Language-learning impairments: a 30-year follow-up of language-impaired children with and without psychiatric, neurological and cognitive difficulties

Abstract

Aims: This study investigated the long-term consequences of language impairments for academic, educational and socio-economic outcomes. It also assessed the unique contributions of childhood measures of speech and language, non-verbal IQ, and of psychiatric and neurological problems.

Aims:

Methods & Procedures: The study was a 30-year follow-up of 198 participants originally diagnosed with language impairments at 3–9 years. Childhood diagnoses were based on language and cognitive abilities, social maturity, motor development, and psychiatric and neurological signs. At follow-up the participants responded to a questionnaire about literacy, education, employment, economic independence and family status. The response rate was 42% (198/470).

Outcomes & Results: At follow-up a majority of the participants reported literacy difficulties, unemployment and low socio-economic status—at rates significantly higher than in the general population. Participants diagnosed as children with specific language impairments had significantly better outcomes than those with additional diagnoses, even when non-verbal IQ was normal or statistically controlled. Childhood measures accounted for up to 52% of the variance in adult outcomes.

Conclusions & Implications: Psychiatric and neurological comorbidity is relevant for adult outcomes of language impairments even when non-verbal IQ is normal.

What this paper adds

What is already known

Many children with language disorders do not grow out of their difficulties and face poorer educational and socio-economic status outcomes as adults. Little is known about how specific childhood speech and language abilities, non-verbal IQ, and comorbid psychiatric and neurological diagnoses contribute to the variation in adult outcomes.

What this study adds

Non-verbal IQ, and psychiatric and neurological diagnoses appear to contribute to adult outcomes independently of early speech and language abilities. This contribution suggests that the distinction between specific language impairment and non-specific language impairment is not only a question of the severity of the language problem.

Introduction

Early language-learning impairments may have serious long-term consequences. In many cases children do not ‘grow out’ of their initial language problems (Conti-Ramsden et al. 2001, Glogowska et al. 2006, Johnson et al. 1999, 2010, Tomblin et al. 2003). If the language problems have not resolved before school, there is a higher than 50% risk that they will continue through the school years and into young adulthood (Stothard et al. 1998); the difficulties with language structure may also spread to difficulties with pragmatic language over time (Whitehouse et al. 2009a). The consequences for the development of academic abilities, literacy in particular, are serious and well-documented (Catts et al. 2002, Hall and Tomblin 1978, Whitehouse et al. 2009a, Young et al. 2002).

The adult outcomes of language-learning impairments are far less thoroughly researched. For example, Beitchman et al. (2001) found that at age 19 only marginally fewer language-impaired children than unimpaired were employed or attending school (83% versus 94% of controls). At age 25, though, the language-impaired group had a lower occupational socio-economic status than the controls (Johnson et al. 2010). Whitehouse et al. (2009b) reported lower levels of education in 18 young adults (aged 16–31 years) who had been diagnosed with SLI in childhood than in a matched control group with no prior diagnosis. Most of the young adults with a history of SLI had gained some form of vocational training, though. On the other hand, in a 30-year follow-up, Clegg et al. (2005) reported much more serious consequences. Of 17 men with childhood diagnoses of severe developmental language disabilities, only 59% were employed (versus 94% of their siblings), only 18% were in continuous employment (versus 94% of the siblings), 65% had received welfare benefits at some point (versus 10% of the control group), and only 7/17 were living independently (versus 15/17 of the control group). It should be noted, though, that the participants in this sample were originally diagnosed with severe receptive (as well as productive) language impairments. However, significantly elevated rates of unemployment in adults with a childhood diagnosis of SLI were also reported by Records et al. (1992).

One aim of the present study was to contribute to the knowledge of adult outcomes of childhood language-learning impairments by studying a large sample. Additionally, a large sample would also allow for more detailed analyses of the possible links between initial characteristics of the children and the outcomes in middle adulthood. Knowledge of such possible links is important because of the great variability in outcomes. Even in studies of severely language-disordered children, some children have overcome their initial problems (e.g. Clegg et al. 2005).

Consequently, another aim was to investigate the more specific links between childhood language abilities, non-verbal abilities, social and motor development, and adult outcomes. Particular attention was given to the long-term consequences of additional problems with non-verbal IQ, a psychiatric diagnosis and/or neurological signs. Hence, the study investigated the long-term external validity of the distinction between specific (SLI) and non-specific language impairments (NSLI) as it was applied when the children were originally seen between 1969 and 1979.

SLI are impairments found in children with at least normal IQ and without other diagnoses that might explain the language impairments (e.g. brain damage, psychiatric disorder, hearing loss, motor dysfunction, etc.) (e.g. Leonard 1998, Stark and Tallal 1981). Traditionally, children with below average IQ are excluded from the SLI group (e.g. Conti-Ramsden et al. 2001), whereas the criteria for brain damage and other reasons for exclusion appear to be rather variable—presumably depending on the detail of the available neurological screening.

Unfortunately, detailed neurological studies of language-impaired children are relatively scarce; some exceptions are mentioned below. The limited number of studies makes it difficult to assess the relevance of excluding LI children with neurological signs from the SLI group. However, the problem appears to be highly relevant as many published studies report a high frequency of occurrence of neurological signs in unselected language-impaired children, e.g. in 40–50% (Dalby 1977, Marschik et al. 2007, Njiokiktjien 1990, Selassie et al. 2005). The neurological findings in this literature are based on a wide variety of signs and clinical techniques, e.g. epileptic syndromes, dysfunctional muscle tone regulation, reflex abnormalities, EEG, CT and MRI abnormalities, and well-documented pre- and peri-natal problems.

