SEARCH

SEARCH BY CITATION

Keywords:

  • Familial aggregation;
  • multifactorial model;
  • proband characteristics;
  • specific language impairment (SLI)

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

There is now little doubt that both environmental factors and genes are likely to make important contributions to the aetiology of specific language impairment (SLI). The most commonly proposed model for understanding these influences is the multifactorial model. In the present study we examine two expectations based on this model: that there will be a systematic relationship between the severity of proband language scores and the rate and severity of SLI in relatives and that relatives will be more strongly affected if they are relatives of a proband of the more rarely affected gender (female) because the latter require a higher genetic liability to become equally impaired. Ninety-three probands and their 300 first-degree relatives participated in this study. Results showed a relationship between proband severity at age 14 and an increased rate of SLI in relatives. This relationship was strong for child siblings and was significant with respect to both rate of SLI and severity over a range of language and literacy measures. In contrast, higher levels of SLI among relatives of female rather than male probands was entirely disproved.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

Specific language impairment (SLI) is a common disorder with current research suggesting that between 3 and 7% of children are affected (American Psychiatric Association 1994; Tomblin et al. 1997). It is a heterogeneous disorder, presenting with a variety of profiles, all of which involve limitations of language learning in the absence of possible explanatory factors such as low non-verbal IQ, hearing impairment or neurological damage (Bishop 1997; Leonard 1998). SLI is likely to be a persistent disorder and affects males more frequently than females (Johnston et al. 1981; Tallal et al. 1989). There is now little doubt that both environmental factors and genes are likely to make important contributions to the aetiology of SLI. Although there is evidence for specific heritability for some quite distinct language impairments (FOX-P2, KE family; Lai et al. 2001), there is still great uncertainty as to the importance of familial factors in the aetiology of the more generic language impairment commonly seen in clinics. There are a number of broadly plausible polygenic models for SLI, for example a threshold model, in which someone becomes affected if he or she carries a number of susceptibility genes. Alternatively there is an epistatic model, in which two (or more) genes jointly act to create a phenotype, but each gene alone has no measurable effect. However, perhaps the simplest and most commonly proposed model for non-articulatory language impairment is the multifactorial model (Falconer 1965), in which the effects of a number of genes, each of relatively small effect, need act only additively and combine with environmental factors to generate an underlying latent or unobserved ‘liability’ to a phenotypic trait, upon which a threshold is imposed that distinguishes those who are affected from those who are unaffected. Family studies typically involve identifying eligible families through an affected family member (the proband) and the assessment of relevant characteristics among the various kinds of relatives found in those families. The multifactorial model leads to several expectations (Melnick et al. 1980). Notable among these are first that severely affected probands with the greatest liability for SLI will have relatives that are more often and more severely affected. Second, because gender differences in prevalence are accounted for by postulating differing thresholds for affectedness, probands of the less frequently affected gender will have on average a greater liability and thus relatives of female probands will be more often and more severely affected.

Family studies so far have been relatively few in number and have generally found mixed results with respect to the expectations of the multifactorial model. In terms of severity of SLI, the early work of Byrne et al. (1974) suggested that proband severity was inversely related to affectedness in relatives. Thus, relatives of the more severely affected probands were less likely to be affected with language-related problems. Two important limitations of this study need to be noted. First, their sample comprised probands with moderate and severe problems, hence there may have not been enough heterogeneity among the participants for a full examination of severity to be considered. Second, and importantly, the definition of language impairment used was very broad and the identification of affectedness in relatives was not systematic across all first-degree relatives. More recently, Choudhury & Benasich (2003) reported contrasting results. They found that probands with low language ability had a greater chance of having a positive family history than children with normal language abilities.

Other researchers have focused more on the type of language difficulty presented by the probands and thus have not examined severity directly. This approach has also resulted in inconsistent findings. Whitehurst et al. (1991) found that probands with expressive language problems (E-SLI) did not show elevated rates of affectedness among their relatives when compared to those relatives of non-impaired children. In contrast, Lahey & Edwards (1995) found that probands with expressive language problems were more likely to have family members with a history of SLI than probands with mixed (expressive and receptive) SLI.

