NEW RESEARCH

Age 31 Mental Health Outcomes of Childhood

Language and Speech Disorders
Joseph H. Beitchman, MD, E.B. Brownlie, PhD, Lin Bao, BA

Objective: Language disorders are associated with emotional and behavioral problems in
childhood and adolescence. Although clinical studies with small samples suggest that psy-
chosocial difficulties continue into adulthood, adult mental health outcomes of childhood
language disorders are not well known. The objective of this prospective longitudinal study is
to determine whether the age 31 mental health outcomes of individuals who had childhood
language disorders differ from the outcomes of typically developing controls. Method: A 26-
year cohort study followed up children with language or speech disorders from age 5 to age 31.
The children were selected from a 1-in-3 random sample of 5-year-olds using a 3-stage
screening and assessment process. A control group matched by sex, age, and classroom or
school was also selected. Diagnoses were assigned with the Composite International Diagnostic
Interview with the additional criterion that Global Assessment of Functioning scores indicated
at least mild impairment. Dimensional psychosocial self-report measures were also adminis-
tered. Results: Rates of diagnosis at age 31 years were equivalent between participants who
had childhood language disorders and controls, with and without multiple imputation to
estimate missing outcomes. Differences in rates of affective and substance use disorders could
not be ruled out because of attrition in the cohort with language disorders, who were less
likely to participate at age 31. Psychosocial scores for both cohorts were in the normal range.
The cohort with language disorders had poorer self-rated physical health than con-
trols. Conclusion: Mild/moderate language disorders may not have significant long-term
mental health consequences in early adulthood. J. Am. Acad. Child Adolesc. Psychiatry,
2014;53(10):1102–1110. Key Words: language disorder, longitudinal, adult outcomes, speech
disorders, physical health

T
here is substantial overlap between child- Language disorder (LD) includes difficulties
hood language difficulties and emotional/ with expressive and/or receptive grammar, vo-
behavioral problems.1-4 Rates of co-occurring cabulary, or discourse; the DSM-5 diagnosis is
language disorders in children’s mental health language disorder.11 Speech disorders include
and youth forensic settings approach 50%.5,6 difficulties with articulation, speech sounds,
Conversely, children and adolescents receiving fluency, or voice; corresponding DSM-5 di-
speech/language services often have concomitant agnoses are speech sound disorder, childhood
emotional, behavioral, or social difficulties.4,7,8 onset fluency disorder, and voice disorder. This
Findings from prospective community studies study examines the age 31 mental health out-
confirm the increased risk for psychiatric di- comes of children with communication disorders
agnoses and psychosocial difficulties among youth in comparison with typically developing controls.
with language disorders.1-3,9,10 In contrast to the extensive literature on
emotional/behavioral correlates of LD in childhood
and adolescence, adult mental health outcomes are
This article is discussed in an editorial by Dr. Claudio O.
Toppelberg on page 1050.
largely unknown. Clinically, accurate information
on prognosis is important for youth with commu-
Clinical guidance is available at the end of this article. nication disorders and their families. Theoretically,
examining comorbidities across developmental
Supplemental material cited in this article is available online. contexts can inform our understandings of how
communication relates to psychosocial functioning.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1102 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

A few studies have reported mental health LD at age 5 years showed higher rates of psychi-
difficulties, psychosocial impairment, social dif- atric disorders at ages 5, 12, and 191,3,9; poorer ac-
ficulties, and reduced social participation in ademics; elevated delinquency symptoms; and
adulthood, consistent with child and adolescent difficulties with social and adaptive func-
findings.12-19 However, these studies have meth- tioning.28,32,33 In contrast, individuals with speech
odological problems including small sample disorders without LD resembled controls by late
sizes, retrospective designs, and measurement childhood.3,9,28,34 Accordingly, participants with
limitations. One prospective community study LD with or without co-occurring speech disorders
reported poorer age 34 mental health outcomes of (cohort with LD) were analyzed separately from
childhood LD. However, mental health was participants with speech disorders only (SD-only
defined by constructs including perceived control cohort). We hypothesized that the cohort with LD
and life satisfaction, limiting comparability with would have poorer mental health outcomes than
related research.16 Because youth with complex controls at age 31 years. We compared rates of
problems may be more likely to be referred for psychiatric disorders and scores on dimensional
treatment, studies using referred samples may measures, which may reflect subtler differences in
overestimate the co-occurrence of mental health psychosocial outcomes.
and language disorders. Prospective longitudinal
studies following nonreferred community sam-
ples are essential to avoid referral bias. METHOD
Given the high psychiatric comorbidity of LD in Participants
childhood and adolescence, continued psychosocial Wave 1 Sample. In 1982, a 1-in-3 sample of English-
vulnerability might be expected. First, LD persists speaking kindergarten children in the Ottawa–Carlton
into adulthood. Three-fourths of late adolescents region of Canada (n ¼ 1,655) was screened for language
with childhood LD continued to meet criteria in late and speech disorders. The 301 children who failed the
adolescence and to lag in vocabulary growth.20,21 screening were assessed by speech–language patholo-
gists, and 180 were diagnosed with a communication
Second, correlates and sequelae of LD may main-
disorder; parental consent for psychosocial assessment
tain vulnerability. Poor academic achievement and was obtained for 142. A control group, matched on age,
continuing communication difficulties limit educa- sex, and school (n ¼ 142) was selected from those who
tional participation and decrease vocational op- passed the initial screening (further methodological in-
portunities. Attained education is associated with formation is available elsewhere).35 The 284 children
employment, income, and physical and mental who constituted the wave 1 sample were reassessed at
health; low socioeconomic status (SES) and associ- ages 12, 19, 25, and 31 years; retention rates were 86%,
ated disadvantage may convey additional risk for 91%, 85%, and 80%, respectively. The sample is not
adverse outcomes.22,23 Poorer linguistic environ- ethnically diverse due to local demographics in the ini-
ments resulting from limited parental resources tial wave; at wave 5, 91.5% identified their ethnicity as
and/or education, including decreased quantity Caucasian, European, or Canadian.
Ethical approval was obtained from the research
and quality of child-directed speech, may con-
ethics boards of the Royal Ottawa Hospital (wave 1),
tribute to the development of LD.24-26 In addition,
Clarke Institute of Psychiatry (waves 2 and 3), Hospital
social difficulties associated with LD may have a for Sick Children (wave 4), and Centre for Addiction
negative impact on mental health.7,8 Third, factors and Mental Health (wave 5).
predicting resilience after adolescence such as Definitions of Language and Speech Disorders. LD was
academic achievement, financial resources, and defined as any 1 of the following: 1 SD below the mean
self-perceived competence are likely to be less on the Test of Language Development (TOLD)36
developed in individuals with LD because of Spoken Language Quotient (SLQ) or Peabody Picture
ongoing academic difficulties and other areas of Vocabulary Test (PPVT)37; 2 SD below the mean on a
disadvantage.27-29 Adult mental health outcomes of language subtest of the TOLD; or 2 SD below the mean
speech disorders are also largely unknown. Only 1 on both Content and Sequence subtests of the
Goldman-Fristoe-Woodcock Test of Auditory Memory
longitudinal study has reported psychosocial out-
(GFW).38 Of the 103 participants who met criteria for
comes in a prospectively identified sample with
LD, 45 met 1 criterion only: PPVT (13), TOLD SLQ (24),
childhood speech disorders; however, mental TOLD subtest (1), GFW (7); 58 met 2 or more criteria.
health outcomes were not reported.30,31 Speech disorders were defined as 2 SD below the mean
This article reports on age 31 mental health on the TOLD Word Articulation or Word Discrimina-
outcomes of the Ottawa Language Study, a 5-wave tion subscales or clinical diagnosis of voice disorder,
prospective longitudinal study. Participants with stuttering, or dysarthria. Exclusionary criteria were not