By definition, neurological signs are rare in children with specific language impairments. Yet, Robinson (1991) conducted a detailed neurological examination of 82 adolescents (mean = 12 years, range = 9–17 years) with severe and persistent speech and language disorders who fit reasonably closely with the SLI profile. The sample did not include participants with low IQ, physical disabilities or behaviour disturbances. The examination revealed a high proportion (26–33%) of participants with pre-, peri- and postnatal causes. The examination also identified 21% (and possibly a further 11%) with a history of seizure disorder, and 17% with epileptic EEG abnormality. More commonly, sensori-motor abnormalities are found in many or even most of children with developmental language disorder—even when children with ‘hard’ neurological findings, such as cerebral palsy, have been excluded from the group (Rapin 1996). In contrast, Shevell et al. (2000) reported virtually no (4%) aetiologically relevant abnormalities in a group of 72 children (age 3;7 years) with developmental language disorder. However, these children were young and the vast majority (97%) were diagnosed with a mild-to-moderate language delay. While motor abnormalities reflect some kind of neurological variance, the relevance to speech-and-language disorders remain unclear and they are not commonly used to exclude children from SLI diagnosis.

The present study addressed the relevance of neurological signs for the diagnosis and adult outcomes of language impairments in several ways. It investigated the correlations in childhood between neurological signs and several measures of speech and language abilities, measures of motor development, and social development. It also investigated the very long-term correlations between neurological signs and adult outcomes both independently and in combination with other childhood measures.

Psychiatric and behavioural disorders are also frequent in LI children (Beitchman et al. 2001, Glogowska et al. 2006) and may be seen as possible comorbid factors which rule the child out for SLI. For example, Willinger et al. (2003) found behavioural problems in the clinical range in about one-third of 94 language-impaired children (of at least normal IQ) against only 6% in a control group. However, little is known of the possible influences of such comorbid disorders on adolescent and adult outcome.

The long-term prognosis for children with NSLI is worse than that for SLI children (e.g. Johnson et al. 1999). This is understandable because children with NSLI may have other problems in addition to their speech and language difficulties. These children are also likely to have fewer resources to compensate for their language difficulties. It is less clear which additional brain damage or psychiatric disorders are responsible for the worsening of the prognosis—and how they contribute. There is a shortage of longitudinal studies of LI children into adulthood with a focus on the influence of comorbid disorders in the psychiatric and neurological domains. However, insight into these contributions is of potential importance both to diagnostic procedures and to prevention. Consequently, a study was conducted to shed light over the following research questions:

• • 
  What are the adult outcomes of severe language-learning impairments in childhood in terms of literacy development, education, employment, economic and family status?
• • 
  To what extent are the adult outcomes predicted by the original diagnostic classification in childhood (SLI or NSLI with different types of comorbid problems: low IQ, psychiatric diagnosis, neurological signs)?
• • 
  To what extent are the adult outcomes predicted by childhood measures language, non-verbal IQ, social and motor development?
• • 
  Once childhood language abilities and non-verbal IQ are controlled, to what degree do comorbid problems—psychiatric diagnosis and/or neurological signs—contribute to adult outcomes?

Method

Design

The participants for the study were selected from clinic files from an observation department at the Speech and Hearing Institute in Aarhus, Denmark, from the years 1969 to 1979. The participants were admitted in childhood to the observation department because of severe speech–language problems—typically between the ages of 3 and 9 years. For the follow-up study 30 years later, a questionnaire was sent to 470 randomly selected LI children—now adults—who were contacted through the Danish social security system. The adults were then aged 35–40 years. Of the 470 questionnaires, 198 (42%) were returned from the adults. There were no significant differences on any of the childhood measures between those 198 adults who responded and those 272 who did not. For example, the male/female ratio was 1.71/1 among the respondents and 2.25/1 among the non-respondents (Chi²(1) = 1, 83, p > 0.15). The productive vocabulary quotient was 0.72 among the respondents and 0.73 among the non-respondents (t(469) = 0.48, n.s.). The respondents thus appeared to be representative of the whole population of language-impaired children seen at the Speech and Hearing Institute in Aarhus during the years 1969–79. The questionnaire about adult outcomes was concerned with reading and spelling, education, and other indicators of socioeconomic status and quality of life (cf. Records et al. 1992). A post-hoc assessment of the adult outcomes was done by comparison with national census data for the same age group. Adult outcomes were also compared between subgroups with different childhood diagnoses. Finally, multivariate regression analyses were conducted to assess the unique contributions of each childhood measure to the prediction of adult outcomes.

Participants

All 198 participants with childhood speech–language impairments spoke Danish as their first language and had no record of hearing impairments or uncorrected visual problems. Average age at childhood testing was 6;4 years (range = 3–9 years). As children, the participants were typically those with such severe speech–language problems that caused local speech therapists to refer them for a comprehensive examination at the Speech and Hearing Institute. Hence, the participants were likely to suffer from more serious speech–language problems and potentially complicating problems than a random sample of language-impaired children.

Following testing (see below), the children were involved in daily observational teaching at the Speech and Hearing Institute for 2–4 weeks. After this period the results were discussed with local speech therapists who then took over the teaching of the children. Hence, it is unlikely that other children with similarly serious impairments would have had significantly better teaching conditions.

