Pragmatic language and associations with externalizing behaviors in autistic individuals and those who have lost the autism diagnosis

Abstract

Background:

Pragmatic language weaknesses, a core feature of autism spectrum disorder (ASD), are implicated in externalizing behavior disorders (Gremillion & Martel, 2014). Particularly in a clinical setting, these co-occurring externalizing disorders are very common in autism; rates of Attentional Deficit-Hyperactive Disorder (ADHD) and Oppositional Defiant Disorder (ODD) are as high as 83% (ADHD) and 73% (ODD; Joshi et al., 2010). It is possible that pragmatic language weaknesses impact the ability to effectively communicate one’s needs, which may lead autistic children to utilize externalizing behaviors in order to achieve a desired outcome (Ketelaars et al., 2010; Rodas et al., 2017).

Methods:

The aim of the current study is to investigate the relationship between pragmatic language, assessed via multiple modalities, and externalizing behaviors, assessed by parent interview, in youth with autistic (n=33) or neurotypical (NT; n=34) developmental histories, along with youth diagnosed with autism, who lost the diagnosis (LAD) by adolescence (n=31).

Results:

The autism group had significantly more pragmatic language difficulties, and more externalizing behaviors and disorders; ADHD symptoms were particularly more prevalent, while LAD and NT groups did not differ. Challenges in pragmatic language abilities were associated with more externalizing symptoms when controlling for other facts that typically influence such symptoms, including nonverbal cognition, structural language, executive functioning, and autistic characteristics, but did not remain when age was included in the model.

Conclusions:

We discuss the mechanisms underlying difficult-to-manage externalizing behaviors and implications for interventions and long-term outcomes for youth with and without a history of autism.

Introduction

Pragmatic language encompasses multiple facets of social communication: content (i.e., word choice, figurative language); speech characteristics (i.e., prosody, repetitions, filled pauses), non-verbal communication (i.e., co-speech gestures, facial expressions, eye contact); and adjusting to context (i.e., turn-taking, register, perspective taking; Adams et al., 2005; Parsons et al., 2017). Challenges in social communication, including pragmatic language, are a central feature of autism spectrum disorder (ASD; American Psychiatric Association, 2013) and are also implicated in externalizing disorders, namely attentional deficit-hyperactive disorder (ADHD), oppositional defiant disorder (ODD; Gremillion & Martel, 2014; Ketelaars et al., 2010) and conduct disorder (CD). The current study examines the association of pragmatic language abilities with externalizing symptoms and disorders in a sample of youth with a history of autism.

Autism is associated with heightened risk for co-occurring psychiatric disorders; as many as 70% of autistic individuals have at least one additional psychiatric diagnosis (Simonoff et al., 2008). Externalizing behavior disorders are particularly frequent, with reported co-occurrences of 28% for Attentional Deficit-Hyperactive Disorder (ADHD), 28% for Oppositional Defiant Disorder (ODD), and 3% for Conduct Disorder (CD) in epidemiological studies (Simonoff et al., 2008). Rates in clinically-referred samples are even higher, with co-occurrences of 83% for ADHD, 73% for ODD, and 22% for CD (Joshi et al., 2010). More recent work has found that the increased rate of co-occurrence between autism and ADHD is associated with increased autism symptomology in both the social communication and restrictive and repetitive interests domains, especially in autistic girls (Wodka et al., 2022). The high prevalence suggests that there may be a relationship between specific features of autism and externalizing disorders.

Specific symptoms and features associated with ADHD, ODD, and Conduct Disorder (CD) such as aggression, hyperactivity, rule-breaking, oppositionality, aggression, and emotional regulation difficulties are common in autism (Baker & Blacher, 2015; Lindor et al., 2019; Matson et al., 2009; Vogan et al., 2018). More autism characteristics (e.g., higher scores on the ADOS) are related to more externalizing problems such as oppositionality and aggressiveness, and poorer emotional regulation abilities (Ros et al., 2018). Mild to borderline intellectual deficits are correlated with aggression and externalizing behaviors (Baker et al., 2002; Dekker et al., 2002). Sleep disturbance, common in autism, is also implicated in problem behaviors (Allik et al., 2006; Lindor et al., 2019). Executive dysfunction correlates with externalizing behaviors in autism (Maddox et al., 2018; Vogan et al., 2018), and in one longitudinal study, improvements in metacognitive skills (one aspect of executive functioning) in autism were associated with a reduction in social difficulties two years later (Vogan et al., 2018). Altogether, there are multiple points of overlap between autism and externalizing disorders.

Language plays a significant role in cognitive and socioemotional development; the ability to understand and manage emotions and behavior hinges, in part, on the ability to explicitly represent abstract concepts and to share emotions and thoughts with others (Nelson et al., 2011). Multiple studies have found that early language weaknesses predict later school achievement challenges, social difficulties, and behavior problems (Salmon et al., 2016). A study of 1,314 children found that expressive vocabulary at age two and receptive vocabulary at age four (measured by the MacArthur Communicative Development Inventory) accounted for 12% of the variance in later emotional and behavioral problems at age six, controlling for nonverbal IQ and biological risk (Clegg et al., 2015). A large study of 8,650 children found that parent-reported expressive vocabulary at 24 months was linked to academic achievement (reading and mathematics scores) and to teacher-reported self-regulation and emotional and behavioral problems at age five years (Morgan et al., 2015). Further, while structural language skills are consolidated relatively early in development (Salmon et al., 2016), the development of pragmatic language skills continues into adolescence (Nippold, 2000; Ryder & Leinonen, 2014).