In terms of gender of proband, differences in rate of impairment among relatives of male versus female probands has once again not provided a consistent picture. Beitchman and colleagues (1992) found that in line with the expectations of the multifactorial model, female probands in their study had more affected relatives than male probands. Tomblin (1989) also found a slightly higher rate of SLI for family members of female probands but the difference in rate was not significant (25.6% vs. 22.8%). Other researchers have not found differences in the affectedness of relatives in relation to proband gender and hence have not replicated the above findings (e.g. Rice et al. 1998). To be able to exclude the simple liability-threshold multifactorial model would prompt consideration of alternative explanations for differences that could involve X-linked genes, epigenetic effects, and interactions with the family and social environment. Thus, although some support for the multifactorial model has been forthcoming in relation to gender, it is important to examine in more detail evidence in relation to this expectation.

Undertaking family studies is a complex task. Those that have been carried out in SLI have generally involved small samples which may have contributed, at least partly, to the inconsistency of the findings reported so far. Furthermore, it is important to note that family studies are complicated by the issue of what constitutes the expression of the pathology among relatives of very different ages (for example, parents vs. siblings), and how this expression should be measured. In the great majority of studies, measures are chosen which are thought to ‘iron out’ such potential variation. For example, standardized scores could be used, although if adults are to be included this severely restricts the choice of measures. Alternatively, ‘lifetime’ diagnoses could be used, although these have been criticized for their questionable validity (Choudhury & Benasich 2003; Flax et al. 2003; Tallal et al. 2001). Similar problems and proposed solutions arise with respect to the characterization of probands – put crudely, at what age should they be considered affected? A thorough investigation would require longitudinal data across development of both probands and relatives – a study design that presents such enormous practical challenges that we are unlikely to see anything but a few partial attempts. We present here findings in relation to SLI from a more feasible design, one that involves a relatively large sample of probands as well as prospective direct measurement of probands but in which we rely on cross-sectional data for first-degree relatives.

The aim of the present study was to examine familial expression of the SLI phenotype with particular reference to the multifactorial model. We examine the relationship between the severity and gender of the proband and the rate and severity of SLI among relatives. To our knowledge, no paper in the literature has yet reported the use of proband severity to explain the relationship between heterogeneity of the phenotype and familial loading in SLI. We examine the extent to which these expectations are met when probands are characterized at ages 7 and 14 years and whether familial liability is more strongly reflected by probands with expressive and/or receptive language difficulties. For relatives under 16 years of age we are able to compare findings when using a more generic battery of tests standardized for children and adults, to findings from a specific language test suitable for children but not standardized for adults.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

Participants

The probands were drawn from the Conti-Ramsden Manchester Language Study. This cohort of children was recruited from 118 language units attached to English mainstream schools. All language units catering for primary school year 2 children were contacted and any centres enrolling children with global delay or hearing impairments were excluded. The remaining language units provided a list of year 2 children attending for at least 50% of the week. Across England approximately 500 children fitted this criterion. All language units were asked to participate and two declined this invitation. Subsequently, approximately half of the eligible children in each unit were randomly sampled. The age range of the 242 children who entered into the sample was 6 years 2 months (6;2) to 7;10 years and consisted of 185 males and 57 females (Conti-Ramsden et al. 1997). These children were reassessed at 8 years (Conti-Ramsden & Botting, 1999), 11 years (Conti-Ramsden et al. 2001) and 14 years of age (the present study). The number of children who participated in these three subsequent waves of data collection were respectively 234, 200 and 113 and they ranged in age between 7;5 and 8;9 years; 10;1 and 11;10 years and 13;1 and 16;2 years respectively. In the last wave of data collection, that is when these children were 14 years old, data on first-degree relatives were also collected. All 242 children who participated in the original study at 7 years were included in the list of potential participants. Of these, four families (2%) were not contacted because the proband was adopted, 59 (24%) families did not respond to the invitation to participate in the study and 55 (23%) families refused consent. Of the 124 (51%) families who agreed to take part in the present phase of the study, 113 (91%) were assessed and 11 (9%) were not assessed because of alterations in family circumstances. From this pool of 113 consenting, assessed families, 93 were selected for participation in the present aggregation study based on examination of longitudinal data at 7, 8, 11 and 14 years using the following proband (i.e. child in the original study) criteria:

  • 1
    Performance IQ (PIQ) of 80 or more and a minimum of one concurrent standardized language test score which fell at least 1SD below the population mean at one of the longitudinal assessment stages (i.e. 7, 8, 11, or 14 years).
  • 2
    No sensory-neural hearing loss.
  • 3
    English as a first language.
  • 4
    No record of a medical condition likely to affect language.
  • 5
    No record of a co-morbid diagnosis of autism.