Data on the background of the participants were obtained from the original clinic files. Information on parents’ education was available for 171 of the fathers and 162 of the mothers. No further education or vocational training was reported by 54% of the parents, while 41% reported vocational training or short to medium-long further education (1–4 years), while the remaining 5% held an academic degree (longer than 4 years, MA or higher). These figures suggest an average educational level somewhat below the national figures in the same age group (people born between 1945 and 1950) which are 41%, 54% and 5%, respectively (Statistics Denmark 2008).

Childhood measures (dates 1969–79)

All participants were originally seen as children by the same groups of professionals: speech therapists, psychologists, psychiatrists, social workers and a trained neurologist. For example, the language measures were taken by only one group of trained speech therapists with all children. The measures taken varied to some extent between children because of the clinical aim of the work and because of the children's varying ability to cooperate. However, the following core measures were obtained from a great majority of the children (the first column of table 2 provides the n's). The measures were state of the art in Danish clinical practice 30 plus years ago (Epstein 1978). Most of the measures have since been updated.

Table 2.  Language and developmental abilities of the participants at the initial testing in childhood

	SLI (LI only), n= 53	LI + neuro, n= 33	LI + psych, n= 21	LI + neuro + psych, n= 41	LI + low IQ + neuro + psych, n= 35	Total	F, p
Notes: Values are means (standard deviations). Group is significantly different from SLI group *p < 0.05, **p < 0.01 and ***p < 0.001 (Scheffe post-hoc group comparisons).
Vocabulary quotient (n= 145)	79.4 (17.2)	74.9 (18.7)	70.0 (22.1)	70.7 (16.4)	60.0** (15.1)	71.9 (18.8)	5.3, = 0.001
Non-verbal IQ (n= 183)	105.9 (13.0)	99.9 (9.8)	100.9 (13.7)	96.6** (8.0)	69.3*** (11.0)	95.2 (17.1)	61.7, < 0.001
Phoneme discrimination (n= 148)	16.5 (5.0)	12.7 (6.7)	15.2 (3.9)	13.2 (6.5)	11.7* (6.3)	14.1 (6.1)	3.9, < 0.01
Productive phonology (n= 135)	36.3 (18.5)	36.4 (24.3)	33.5 (19.0)	37.7 (19.3)	33.8 (18.8)	35.7 (19.8)	0.9, n.s.
Verbal memory (n= 145)	9.2 (5.0)	8.5 (6.1)	6.8 (4.4)	6.9 (4.7)	5.0* (3.1))	7.5 (5.0)	3.4, < 0.05
First words (months) (n= 175)	27.1 (9.3)	31.4 (9.9)	27.5 (10.5)	29.9 (12.1)	31.1 (12.8)	29.3 (10.9)	1.2, n.s.
Social maturity (quotient) (n= 128)	90.6 (12.2)	85.4 (15.4)	86.6 (15.3)	76.5 (18.6)	67.6 (12.9)	81.3 (17.0)	11.2, < 0.001
Motor development (quotient) (n= 169)	97.5 (10.0)	87.5* (14.8)	92.3 (10.3)	85.9** (15.5)	72.6*** (16.2)	87.8 (15.9)	16.2, < 0.001
First walk (months) (n= 192)	14.2 (3.2)	18.3 (5.6)	14.2 (3.7)	18.5 (6.9)	17.9 (4.8)	16.6 (5.3)	7.2, < 0.001

• • 
  Productive vocabulary was measured in a children's picture-naming task with 22 pictures (Ege 1969). Each response is scored on a three-point scale according to semantic precision. For example, responses to a picture of a rifle are scored thus: 0 points for no response, 1 point for bang-bang, 2 points for gun, and 3 points for rifle. The test has age norms from 3 to 7 years. The score used in the present study was the vocabulary age quotient, i.e. the age-scaled score divided by the chronological age of the child.
• • 
  Productive phonology was measured in a picture-naming test (Kristensen and Kristensen 1972). The test assesses the pronunciation of all Danish vowels, consonants and 38 consonant clusters—with one critical segment in each word. Since the test is not scaled, the score was number of correctly pronounced critical segments. The maximum score is 80.
• • 
  Phoneme discrimination. The child was shown pairs of pictured objects and asked to point out the one object named (Epstein 1978). The pairs were minimal, i.e. only one segment distinguish them, such as in masks—masts. The spoken words were tape-recorded and played to the children at a comfortable level. The test is not scaled, and the score was number correct of 25 pairs.
• • 
  Verbal memory was tested in a repetition task with nonsense syllables, a task similar to modern non-word repetition tests. Items varied from two to ten syllables, with five items of each length. Some examples are pa-da, la-ki-bo, and ti-po-du-na-mou (Epstein 1978). The test is not scaled, and the score was number of items correct (maximum 45).
• • 
  First words. Parents were asked at which age their child said his or her first words. The score was age (months).
• • 
  Non-verbal IQ was measured with the performance part of the Binet–Simon intelligence scale for children. A Danish standardization of the second edition was employed (Kirkelund and Rifbjerg 1945). The standard age-adjusted score was used (mean = 100, SD = 15).
• • 
  Social maturity was measured in months by means of the Vineland Social Maturity Scale (Doll 1953). The test is a structured questionnaire with 117 items that cover levels of independence, social communication and responsibility. For example, at 6 years the child is expected to use pencils or similar drawing tools, a knife to spread bread and to be able to go independently to known places. Age norms exist from 6 months to 25 plus years. The standard age-adjusted score was used.
• • 
  Motor developmental age was measured by means of the Ozeretzky test (Ozeretzky 1931, Danish translation by Epstein 1978). The test has increasingly difficult tasks that require fine and gross motor control and coordination. For example, at 6 years children are expected to be able to jump feet together over a string 25 cm above the floor, and draw at least 20 horizontal lines in 30 s, etc. Age norms are provided from 4 to 16 years. The standard age-adjusted score was used.
• • 
  First walk. Parents were asked at which age their child could walk. Responses were scored in months.
• • 
  Psychiatric examination. The child was observed by psychologists during play and in interaction with adults. Social workers paid visits to the child's home and family, and their reports were part of the material on which the psychologists based their diagnoses. The child's behaviour was described as either normal or primarily anxious, aggressive, passive, or with attention deficits, motor unrest or peculiarities of preferences. At the time of testing it was not assumed whether a psychiatric diagnosis might be causally implicated in a child's language impairments or constitute a separate complication. Thus, all children with a psychiatric diagnosis were treated in the analyses separately and not as a part of the SLI group. A fine-grained study of the influences of each of the specific psychiatric diagnoses was not possible in the present material because of limited numbers of children with each diagnosis.
• • 
  Neuropediatric examination sought to establish the neurological status of the child. Neurological signs included uni- or bilateral reflex disturbances, ataxia, pareses, disturbances of tonus or gait, as well as disturbances of sense of touch, prick, vibration and/or position. Additional EEG and neuroimaging measures were used in some cases for further diagnostic purposes. Based on the examination, children were characterized as either showing abnormal neurological signs or not. Clearly, some signs were more severe than others, and some signs were presumably closer linked to language sensitive areas of the brain than other signs. However, it was not assumed which neurological signs might be causally implicated in the language impairments. Such assumptions would require separate analyses beyond what was possible in the present data.