The current study focuses on a specific language domain that might contribute to externalizing behaviors in autism: pragmatic language. Studies have identified a relationship between pragmatic language abilities and behavioral regulation in various clinical populations. For example, school-age children with ADHD have weaknesses in multiple components of expressive language including phonology, vocabulary, sentence structure, and pragmatics (Boo et al., 2022; Kim & Kaiser, 2000). A study of children ages 8–13 years who were referred to a child psychiatric outpatient clinic (Brenne & Rimehaug, 2019) indicated that 28 of 73 (38%) had pragmatic language impairments, which were associated with clinical elevations on the anxious/depressed, social problems, and rule breaking behavior domains of the Child Behavior Checklist (CBCL; Achenbach 2009). Gremillion and Martel (2014) studied 109 preschoolers with ADHD, ODD, ADHD and ODD, or no disruptive behaviors; those with disruptive behavior disorders were more prone to weaknesses in expressive and receptive structural language and pragmatic language. Furthermore, pragmatic language ability was significantly associated with disruptive behavior symptoms, an association that was strongest in the ADHD-only and ADHD plus ODD groups. Conversely, a study of 111 4-year-old children with pragmatic language impairments revealed that externalizing behavioral problems, including hyperactivity and reduced prosocial behaviors, were correlated with pragmatic weaknesses (Ketelaars et al., 2010). These findings suggest that pragmatic language (dis)abilities appear to play an important role in disruptive behaviors.

The protracted development of pragmatic skills in autism (Eigsti et al., 2011; Reindal et al., 2021) is a potential risk factor for externalizing symptoms, with more pronounced relationships emerging later in childhood. Rodas and colleagues (2017) followed 159 autistic children ages 4–7 years and reported that challenges in structural and pragmatic language abilities, assessed at the start of the school year via the Children’s Communication Checklist (CCC-2; Bishop 2006), predicted anxiety and externalizing symptoms at the end of the school year (CBCL scores). Specifically, structural language challenges weakly predicted anxiety but not externalizing symptoms; pragmatic language challenges strongly predicted both. In another study, pragmatic language difficulties made a significant, unique contribution to externalizing symptoms in both autistic (n=206) and typically developing (NT; n=187) participants, while structural language difficulties was correlated with externalizing symptoms only in the NT group (Boonen et al., 2014). Both the Boonen et al. and the Rodas et al. studies measured pragmatic language using the CCC-2 parent-report questionnaire. Because of the complex nature of pragmatic language, a single measure may not fully characterize these complex abilities; indeed, the CCC-2’s internal consistency (Cronbach’s alpha) ranges from .4 to .6, considered low (Bishop, 2006). A more nuanced assessment of pragmatic language abilities, and their impact on disruptive behavior disorders in autism, could illuminate the relationship between these aspects of development.

Pragmatic language abilities are notoriously difficult to assess because they are contextually dependent, reflecting both situational variables and the input from interaction partners (Norbury, 2014). As such, multiple distinct cognitive processes impact pragmatic language, including attention, working memory, structural language, and theory of mind (Rasgado-Toledo et al., 2021). These skills can be assessed via multiple methods including: Direct standardized clinician-led assessment such as the Test of Language Competence (Wiig and Secord, 1989) or Test of Pragmatic Language (Phelps-Terasaki & Phelps-Gunn, 2007), informant report (parent or teacher questionnaires, such as the CCC-2), and semi-structured interviews such as clinician scoring of a dyadic interaction such as the ADOS (Lord et al. 2012), or Yale in vivo Pragmatic Protocol, (Simmons et al., 2014). Each approach has its own pros and cons. While questionnaires like the CCC-2 have good ecological validity, some informants may be unable to identify subtly atypical aspects of pragmatics, and may be influenced by salient recent experiences. While standardized testing provides normed scores, the highly structured (and often visually-supported nature of standardized tests may fail to social communication challenges that may arise in everyday interactions, where the rules of engagement are implicit and highly dynamic (Eigsti & Schuh, 2016; Schuh & Eigsti, 2012). Given these limitations, pragmatic language was assessed in the current study using a composite score of multiple pragmatic language measures that included both semi-structured interviews and formal assessments of pragmatic language. This approach allows us to capture pragmatic language abilities identified through both normed, clinician-led assessments and more dynamic, spontaneous approaches.

Theoretically, the co-occurrence of pragmatic language weaknesses and externalizing symptoms may reflect a prior factor, such as genetic risk. Indeed, there is a strong genetic overlap among disruptive behavior disorders (Tuvblad et al., 2009), and possible genetic overlap of both domains of ADHD (inattentive and hyperactive) with the social and communicative impairments of autism (Pinto et al., 2016). Under a “general risk” hypothesis, the presence of autism predicts both pragmatic language difficulties and externalizing symptoms in affected individuals, with more-strongly-affected individuals showing greater difficulties in both domains. Alternatively, the link could be mechanistic, such that the presence of pragmatic language impairments specifically increases the risk of externalizing symptoms. In a community sample of 54 children, pragmatic language skill accounted for all the variance in the relationship between hyperactivity and social skills; it was also a partial mediator of the relation between inattention and social skills (Leonard et al., 2011). The current study’s unique diagnostic groups provide an opportunity to test some mechanisms underlying associations between pragmatic language abilities and externalizing behaviors by examining how they are related in individuals who were once diagnosed with autism, but no longer meet diagnostic criteria.

Loss of the ASD Diagnosis (LAD).

A series of studies by Fein and colleagues (2013) provided a detailed examination of 44 LAD individuals, compared to peers with current autism and a typically developing comparison group. The LAD sample had IQs in the average or higher range and was indistinguishable from typically developing peers on most dimensions, including adaptive skills (Fein et al., 2013), social communication skills (Orinstein et al., 2015), academic abilities (Troyb, Orinstein, et al., 2014), executive functioning (Troyb, Rosenthal, et al., 2014) and standardized tests of language (Tyson et al., 2014). Notably, this group was still more likely to have co-occurrence ADHD and phobias, but not other psychiatric disorders (Orinstein et al., 2015). By definition, individuals with LAD had a history of pragmatic language weaknesses at the time of diagnosis; in contrast, they had no or minimal weaknesses in pragmatic language skills at the time of the study, according to DSM-IV-TR criteria, although, some prior work has demonstrated subtle pragmatic language weaknesses in LAD in studies of narrative structure (Canfield et al., 2016), idiosyncratic language (Suh et al., 2014), and semantic category structure (Naigles et al., 2013). While the current study is cross-sectional, the history of autism in the LAD group, coupled with the absence of current autism symptoms, provides insight into how improvements in pragmatic language skills over time are related to externalizing symptoms. If indeed the resolution of pragmatic language weaknesses is associated with fewer externalizing symptoms, this suggests that interventions that serve to strengthen pragmatic language skills would also reduce co-occurring externalizing symptoms.