In the present study, there were 68 male and 25 female probands and they had a mean age of 14;5 years (age range 13;1 to 16;2 years). Minimum age for participation for first-degree relatives was 6 years. In terms of total family members, there were 300 first-degree relatives. More specifically, there were 93 fathers, 93 mothers, 35 male/24 female siblings over the age of 16 years and 26 male/29 female siblings between the ages of 6 and 16 years. Total sample size (including probands) was 393 relatives.

Procedure

Consenting families were telephoned to complete a pedigree and arrange assessment. Where possible, children were tested at school on the battery of language, literacy and general cognitive measures for either a morning or afternoon session. In the majority of cases, testing took place in a room with just the examiner and child present. Where access to empty rooms was not possible, testing took place in a quiet area such as the library. If the parents had requested that the child be assessed at home, testing took place in a quiet room with only the child and examiner present. Parents were seen at home for a minimum of two visits. Each session usually lasted for a maximum of 3 h. Psychometric tests and interviews were carried out on all consenting first-degree relatives. In all cases, testing was started and completed within 1 month for each separate family.

Measures for probands at age 7

Test for Reception of Grammar (TROG; Bishop 1989)

A multiple-choice test designed to assess understanding of grammatical constructions. Children are shown four pictures while the examiner reads a sentence. The child is required to select the picture that illustrates the sentence. This test has norms for ages 4;0 to 12;11 based on a UK sample of 2112 in 1983.

British Ability Scales – naming vocabulary subtest (BAS-nv; Elliot 1983)

A test of expressive vocabulary in which the child is asked to name a series of pictures of everyday items. This test has norms for ages 2;0 to 17;10 based on a UK sample of 3435 in 1976.

The Bus Story Test (BS; Renfrew 1991.

In this assessment, the examiner tells the child a short story about a bus while the child looks through a book of pictures illustrating the story. The child then retells the story as accurately as possible, using only the pictures as cues. Stories are audiotaped, transcribed and scored for the amount of correct information given. This test has norms for ages 3;9 to 8;3 years based on a UK sample of 573 in 1991.

Illinois Test of Psycholinguistic Ability: grammatic closure subtest (ITPA; Kirk et al. 1968).

A test of expressive syntax in which children are given picture prompts. The examiner reads an incomplete sentence that the child must finish in a grammatically correct manner. Items cover a range of grammatical knowledge and include plurals and past-tense items. This test has norms for ages 2;4 to 10;3 based on an American sample of approximately 1000 in 1968. UK norms are not available for this test.

Measures for probands at age 14 and their relatives

This battery included language, literacy and general cognitive measures. We were interested in including literacy measures in this investigation because a number of previous family studies have included literacy difficulties as an indicator of language problems (Neils & Aram 1986; Whitehurst et al. 1991). In addition, it is now being suggested that reading difficulties, i.e. dyslexia, may be a less severe form of SLI (Snowling et al. 2000) and that these two disorders may not be as distinct as initially conceptualized (Bishop & Snowling 2004). Having said this, the instrument that was used, the WORD, does not have adult norms available (beyond 16;11 years). Thus, there is a need to be cautious in the interpretation of the literacy standardized scores when they are used with adult relatives.

We also would like to note that the use of direct assessment (albeit with its own limitations given the instruments that were available) is a relative strength of the present study. The usual methodology used in family studies is parental report with very few investigations using direct assessment of relatives (although see Flax et al. 2003; Tallal et al. 2001).

Adult language measures

Wechsler Adult Intelligence Scale – Revised (WAIS-R VIQ; Wechsler 1986). Adult relatives (parents and siblings 16 years of age or over) were administered the verbal short form of this test. This test has norms for ages 16;0 to 74;11 years based on an American sample of 1880 in 1976–1980. UK norms are not available for this test.

Child language measures

Wechsler Intelligence Scale for Children – Third Edition UK (WISC-III VIQ; Wechsler 1992). Child relatives (probands and siblings between 6 and 16 years of age) were administered the full form of this test. This test has norms for ages 6;0 to 16;11 years based on an American sample of 2200 in 1991. A UK validation study was carried out in 1991 comprising 814 British children of the same age range.