Follow-up questionnaire (30 years later)

A questionnaire was sent to the adults who were initially seen as children. The questionnaire has 17 questions concerned with eight themes; most of the questions are yes/no questions. (For an English translation, see Appendix 1.) The themes are initial difficulties learning to read, remedial teaching in school, reading and spelling difficulties at present, education, occupation, economic independence, housing, and children. A lower-bound estimate of the reliability of the questionnaire data was provided by the correlations with the childhood measures. As indicated by the multivariate regression analyses (table 5) more than 50% of the variance in some of the questionnaire responses could be accounted for by childhood measures; the corresponding multiple r of 0.71 is high.

Table 5.  Long-term outcomes (‘yes’ responses) predicted by measures at initial testing. Each column presents effect sizes (Nagelkerke R²) from a hierarchical logistic regression analysis

Initial testing	Responses at follow-up (‘yes’)
	Difficulties in learning to read	Remedial teaching	Reading difficulties as an adult	Spelling difficulties as an adult	Began education or training	Completed education or training	In a paid job	Receives a pension
Note: *p < 0.05, **p < 0.01 and ***p < 0.001. n.s., Not significant.
IQ and age, ΔR^2	0.076*	0.166**	0.389***	0.127**	0.232***	0.301***	0.351***	0.307***
Verbal memory, ΔR^2	0.037*	n.s.	0.029*	n.s.	n.s.	n.s.	n.s.	0.060**
Psychiatric diagnosis, ΔR^2	n.s.	n.s.	n.s.	n.s.	n.s.	0.041*	0.055**	0.025*
Neurological signs, ΔR^2	n.s.	0.058*	n.s.	n.s.	n.s.	n.s.	0.090***	0.125***
Total R^2	0.113*	0.224***	0.417***	0.127**	0.232***	0.342***	0.496***	0.517***

Results

Childhood measures (1969–79)

Table 1 displays characteristics of the children in five subgroups at the initial examination during the years 1969–79. Children were considered specifically language impaired (SLI, n= 53) if they had at least normal IQ (non-verbal IQ > 85), no neurological signs, and no psychiatric diagnosis. The other four subgroups comprise non-specific LI children (NSLI), i.e. children with other diagnoses in addition to their language impairments: neurological signs only (n= 33), a psychiatric diagnosis only (n= 21), both neurological signs and a psychiatric diagnosis (n= 41), and a combination of low IQ, neurological signs and a psychiatric diagnosis (n= 35). In addition to these five subgroups there were children with other combinations of disabilities in addition to their language impairments: four children with low IQ only, seven children with low IQ and neurological signs, and four children with low IQ and a psychiatric diagnosis. These children formed subgroups too small to provide reliable groups means, so they are omitted from the simple group comparisons in tables 2 and 3. However, these children were included in the multivariate analyses mentioned later (tables 4 and 5).