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1103
BEITCHMAN et al.

used. One participant’s nonverbal IQ score was <70, (dependence or abuse; 9 substances and not otherwise
and 4 were missing nonverbal IQ scores due to low specified [NOS]); and schizophrenia. Posttraumatic stress
functioning; none participated at wave 5. The cohort disorder was assessed using DSM IV criteria.42 Antisocial
with LD consisted of 62 who met criteria for LD only personality disorder was omitted because of lack of
and 41 who met criteria for LD and a speech disorder. recency data. The CIDI has good interrater reliability,
The SD-only cohort consisted of 39 children who met test–retest reliability, and validity for most diagnoses.39,43
criteria for a speech disorder but not LD. Global Assessment of Functioning. Trained in-
Wave 5 sample. In 2008 to 2010, a total of 226 (79.6%) terviewers completed the Global Assessment of Func-
of the original sample (N ¼ 284) participated at wave 5 tioning (GAF) after participants completed the protocol
(Table 1). Attrition was greater in the wave 1 cohort with including the UM-CIDI. To ensure that only in-
LD (34.0%) than with controls ([13.4%], c2 [1, n ¼ 245] ¼ dividuals at least mildly affected by their symptoms
14.74, p < .001), because of differences in proportion were classified with psychiatric disorders, diagnostic
located (c2 [1, n ¼ 245] ¼ 15.44, p < .001). At wave 1, criteria included a GAF score of less than 70 (i.e., at
follow-up participants had higher family SES (t[272] ¼ least mild functional impairment).44
0.294, p ¼ .016), and were less likely to live in single-parent Brief Symptom Inventory. The Brief Symptom In-
households than nonparticipants (c2 [1, N ¼ 284] ¼ 10.38, ventory (BSI) is a 53-item short form of the SCL-90
p ¼ .001). Wave 5 participants did not differ from non- general psychopathology measure.45 Two-week test–
participants in ratings of behavior problems at age 5 or 12 retest reliability for the Global Severity Index is 0.90.
years, or rates of psychiatric diagnoses or arrests at age 19 Convergent validity has been demonstrated with other
years. Within the cohort with LD, there were no differ- psychopathology measures.45
ences in severity of LD by wave 5 participation. Center for Epidemiologic Studies Depression Scale. The
Wave 5 participants completed a 4- to 5-hour Centre for Epidemiological Studies Depression Scale
assessment; 42 participants (7 SD-only, 9 LD, and 26 (CES-D) is a 20-item self-administered depression
controls) were interviewed by telephone and returned symptom scale.46 Measures of internal consistency
questionnaires by mail. Interviewers had no access to (split half and Cronbach’s a) in the standardization
participants’ cohort. sample were approximately 0.85. Test–retest correla-
tions (2–8 weeks) range from 0.51 to 0.67.46
Short Form–12 Health Survey. The Short Form–12
Measures Health Survey version 2 (SF-12v2) contains 12 items
Composite International Diagnostic Interview. The Uni- that assess perceived physical and mental health and
versity of Michigan Composite International Diagnostic health-related role functioning and limitations.47
Interview (UM-CIDI), a highly structured diagnostic in- Physical component scores differentiate individuals
strument, was administered by trained interviewers.39,40 with physical conditions from healthy controls and
DSM III-R diagnoses were used to allow comparison according to severity; reported short-term (1–2 weeks)
with previous waves.41 Current (12-month) diagnoses test–retest correlations range from 0.75 to 0.84 for the
assessed included affective disorders (major depression, SF-12; estimated reliability for the SF-12v2 physical
dysthymia, bipolar disorder); anxiety disorders (social component score is 0.89.47
phobia, simple phobia, panic disorder, agoraphobia, Adult Self-Report Antisocial Personality Problems. The
generalized anxiety disorder); substance disorder Adult Self-Report measures behavior problems and

TABLE 1 Participation and Attrition by Cohort and Age 19 Psychiatric Status

Control Language Disorder Speech Disorder
(n ¼ 142) (n ¼ 103) Only (n ¼ 39)

Characteristic n % n % n %
Age 31 Participation
Known deceased 2 1.4 5 4.9 0 0.0
Not located (excluding known deceased) 5 3.5 17 16.5 0 0.0
Located 135 95.1 81 78.6 39 100.0
Declined / unable 12 8.5 13 12.6 4 10.3
Participated 123 86.6 68 66.0 35 89.7
Age 31 Participation by Age 19 Psychiatric Diagnosis
Age 19 psychiatric disorder 19/24 79.2 22/28 78.6 7/7 100.0
No age 19 psychiatric disorder 95/105 90.5 39/49 79.6 28/31 90.3
Age 31 Participation by Age 19 Arrest Status
Arrested by age 19 14/18 77.8 17/20 85.0 9/10 90.0
Not arrested 104/115 90.4 45/64 70.3 26/29 89.7

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1104 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

psychosocial functioning for adults aged 18 to 59 mechanisms.53 The full imputation model included
years.48 The DSM-oriented Antisocial Personality variables in the analysis model, auxiliary variables
Problems scale contains 20 items that were rated by a correlated with the analysis variables, and auxiliary
panel of psychiatrists as “very characteristic” of the variables correlated with attrition. The number of vari-
DSM IV antisocial personality disorder diagnosis. The ables in each imputation model was limited to 25 or one-
scale differentiates clinically referred from unreferred third of the sample size of the imputed variable.54,55 The
samples and is correlated with other measures of psy- variables used in the imputation models are shown in
chopathology. One week test–retest correlation is 0.84.48 Tables S1 to S8 (available online). Results are also shown
Demographics. At wave 5, participants completed a for analyses restricted to participants with non-missing
brief demographics questionnaire. Ethnicity was dependent variables (imputing missing age 5 SES only).
defined using Statistics Canada categories. Low income Multiple imputation procedures were conducted
was defined using Statistics Canada 2007 household under the missing at random (MAR) assumption, i.e.,
pretax low income cut-offs.49 At wave 1, demographic that the probability of missing data is unrelated to the
data were collected in semi-structured interviews. Age missing values, over and above variance explained by
5 SES was coded using Blishen occupational coding, the observed data. Longitudinal studies have shown
which takes into account education, income, and evidence for selective drop-out with negative bias (i.e.,
prestige associated with occupations.50 attrition associated with poorer outcomes).56,57
Missing data mechanisms with positive bias are also
Data Analysis possible; either can result in biased conclusions if not
Logistic regression models were used to compare rates fully accounted for by the variables in the imputation
of psychiatric disorders in the LD and control cohorts. model.
Sample size was insufficient for statistical comparisons Accordingly, sensitivity analyses were conducted to
using the SD-only cohort; rates of diagnoses are shown. examine the robustness of the results (i.e., relation be-
For logistic regression with full LD and control samples tween LD and diagnosis), should the assumption of
of 142 controls and 103 with LD, power is 79% to detect B MAR not be met. Specifically, the multiply imputed
¼ 1 (moderate effect size). Wave 1 SES was higher among datasets were used to estimate the values of the
the controls than the cohort with LD,29 consistent with dependent variable if data were not missing at random.
other research showing an association between LD Positive and negative biases were considered in sepa-
and social disadvantage.24-26 Accordingly, analyses are rate analyses. Parameter estimates from the 500
shown with and without wave 1 SES as a covariate. imputed datasets generated under the MAR assump-
General linear models were used for dimensional tion were re-weighted according to their extent of de-
outcomes. Missing values were estimated by multiple viation from MAR. The weights were a function of the
imputation by chained equations using the Stata (release odds of missing data given changes in imputed vari-
12) mi impute chained command.51 A total of 500 imputed ables (controlled for imputation model variables). This
datasets were created for each analysis.52 Auxiliary method imputes outcomes based on observed data,
variables were selected from mental health, language, using datasets consistent with bias, thereby allowing
cognitive, and family demographic domains. Data were assessment of the robustness under deviation from
imputed separately by cohort to avoid confounding MAR. To prevent disproportionate influence of indi-
because of potential differences in missing data vidual datasets, no single imputed dataset was