Clinical Evaluation of Language Fundamentals – Revised UK (CELF-R; Semel et al. 1987). Child relatives were also assessed using this test, which yielded a Receptive Language Score (rls), an Expressive Language Score (els) and a Total Language Score (tls). This test has norms for ages 5;0 to 16;11 years based on an American sample of 2426 in 1986. Although the test has been modified for administration to British children, UK norms are not available for this test.

Adult Literacy Measures

Wechsler Objective Reading Dimensions (WORD; Rust et al. 1993). Adults relatives were administered the short form of this test which comprised Basic Reading and Spelling subtests. This test has norms for ages 6;0 to 16;11 years based on an American sample of 4252 in 1990. A UK validation study was carried out in 1991 comprising 814 British children of the same age range.

Child literacy measures

Child relatives were administered the WORD Basic Reading and Spelling subtests (as above for adults) and additionally the WORD Reading Comprehension subtest. This test has norms for ages 6;0 to 16;11 years based on an American sample of 4252 in 1990. A UK validation study was carried out in 1991 comprising 814 British children of the same age range.

Adult general cognitive measures

Wechsler Adult Intelligence Scale – Revised (WAIS-R PIQ; Wechsler 1986). Adult relatives were administered the performance short form of the WAIS-R. This gave an overall performance IQ score. This test has norms for ages 16;0 to 74;11 years based on an American sample of 1880 in 1976–1980. UK norms are not available for this test.

Child general cognitive measures

Wechsler Intelligence Scale for Children – Third Edition UK (WISC-III PIQ; Wechsler 1992). Children in the study completed all the WISC-III performance subtests. This gave an overall performance IQ score. This test has norms for ages 6;0 to 16;11 years based on an American sample of 2200 in 1991. A UK validation study was carried out in 1991 comprising 814 British children of the same age range.

Establishing proband severity at age 14

To measure proband severity at age 14, a factor analysis of WISC verbal and CELF subtests scores (WISC subtests on information, similarities, arithmetic, vocabulary, comprehension and digit span and CELF subtests on oral directions, word classes, semantic relationships, formulated sentences, recalling sentences and sentence assembly) provided strong evidence for the existence of a major first factor. However, in view of the very high correlation (0.925; 95% CI: 0.89–0.95) of this factor score with the much more directly interpretable CELF total language score (CELF tls), we used the CELF tls alone as a measure of proband severity. The negative part of the CELF tls converted into a z-score was used as both a continuous measure of proband severity (low scores of CELF tls corresponding to high severity) and a binary indicator based on above or below a threshold. Using the −1.25 SD cut-off commonly used in clinical practice for defining SLI (Tomblin et al. 1996) 69 out of 93 probands were classified as persistent SLI. The analyses reported later in this paper also include examination of other cut-offs (−2 SD, −1.5 SD) when examining severity/persistence of SLI. For ease of reporting, when applying these binary classification systems probands will be referred to as belonging to the severe vs. non-severe groups although we are aware that when the cut-off is −1.25 SD below the mean, this terminology is not transparent as – 1.25 is usually taken as the cut-off for SLI and not the cut-off for severe SLI. Three probands could not be classified or scored because of their missing CELF total language score. Measures of expressive and receptive severity were obtained using the negative, respectively, of the CELF expressive language score (CELF els) and CELF receptive language score (CELF rls).

Establishing proband severity at age 7

Proband severity at age 7 was measured as the average of the negative of the probands’ scores on two tests assessed at that age: the TROG and the BS. The former provided a measure of proband receptive severity, whereas the latter gave a measure of proband expressive severity.

Classification of probands according to type of language problem

Probands were classified into four categories reflecting different types of language problems on the basis of their scores in a battery of expressive and receptive tests at ages 7 and 14 years.

At age 7, the probands were classified on the basis of their performances on the TROG test and three expressive tests (BS, BAS-nv and ITPA) according to the following criteria:

  • 1
    Expressive SLI (E-SLI) if TROG was in the normal range (better than or equal to 1 SD below the mean) but at least one of the expressive tests was not in the normal range (more than 1 SD below the mean) or TROG was at least 10 centiles above one or more of the expressive tests.
  • 2
    Receptive SLI (R-SLI) if TROG was more than 1 SD below the mean but all the expressive tests were in the normal range.
  • 3
    Expressive-Receptive SLI (ER-SLI) if TROG and at least one of the expressive tests were more than 1 SD below the mean.
  • 4
    Resolved if TROG and all the expressive tests were in the normal range and TROG was not 10 centiles or more above any of the expressive tests.