Table 1.  Characteristics of participants at the initial testing in childhood, grouped by the original diagnoses

	SLI (LI only), n= 53	LI + neuro, n= 33	LI + psych, n= 21	LI + neuro + psych, n= 41	LI + low IQ + neuro + psych, n= 35	Other NSLI, n= 15	Total, n= 198
Note: *Group is significantly different from SLI group, p < 0.05.
Female/male	20/33	9/24	7/14	16/25	15/20	6/9	73/125
Right-handed (%)	88.7	87.9	85.7	65.9*	75.0	73.3	80.0
Age (months)	76.9	73.9	71.6	73.6	79.8	78.0	75.8

Table 3.  Long-term outcomes: percentage of ‘yes’ responses in adults. Percentages from national data from the same age group are added for comparison in the last column

Responses at follow-up	SLI (LI only), n= 53	LI + neuro, n= 33	LI + psych, n= 21	LI + neuro + psych, n= 41	LI + low IQ + neuro + psych, n= 35	Total, n= 183	National data
Note: Group is significantly different from SLI group *p < 0.05, **p < 0.01 and ***p < 0.001.
Difficulties in learning to read	68.0*	87.5*	80.0	73.2	100.0***	80.3
Remedial teaching in school	64.2	96.8**	84.2	89.7**	97.1***	84.2	12.0
Reading difficulties as an adult	35.8	51.5	47.6	46.3	91.4***	53.0	4.9
Spelling difficulties as an adult	60.4	81.8*	66.7	75.6	97.1***	75.4
Entered further education or vocational training	67.9	39.4**	38.1*	26.8***	8.6***	38.8	92.4
Completed further education or vocational training	54.7	36.4	33.3	12.2***	2.9***	29.5	79.5
In a paid job	80.8	50.0**	55.0*	22.0***	2.9***	56.1	84.1
Receives a pension	24.5	66.7***	47.6	90.2***	97.1***	63.4	12.9
Lives in an independent house or flat	96.2	84.4	85.7	61.0***	20.0***	70.3
Has children	60.4	27.3**	19.0**	7.3***	5.9***	27.5	71.8
Children with speech–language impairments**	22/66	8/23	0/11*	2/7	2/3	34/110
Children with reading difficulties**	24/66	10/23	3/11	4/7	2/3	43/110

Table 4.  Prediction of outcomes (‘yes’ responses) at follow-up from initial language abilities. The direction (±) and strength (Eta) of the association and significance level are given for each prediction. For clarity, only significant associations are displayed

Initial testing	Responses at follow-up (‘yes’)
	Difficulties in learning to read	Remedial teaching	Reading difficulties as an adult	Spelling difficulties as an adult	Began education or training	Completed education or training	In a paid job	Receives a pension
Note: *p < 0.05, **p < 0.01 and ***p < 0.001.
Vocabulary	–	−0.17*	−0.32***	−0.16**	0.35***	0.34***	0.36***	0.39***
Phonological discrimination	–	−0.28***	−0.22**	–	–	–	0.23**	0.30***
Prod. phonology	–	–	–	–	–	–	–	0.25**
Verbal memory	−0.22**	−0.26**	−0.37***	−0.24**	0.28**	0.30**	0.31***	0.40***
First words	–	0.21**	0.23**	–	−0.22**	−0.18*	–	−0.23 **
Non-verb IQ	−0.21**	−0.31 ***	−0.48***	−0.28***	0.43***	0.44***	0.50***	0.46***
Social maturity	−0.18*	−0.30***	−0.38***	−0.20*	0.47***	0.43***	0.50***	0.54***
Motor development	−0.20*	−0.28***	−0.36***	−0.25**	0.43***	0.39***	0.47***	0.53***
First walk	–	0.15*	0.21**	0.17*	−0.20**	−0.20**	−0.20**	−0.31***
Age	0.18*	0.21**	–	–	–	−0.15*	–	–

The male/female ratio was 1.7/1, which is fairly typical for language-delayed children, and the ratio did not differ significantly between the subgroups (Chi²(4) = 2.1, n.s.). There were no significant differences between the mean age of the children in the five subgroups at the time of the initial testing (F(4, 182) = 1.7, n.s.). However, there was a difference in hand preference (Chi²(4) = 10.0, p < 0.05); post-hoc pairwise comparisons indicated that NSLI children with both neurological signs and a psychiatric diagnosis were significantly more often left-handed or ambidextrous than were the SLI children.

Neurological signs were found in 59% (116/198) of the total group of children. The most frequent diagnoses were cerebral palsy (in 28%, often mild and mostly attributed to asphyxia at birth), and focal or generalized epilepsy (11%). Other diagnoses were encephalitis (5%), malformation (5%), arrested hydrocephalus and various rare syndromes. The proportion of children with neurological signs was high, but it should be noted that the proportion was lower (33/86 = 38%) in the subgroups of children with at least normal IQ. This proportion is in line with findings from other studies of severely language-disordered children (see the Introduction). Almost 25% of the children with neurological signs were left-handed or ambidextrous, which is approximately twice the proportion found in a normal population. This finding of an association between neurological signs and non-right-handedness is consistent with predictions by Bishop (1990, ch. 7).

As many as 51% (101/198) of the children received a psychiatric diagnosis. The most frequent was DAMP (deficit in attention, motor control and perception), which was found in 18% of all the children. Some of these children would probably have received an ADHD diagnosis, if it had been an option at the time of the initial testing. A further 8.5% of the children were characterized by primarily attention deficits, while 7.5% were exceedingly ‘passive and clinging’, and 6.5% had primarily aggressive behaviour.

Strong associations were found between low IQ and neurological signs (Chi²(1) = 17.8, p < 0.001) and between low IQ and a psychiatric diagnosis (Chi²(1) = 19.5, p < 0.001).

As would be expected, the language scores were much below age norms (Table 2). The average vocabulary age quotient was 72, i.e. about –2 SD from age expectancy. This was in contrast to an average non-verbal IQ of 95, which is near normal. On average, the children were almost 30 months when they spoke their first word. They would normally be expected to do so before the age of 18 months (typically at 11–13 months). As a group the children also had delayed social and motor developments (average scaled scores of 81 and 88, respectively).