TABLE 2 Demographic Characteristics at Age 31 Follow-Up by Cohort

Language
Control Disorderb Speech Disorder Language Disorder
(n ¼ 123)a (n ¼ 68)a Only (n ¼ 35)a vs. Control

Characteristic n % n % n % OR 99% CI
Male 75 61.0 39 57.4 26 74.3 0.86 0.39, 1.90
Low income 7 5.7 18 26.9 3 8.6 6.03*** 1.77, 20.62
Marriedc 84 68.9 35 53.0 22 62.9 0.51 0.23, 1.15
Parent 44 36.4 35 53.0 11 32.4 1.98 0.89, 4.40
Note: OR ¼ odds ratio.
a
Sample size, control: low income, married (122); parent (121); language disorder: low income (67), married, parent (66); speech disorder only:
parent (34).
b
Language disorder includes comorbid language and speech disorders.
c
Includes marriage and common law/cohabitation.
***p < .001; after Bonferroni correction for multiple comparisons (corrected testwise a: ***p < .0003).

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1105
BEITCHMAN et al.

weighted more than 50%, and at least 5 of the 500 With substantial deviation from MAR with
weights had substantial weights (>1/500).58,59 negative bias (i.e., disorder associated with lower
participation, controlled for imputation model
variables), significantly higher rates of affective
RESULTS disorder and SUD in the cohort with LD could
Table 2 shows demographic characteristics of the not be ruled out. However, the substantial
LD, SD-only, and control cohorts. Participants in negative bias that would yield significant dif-
the cohort with LD were more likely to have low ferences corresponded to high prevalence in the
income than were controls. cohort with LD for affective disorder (33.6%)
Table 3 presents rates of any psychiatric diag- and SUD (33.5%). Given the low rates of diag-
nosis and of anxiety, affective, and substance use nosis in the observed data, these estimates
(SUD) disorders for LD, SD-only, and control correspond to high rates of affective disorder
cohorts. No participants met criteria for schizo- (74.6%) and SUD (79.4%) among the missing
phrenia. Logistic regression models showed no dif- cohort with LD participants. Multiple imputa-
ferences in rates of diagnoses between the LD and tion was not used in the SD-only cohort because
control cohorts, with and without wave 1 SES of small sample size. However, if all 4 of the
controlled. Results were consistent including par- missing participants in the SD-only cohort had a
ticipants with complete data only, using multiple diagnosis, rates would equal rates for control
imputation for participants with non-missing diag- participants (15%).
nostic outcome data (imputing missing SES), and Table 5 shows group mean scores on dimen-
for all participants (imputing outcome and/or SES). sional measures of psychosocial outcomes. Means
Sensitivity analyses were conducted to assess for all cohorts were in the normal range. How-
whether the logistic regression coefficients ever, the cohort with LD reported poorer physical
derived using multiple imputation of psychiatric well-being and had higher depression scores
diagnoses were sensitive to deviations from the than controls. Using all participants (imputing
MAR assumption. Table 4 shows odds ratios of missing outcomes), we calculated that BSI psy-
psychiatric disorders by cohort and correspond- chopathology symptom scores were also higher
ing age 31 prevalence estimates, assuming sub- than for control participants. Controlled for wave
stantial deviation from the MAR assumption. 1 SES, all differences became nonsignificant
Results were consistent, with no differences except physical well-being. There were no gender
between the LD and control cohorts for any effects or interactions in the mental health
psychiatric disorder and for anxiety disorder. outcomes.

TABLE 3 Age 31 Psychiatric Diagnoses by Cohort

Language Disorder
Language Language Disorder vs. Control:
Control Disorder vs. Control SES-Controlled
Age 31
Diagnosis n % n % OR 99% CI OR 99% CI
a
All participants
Any disorder 20 19.5 22 15.3 1.33 0.48, 3.71 0.92 0.31, 2.77
Anxiety disorder 20 19.5 17 11.7 1.84 0.60, 5.62 1.34 0.41, 4.33
Affective disorder 14 13.1 13 9.5 1.37 0.29, 6.55 1.03 0.20, 5.42
SUD 9 8.5 14 9.6 0.66 0.05, 8.23 0.45 0.03, 6.66
Participants with outcome datab
Any disorder 9 13.8 18 15.1 0.90 0.29, 2.81 0.55 0.16, 1.87
Anxiety disorder 9 13.8 13 10.9 1.31 0.40, 4.33 0.86 0.24, 3.05
Affective disorder 4 6.2 10 8.3 0.72 0.15, 3.50 0.49 0.09, 2.57
SUD 2 3.1 9 7.5 0.39 0.05, 3.05 0.23 0.03, 1.93
Note: OR ¼ odds ratio; SES ¼ socioeconomic status; SUD ¼ substance use disorder.
a
Missing data estimated using multiple imputation. All participants; sample size, control: (142), language disorder: (103); missing outcome and/or
age 5 SES estimated using multiple imputation.
b
Sample size, control: any diagnosis and anxiety (119), affective and substance use (120); language disorder: (65). Age 5 SES was estimated using
multiple imputation for 4 (3 with language disorder, 1 control).

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1106 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

TABLE 4 Sensitivity Analyses: Estimated Rates of Diagnosis Assuming Data Missing Not at Random
Prevalence of Diagnosisa
(Hypothetical) Language Disorder vs. Control
b c
Age 31 Diagnosis Bias Direction OR: Dropout Control (%) Language Disorder (%) OR: Disorderd 99% CI
Any Axis-I disorder Positive 0.305 15.29 12.46 0.55 0.18, 1.69
Negative 2.562 15.69 22.73 0.71 0.16, 3.12
Anxiety disorder Positive 0.004 10.57 12.62 0.90 0.31, 2.57
Negative 2.365 22.98 21.97 0.69 0.28, 1.71
Affective disorder Positive 0.428 7.64 6.37 0.60 0.14, 2.69
Negative 1.380 11.90 33.59 2.98* 1.00, 8.87
SUD Positive 0.087 7.19 2.30 0.20 0.03, 1.44
Negative 1.303 10.66 33.53 3.32* 1.00, 11.04
Note: OR ¼ odds ratio; SUD ¼ substance use disorder.
a
Hypothetical prevalence in sensitivity analysis models assuming data missing not at random.
b
Psychiatric diagnosis modeled as positively vs. negatively associated with age 31 participation.
c
Odds ratio of missing age 31 diagnostic data when disorder is present vs. absent in sensitivity analysis models.
d
Hypothetical odds ratio of any psychiatric disorder by cohort (age 5 socioeconomic status controlled) in sensitivity analysis models.
*p < .05; after Bonferroni correction for multiple comparisons (corrected test-wise a: *p < .0125).