This identified 35 E-SLI, 2 R-SLI, 41 ER-SLI and 10 resolved probands. Five probands could not be classified because of missing values on their language expressive and/or receptive scores.

At age 14, the probands were classified on the basis of their CELF els and CELF rls scores:

  • 1
    E-SLI if CELF rls was in the normal range (better than or equal to 1 SD below the mean) but CELF els was impaired (more than 1 SD below the mean).
  • 2
    -SLI if CELF els was in the normal range but CELF rls was impaired (more than 1 SD below the mean).
  • 3
    ER-SLI if both CELF els and CELF rls were more than 1 SD below the mean.
  • 4
    Resolved if both CELF els and CELF rls were better than or equal 1 SD below the mean.

According to these criteria 29 probands were E-SLI, 53 were ER-SLI and eight were resolved. None was identified as R-SLI. The type of language problem could not be determined for three probands because of missing values on CELF els and/or rls.

Classification of relatives

A basic battery of the WAIS VIQ, WORD basic reading and WORD spelling was available for all first-degree relatives. Relatives were classified as ‘affected’ if their scores on any of these three measures were more than 1 SD below the mean. For child siblings an additional, more specific, language battery was also available based on the CELF els, CELF rls scores and WORD reading comprehension. These batteries will be referred to as the basic and language batteries respectively.

Analysis

Logistic regressions were used to model the probability of a subject being affected as a function of the proband and subject characteristics. Since sample units were clustered in families and data on related individuals are likely to be correlated because of additional shared genetic and environmental factors, valid inference required adjustments for lack of independence of observations. We used a population-averaged/generalized estimating equations approach based on an independence working assumption (see Liang & Zeger 1986; Neuhaus et al. 1991). This involved fitting models as if the observations were independent but then accounting for the within-family correlation by using a robust or sandwich estimator for the standard errors (Huber 1967; Royall 1986). Both continuous measures (the negative of CELF total language z-score) and binary indicators of proband severity were considered to investigate the effect of severity as a determinant of familial aggregation. Analyses were also undertaken to understand the relationship between relatives’ risk of SLI and proband type of language problem (ER-SLI versus E-SLI).

Differences among relatives in the pattern or profile across a set of measures required multivariate analysis of the battery of assessments. Analyses were undertaken using a response-appropriate generalized linear model (usually regression), involving one record for each measure in the battery for each relative and a design matrix of dummy variables, covariates and their interactions that distinguished effects (contrasts) for each measure. In addition, the precision of these effects (contrasts) was estimated by the use of the robust or sandwich estimator. This recognized the possible correlation across measures and individuals within a family. This parameter covariance matrix was used as the basis for calculating Wald-based t- and F-tests for single and multi-parameter contrasts respectively. Reported confidence intervals are for the nominal 95% probability interval.

Almost all the analyses in this paper were performed using the statistical package STATA, version 8 (StataCorp 2003). For analyses of binary outcomes, where the small sample size or the sparseness of the data was likely to make asymptotic approximations inadequate, exact logistic regression was used (see Mehta & Patel 1995; LogXact version 6, Cytel Software Corporation 2004). We know of no procedure that adjusts such exact methods for non-independent observations, but the latter becomes of less importance with low outcome rates typical of many sparse data analyses.

Participation and non-response

Of the initial cohort of 242 children with SLI at age 7, 113 children participated at 14 years and of these, 93 met proband criteria and thus participated in the present study.

The 93 proband-ascertained families included 300 relatives (245 adult relatives and 55 child siblings). Rates of non-response among child siblings were very low, between 3.6 and 7.3% for different instruments. However, among adult relatives rates ranged between 31.4 and 33.5%. Non-response rates in adult relatives were roughly twice as common among males compared to females (P < 0.001), around three times more common if the proband was female (P = 0.02), and marginally associated with proband severity (P = 0.1). Selection biases associated with these variables were accounted for by including the type and gender of relatives, and severity and gender of proband in many analyses (i.e. covariate dependent missingness –Little & Rubin 1987).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

Descriptive statistics for probands and relatives

Table 1 gives descriptive statistics for probands and their relatives. On average, adult relatives had performance IQ scores above what would be expected from the population norm (where the expected mean is 100 with an SD of 15). Not surprisingly, probands showed a depressed profile compared to age norms (where again the expected mean is 100 with a SD of 15) across most measures.

