in the home. Exclusion of any particular group (bilingual speakers) can be viewed as a limitation for an epidemiologic investigation that seeks to estimate prevalence of a certain condition. However, exclusion of children from homes where the primary language was not English was necessary not only so that parents of the children could accurately complete the English language surveys but because research indicates that there are differences in the patterns of language development in monolingual versus bi/multilingual speakers (Bialystok, 2001; de Houwer 2017), making it difficult to distinguish language disorders from the influences of second language learning. Therefore, it is common practice to limit investigations of language disorders, including epidemiologic studies, to children who are primarily exposed to the language under investigation (see Ash et al. 2017, Norbury et al. 2016).

There is no gold standard measure for identifying SCD. Although the measure we used to identify SCD (CCC-2) is widely used for research and clinical purposes (Norbury 2014), it was not developed specifically for SCD and does not entail direct clinical assessment of behaviors. Given these various limitations, epidemiologic investigation of SCD might be viewed as premature; however, we would argue that this is an iterative process in which data from population-based samples can assist in characterizing the SCD phenotype and designing better assessment tools to use in subsequent clinical and epidemiologic research. In the absence of clear-cut guidelines, we offered two case definitions of SCD and provided preliminary estimates of prevalence, acknowledging that prevalence depends on the cut-point selected. These tentative prevalence estimates are intended to provide initial guideposts until prospective epidemiologic studies of SCD can be completed using a psychometrically-sound SCD diagnostic tool.

Two other limitations of the study pertain to the narrow age range and lack of inclusion of a sample with ASD along with DLD sample. We examined SCD within a narrow age range such that different results might be found in a younger or older sample of children/adolescents. From a practical perspective, we were limited in that the EpiSLI database only included CCC-2 assessments at 8^th grade; however, by analyzing performance of adolescents, we anticipated that we would be able to capture even subtle cases of SCD (see Topal et al. 2018). Future epidemiologic research is needed that is designed specifically to investigate the SCD phenotype in relation to other language disorders as well as ASD and to establish a definitive prevalence estimate of this condition to assist in planning educational supports and intervention services.

Conclusion

This investigation used existing data from the EpiSLI Database (Tomblin 2010), a longitudinal epidemiologic study of language disorder in children from the US. The current study focused on 8^th grade social communication skills, using two case definitions of SCD. We were able to determine the percentage of children with/without a history of DLD who exhibited SCD, establish preliminary estimates of SCD prevalence, and identify risk factors for SCD.

These findings add support for the utility of the CCC-2 in assessing SCD (Norbury 2014), though we acknowledge the need for clinical measures designed specifically for SCD. Understanding the association between DLD and SCD is important for refining our diagnostic categories and recognizing varying profiles of social communication problems. The availability of preliminary prevalence estimates for SCD is useful as a general index of how many children we might expect to require services. If the number of children being identified and enrolled in intervention is substantially below the anticipated number, this may signal the need for more comprehensive screening of pragmatic and social abilities. There is evidence, including randomized clinical trial results, to indicate that intervention directed at improving pragmatic/social communication skills is effective (Adams et al. 2012, Gail and Adams 2018); therefore, identification of SCD appears to be a worthwhile goal. The current study focused on the link between SCD and DLD but there is also evidence that SCD has overlapping features with ASD; therefore, SCD may entail service needs extending beyond a narrow focus on treatment of social communication (Brukner-Wertman et al. 2016; Ellis Weismer et al. 2020).

What this paper adds.

What is already known on this subject

There is considerable debate about the diagnostic category of SCD and its relation to other neurodevelopmental disorders.

What this study adds to existing knowledge

Using data from a US-based epidemiologic sample of DLD, this study offers new information about the association between SCD and DLD, provides preliminary estimates of SCD prevalence, and identifies risk factors for SCD.

Clinical Implications of this study

Improved understanding of possible profiles of pragmatic and social communication deficits will help to clarify diagnostic categories and preliminary prevalence estimates may assist with ensuring availability of adequate intervention services.