The current study analyzed pragmatic language abilities and their association with externalizing symptoms in LAD, autistic, and NT youth. We measured pragmatic language skills using a novel metric, a composite score derived from multiple pragmatic language measures. The first aim was to compare differences in pragmatic language skills across three groups (LAD, autism, NT). Given the possibility of residual subtle differences in pragmatic language for the LAD group compared to the NT group, and the use of a novel composite of pragmatic language, we planned to test for group differences in pragmatic language. We predicted significantly greater pragmatic language weaknesses and greater current externalizing symptoms in the autism group compared to the LAD and NT groups. Further, we hypothesized that greater pragmatic language weaknesses would be associated with higher rates of externalizing behaviors in all three diagnostic groups and that this relationship would hold when controlling for age, nonverbal IQ, autism symptomatology, structural language abilities, and executive function, which would suggest a mechanistic link between pragmatic language and externalizing symptoms.

Methods

Participants.

Recruitment was done through media outlets (newspaper stories, radio interviews), private practices, and clinic referrals in Connecticut. In some cases, therapists contacted parents of children known to have LAD, while in other cases, parents contacted investigators after seeing media reports. This study was a secondary analysis of data that reported by Fein and colleagues in 2013. A full description of participants, recruitment and procedures is presented in Fein et al. (2013). The original Fein 2013 study had a total of 112 youth participants (LAD = 44, autism = 34, NT = 34). This original sample was matched on age, sex assigned at birth, and nonverbal IQ. Of the 112 individuals, nine autistic participants and three LAD participants were evaluated by a study collaborator at Queens University in Ontario, Canada. Due to missing data, the Canadian sample is not included here. For the current study 98 youth (LAD = 31, autism = 33, NT = 34) were included in the final analysis. Groups did not differ on age, gender, or nonverbal IQ; groups differed on verbal IQ, with higher scores in the LAD and NT groups relative to the autism group; see Table 1. Participants were mostly White, with three LAD individuals, two autistic individuals, and three NT individuals reporting other races or ethnicities. The study was approved by the IRB and all participants provided written parental permission and verbal assent.

Table 1.

Descriptive Statistics

	Autism Spectrum Disorder	Loss of ASD Diagnosis	Typical Development	χ2 or F	p	Post-hoc tests
N (male)	33 (30 M)	31 (24 M)	34 (31 M)	3.42	0.18
Age (yrs)	13.4 (2.8) 8.8 – 20.3	13.0 (3.5) 8.6 – 21.6	14.0 (2.6) 10.1 – 22.1	0.91	0.41
ADOS CSS	5.6 (18.7) 3–9	1.4 (0.5) 1–2	1.1 (0.3) 1–2	152	<0.01	autism < LAD, NT
WASI NVIQ	112 (13) 78–147	112 (14) 87 – 142	113 (11) 89 – 139	0.67	0.94
WASI VIQ	103 (13) 81 – 133	113 (14) 80 – 137	112.0 (11) 93 – 138	5.69	<0.01	autism < LAD, NT
SCQ	23 (6) 10–33	17 (7) 5–30	2 (2) 0–12	140	<0.01	autism > LAD > NT
CELF Core Lang	99 (13) 70 – 124	110 (11) 79 – 126	117 (7) 106 – 132	23.69	<0.01	autism < LAD < NT
Tower	9.7 (2.6) 3 – 17	9.4 (2.2) 5 – 14	11.0 (2.3) 6 – 17	4.27	0.02	autism < NT, LAD

Note: Data presented asM(SD), range. ADOS CSS= Autism Diagnostic Observation Schedule Calibrated Severity Score; WASI = Wechsler Abbreviated Scales of Intelligence; NVIQ = Nonverbal IQ (standard score); VIQ = Verbal IQ (standard score); SCQ = Social Communication Questionnaire, Lifetime; CELF = Clinical Evaluation of Language Fundamentals, Core Language (standard scores); Tower = D-KEFS Tower, Total Achievement (scaled score). Standard scores: M(SD) = 100(15); scaled scores: M(SD) = 10(3).

Inclusion criteria.

All participants had IQ scores greater than 77 (i.e., within 1.5 SD of the mean of 100), to ensure that group differences in pragmatic language were not due to broad, domain-general cognitive abilities. Specific criteria for participants in the LAD group included a documented history of early language delay (no words by 18 months or no phrases by 24 months); an ASD diagnosis made by age five years by a physician or psychologist specializing in autism, documented in a written report; Vineland Adaptive Behavior Scales (Sparrow et al. 1985) Communication and Socialization scores greater than 77 at evaluation; at current evaluation, participants could not meet criteria for ASD on the Autism Diagnostic Observation Schedule (ADOS; Lord et al. 2000); and had to have at least two typically developing friends of any age at the time of testing. The early diagnostic reports were reviewed by a clinical psychologist with autism diagnostic expertise who was naive to early diagnosis and group membership, to verify that early symptoms were consistent with autism. Because participants were not seen periodically over time, the age at which individuals no longer displayed ASD characteristics is not known. LAD individuals had to be fully included in regular education classrooms with no one-on-one assistance and no special education services to specifically addressed autism (e.g., no social skills training). However, participants could receive special education services or psychological support to address impairments not specific to autism, such as attention or academic difficulties. The rationale for the “friendships” and “special educational services” criteria was to ensure that LAD participants had age-appropriate social skills. Participants in the autism group had to meet criteria for autism on the ADOS (Social and Communication domains, and total score) and according to best estimate clinical judgment. There was no Vineland cutoff for the autism group. Participants in the NT group had to have scores greater than 77 on the Vineland Communication and Socialization scales, could not have met criteria for autism at any time per parent report, could not meet current criteria according to the ADOS and clinical judgment, and could not have a first-degree relative with an autism diagnosis.

Exclusion criteria.

Potential participants were excluded if they exhibited symptoms of major psychopathology (e.g., active psychotic disorder), uncorrected visual or hearing impairments, seizure disorder, Fragile X syndrome, or significant head injury with loss of consciousness. Two potential NT and two autism group participants were excluded due to seizure disorder; none were excluded for other reasons. To avoid a hyper-normal sample, there were no exclusions for other learning or psychiatric disorders (ADHD, dyslexia, etc.). NT participants could not have a first-degree relative with an autism diagnosis.