Table 1.  Descriptive statistics for standard scores on the battery of tests
 TestnMeanSDMinMax
Adult relatives 
 WAIS PIQ168110.13718.74772150
 WAIS VIQ16799.88614.46172144
 WORD spelling16396.22117.90049128
 WORD basic reading167100.56313.67054117
Child siblings
 WISC PIQ5192.52919.64146131
 WISC VIQ5393.15119.32546127
 WORD spelling5293.76918.76555130
 WORD basic reading5396.24518.24149120
 WORD reading compreh.5287.7517.99041124
 CELF els5284.26917.23650126
 CELF rls5192.76522.11550136
 CELF tls5187.62719.94850123
Probands
 WISC PIQ9084.18919.33348134
 WISC VIQ9278.13016.97346126
 WORD spelling9280.01115.94748113
 WORD basic reading9283.43516.20446120
 WORD reading compreh.9176.39614.40540114
 CELF els9168.03312.11750108
 CELF rls9178.53819.35650131
 CELF tls9071.62215.42850115

Probands were classified by type of language problem at ages 7 and 14. Omitting the R-SLI category (no proband fell into this category at age 14 and only two were identified at age 7), Figs 1 and 2 show the probands’ profile of mean z-scores by type of language problem. For the classification made when the probands were 7 years of age, ER-SLI probands present a profile that remains uniformly lower than those of resolved and E-SLI probands. For the classification made at age 14, the profiles for the different types of probands were significantly different from each other, and ER-SLI probands were still the poorest performers on both the basic and the language batteries. It is also important to note that for the resolved probands the CELF els remained below expectation.

image

Figure 1. Mean z-score profiles for probands at age 14 by type of language problem at age 7. Vertical dashed bars represent 95% confidence intervals.

Download figure to PowerPoint

image

Figure 2. Mean z-score profiles for probands at age 14 by type of language problem at age 14. Vertical dashed bars represent 95% confidence intervals.

Download figure to PowerPoint

Rate of SLI in relatives: examining affectedness

Affectedness in relatives by relative type

On the basic battery 36% of relatives of SLI probands were affected (35% of adult relatives and 40% of child siblings). This difference was found to be non-significant (OR = 1.24, CI 0.68–2.27, P = 0.5).

The additional language battery was available for 53 of the 55 child siblings. Among these 53 the language battery identified all 20 children that were affected according to the basic battery, plus a further 11 not previously identified, giving a prevalence of 61%.

Affectedness in relatives and proband severity

First, probands were classified as severe or non-severe on the basis of their CELF tls at age 14 using a range of cut-offs (−2, −1.5 and −1.25 SD below the mean). Figure 3 shows the odds ratios for risk of affection among adult relatives and child siblings (for both basic and language batteries) for these three binary thresholds of proband severity. There is no single cut-off point of proband severity that showed uniformly more significant association across both adult relatives’ and child siblings’ measures. We therefore focused on the −1.25 SD cut-off, which is commonly used in research with SLI (Tomblin et al. 1996). Recall that for this sample, this threshold reflects more accurately the persistence of SLI (rather than severity per se) but for ease of reporting the severity terminology will be used consistently across the results. Considering the basic battery of tests we found an overall marginal significant association between rate of SLI in relatives and proband severity (OR = 2.06, CI 0.90–4.75, P = 0.09). While estimated differences for adult relatives were modest (OR = 1.7, CI 0.67–4.30, P = 0.3), for child siblings there was a stronger but still marginal association (OR = 3.7, CI 0.81–23.78, P = 0.1). Using the affectedness criterion from the language battery, child siblings of severe probands were found to be significantly more likely to be affected than those of non-severe probands (OR = 6.46, CI 1.46–35.11, P = 0.01). This corresponds to rates of 73% and 29% of affected siblings of severe and less severe probands, respectively.

image

Figure 3. Odds ratio for risk of SLI among adult relatives and child siblings for three binary thresholds of proband severity (−2, −1.5 and −1.25 SD below the mean of proband CELF total language score). Vertical dashed bars represent 95% confidence intervals.