The SLI children outperformed the NSLI groups at initial testing (Table 2; the very small groups, n= 15, of NSLI children are not included). Significant effects of group were found for all language and developmental abilities except productive phonology and the occurrence of the first words (see Table 2, last column for statistics). However, only the NSLI group with multiple comorbidity (low IQ, neurological signs and a psychiatric diagnosis) had significantly lower scores than the SLI children across the board. The two NSLI groups with only neurological signs or a psychiatric diagnosis did not score significantly lower than the SLI group on any of the language or cognitive tests. This suggests that these children—with some but limited problems in association with or addition to their language problems—may not appear much different from the SLI children in terms of their performance on standard language tests. More rigorous definitions of neurological signs and psychiatric diagnoses may have yielded significant group differences. However, the subgroups of children with different psychiatric diagnoses tended to be too small for reasonably sensitive analyses.

Follow-up 30 years later

Follow-up results are displayed in Table 3 with national data from the same age group for comparison. These outcomes should be seen in the light of the multiple difficulties which characterized the majority of the children, as discussed in the next section. About 80% of the adults reported difficulties learning to read, and a similar proportion of the adults reported that they had received remedial teaching. About half of the total sample received remedial teaching from grade 1. A post-hoc assessment indicated that these figures were much higher than those in the general population. A national survey of adult literacy reported that only about 12% of the adults in the same age range had received remedial teaching for reading and/or spelling difficulties at school (Elbro et al. 1995); the national survey was based on a representative sample of 1124 adults aged 18–67 years. More than half of the participants in the present study reported reading difficulties persisting into adulthood. By comparison, only 4.9% of the Danish adults reported reading difficulties in the Danish part of the Second International Adult Literacy Study (based on a representative sample of 3000 adults; Jensen and Holm 2000). Given the sample sizes, the group differences here are sizeable—at least ten times the sampling error in the smallest group (the follow-up sample).

Of the follow-up sample, 39% reported that they had entered into post-compulsory education or vocational training. This is a much lower proportion than for the population in general (92.4% of all 35–39 year olds in 2005; Statistics Denmark 2008. The population data from Statistics Denmark are census data, not estimates based on samples). Thirty per cent of the participants reported having finished education or vocational training. The corresponding national figure is 82.5% (Statistics Denmark 2008); the figure was 46% among the parents of the participants.

Concerning socio-economic outcomes, 56% reported that they held a paid job—comparable with 84.1% in the same age band in the general population; Statistics Denmark 2008). Of those in job the majority reported that they were unskilled (35%) or skilled (40%) workers. Almost two-thirds of the whole follow-up sample reported that they received a pension (i.e. a long-term state allowance). The corresponding national figure is 12.9% including individuals on a yearly state allowance (Statistics Denmark 2008).

Twenty-eight per cent of the participants had children of their own, where the national count for the age group is 71.8% (Statistics Denmark 2008). Thirty-one per cent of the children had speech difficulties, and 39% had reading difficulties. These last-mentioned proportions include all offspring, hence they are a lower-bound estimate of the proportion who will develop such difficulties as they grow up. These figures are much higher than the national figure; a total of 8.4% of all school children received some kind of special teaching for any reason including language and reading difficulties (in class or in special classes or special schools) in the school year 2008/09 (Ministry of Education 2010).

Predictions from childhood into adulthood

Adult outcomes for children diagnosed with SLI were significantly better than for NSLI children originally diagnosed with additional impairments in non-verbal IQ, neurological signs and/or a psychiatric diagnosis (Table 3). The effect of diagnostic group was significant for all follow-up measures—Chi² values ranged from 15.8 (p < 0.01) for difficulties learning to read to 66.8 (p < 0.001) for reception of a pension (long-term dependence on public support). Post-hoc comparisons between SLI participants and each of the NSLI groups revealed the following significant differences:

Participants with normal non-verbal IQ but originally diagnosed with neurological signs had significantly poorer adult outcomes than the SLI group. This was true with respect to difficulties learning to read, a higher frequency of remedial teaching in school, poorer spelling in adulthood, a lower educational level, a lower employment rate, a higher rate of economic dependence (receives a pension) and a lower likelihood of having children.

Similarly, participants with normal non-verbal IQ but with a psychiatric diagnosis in childhood had a poorer prognosis than SLI children, but only marginally so in terms of reading and spelling. The group differences were significant with respect to enrolment in further education or vocational training, employment, and the likelihood of having children. Because of the small groups, it is difficult to provide reliable predictions of subgroups with specific psychiatric diagnoses. However, it can be noted that all of the 15 children diagnosed with severe attention disorders were referred to remedial teaching in school. The same was true for all 13 children with an excessively ‘passive, clinging’ behaviour.

The adult outcomes for children originally diagnosed with multiple comorbidity (low IQ, neurological signs and a psychiatric diagnosis) were particularly poor—and significantly poorer in all of the outcome domains studied than for participants originally diagnosed with SLI.

The predictive power of each of the childhood measures is given in Table 4 for all 198 participants. Because of the dichotomous outcome variables, the measure of the strength of the prediction is the Eta coefficient after controlling for age (except ‘first words’). The Eta coefficient indicates the direction and magnitude of the partial correlations between each childhood measure and each outcome. Verbal memory was the strongest predictor among the language measures and contributed significantly to the variance in all follow-up measures. Productive vocabulary was another significant predictor of all adult measures with the exception of difficulties learning to read. The strength of the associations hovered around the 10% level for the socio-economic outcomes. Productive phonology, on the other hand, did not correlate with outcomes, except economic dependence (receives a pension).