DISCUSSION with the control cohort, whose rates remained

To our knowledge, this is the first prospective stable and relatively low. What are potential
study of children with LD from a community reasons for the apparent decrease in rates of
sample followed prospectively to adulthood to diagnosis among the cohort with LD? At age 19
show similar rates of psychiatric disorders years, three-fourths of participants in the cohort
compared to controls. These findings are espe- with LD still met criteria for LD.20 At age 25
cially of interest, as significant differences in rates years, the cohort with LD continued to lag behind
of psychiatric disorders were found at previous controls in vocabulary scores,21 which were
waves,1,3,9 and as these findings stand in stark correlated 0.87 with a composite language mea-
contrast to the body of research literature doc- sure at age 19. Consequently, it is unlikely that
umenting elevated psychiatric risk among chil- the similarity in rates of disorder at age 31 was
dren and adolescents with LD.1-10 due to converging language abilities in the LD
Could these findings be an artifact of our and control cohorts. The contrast between these
methods? Selective attrition with decreased findings and existing research on adult outcomes
participation among individuals with mental of LD from clinical settings12-19 may be partly due
health problems could obscure differences be- to differences in severity; these results represent
tween the LD and control cohorts, particularly as outcomes of primarily mild/moderate LD.20
individuals in the cohort with LD were less likely The changing results from childhood and
to be located and to participate at age 31 years. adolescence to adulthood suggest varying impact
However, results were consistent using multiple of communication in different developmental
imputation and dimensional mental health mea- periods and contexts. The impact of LD may be
sures. Furthermore, although sensitivity analyses more severe in early developmental periods,
did not rule out cohort differences in rates of SUDs when language plays a key role in emerging
or affective disorders, these differences would behavior and emotional regulation skills. Matu-
correspond to rates of disorder among non- ration, continued language development, and the
participants of approximately 75%. Missing out- development of compensatory strategies may
comes are unknown; however, our data do not mitigate the impact of LD with age. Importantly,
suggest that this was the case. At age 19 years, half participants exited compulsory education in late
of the cohort with LD not participating at age 31 adolescence. For many youth with LD, school
had diagnostic data, and 22% had a diagnosis. involved academic difficulties, daily challenges to
From age 19 to 31, rates of psychiatric disorder their language skills, and social difficulties that
decreased among the cohort with LD to converge may have decreased confidence, increased stress,

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1107
BEITCHMAN et al.

TABLE 5 Age 31 Psychosocial Outcomes by Cohort

Language Disorder
Language Language Disorder vs. Control:
Control Disorder vs. Controla SES-Controlleda

Outcomes Mean SD n Mean SD n B SE B SE

b
All Participants
BSI Global Severity Index 57.77 14.60 118 53.38 11.91 64 4.39* 1.73 3.50 1.80
CES-D total score 12.89 13.78 119 8.27 9.27 64 0.73** 0.23 0.59 0.24
SF-12 Physical health 50.42 11.77 118 54.51 6.41 64 27,178.67** 8,423.64 26,142.45* 8,812.10
ASR Antisocial T score 53.43 9.09 114 53.32 6.32 61 0.04 0.02 0.03 0.02
Participants With
Outcome Datac
BSI Global Severity Index 58.11 12.28 142 53.33 11.31 103 4.78 1.81 3.96 1.89
CES-D total score 12.23 10.62 142 8.14 8.24 103 0.67* 0.23 0.54 0.24
SF-12 Physical health 50.73 9.23 142 54.70 5.61 103 26,497.36** 7,929.80 25,711.62* 8,351.28
ASR Antisocial T score 54.59 6.40 142 53.03 5.32 103 0.03 0.02 0.02 0.02
Note: ASR ¼ Adult Self Report; BSI ¼ Brief Symptom Inventory; CES-D ¼ Centre for Epidemiological StudieseDepression; SE ¼ standard error; SES ¼
socioeconomic status; SF-12 ¼ Short forme12. Model statistics were obtained using transformed data for CES-D, SF-12, and ASR. Means and SDs are
shown in original units.
a
Scores were transformed (CES-D: square root; SF-12: cube; ASR: log) to correct non-normality.
b
Missing data estimated using multiple imputation. Missing psychiatric outcome and/or age 5 SES estimated using multiple imputation.
c
Age 5 SES estimated using multiple imputation for 4 (3 with language disorder, 1 control).
*p < .05; **p < .01; after Bonferroni correction for multiple comparisons (corrected test-wise a: *p < .0125; **p < .0025).

and possibly contributed to the development of The disproportionately high dropout rate
psychiatric symptoms.28,33 Youth attending spe- among the cohort with LD compared to the
cial education classes may have experienced control participants increased the uncertainty of
additional stigma. The end of compulsory edu- the results and conclusions, which could be only
cation, with accompanying changes in vocational partially addressed by imputation and sensitivity
and social contexts, likely increased opportunities analyses. For the SD-only cohort, the sample size
for individuals with LD to pursue success on was not large enough to permit adequate test of
their own terms and to capitalize on personal the null hypothesis. For the cohort with LD, po-
strengths, given conducive social conditions. wer was sufficient to detect a moderate effect;
Attention to the educational experiences of in- smaller effects may have been undetected. Full
dividuals with LD could help in prevention. speech/language assessments would have pro-
Although the mean psychosocial scores for all vided more information on the co-occurrence of
cohorts were in the normal range, the cohort with communication disorders and mental health in
LD reported lower levels of perceived physical adulthood. The impact that treatment might have
health compared to controls. They also reported had on psychiatric outcomes is also unknown. In
lower income levels consistent with cohort dif- addition, these outcomes are situated in specific
ferences in attained education.34 In addition, they social contexts and depend on social and eco-
had somewhat higher depression scores; results nomic factors, including employment patterns
were mixed for BSI psychopathology symptoms. and the availability of social safety nets; outcomes
All differences except physical health were may be less positive in contexts with fewer sup-
nonsignificant with SES controlled. Given pre- ports in comparison to this Canadian sample.
existing cohort differences in SES, these results Our sample was not ethnically diverse, limiting
may reflect social disadvantage more than lan- generalizability; outcome studies are also needed
guage difficulties. Although the decreased rate of on dual language and language-minority chil-
psychiatric diagnoses in young adulthood among dren. These findings should be confirmed in
the cohort with LD is encouraging, these subtle, future studies including full communication
nonclinical differences could indicate increased assessment; however, they present a hopeful
risk for future mental health disorders or de- possibility for youth with mild to moder-
creased well-being. ate LD. &

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1108 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

Accepted July 24, 2014.

Clinical Guidance
Drs. Beitchman and Brownlie and Ms. Bao are with the Centre for
Addiction and Mental Health, Toronto, ON, Canada. Drs. Beitchman
 Clinicians need to be aware that, in contrast to the and Brownlie are also with University of Toronto.
increased risk for significant difficulties in childhood This research was funded by the Canadian Institutes of Health Research
and adolescence, the young adult mental health (MOP 84421).
outcomes of youth with mild and moderate Portions of this paper were presented at the American Academy of
language disorders may be comparable to those of Child and Adolescent Psychiatry 60th Annual Meeting, October
youth with typical language development. 22nd-27th 2014, Orlando, FL.
Dr. Brownlie served as the statistical expert for this research.
 Language disorder appears to have its greatest
impact on psychiatric symptoms in the context of the The authors thank Brenda Cavanagh, BA, of the Centre for Addiction
and Mental Health, for additional support in data collection and
youths’ school experience. Clinicians need to interviewing. The authors also thank the study participants and families
carefully evaluate the extent to which psychiatric for making this research possible.
symptoms are secondary to or exacerbated by Disclosures: Dr. Beitchman has received research grant funding from
youths’ experiences at school. the Canadian Institutes for Health Research, the Ontario Ministry of
Health and Long-Term Care, and the Howitt/Dunbar Foundation.
 Greater focus on developing alternatives to Dr. Brownlie has received research grant funding from the Canadian
traditional educational models for youths with Institutes for Health Research. Ms. Bao reports no biomedical financial
language disorders and/or adjusting the school interests or potential conflicts of interest.
environment may help to mitigate the development Correspondence to Dr. Joseph Beitchman, Child, Youth and Family
and severity of comorbid psychiatric symptoms. Services, Centre for Addiction and Mental Health, 80 Workman
Way, Toronto, ON, Canada, M6J 1H4; e-mail: joe.beitchman@
 Current evidence suggests that improvement in camh.ca
psychiatric symptoms in young adulthood may not 0890-8567/$36.00/ª2014 American Academy of Child and
require substantial improvement in language Adolescent Psychiatry
competence. http://dx.doi.org/10.1016/j.jaac.2014.07.006