Download figure to PowerPoint

Second, a continuous measure of proband severity (the negative of the z-scores of CELF tls) was used. We found a marginal (linear on the log-odds scale) association between proband severity and rate of SLI in relatives on the basic battery (OR = 1.38, CI 0.95–1.99, P = 0.09). Distinguishing by relative type, the odds ratios were marginally significant for adult relatives (P = 0.1) and not significantly different from 1 for child siblings (P = 0.4).

Affectedness of relatives and gender of proband

Interestingly, and against the expectation of the multifactorial model, we found that relatives of female probands were overall at a lower risk (OR = 0.31, CI 0.12–0.81, P = 0.02), a difference that was significant for both adult relatives (OR = 0.4, P = 0.05) and child siblings (OR = 0.21, P = 0.04). Being a female relative did not significantly lower risk either overall (P = 0.4), separately among adult relatives (P = 0.2) or among child siblings (P = 1). The incorporation of an interaction term between gender of relative and gender of proband was not significant for either group. After controlling for gender of proband and gender of relative, proband severity was marginally significant for adult relatives (P = 0.1) but not significant for child siblings (P = 0.3).

Controlling for IQ

A further important control involved performance IQ. We found that the performance IQ of probands was not associated with affectedness on the basic battery among adult relatives (P = 1) or child siblings (P = 0.3). Including performance IQ of the relative changed the analysis to one more akin to an analysis of verbal/non-verbal discrepancy. Performance IQ was significantly associated with language problems with an odds ratio of 0.95 (P < 0.001) for adult relatives and 0.86 (P < 0.001) for child siblings. While covariation for performance IQ increased the odds ratio for the association between proband severity and the rate of affectedness among adult relatives and child siblings, these remained non-significant (adult P = 0.4, child P = 0.2).

Affectedness in child siblings

To further explore the significant findings relating to child siblings (using the basic battery and the binary categorization of severity), we examined associations between proband severity and rate of affectedness in child siblings classified on the language battery. A significant simple association between the continuous measure of proband severity and affectedness rate was found (OR = 2.03, P = 0.01). When controlling for gender of relative and gender of proband we found no significant effect of gender of relative (P = 1) but a marginal association with gender of proband (OR = 0.25, P = 0.08). A marginally significant interaction term between gender of proband and gender of relative highlighted a much lower risk of language impairment for child sisters of female probands (OR = 0.04, P = 0.08). This result further underlines our finding of an associated lower risk among relatives of female probands. Furthermore, it is important to note that the relationship between proband severity and affectedness rate in child siblings remained significant even after controlling for gender of relative (P = 1), gender of proband (P = 0.8), proband performance IQ (P = 0.06) and relative performance IQ (P < 0.009).

Affectedness and proband severity at age 7 years

Proband severity score at age 7 showed no significant association with the rate of affectedness of adult relatives (P = 1) or child siblings (P = 0.9).

Affectedness and proband type of language problem

Considering first the classification of proband type of language problem at age 14, Table 2 reports the percentages of affected relatives by proband type of language problem (E-SLI and ER-SLI) and gender. Three sets of percentages (with frequency ratios) are reported: affected adult relatives, affected child siblings (both using the basic battery) and affected siblings when using the more specific language battery. An examination of the combined proportions (for both male and female probands across relative types) revealed that overall, relatives of ER-SLI probands were about twice as likely to be affected as relatives of E-SLI probands (OR = 1.88, CI 0.85–4.13, P = 0.1). Disaggregating by relative type and controlling for gender of relative and proband, the ER-SLI: E-SLI odds ratio was found to be significant for adult relatives (OR = 2.43, CI 1.05–5.62, P = 0.04) but not significant for child siblings (OR = 1.41, CI 0.33–6.36, P = 0.8). When child siblings were classified on the language battery the odds ratio was 3.12 but still not significant (P = 0.2).

Table 2.  Percentage of affected relatives (with frequencies in parentheses) by proband type at age 14
 Affected adult relativesAffected child siblingsAffected child siblings (classified on language battery)
  • *

    From logistic regression for each relative type which include control for gender of proband and also gender of relative. Disaggregating the table by gender of relative made it too sparse to easily interpret.