Non-verbal IQ (Binet) in childhood accounted for between 4% and 25% of the variation in adult outcomes. Hence, non-verbal IQ appeared to be a better predictor than verbal abilities. The other developmental measures, social maturity and motor development in childhood, provided prediction rates which were very similar in strength to non-verbal IQ.

Age at initial testing was negatively related to long-term outcomes (Table 4, last line), indicating that children referred at an older age had poorer outcomes than children referred at a younger age.

The final series of analyses addressed the research question about the unique contributions of a psychiatric diagnosis and/or neurological signs to the adult outcomes after controlling for age and non-verbal IQ. This question could not be answered reliably by means of simple group comparisons (Table 3) because of some very small groups. Instead, and because of the different properties of the childhood measures, logistic regression analyses were applied. In all analyses, IQ and age at initial referral were entered first, because these measures are standard, followed individually by verbal memory, psychiatric diagnosis and neurological signs. These three independent variables were entered according to their relative strengths as predictors in the individual analysis. Verbal memory was selected to represent the verbal measures because it provided the strongest single correlate with the follow-up data across the board (Table 4).

In Table 5, each column presents results from a separate analysis of the responses to a question at follow-up (shown in the column heading). The strength of the predictions of adult outcomes ranged from moderate to strong with socio-economic outcomes among the outcomes that were strongly predicted by childhood measures (Table 5 bottom row). Up to around 50% of the variance in socio-economic status (job and pension) was predicted. The magnitude of these correlations (r > 0.7) is noteworthy given the 30-year interval from childhood testing to follow-up, and it suggests a high reliability of the measures at both ends of the study.

Once non-verbal IQ and age were controlled, the language measures did still contribute variance to the adult outcomes. Verbal memory contributed unique variance over and above non-verbal IQ and age to the reading outcomes in school and adulthood and to economic independence (‘receives a pension’). A psychiatric diagnosis in childhood contributed unique variance to adult outcomes in education, employment and economic independence. Finally, neurological signs in childhood contributed unique variance to remedial teaching in childhood and to employment, and economic independence in adulthood.

Discussion and conclusion

In response to the first research question, the 30-year follow-up results indicated very poor adult academic and socio-economic outcomes. No indications were found that the long-term adult consequences of LI were any less severe than consequences reported earlier in life by other studies. The language-impaired children did not appear to have grown out of their difficulties—not even later in adulthood. The results thus replicated the worryingly poor outcomes reported by Clegg et al. (2005) in a larger sample. Obviously, a full evaluation of the adult outcomes would require a control group. However, there is little doubt that the present group was far from national averages in every aspect studied.

Clearly, the adult outcomes depend on the original population referred to the Speech and Hearing Institute in Aarhus. It is quite possible that a majority of the children in this study were more affected by language and other disabilities than the average child seen by local speech therapists. Therefore, it is difficult to generalize the outcomes to the whole population of children diagnosed with speech impairments. However, even though the population under study may represent the severe end of language impairments and a relative high frequency of comorbidity, it may still be the case that the relationships between childhood measures and adult outcomes hold up in a more general population of speech-impaired children. Anyway, this is the kind of assumption behind the expectation that adult outcomes are particularly poor in the population in the present study.

In response to the second research question about the effects of comorbidity, the results of the follow-up study stress the importance of comorbid problems with low IQ, neurological signs and/or psychiatric disorders for long-term outcomes. More specifically, the results underline the independent importance of psychiatric and neurological disorders—even though they may not be reflected in standard verbal or non-verbal measures of abilities (e.g. IQ measures). NSLI children with either isolated neurological signs or psychiatric problems in addition to their language impairments may not differ from SLI children in terms of speech, language or non-verbal IQ. But their social and educational outcomes are likely to be worse than those of SLI children.

Naturally, the proportion of language-impaired children with comorbid neurological signs or psychiatric problems may vary between clinical populations. In the present population, the proportion may have been relatively high because the speech therapists who admitted children to the Speech and Hearing Institute, and thereby to the study, may have referred a high proportion of children with signs that were unfamiliar to the therapists. However, there are previous reports of a high comorbidity between LI and psychiatric disorders (van Daal et al. 2007) and LI and neurological signs (e.g., Marschik et al. 2007, Njiokiktjien 1990, Selassie et al. 2005).

As for the predictive power of the childhood speech and language measures (cf. the third research question), the strongest verbal predictor of adult outcomes across the board was verbal memory. It was significantly correlated with every one of the outcome measures. The verbal memory task was given in the form of a non-word repetition task, and the results are in accordance with those of recent research. Since the time of the initial examination of the children, non-word repetition tasks have been used in many studies and are considered effective in identifying language impairments (Estes et al. 2007).

The poorest verbal long-term predictor was productive phonology. This is worth noting because poor productive phonology was (and is) among the main reasons for referral to the Speech and Hearing Institute. Productive phonology is an important measure for the intelligibility of the child. The present study did not record the various kinds of interventions that the children were offered. Therefore, it is impossible to say to what degree intervention may have helped some children more than others. Nonetheless, the results are in line with those from previous studies of both unselected and populations at risk for reading failure (e.g. Catts et al. 2002).