Measures

Psychiatric diagnoses.

ASD diagnoses were made using Autism Diagnostic Observation Schedule (ADOS; Lord et al. 2000) Module 3 or Module 4, depending on the participant’s age, interests and level of independence. To determine lifetime presence of autism symptoms, parents completed the Social Communication Questionnaire (SCQ; Berument, Rutter, Lord, Pickles, & Bailey, 1999), a 40-item parent questionnaire based on the ADI-R. Symptoms of current and past externalizing disorders (ODD, CD, and ADHD) were assessed via the parent interview Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS-PL; Kaufman et al., 2000), which has excellent test-retest (.77 to 1.0) and inter-rater (93–100%) reliability (Kaufman et al. 1997). Few participants met full diagnostic criteria for any of these diagnoses; therefore, in the current study, individual symptoms of ADHD, ODD, and CD were tallied for a total Externalizing Symptoms Score. Parent responses for externalizing symptoms ranged from 0 to 2 (0 = not present, 1 = subthreshold, 2 = threshold). There were four ADHD symptom questions (range = 0 – 8), three ODD questions (range = 0 – 6), and five CD questions (range = 0 – 10), for a total of 12 externalizing symptom questions (range = 0 – 24) across three conditions.

IQ, EF, and language assessments.

The Wechsler Abbreviated Scales of Intelligence (WASI; Wechsler 1999) was used to assess verbal and nonverbal cognitive abilities. The Delis-Kaplan Executive Function System (D-KEFS; Delis, Kaplan, & Kramer, 2001) Tower subtest provided a clinician-administered assessment of executive functioning skills. The Tower subtest asks the participant to move a series of wooden discs on pegs in order to replicate a presented figure, following specific rules, requiring planning, working memory, and inhibition.

Participants completed subtests of the Clinical Evaluation of Language Fundamentals-IV Core Language score (CELF-IV; Semel, Wiig, & Secord, 2003), including Concepts and Following Directions (up to age 12), Word Definitions (ages 13 and older), Recalling Sentences, Formulated Sentences, and Word Classes. The CELF has excellent test-retest reliability (ranging from .71 to .86 for subtests, and .88 to .92 for the Core Language composite) and inter-rater reliability of .88 to .99 (Semel, Wiig, & Secord, 2003).

Pragmatic language assessments.

Pragmatic language skills were evaluated using multiple instruments. The Pragmatics Profile of the CELF-IV (Semel, Wiig, & Secord, 2003) is a clinician rating of 52 items in three domains (Conversation, Information, Nonverbal) on a 1–4 Likert Scale (1 = “never observed;” 4 = “always observed”). Internal reliability is .98; test-retest reliability is .93. The Rituals and Conversational Skills scale (Conversation) assesses turn-taking, introduction and maintenance of topics, and appropriate strategies for getting attention; Asking For, Giving, and Responding to Information (Information) assesses giving and responding to advice or suggestions appropriately, agreeing and disagreeing using appropriate language, and initiating and responding to verbal and nonverbal negotiations; and Nonverbal Communication Skills (Nonverbal) assesses facial cues, body language, and tone of voice. The Pragmatic Profile was scored using videotaped ADOS evaluations by a trained graduate student rater (J.C.) who was naïve to group membership. Twelve items (out of 52 total) were excluded because the items were not applicable to the ADOS (e.g., “participates/interacts appropriately in structured group activities”) or could not be coded from video (e.g., “participant appropriately reads and interprets facial cues from assessment administrator”). An item could also be scored as “Not Observed,” which means that although the behavior could be rated, the participant did not display the behavior. The distribution and number of “Not Observed” items varied across participants. Therefore, mean scores were calculated across coded items for each participant within each domain (Conversation, Information, and Nonverbal). Fifteen percent of the videos were double scored by a trained undergraduate coder to establish inter-rater reliability; intraclass correlations (ICCs) were high: Conversation = 0.99, Information = 0.87, and Nonverbal = 0.97. Although the reliability and validity of this scoring approach have not been evaluated, this methodology has previously been used to assess pragmatic abilities in autistic children (Orinstein et al., 2015).

A second clinician-rating measure was the Pragmatic Language Scale (PLS; de Villiers et al. 2007), designed to assess pragmatic conversational abilities such as “atypical intonations”, “abrupt topic shift”, and “perseveration” in individuals with communication challenges. Children were recorded during a 10-minute semi-structured conversation with a graduate clinician. The same clinician conversed with all participants and was blind to the purposes and objectives of the study. Two independent raters naïve to diagnostic status independently scored features on a 3-point Likert scale (0 = did not occur; 1 = occurred sometimes; 2 = occurred frequently or very salient). Inter-rater reliabilities for the nine features ranged from .34 to .83 (de Villiers et al., 2007). Of the nine features, two had Kappa scores in the “fair” range (Attention to outside environment = .34; Terseness = .38). All other inter-rater reliability scores were in the moderately high to substantial range (.41 - .82).

As a direct assessment of pragmatic language, participants completed subtests of the Test of Language Competence (TLC-E; Wiig and Secord 1989). In the Making Inferences subtest, participants hear descriptions of a situation and generate plausible inferences. In the Figurative Language subtest, participants are asked to interpret non-literal phrases. These subtests have excellent interrater reliability (97%, 98%) and low to adequate test-retest reliabilities (.34-.78).

Another direct assessment was comprised of three subtests of the Test of Written Language (TOWL-3; Hammill and Hresko, 1994). Story Construction evaluates the quality of a written story and its interest to readers; Contextual Language evaluates the quality of language used in the story; and Contextual Conventions assesses adherence to spelling and grammatical writing conventions, as well as written comprehension within a specific context. The subtests have excellent test-retest reliability (.70 to .80) and inter-rater reliability (.83 to .95). These subtests of written language were included in the pragmatic language composite because of the relevance of narrative (written or spoken) comprehension for pragmatic language skills.

Given the variable reliability of pragmatic language assessments in this study (ranging from .34 to .92), we calculated a pragmatic language composite as the average of the z-scores for each of six measures: CELF Pragmatics Profile, the PLS, TLC Inferences and Figurative language, and TOWL Story Construction and Contextual Conventions. Z-score composite scores have been used in other studies to test correlations between factors such as pragmatic language and personality traits (e.g., Suh et al., 2016).