REFERENCES
1. Beitchman JH, Nair R, Clegg M, Ferguson B, Patel PG. Prevalence 12. Arkkila E, R€as€anen P, Roine RP, Vilkman E. Specific language
of psychiatric disorders in children with speech and language impairment in childhood is associated with impaired mental and so-
disorders. J Am Acad Child Psychiatry. 1986;25:528-535. cial well-being in adulthood. Logop Phoniatr Voco. 2008;33:179-189.
2. Yew SG, O’Kearney R. Emotional and behavioural outcomes later 13. Clegg J, Hollis C, Mawhood L, Rutter M. Developmental language
in childhood and adolescence for children with specific language disorders—a follow-up in later adult life: cognitive, language and
impairments: meta-analyses of controlled prospective studies. psychosocial outcomes. J Child Psychol Psychiatry. 2005;46:
J Child Psychol Psychiatry. 2013;54:516-524. 128-149.
3. Beitchman JH, Brownlie EB, Wilson B, et al. Seven year follow-up 14. Howlin P, Mawhood L, Rutter M. Autism and developmental
of speech/language impaired and control children: psychiatric receptive language disorder—a follow-up comparison in early
outcomes. J Child Psychol Psychiatry. 1996;37:961-970. adult life. II: Social, behavioural, and psychiatric outcomes. J Child
4. Conti-Ramsden G, Botting N. Emotional health in adolescents Psychol Psychiatry. 2000;41:561-578.
with and without a history of specific language impairment (SLI). 15. King RR, Jones C, Lasky E. A fifteen-year follow-up report of
J Child Psychol Psychiatry. 2008;49:516-525. speech-language-disordered children. Lang Speech Hear Ser. 1982;
5. Cohen NJ, Menna R, Vallance DD, Barwick MA, Im N, 13:24-32.
Horodezky NB. Language, social cognitive processing, and 16. Law J, Parsons S, Rush R, Schoon I. Modeling developmental
behavioral characteristics of psychiatrically disturbed children language difficulties from school entry into adulthood: literacy,
with previously identified and unsuspected language impair- mental health, and employment outcomes. J Speech Lang Hear R.
ments. J Child Psychol Psychiatry. 1998;39:853-864. 2009;52:1401-1416.
6. Giddan JJ, Milling L, Campbell NB. Unrecognized language and 17. Mawhood L, Howlin P, Rutter M. Autism and developmental
speech deficits in preadolescent psychiatric inpatients. Am J receptive language disorder—a comparative follow-up in early
Orthopsychiatry. 1996;66:85-92. adult life. I: Cognitive and language outcomes. J Child Psychol
7. Botting N, Conti-Ramsden G. The role of language, social cogni- Psychiatry. 2000;41:547-559.
tion, and social skill in the functional social outcomes of young 18. Records NL, Tomblin JB, Freese PR. The quality of life of young
adolescents with and without a history of SLI. Br J Dev Psychol. adults with histories of specific language impairment. Am J
2008;26:281-300. Speech-Lang Pat. 1992;1:44-53.
8. Durkin K, Conti-Ramsden G. Young people with specific language 19. Whitehouse AJ, Watt HJ, Line EA, Bishop DV. Adult psychosocial
impairment: a review of social and emotional functioning in outcomes of children with specific language impairment, prag-
adolescence. Child Lang Teach Ther. 2010;26:105-121. matic language impairment and autism. Int J Lang Comm Dis.
9. Beitchman JH, Wilson B, Johnson CJ, et al. Fourteen-year follow-up 2009;44:511-528.
of speech/language-impaired and control children: psychiatric 20. Johnson CJ, Beitchman JH, Young A, et al. Fourteen-year follow-up
outcome. J Am Acad Child Psychiatry. 2001;40:75-82. of children with and without speech/language impairments:
10. Silva PA, Williams S, McGee R. A longitudinal study of children speech/language stability and outcomes. J Speech Lang Hear R.
with developmental language delay at age three: later intelligence, 1999;42:744-760.
reading and behaviour problems. Dev Med Child Neurol. 1987;29: 21. Beitchman JH, Jiang H, Koyama E, et al. Models and determinants
630-640. of vocabulary growth from kindergarten to adulthood. J Child
11. American Psychiatric Association. Diagnostic and Statistical Psychol Psychiatry. 2008;49:626-634.
Manual of Mental Disorders (5th ed.). Arlington, VA: American 22. Kurtze N, Eikemo TA, Kamphuis CB. Education inequalities in
Psychiatric Publishing; 2013. general and mental health: differential contribution of physical

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1109
 BEITCHMAN et al.