However, the speech and language measures were generally less predictive of long-term social and educational outcomes than were non-verbal IQ and neurological and psychiatric diagnoses. This is a common finding in longitudinal studies of language-impaired children (e.g. Catts et al. 2002).

Indeed, the present study indicated that the non-verbal measures and diagnoses provided relatively strong contributions to the adult outcomes even after controlling for the stronger of the verbal measures (cf. the fourth research question). This result is in line with recent findings from a 20-year follow up (Johnson et al. 2010) and suggests that the distinction between SLI and NSLI is not only a question of the severity of the language problem. Further support for the hypothesis of multiple, independent sources of variation in adult outcomes stem from the childhood results. The language abilities in the SLI children were not all that different from those of the NSLI children with only one additional diagnosis (an additional psychiatric diagnosis or neurological signs) (Table 3). Only children with multiple additional problems—including low non-verbal IQ—had significantly lower language scores than the SLI children.

Alternatively, the verbal measures may not have been sufficiently sensitive. One reason may be a limited variability in verbal abilities. The children in the study were all language impaired, whereas the variability in other domains was unrestricted (see also the second limitation mentioned below). On the other hand, more time and effort were spent on the assessment of the children's verbal abilities than on other abilities and signs. Therefore it is unlikely that major, strongly predictive differences in verbal abilities went unnoticed.

Non-verbal abilities may be important predictors of adult outcomes because they tap the potentials for compensation for speech and language impairments. The results of the present study are in concord with this hypothesis. After all, considerable proportions of the variance in adult outcomes were predicted by non-verbal abilities such as non-verbal IQ, social maturity, psychiatric disorders and motor development. This was true for all children, not just for those with low IQ or neurological signs. It is a possibility that the very poor adult outcomes may be an indication of a negative spiral that is the result of a bidirectional relationship between non-verbal IQ and verbal abilities (Botting 2005).

There may be at least two other reasons for the limited predictive strength of the childhood language abilities, however. One reason is that the language-impaired children received regular speech and language therapy which may have been effective to a varying degree against speech and language impairments but not very effective against general cognitive and/or psychiatric problems. A second reason is, simply, that individual educational and social development require more than speech and language proficiency.

There are a number of limitations of this study. One is time. The study does not and cannot provide longitudinal results for today's measures and clinical practices. It is limited by the selection and quality of the measures and diagnoses in use 30 plus years ago. However, the abilities and problems targeted are still current even though the measures may have been revised.

A second source of limitations is the clinical nature of the initial examination of the children. The clinical examination, although thorough and multidisciplinary, did not assess all children on all measures. In some cases, children refused or were unable to cooperate; in other cases, a test was not given because there was no clinical indication why it should be. Realistically, some missing values may really be zero scores. The consequence is that the initial speech and language measures may be less sensitive and possibly less predictive than they really are.

The third source of limitations is the use of a questionnaire rather than a proper individual assessment at follow-up. Questionnaire data are notoriously difficult to interpret. Clearly, many of the participants may have forgotten their initial problems learning to read. On the other hand, previous studies have reported significant adverse long-term effects of SLI based on questionnaire data (e.g. Arkkila et al. 2008). More importantly, the moderate-to-strong correlations between childhood measures and diagnoses and the adult responses to the questionnaires (with up to 50% of the variance accounted for) are noteworthy and suggest a high reliability of the questionnaire data in the present study. Even so, the predictions would probably have been even stronger had the questionnaire been supplemented by a proper assessment.

Despite the limitations, the findings have several practical implications. The very poor adult outcome is an important challenge. Given the enormous expenses at personal and all other levels, research into prevention and intervention should be intensified. Bearing in mind the relative independence of speech and language impairments and poor abilities in other domains, it is unrealistic to expect that the treatment of one impairment will lead to improvements across the board. Many children will need support in several other areas in addition to the speech and language support that they may receive. It is also important to include more than just a measure of non-verbal abilities (IQ) for the diagnosis and long-term prognosis of children with language impairments. Neuropediatric and psychological and/or psychiatric expertise may be called for.

Acknowledgements

The study was approved by the Committee on Biomedical Research Ethics, Region Aarhus, Denmark. The authors would like to thank an anonymous reviewer for many valuable suggestions that improved an earlier version of this paper. The authors are particularly indebted to the participants for their time and care in responding to the questionnaire. Declaration of interest: The authors report no conflicts of interest. The authors are responsible for the content and writing of the paper.

Appendix

Appendix: Questionnaire from the follow-up study

If you have difficulties reading this, please ask your family or a friend to help you.

• 1
  Do you remember your stay at The Speech Institute in your childhood?
• 2
  Do you remember receiving speech therapy after your visit to The Speech Institute?
• 3
  Did you have problems learning to read when you started school?

  • Did you receive special teaching to learn to read?

  • At what grade level did you start special teaching?

• 4
  Do you still experience some difficulties

  • with reading?

  • with spelling?

• 5
  At which grade level did you finish school?
• 6
  Did you enter further education or vocational training after you finished school?

  • What sort of education or training was it?

  • Have you started more than one education or training?

• 7
  Did you complete your education or training?
• 8
  Did you get a job that required your education or training?
• 9
  What is your present job?
• 10
  Do you receive state benefits?
• 11
  Do you live in an apartment, a house, in shared, state provided accommodation, in a flat with special support, with your parents or guardians?
• 12
  Do you have any children? How many?

  • Do any of your children have difficulties

  • with speech?

  • with reading?