Statistical Analyses.

Descriptive and ANOVA analyses were performed in SPSS (IBM Corp. 2017), regression analyses were performed in Stata 14.2 (StataCorp 2015), and figures were created in R (Core Team, 2013). ANOVAs were used to test for group differences in pragmatic language abilities, current externalizing symptoms, and the lifetime presence of autism characteristics. The relationship between the pragmatic language composite score and externalizing symptoms was assessed using Poisson regression, with significant results followed by planned post-hoc analyses to test specific co-occurrences (ADHD, ODD, and CD). Poisson regression models were used because they are the most appropriate for count variables (e.g., number of externalizing symptoms). First, the regression model was implemented with Group (LAD, autism, NT) entered as a dummy variable, to test for group differences in the relationship between externalizing symptoms and pragmatic language skills; see Table 3. Second, all covariates were entered into the regression models in a single block, which included age, Full-Scale IQ, CELF-Core Language Score, and Tower, to test whether these factors impact both pragmatic language and externalizing symptoms; see Table 4. Third, due to strong correlations between age with both pragmatic language and externalizing symptoms revealed in post-hoc correlations, regression models were rerun without the inclusion of age in the model; see Table 4–5 and Figure 3. Separately, diagnostic group (LAD, autism, NT) was replaced with SCQ scores as a continuous variable (groups combined) in order to test an alternative to diagnostic group that instead measures lifetime autistic symptoms. Results yielded largely similar results, therefore diagnostic group was reported; see Supplemental Materials for SCQ results. Because pragmatic language was expected to be highly correlated with autism status, multicollinearity was assessed in each regression model by evaluating the Variance Inflation Factor (VIF). Conservative cutoffs for each predictor (VIF < 4.0) and overall model (VIF < 10) were used.

Table 3.

Poisson regression: Externalizing symptoms as a function of pragmatic language and group

	Standardized Coefficient	SE	z	P>|z|	95% Confidence Interval
Externalizing Symptoms
Pragmatic Language Composite	−0.123	.0489	−2.51	0.012	−.219 – −0.027
Diagnostic Group	−0.098	.0339	−2.90	0.004	−.165 – −0.032
Constant	2.950	.071	41.69	0.000	2.811 – 3.089

Note. Constant = the predicted value of externalizing symptoms when all other variables are held constant; significant variables are shown in bold font.

Table 4.

Poisson regression: Externalizing symptoms as a function of pragmatic language, diagnostic group, NVIQ, structural language, executive functioning, and age

Externalizing Symptoms	Standardized Coefficient	SE	z-score	P>|z|	95% Confidence Interval
Prag. Lang. Composite	−0.133	0.052	−2.57	0.010	−.234 – −.032
Diagnostic Group	−0.079	0.040	−1.97	0.049	−.158 – −0.000
NVIQ SS	−0.001	0.002	−0.40	0.676	−.006 – .004
CELF Core Language SS	−0.002	0.004	−0.53	0.593	−.010 – .006
Tower	−0.009	0.011	−0.78	0.437	−.030 – .013
Constant	3.204	.238	13.48	0.000	2.738 – 3.67
Model including age
Prag. Lang. Composite	−0.058	0.059	−0.99	0.332	−.173 – −.057
Diagnostic Group	−0.083	0.040	−2.05	0.040	−.162 – −.004
Age	−0.029	.010	−2.76	0.006	−.049 – −.008
NVIQ SS	−0.002	0.002	−0.84	0.403	−.006 – .003
CELF Core Language SS	−0.003	0.004	−0.53	0.392	−.011 – .004
Tower	−0.010	0.011	−0.87	0.383	−.031 – .012
Constant	3.771	.310	12.16	0.000	3.163 – 4.379
Pairwise DX*Prag. Lang Interaction terms
autism vs NT	0.044	0.140	0.33	0.741	−.218 – .307
autism vs LAD	0.289	0.137	2.12	0.034	.022 – .557
LAD vs NT	0.333	0.167	2.00	0.046	.006 – .661

NVIQ = WASI Nonverbal IQ; CELF = Clinical Evaluation of Language Fundamentals; Tower = Delis-Kaplan Executive Function System Tower subtest; P>|z| = 2-tailed p-value. Constant = the predicted value of externalizing symptoms with other predictors held constant. Significant predictors are shown in bold font.

Table 5.

Correlations of Age, Pragmatic Language and Externalizing Symptoms

		r	t	p
autism	Age and Pragmatic Language	.56	3.77	<.001
	Age and Externalizing Symptoms	−.63	−4.56	<.001
LAD	Age and Pragmatic Language	.61	4.11	<.001
	Age and Externalizing Symptoms	−.19	−1.02	=0.31
NT	Age and Pragmatic Language	.29	1.71	=0.10
	Age and Externalizing Symptoms	−.26	−1.55	=0.13

Figure 3.

Relationship of age to pragmatic language and externalizing symptoms

Sample and multiple imputation.

Of the 112 original participants, data were missing from n=12 (the Canadian sample) and the K-SADS was missing for n=2; these participants were excluded from all analyses. In the remaining 98 (see Table 1), individual data (n=13, CELF Pragmatic Profile; n=2, TLC; n=15, TOWL; and n=8, PLS) were missing due to examiner error, equipment failure, or time constraints. Multiple imputation was used to replace this missing data. Analyses were repeated with the participants with complete data (n= 67), with similar results; therefore, the reported results of the full sample.

Results

Group differences.

The autism group had significantly lower Pragmatic Language Composite scores compared to both the LAD and NT groups; the latter groups did not differ (see Figure 1 and Table 2). Similarly, the autism group had significantly more externalizing symptoms than the LAD and NT groups, and the latter groups did not differ. The autism group had significantly more ADHD symptoms than both the LAD and NT groups, and the LAD group had significantly more symptoms than the NT group. Groups did not differ for symptoms of ODD or CD, reflecting in part the low frequency of these symptoms. Given this pattern, subsequent analyses focused on total externalizing symptoms; findings for ADHD, CD and ODD separately are presented in Supplemental Materials.

Figure 1.

Group Differences in Pragmatic Language and Externalizing Symptoms

Table 2.