activity, smoking, alcohol consumption and diet. Eur J Public 41. American Psychiatric Association. Diagnostic and Statistical
Health. 2013;23:223-229. Manual of Mental Disorders, 3rd rev ed. Washington, DC:
23. Gilman SE. Childhood socioeconomic status, life course pathways American Psychiatric Association; 1987.
and adult mental health. Int J Epidemiol. 2002;31:403-404. 42. American Psychiatric Association. Diagnostic and Statistical
24. Ginsborg J. The effects of socio-economic status on children’s manual of Mental Disorders. 4th ed. Washington, DC: American
language acquisition and use. In: Clegg J, Ginsborg J, eds. Lan- Psychiatric Association; 1994.
guage and Social Disadvantage: Theory into Practice. West Sussex, 43. Wittchen HU. Reliability and validity studies of the WHO-
England: John Wiley and Sons; 2006:9-27. Composite International Diagnostic Interview (CIDI): a critical
25. Hoff E. How social contexts support and shape language devel- review. J Psychiatr Res. 1994;28:57-84.
opment. Dev Rev. 2006;26:55-88. 44. Bird HR, Canino G, Rubio-Stipec M, et al. Estimates of the prev-
26. Hart B, Risley T. Meaningful Differences in the Everyday Experi- alence of childhood maladjustment in a community survey in
ence of Young American Children. Baltimore, MD: Paul H. Puerto Rico: the use of combined measures. Arch Gen Psychiatry.
Brookes; 1995. 1988;45:1120-1126.
27. Masten AS, Burt KB, Roisman GI, Obradovic J, Long JD, 45. Derogatis LR. Brief Symptom Inventory: Administration, Scoring,
Tellegen A. Resources and resilience in the transition to adulthood: and Procedures Manual. Minneapolis, MN: National Computer
Continuity and change. Dev Psychopathol. 2004;16:1071-1094. Systems; 1993.
28. Young AR, Beitchman JH, Johnson C, et al. Young adult academic 46. Radloff LS. The CES-D scale: a self-report depression scale for re-
outcomes in a longitudinal sample of early identified language search in the general population. Appl Psych Meas. 1977;1:385-401.
impaired and control children. J Child Psychol Psychiatry. 2002;43: 47. Ware JE, Kosinsky M, Turner-Bowker DM, Gandeck B. User’s
635-645. Manual for the SF-12v2 Health Survey. Lincoln, RI: QualityMetric
29. Beitchman JH, Petersen M, Clegg M. Speech and language Incorporated; 2007.
impairment and psychiatric disorder: the relevance of family 48. Achenbach TM, Rescorla LA. Manual for the ASEBA Adult Forms
demographic variables. Child Psychiatry Hum D. 1988;18: and Profiles. Burlington, VT: University of Vermont, Research
191-207. Center for Children, Youth, and Families; 2003.
30. Felsenfeld S, Broen PA, McGue M. A 28-year follow-up of adults 49. Statistics Canada. Low Income Cut-Offs for 2007 and Low Income
with a history of moderate phonological disorder: linguistic and Measures for 2006. Ottawa: Statistics Canada Catalogue no.
personality results. J Speech Hear Res. 1992;35:1114-1125. 75F0002MWE2008004; 2008.
31. Felsenfeld S, Broen PA, McGue M. A 28-year follow-up of adults 50. Blishen BR, Carroll WK, Moore C. The 1981 socioeconomic index
with a history of moderate phonological disorder: educational and for occupations in Canada. Can Rev Soc Anthrop. 1987;24:465-488.
occupational results. J Speech Hear Res. 1994;37:1341-1353. 51. Stata [computer program]. Release 12. College Station, TX: Stata-
32. Brownlie EB, Beitchman JH, Escobar M, et al. Early language corp LP; 2011.
impairment and young adult delinquent and aggressive behavior. 52. Azur MJ, Stuart EA, Frangakis C, Leaf PJ. Multiple imputation by
J Abnorm Child Psychol. 2004;32:453-467. chained equations: what is it and how does it work? Int J Methods
33. Beitchman JH, Brownlie EB, Wilson B. Linguistic impairment and Psychol Res. 2011;20:40-49.
psychiatric disorder: pathways to outcome. In: Beitchman JH, 53. National Center for Education Evaluation and Regional Assis-
Cohen N, Konstantareas M, Tannock R, eds. Language, Learning tance. Technical methods report: What to do when data are
and Behavior Disorders: Developmental, Biological, and Clinical missing in group randomized controlled trials. 2009. Available at:
Perspectives. New York, NY: Cambridge University Press; 1996: http://ies.ed.gov/ncee/pubs/20090049/index.asp. Accessed
493-514. August 29, 2013.
34. Johnson CJ, Beitchman JH, Brownlie EB. Twenty-year follow-up of 54. Hardt J, Herke M, Leonhart R. Auxiliary variables in multiple
children with and without speech-language impairments: family, imputation in regression with missing X: a warning against
educational, occupational, and quality of life outcomes. Am J including too many in small sample research. BMC Med Res
Speech-Lang. 2010;19:51-65. Methodol. 2012;12:184.
35. Beitchman JH, Nair R, Clegg M, Patel PG. Prevalence of speech 55. vanBuuren S, Boshuizen HC, Knook DL. Multiple imputation of
and language disorders in 5-year-old kindergarten children in the missing blood pressure covariates in survival analysis. Stat Med.
Ottawa-Carleton region. J Speech Hear Res. 1986;51:98-110. 1999;18:681-694.
36. Hammill DD, Newcomer PL. Test of Language Development: 56. Wolke D, Waylen A, Samara M, et al. Selective drop-out in lon-
Primary. 3rd ed. Austin, TX: PRO-ED; 1997. gitudinal studies and non-biased prediction of behaviour disor-
37. Dunn LM, Dunn LM. Manual for Peabody Picture Vocabulary ders. Br J Psychiatry. 2009;195:249-256.
Test–Revised. Circle Pines, MN: American Guidance Service; 1981. 57. Touloumi G, Pocock SJ, Babiker AG, Darbyshire JH. Impact of
38. Goldman R, Fristoe M, Woodcock RW. Goldman-Fristoe-Wood- missing data due to selective dropouts in cohort studies and
cock Test of Auditory Discrimination. Circle Pines, MN: American clinical trials. Epidemiology. 2002;13:347-355.
Guidance Service; 1970. 58. Heraud-Bousquet V, Larsen C, Carpenter J, Desenclos JC, Strat YL.
39. Kessler RC, Wittchen HU, Abelson JM, et al. Methodological Practical considerations for sensitivity analysis after multiple
studies of the Composite International Diagnostic Interview imputation applied to epidemiological studies with incomplete
(CIDI) in the US National Comorbidity Survey. Int J Methods data. BMC Med Res Methodol. 2012;12:73.
Psychol Res. 1998;7:33-55. 59. Carpenter JR, Kenward MG, White IR. Sensitivity analysis after
40. World Health Organization. Composite International Diagnostic multiple imputation under missing at random: a weighting
Interview. Geneva, Switzerland: World Health Organization; 1990. approach. Stat Methods Med Res. 2007;16:259-275.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1110 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

TABLE S1 Variables in Multiple Imputation Models Predicting Age 31 Psychiatric Diagnosis

Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
Any psychiatric disorderd 5 Yes 83.8
Family SES 1 Yes 98.6
GAF 2, 3, 4, 5 (mean) Yes 97.2
TRF delinquent behavior T score 3 Yes 65.5
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 93.7
PPVT standard score 2, 3, 4 (mean) Yes Yes 95.8
Learning disability at age 12 or 19 2, 3 Yes 91.5
WRAT spelling standard score 3 Yes 84.5
Academic achievement scoree 3 Yes 85.2
Language Disorder
Any psychiatric disorder diagnosisd 5 Yes 63.1
Family SES 1 Yes 93.2
GAF 2, 3, 4, 5 (mean) Yes Yes 87.4
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
PPVT standard score 2, 3, 4 (mean) Yes 84.5
SRQ clinical probability 2 Yes 74.8
Academic achievement scoree 3 Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoref 1,2,3,4 (mean) Yes 100.0
Participants’ number of childreng 3,4,5 Yes 76.7
Physical abuse by parentsh 4 Yes 75.7
CBCL delinquent behavior T score 1, 2, 3 (mean) Yes 93.2
Note: CBCL ¼ Child Behavior Checklist; CTRS ¼ Conners Teacher Rating Scale; GAF ¼ Global Assessment of Functioning; PIQ ¼ Performance IQ;
PPVT ¼ Peabody Picture Vocabulary Test-Revised (Wave 1, 2 and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼ socioeconomic status;
SRQ ¼ Self Report Questionnaire; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate; wave 2); TRF ¼
Teacher’s Report Form; WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
12-Month diagnosis with GAF score <70.
e
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
f
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
g
Most recent available information from the 3 waves was used.
h
Physical abuse during childhood/adolescence was reported retrospectively at age 25.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1110.e1
BEITCHMAN et al.

TABLE S2 Variables in Multiple Imputation Models Predicting Age 31 Anxiety Disorder

Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
Any anxiety disorder diagnosisd 5 Yes 83.8
Family SES 1 Yes 98.6
GAF 2, 3, 4, 5 (mean) Yes 97.2
Physical abuse by parentse 4 Yes 94.4
Clinical-level CTRS overall score 1 Yes 97.2
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 93.7
Learning disability at age 12 or 19 2, 3 Yes 91.5
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes Yes 95.8
Academic achievement scoref 3 Yes 85.2
Language Disorder
Any anxiety disorder diagnosisd 5 Yes 63.1
Family SES 1 Yes 93.2
Affective disorder at age 19 and 25d 3, 4 Yes 79.6
GAF 2, 3, 4, 5 (mean) Yes Yes 87.4
TOLD/TOAL spoken language quotient 2,3 (mean) Yes 83.5
SRQ clinical probability 2 Yes 74.8
Academic achievement scoref Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoreg 1, 2, 3, 4 (mean) Yes 100.0
ADHD symptom score 3 Yes 80.6
Physical abuse by parentse 4 Yes 75.7
Note: ADHD ¼ attention-deficit/hyperactivity disorder; CTRS ¼ Conners Teacher Rating Scale; GAF ¼ Global Assessment of Functioning; PIQ ¼ Per-
formance IQ; PPVT ¼ Peabody Picture Vocabulary TesteRevised (waves 1, 2 and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼ socio-
economic status; SRQ ¼ Self Report Questionnaire; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate)
(wave 2); WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
12-Month diagnosis with GAF score <70.
e
Physical abuse during childhood/adolescence was reported retrospectively at age 25.
f
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
g
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1110.e2 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