Group means for Pragmatic Language and Externalizing Symptoms

	autism (n=33)	LAD (n=31)	NT (n=34)	F	p	Post-hoc tests	Cohen’s d
Pragmatic Language Composite	−0.34 (0.62) −2.23 – 0.59	0.18 (0.45) −0.72 – 1.03	0.20 (0.40) −0.58 – 0.99	12.48	<0.01	autism <LAD, NT	autism/LAD: −0.96 autism/NT: −1.04
No. ext. symptoms	6.18 (4.38) 1 – 21	3.23 (3.23) 0 – 11	1.97 (2.61) 0 – 12	12.64	<0.01	Autism <LAD, NT	autism/LAD: 0.76 autism/NT: 1.17
No. ADHD symptoms (Full diagnosis)	4.39 (2.4) 0 – 8 (8)	2.39 (2.7) 0 – 8 (8)	0.82 (1.6) 0 – 6 (1)	21.02	<0.01	autism <LAD< NT	autism/LAD: 0.79 autism/NT: 1.76 LAD/NT: 0.72
No. ODD symptoms (Full diagnosis)	1.30 (1.9) 0 – 3 (1)	0.71 (1.2) 0 – 5 (1)	1.03 (1.5) 0 – 6 (3)	1.17	.31
No. CD Symptoms (Full diagnosis)	0.55 (1.39) 0 – 7 (2)	0.13 (0.34) 0 – 1 (0)	0.12 (0.41) 0 – 2 (0)	2.76	0.07

Note: Data shown as M(SD), range. Pragmatic Language composite was calculated as the average of z-scores across the six pragmatic assessments.

Regression models of pragmatic language, externalizing symptoms, and diagnostic group.

Poisson regression models were implemented to test the relationship between pragmatic language and externalizing symptoms. Results indicated that both poorer pragmatic language ability and diagnostic group (dummy-coded) were associated with total externalizing symptoms; see Table 3. Note that a regression with SCQ score as a measure of lifetime autism symptom severity yielded an identical pattern of results; see Supplemental Materials. Goodness of fit measures of pragmatic language as a predictor of total externalizing symptoms was calculated for the three diagnostic groups individually. These relationships are shown in Figure 2. Pragmatic language was significantly negatively correlated with externalizing symptoms in the autism, R = −0.43, F = 6.91, p = .013, and NT, R = −.39, F = 5.76, p = .02, groups, but not the LAD group, R = −.08, F = .18, p = .68. Scatter plots (see Figure 2) indicated two outliers in the dataset, based on either a lower pragmatic language z-score or more elevated K-SADS score compared to the rest of the population. Analyses were conducted without these outliers with no change in results. Subsequent analyses thus retained these participants.

Figure 2.

Pragmatic Language composite scores and Externalizing Symptoms (count)

Regression models including other predictors.

Poisson regression models were implemented to test the relationship between pragmatic language score and externalizing symptoms, accounting for NVIQ, structural language (CELF Core), and executive functioning (DKEFS-Tower). Results showed that both poorer pragmatic language and diagnostic group remained significantly and independently associated with externalizing symptoms; see Table 4. When age was added to the regression, only diagnostic group and age accounted for significant variability in total externalizing symptoms. Participants who were older, and those in the LAD or NT groups, had fewer total externalizing symptoms, after accounting for a host of relevant predictors. The interaction terms between each diagnostic group and pragmatic language were tested as well. Pairwise comparisons of the interaction terms suggested a different relationship for the LAD group between pragmatic language and externalizing symptoms, compared to the autism group, p = .034, and the NT group, p = .046, each individually. Results that included other predictors in the regression model for ADHD and ODD are presented in Supplemental Materials.

Effects of Age.

Given the significance of age in the regression models, post-hoc correlations assessed the impact of age on pragmatic language and externalizing disorder within groups; see Table 5 and Figure 3. For the autism group, age was correlated with pragmatic language, r = .56, and total externalizing symptoms r = −.63; for the LAD group, age was correlated with pragmatic language, r = .61, but not with total externalizing symptoms; for the NT group, age was uncorrelated with either domain. The relationships between age, pragmatic language, and externalizing symptoms for ADHD and ODD are presented in Supplemental Materials.

Discussion

This study examined the associations between pragmatic language and externalizing symptoms of Attentional Deficit-Hyperactive Disorder (ADHD), Oppositional Defiant Disorder (ODD), and Conduct Disorder (CD) in adolescents with a history of autism or of typical development. Among the two groups with a history of autism, the LAD group had no current symptoms of autism, permitting an analysis of whether loss of the diagnosis, which entails improvement of pragmatic language, would be associated with differential prevalence of externalizing behaviors. Pragmatic language was assessed in detail using a composite of standardized measures (Test of Language Competence Making Inferences & Figurative Language, Test of Written Language Story Construction, Contextual Language, & Contextual Conventions) and clinician observer ratings (CELF-IV Pragmatic Profile, Pragmatic Language Scale). Externalizing symptoms were assessed using a gold-standard research interview, the K-SADS.

Results indicated that the autism group had significantly more pragmatic language difficulties than participants in the LAD and NT groups, consistent with current diagnostic status; the latter two groups did not differ. Participants in the ASD group had cognitive abilities in the average range, but still struggled with pragmatic language. This group had lower scores on both verbal IQ and structural language measures; weaknesses in pragmatic language remained even when scores on the former measures were included as covariates. This supports the validity of the pragmatic language composite, comprised of scores from nine distinct measures; this detail is critical, given the challenges inherent in assessing a domain that depends heavily on context, rapport, and participant engagement (Simmons et al. 2014).

Challenges in pragmatic language, one aspect of social communication, are one of the primary diagnostic components of autism; by definition, individuals with no current diagnosis should show no or minimal pragmatic language weaknesses (Fein et al., 2013; Orinstein et al., 2015). The finding that LAD and NT groups did not differ in pragmatic language abilities lends further support to the claim that the LAD children no longer present with characteristics of autism. While work by our group suggests some subtle pragmatic language weaknesses in LAD in narrative structure (Canfield et al., 2016), idiosyncratic language (Suh et al., 2014), and semantic category structure (Naigles et al., 2013), the current finding suggests that youth in this group attained appropriate and functional social communication skills. The autism group also displayed more current externalizing symptoms overall, and more ADHD symptoms specifically, relative to participants in the LAD and NT groups. The LAD group had more ADHD symptoms than the NT group. This is consistent with previous findings showing that LAD individuals have increased psychiatric symptoms overall, compared to their peers without a history of autism (Orinstein et al., 2015). It is possible that those with a history of autism with a loss of diagnosis remain more susceptible to other psychiatric diagnoses later in adolescence and adulthood; data suggests that ADHD is a particularly common outcome (Fein et al., 2005). While the precise mechanism underlying LAD outcomes is still unknown, predictors include higher IQ, higher receptive language, earlier diagnosis, earlier interventions, and more intensive intervention (Fein et al., 2013; Helt et al., 2008; Orinstein, et al., 2014). However, many autistic children who have all these predictor factors have a lifelong struggle with social communication and restrictive and repetitive behaviors. Ongoing studies of LAD aim to elucidate the cognitive and neuropsychological underpinnings of this outcome and to describe the heterogeneity of long-term outcomes.

We evaluated the association between pragmatic language and externalizing symptoms, and whether that association was a simple outcome of covariation with other constructs that are also important for pragmatic language and externalizing symptoms – age, nonverbal cognitive ability, structural language ability, and executive functioning – as well as autism symptomatology and diagnostic group. Other than age, the significant relationship between pragmatic language and externalizing symptoms held over and above these variables, suggesting that pragmatic language abilities (or lack thereof) independently contribute to mitigate these difficult behaviors. This may be especially true for autistic adolescents, who struggle with pragmatic language skills.

However, when age was included in the regression model, pragmatic language skills were no longer associated with externalizing behaviors. Only age and diagnostic group significantly contributed to externalizing symptoms, such that younger children and those with autism had more externalizing symptoms. This was driven primarily by the autism group, which had a strong negative correlation, r = −.63, while neither the LAD nor NT groups had significant relationships between age and externalizing symptoms. Previous longitudinal studies have shown a strong relationship between age and externalizing symptoms in ADHD, ODD, and CD, but different developmental trajectories for each (Atherton et al., 2019; Kuja-Halkola et al., 2015). Across the ages of 10–17, rates of ADHD symptoms were highest at 10 and slowly decreased over time, while in ODD, rates increased until the age of 13 and then steadily declined (Atherton et al., 2019). The current study sample primarily fell within the range highlighted in Atherton (2019), with a mean age of 13–14 years, but had a wider age range (8–22 years). Due to this wide age range, and the cross-sectional nature of this study, it is not possible to fully investigate the developmental trajectory of the co-occurrences of externalizing disorders with individual diagnostic groups. However, the present findings are generally consistent with prior work showing that as children get older, their externalizing symptoms decrease. This result also suggests that complex social communication skills undergo continued development, particularly in children with autism and LAD, who develop these skills at a later age and primarily as a result of intervention.

Not only did age have a strong relationship with externalizing symptoms, but age was also significantly correlated with pragmatic language abilities in the autism and LAD but not NT groups. This is in line with previous research by Whyte and Nelson (2015), who found that verbally autistic children (ages 5–12) exhibited a slower and delayed progression of development for pragmatic language and nonliteral language compared to the NT children, but similar rates of development for vocabulary and syntax abilities. The lack of a relationship between age and pragmatic language in the NT group may be due to ceiling effects, where the entire sample of NT individuals plateaued in developing these skills; in contrast, both clinical groups appear to display developmental progression even at this stage of adolescence. This provides further motivation for continued intervention in pragmatic language skills for autistic individuals. Development of pragmatic language skills appears to benefit from peer socialization and conversation (Nippold, 2000). Adolescents with autism often have reduced socialization experiences in group settings or individually with friends (Orsmond et al., 2004, 2013; Shattuck et al., 2011). Reduced social opportunities can negatively impact the rate of development of pragmatic language in autism and may explain why the autism group displayed particularly strong relationships among age, pragmatic language, and externalizing symptoms.

When not including other variables in the regression model, there was a significant negative association between pragmatic language abilities and externalizing symptoms, across groups and within the autism group in particular; this relationship was largely absent in the LAD group. Pairwise tests of the interaction term indicated no strong association between pragmatic language and externalizing symptoms in the LAD group, in contrast to the autism and NT groups. This is a salient finding, suggesting that the ability to use language socially is strongly linked to difficult-to-manage aggression and other acting-out behaviors, over and above the impact of diagnostic group. Individuals with LAD received earlier and more behavioral intervention that directly or indirectly targeted social communication skills, such that pragmatic language weaknesses may have had a reduced influence on externalizing symptoms (Orinstein et al., 2014; Orinstein et al., 2015). This may indicate that improved pragmatic language functioning during adolescence is tied to fewer externalizing symptoms; when children can communicate more effectively, they may rely less on maladaptive strategies and behaviors to get their needs met. However, due to the cross-sectional nature of this study, the directionality of this relationship between pragmatic language and externalizing symptoms remains uncertain. It is possible that having fewer maladaptive behaviors leads to more opportunities to build pragmatic language skills. It is also possible that factors such as age, IQ, or executive functioning, play a moderating role in the LAD group specifically; intervention studies are needed on how pragmatic language remediation influences externalizing symptoms.

In addition to hypothesized links between externalizing symptoms, autism characteristics, and pragmatic language, analyses probed the impact of other important domains that could influence externalizing symptoms. Several predictors were significant in regression models (age, autism symptom severity), while others were not (IQ, structural language, and executive functions). Some work reports the importance of receptive and expressive structural language in behavioral problems in typical populations (Clegg et al., 2015; Morgan et al., 2015; Salmon et al., 2016) and in autism (Matson et al., 2009). When studies include both structural and pragmatic language measures, pragmatic language consistently explains more of the variance in externalizing symptoms (Hollo et al., 2019; Miranda et al., 2020; Petranovich et al., 2017; Rodas et al., 2017). Structural language is also typically highly correlated with pragmatic abilities in autism (Norbury, 2014). It is likely that the present study was underpowered to detect the role of structural language, given that participants’ skills were generally in the typical range across groups, making it more difficult to detect subtle effects. Similarly, nonverbal intellectual functioning did not account for unique variance in any of the regression models, despite previous work indicating that IQs in the borderline to mild ID range are related to higher externalizing symptoms and aggression (Baker et al., 2002; Dekker et al., 2002). Our participants had IQs in the normal range, making it difficult to draw firm conclusions from this finding.