TABLE S3 Variables in Multiple Imputation Models Predicting Age 31 Affective Disorder

Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
Any affective disorder diagnosisd 5 Yes 84.5
Family SES 1 Yes 98.6
Any postsecondary educatione 4, 5 Yes 94.4
Clinical-level CTRS hyperactivity score 1 Yes 97.2
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 93.7
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes Yes 95.8
Academic achievement scoref 3 Yes 85.2
Language Disorder
Any affective disorder diagnosisd 5 Yes 63.1
Family SES 1 Yes 93.2
Participant has any childreng 3, 4, 5 Yes 87.4
GAF 2, 3, 4, 5 (mean) Yes 87.4
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
PPVT standard score 2, 3, 4 (mean) Yes 84.5
SRQ clinical probability 2 Yes 74.8
Academic achievement scoref 3 Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoreh 1, 2, 3, 4 (mean) Yes 100.0
ADHD symptom score 3 Yes 80.6
Physical abuse by parentsi 4 Yes 75.7
Note: ADHD ¼ attention-deficit/hyperactivity disorder; CTRS ¼ Conners Teacher Rating Scale; GAF ¼ Global Assessment of Functioning; PIQ ¼ Per-
formance IQ; PPVT ¼ Peabody Picture Vocabulary TesteRevised (waves 1, 2, and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼
socioeconomic status; SRQ ¼ Self Report Questionnaire; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development
(Intermediate) (wave 2); WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
12-Month diagnosis with GAF score <70.
e
Postsecondary education included college and university.
f
Mean of Woodcock-Johnson Psychoeducational Battery-Revised broad reading, calculation, WRAT spelling standard scores.
g
Variable is coded as yes if at any of the 3 waves the participant indicated that he/she has a child.
h
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
I
Physical abuse during childhood/adolescence was reported retrospectively at age 25.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1110.e3
BEITCHMAN et al.

TABLE S4 Variables in Multiple Imputation Models Predicting Age 31 Substance Use Disorder (SUD)
Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
Any SUD diagnosisd 5 Yes 84.5
Family SES 1 Yes 98.6
Any postsecondary educatione 4, 5 Yes 94.4
Clinical-level CTRS hyperactivity score 1 Yes 97.2
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 93.7
Learning disability at age 12 or 19 2, 3 Yes 91.5
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes Yes 95.8
Academic achievement scoref 3 Yes 85.2
Language Disorder
Any SUD diagnosisd 5 Yes 63.1
Family SES 1 Yes 93.2
GAF 2, 3, 4, 5 (mean) Yes 87.4
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
PPVT standard score 2, 3, 4 (mean) Yes 84.5
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 95.1
Nonverbal PIQ scoreg 1, 2, 3, 4 (mean) Yes 100.0
Participants’ number of childrenh 3, 4, 5 Yes 76.7
CBCL delinquent behavior T score 1, 2, 3 (mean) Yes 93.2
Note: CBCL ¼ Child Behavior Checklist; CTRS ¼ Conners Teacher Rating Scale; GAF ¼ Global Assessment of Functioning; PIQ ¼ Performance IQ;
PPVT ¼ Peabody Picture Vocabulary TesteRevised (waves 1, 2, and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼ socioeconomic status;
SRQ ¼ Self Report Questionnaire; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate) (wave 2);
WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
12-Month diagnosis with GAF score <70.
e
Postsecondary education included college and university.
f
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
g
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2),
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
h
Most recent available information from the 3 waves is used.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1110.e4 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

TABLE S5 Variables in Multiple Imputation Models Predicting Brief Symptom Inventory Global Severity Index
Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
BSI global severity index score 5 Yes 83.1
Family SES 1 Yes 98.6
BSI global severity index score 3, 4 (mean) Yes 97.2
GAF 2, 3, 4, 5 (mean) Yes 97.2
ADHD symptom score 3,4,5 (mean) Yes 97.2
CES-D score (square root transformed) 3, 4 (mean) Yes 97.2
Any postsecondary educationd 4, 5 Yes 94.4
Clinical level CTRS hyperactivity score 1 Yes 97.2
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes 95.8
Academic achievement scoree 3 Yes 85.2
Language Disorder
BSI global severity index score 5 Yes 62.1
Family SES 1 Yes 93.2
Affective disorder diagnosis at age 19 and 25f 3, 4 Yes 79.6
GAF 2, 3, 4, 5 (mean) Yes Yes 87.4
BSI global severity index score 3, 4 (mean) Yes 83.5
SSA family social support score 3, 4, 5 (mean) Yes 85.4
SEC mean score 3, 4, 5 (mean) Yes 85.4
CES-D score (square root transformed) 3, 4 (mean) Yes 83.5
Any postsecondary educationd 4, 5 Yes 77.7
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
PPVT standard score 2, 3, 4 (mean) Yes 84.5
Academic achievement scoree 3 Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoreg 1, 2, 3, 4 (mean) Yes 100.0
ADHD symptom score 3 Yes 80.6
Physical abuse by parentsh 4 Yes 75.7
Note: ADHD ¼ attention-deficit/hyperactivity disorder; BSI ¼ Brief Symptom Inventory; CES-D ¼ The Center for Epidemiologic Studies Depression Scale;
CTRS ¼ Conners Teacher Rating Scale; GAF ¼ Global Assessment of Functioning; PPVT, Peabody Picture Vocabulary TesteRevised (waves 1, 2, and 3)
or Peabody Picture Vocabulary TesteIII (wave 4); PIQ ¼ performance IQ; SEC ¼ self-evaluation of communication; SES ¼ socioeconomic status; SSA ¼
Social Support Appraisals scale; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate; wave 2); WRAT ¼
Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
Postsecondary education included college and university.
e
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
f
12-Month diagnosis with GAF score <70.
g
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
h
Physical abuse during childhood/adolescence was reported retrospectively at age 25.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1110.e5
BEITCHMAN et al.

TABLE S6 Variables in Multiple Imputation Model Predicting Center for Epidemiological Studies-Depression
Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
CES-D score (square root transformed) 5 Yes 83.8
Family SES 1 Yes 98.6
GAF 2, 3, 4, 5 (mean) Yes 97.2
Marriage or common-law relationship history 3, 4, 5 Yes 97.2
Clinical-level CTRS overall score 1 Yes 97.2
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 93.7
Learning disability at age 12 or 19 2, 3 Yes 91.5
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes Yes 95.8
Academic achievement scored 3 Yes 85.2
Language Disorder
CES-D score (square root transformed) 5 Yes 62.1
Family SES 1 Yes 93.2
GAF 2, 3, 4, 5 (mean) Yes Yes 87.4
BSI global severity index score 3, 4 (mean) Yes 83.5
Any postsecondary educatione 4, 5 Yes 77.7
SSA social support score 3, 4, 5 (mean) Yes 83.5
CES-D score (square root transformed) 3, 4 (mean) Yes 83.5
TRF externalizing T score 2, 3 (mean) Yes 76.7
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
SRQ clinical probability 2 Yes 74.8
Academic achievement scored 3 Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoref 1, 2, 3, 4 (mean) Yes 100.0
ADHD symptom score 3 Yes 80.6
Physical abuse by parentsg 4 Yes 75.7
CBCL delinquent behavior T score 1, 2, 3 (mean) Yes 93.2
Note: ADHD ¼ attention-deficit/hyperactivity disorder; BSI ¼ Brief Symptom Inventory; CBCL ¼ Child Behavior Checklist; CES-D ¼ The Center for
Epidemiologic Studies Depression Scale; CTRS ¼ Conners Teacher rating scale; GAF ¼ Global Assessment of Functioning; PPVT ¼ Peabody Picture
Vocabulary Test-Revised (Wave 1, 2 and 3) or Peabody Picture Vocabulary Test e III (wave 4); PIQ ¼ performance IQ; SEC ¼ self-evaluation of
communication; SES ¼ socioeconomic status; SRQ ¼ Self Report Questionnaire; SSA ¼ Social Support Appraisals scale; TOAL ¼ Test of Adult Lan-
guagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate; wave 2); TRF ¼ Teacher’s Report Form; WRAT ¼ Wide Ranging Achievement
Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
e
Postsecondary education included college and university.
f
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
g
Physical abuse during childhood/adolescence was reported retrospectively at age 25.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1110.e6 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014
 ADULT OUTCOMES OF LANGUAGE IMPAIRMENT

TABLE S7 Variables in Multiple Imputation Models Predicting Short Form–12 Physical Health
Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control
SF-12 physical health score (square transformed) 5 Yes 83.1
Family SES 1 Yes 98.6
Any postsecondary educationd 4, 5 Yes 94.4
Clinical-level CTRS hyperactivity score 1 Yes 97.2
WRAT spelling standard score 3 Yes 84.5
PPVT standard score 2, 3, 4 (mean) Yes 95.8
Academic achievement scoree 3 Yes 85.2
Language Disorder
SF-12 physical health score (square transformed) 5 Yes 62.1
Family SES 1 Yes 93.2
GAF 2, 3, 4, 5 (mean) Yes 87.4
Marriage or common-law relationship history 3, 4, 5 Yes 86.4
Any postsecondary educationd 4, 5 Yes 77.7
TRF externalizing T score 2, 3 (mean) Yes 76.7
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
SRQ clinical probability 2 Yes 74.8
K-TEA battery composite standard score by age 2 Yes 65.0
Academic achievement scoree 3 Yes 73.8
CTRS hyperactivity score (square root transformed) 1, 2 (mean) Yes 98.1
Nonverbal PIQ scoref 1, 2, 3, 4 (mean) Yes 100.0
ADHD symptom score 3 Yes 80.6
Physical abuse by parentsg 4 Yes 75.7
CBCL delinquent behavior T score 1, 2, 3 (mean) Yes 93.2
Note: ADHD ¼ attention-deficit/hyperactivity disorder; CBCL ¼ Child Behavior Checklist; CTRS ¼ Conners Teacher rating scale; GAF ¼ Global
Assessment of Functioning; K-TEA ¼ Kaufman Test of Educational Achievement; PIQ ¼ Performance IQ; PPVT ¼ Peabody Picture Vocabulary Test
eRevised (waves 1, 2, and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼ socioeconomic status; SF-12 ¼ 12-item Short Form Health
Survey; SRQ ¼ Self Report Questionnaire; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate; wave 2);
TRF ¼ Teacher’s Report Form; WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
Postsecondary education included college and university.
e
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
f
Mean PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2); Wechsler
Adult Intelligence ScaleeRevised (waves 3 and 4).
g
Physical abuse during childhood/adolescence was reported retrospectively at age 25.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

VOLUME 53 NUMBER 10 OCTOBER 2014 www.jaacap.org 1110.e7
BEITCHMAN et al.

TABLE S8 Variables in Multiple Imputation Models Predicting Adult Self Report Antisocial Personality T Scores
Correlates
Variables in Multiple Analysis of Analysis Correlates of
Imputation Model Wavea Model Variablesb Missingnessc % Non-missing
Control Cohort
Adult Self Report antisocial personality T score 5 Yes 80.3
(log transformed)
Family SES 1 Yes 98.6
GAF 2, 3, 4, 5 (mean) Yes 97.2
Any SUD diagnosisd 3 Yes 90.8
Any postsecondary educatione 4, 5 Yes 94.4
Clinical-level CTRS hyperactivity 1 Yes 97.2
SSA social support score 4 Yes 92.3
WRAT spelling standard score 3 Yes 84.5
Absent parentf 1, 2, 3 Yes 99.3
Family mental illness historyf 1, 2 Yes 99.3
Family separationf 1, 2, 3 Yes 99.3
Partner had substance use problem 4 Yes 91.5
PPVT standard score 2, 3, 4 (mean) Yes 95.8
Academic achievement scoreg 3 Yes 85.2
Language Impaired Cohort
Adult Self Report antisocial personality T score 5 Yes 59.2
(log transformed)
Family SES 1 Yes 93.2
GAF 2, 3, 4, 5 (mean) Yes Yes 87.4
BSI global severity index scoreh 4 Yes 70.9
ADHD symptom score 3, 4, 5 (mean) Yes 86.4
TRF externalizing T score 2, 3 (mean) Yes 76.7
TOLD/TOAL spoken language quotient 2, 3 (mean) Yes 83.5
K-TEA battery composite standard score by age 2 Yes 65.0
Academic achievement scoreg 3 Yes 73.8
PPVT standard score 2, 3, 4 (mean) Yes 98.1
Nonverbal PIQ scoreh 1, 2, 3, 4 (mean) Yes 100.0
Participants’ number of childreni 3, 4, 5 Yes 76.7
ADHD symptom score 3 Yes 80.6
CBCL delinquent behavior T score 1, 2, 3 (mean) Yes 93.2
Note: ADHD ¼ attention-deficit/hyperactivity disorder; BSI ¼ Brief Symptom Inventory; CBCL ¼ Child Behavior Checklist; CTRS ¼ Conners Teacher Rating
Scale; GAF ¼ Global Assessment of Functioning; K-TEA ¼ Kaufman Test of Educational Achievement; PIQ ¼ performance IQ; PPVT ¼ Peabody Picture
Vocabulary TesteRevised (wave 1, 2, and 3) or Peabody Picture Vocabulary TesteIII (wave 4); SES ¼ socioeconomic status; SSA ¼ Social Support
Appraisals scale; SUD ¼ substance use disorder; TOAL ¼ Test of Adult Languagee3 (wave 3); TOLD ¼ Test of Language Development (Intermediate;
wave 2); TRF ¼ Teacher’s Report Form; WRAT ¼ Wide Ranging Achievement Test (3rd edition).
a
Wave 1 ¼ age 5; wave 2 ¼ age 12; wave 3 ¼ age 19; wave 4 ¼ age 25; wave 5 ¼ age 31.
b
Variables substantially correlated with analysis model variables.
c
Variables substantially correlated with missing vs. non-missing analysis variables.
d
12-Month diagnosis with GAF score <70.
e
Postsecondary education included college and university.
f
Reported by parent at either of the indicated waves.
g
Mean of Woodcock-Johnson Psychoeducational BatteryeRevised broad reading, calculation, WRAT spelling standard scores.
h
Mean of PIQ from Wechsler Primary and Preschool Scale of Intelligence (wave 1); Wechsler Intelligence Scale for ChildreneRevised (wave 2);
Wechsler Adult Intelligence ScaleeRevised (waves 3 and 4).
i
Most recent available information from the 3 waves is used.

JOURNAL OF THE AMERICAN ACADEMY OF C HILD & ADOLESCENT PSYCHIATRY

1110.e8 www.jaacap.org VOLUME 53 NUMBER 10 OCTOBER 2014