Executive functioning difficulties are associated with more externalizing behaviors in autism (Cristofani et al., 2020; Maddox et al., 2018; Schuiringa et al., 2017; Vogan et al., 2018). A review by Friedman and Sterling (2019) highlights the importance of working memory in pragmatic language abilities. In the current study, clinician-assessed executive functioning was not associated with externalizing symptoms. Rather, these findings are consistent with a mechanism in which language weaknesses are a specific risk factor for externalizing behaviors, and less consistent with a general-risk genetic hypothesis, under which weaknesses in other factors like executive functioning also play a larger role. This is supported by the LAD group, which previously had pragmatic language weaknesses but no longer do, having significantly fewer externalizing symptoms than the autism group. This indicates that an improvement in pragmatic language skills may play a mechanistic role in the reduction of externalizing symptoms. Age did erase a significant portion of the pragmatic language variance, so a mechanistic conclusion of these results should be interpreted with caution. That said, age was not significantly correlated with externalizing symptoms in either the LAD or NT group, the two groups that had better pragmatic language ability and fewer externalizing symptoms. The role of age in the model may be a driving force in the autistic group, for which pragmatic language skills are actively developing through activities like social skills groups, and less so for the other groups. Longitudinal studies will be needed to fully explore the mechanistic relationship and interactions between pragmatic language, age, and externalizing symptoms.

Post-hoc exploratory analyses, shown in Supplemental Tables 6 and 7, assessed the relationship with single pragmatic language measures (TLC, TOWL, CELF-PP, and PLS), indicating that each was significantly associated with externalizing symptoms; after accounting for other predictors, only the TOWL subtests remained significant. These TOWL subtests focus on the conventions of written material and the quality of its composition (i.e., development of plot and characters and connecting to interests of the reader), requiring children to adjust their language based on the situation or person and specifically having another person’s perspective in mind when communicating language. In autism, the ability to connect to a reader’s perspective may play a particularly important role in externalizing symptoms. The TOWL also has very high test-rest and interrater reliability, such that this may be a particularly sensitive and stable measure of pragmatic language; however, the CELF pragmatic profile also had high reliability, but was not found to capture significant independent variance over and above other predictors. Clearly, more work is needed on the relationship between specific components of pragmatic language and externalizing symptoms.

The current study provides further support for the detailed assessment of pragmatic language skills in children with autism, even those with cognitive abilities in the average range and strong structural language skills. Such children will likely benefit from assessment and intervention that specifically targets pragmatic language skills. A recent meta-analysis of 21 studies described interventions targeting multiple pragmatic skills (Parsons et al., 2017). Across 21 studies reporting on 18 interventions, there was an overall positive intervention effect of moderate size, particularly for preverbal skills, turn-taking, nonverbal communication, and social emotional attunement. However, only a few studies have assessed the impact of pragmatic language interventions on externalizing symptoms. One study demonstrated that a classroom-based pragmatic language intervention led to improvements in four pragmatic language scales (describing information, expressing judgment, considering the listener, and understanding the listener) on the Test of Pragmatic Language (Hyter et al., 2001); this improvement was associated with reductions of inappropriate classroom behaviors. Similarly, a Social Stories intervention (n=4) to teach social and pragmatic knowledge led to improved pragmatic language and a reduction in problematic behaviors at post-intervention and one-month follow-up (O’Connor & Hayes, 2019). Although there are only a few studies, the current results align with other findings to date, indicating the highly promising impact of pragmatic language interventions on externalizing symptoms.

Limitations.

There are important limitations to consider in this study. Missing data were replaced with multiple imputation, which led to a more robust sample size; however, estimation is less reliable than complete data. However, the fact that our findings were similar when results included only participants with full data provides confidence that the present findings are reliable. Another limitation is the relatively low prevalence of disruptive behavior disorders in this study, which was originally designed to better understand LAD outcomes. This secondary analysis indicated that only 23 of 98 individuals (23%) met full criteria for a disruptive behavior disorder (ADHD, CD, ODD). The current analyses relied on analyses of individual symptoms (four possible symptoms for ADHD, three for ODD, and five for CD), collapsed into a summary. Fewer observations means reduced variability, decreasing our power to detect relationships; furthermore, findings were not informative about the nuanced differences among externalizing disorders. Thus, these findings should be replicated within a sample that has been clinically referred for disruptive behavior disorders, and in children with a wider range of cognitive abilities. Relatedly, the sample comprised a wide age range, spanning school-age to young adulthood; pragmatic language skills vary greatly over development. This variability is addressed to some extent by the use of age-appropriate normed, standardized, measures (e.g., the TLC and TOWL) in the pragmatic language composite. However, a more selective age range in future studies will likely yield more fine-grained conclusions. Finally, while there was a documented change in symptom presentation over time (i.e., in the LAD group), the current dataset is cross-sectional in nature and relies on parent recall of early externalizing symptoms; the results await replication in a true longitudinal study.

In the present study, pragmatic language ability was associated with externalizing symptoms of ADHD, ODD, and CD in a sample of adolescents with a history of autism or of typical development. This relationship remained when controlling for other factors that typically influence externalizing symptoms, including nonverbal cognition, structural language, executive functioning, and autism symptomatology, but did not remain when age was included in the model. These findings were driven primarily by the autism group, which had both greater pragmatic impairments and more disruptive behaviors, both of which were strongly correlated with age. Whereas this association between pragmatic language and externalizing symptoms was not observed in the LAD group, suggesting that amelioration of pragmatic difficulties is not necessarily associated with a concomitant decrease in externalizing symptoms across etiologies. Pragmatic language intervention may play a role in reducing externalizing symptoms in autism.

Highlights.

• Pragmatic language ability was associated with externalizing symptoms, especially ADHD

• Amelioration of pragmatic language was seen in a subset of individuals with an autism history

• Age was strongly correlated with both pragmatic language and externalizing symptoms

• Development of pragmatic language skills may facilitate a decrease in externalizing symptoms