E-SLI
 male proband28.9% (15/52)58.3% (7/12)63.6% (7/11)
 female proband10% (1/10)0% (0/4)25% (1/4)
 combined25.8% (16/62)43.8% (7/16)53.3% (8/15)
ER-SLI
 male proband46.4% (32/69)57.9% (11/19)84.2% (16/19)
 female proband29.4% (5/17)30% (3/10)60% (6/10)
 combined43% (37/86)48.3% (14/29)75.9% (22/29)
Partial odds-ratio*OR = 2.43, P = 0.04OR = 1.41, P = 0.80OR = 3.12, P = 0.20

The negative of the z-scores of the probands’ CELF els and rls at age 14 were then used as continuous measures of proband expressive and receptive severity. Including them as predictors in the logistic model of relatives’ affectedness showed no overall meaningful specificity of effect among relatives. Distinguishing by relative type and controlling for gender of relative and gender of proband, we found that the rate of SLI among adult relatives was associated with proband receptive severity (OR = 1.64, CI 1.02–2.62, P = 0.04) but not expressive severity (P = 0.4). For child siblings both these measures of proband severity were not significant. This was found also when the alternative definition of affectedness for child siblings based on the language battery was used. Furthermore, the significant association between proband receptive severity and rate of SLI in adult relatives remained significant even after including performance IQ of both proband and relative as additional covariates into the model (OR = 2.09, P = 0.04).

Table 3 presents percentages (and frequency ratios) of affected relatives by proband type of language problem and gender when the proband type of language problem was determined at age 7. The format of Table 3 is identical to that of Table 2.

Table 3.  Percentage of affected relatives (with frequency ratios in parentheses) by proband type at age 7
 Affected adult relativesAffected child siblingsAffected child siblings (classified on language battery)
E-SLI
 male proband49.1% (27/55)52.9% (9/17)68.8% (11/16)
 female proband14.3% (2/14)42.9% (3/7)42.9% (3/7)
 combined42% (29/69)50% (12/24)60.9% (14/23)
ER-SLI
 male proband34% (16/47)38.5% (5/13)58.3% (7/12)
 female proband28.6% (4/14)0% (0/8)50% (4/8)
 combined32.8% (20/61)23.8% (5/21)55% (11/20)
Partial odds-ratio*OR = 0.68, P = 0.36OR = 0.35, P = 0.18OR = 0.83, P = 0.99

No association with the rate of affectedness among either adult relatives or child siblings was found for proband expressive and receptive severity measured when probands were 7 years of age.

Analysis of mean score profiles

An exploratory analysis of mean score profiles of relatives was carried out using test norms as comparisons. Ideally, such comparisons should have been carried out in relation to a common control group. The lack of such a control group is a limitation of the present study. Thus, the results presented in this section should be interpreted with caution and be considered suggestive of potential hypotheses that need to be pursued in future research.

Mean scores by relative type

Figure 4 shows mean z-scores by test and type of relative. For performance and verbal IQ, spelling and basic reading the profile for adult relatives was significantly different from what would be expected of population norms (F(4,70) = 28.71, P < 0.001), with performance IQ being 0.68 SD higher (P < 0.001) but spelling 0.25 SD lower (P = 0.03).

image

Figure 4. Mean z-score profiles by type of relative. Vertical dashed bars represent 95% confidence intervals adjusted for the clustering.

Download figure to PowerPoint

The profile for child siblings over the same four-measure basic battery was quite different. All scores, except basic reading (P = 0.2), were significantly lower than expectation (P = 0.02 for performance IQ; P = 0.04 for spelling; P = 0.02 for verbal IQ). The depressed profile also extended to the additional measures not available on adult relatives (the language battery), where expressive language was 1.05 SD below (P < 0.001), receptive language was 0.48 SD below (P = 0.05) and reading comprehension was 0.82 SD below (P < 0.001) population norms. Comparison of child relatives with adult relatives across the three shared language-related measures showed a significant difference (F(3,74) = 3.14, P = 0.03), with performance and verbal IQ being significantly higher in adult relatives than in child relatives (P < 0.001 and P = 0.01, respectively).

Mean score profiles and proband severity

For adult relatives each of the three measures in the basic battery was lower for relatives of probands with higher severity scores (P = 0.02 for verbal IQ, P = 0.1 for spelling, P = 0.07 for basic reading). These associations with severity scores remained after controlling for gender and performance IQ of both relative and proband (P = 0.02 for verbal IQ, P = 0.1 for spelling, P = 0.03 for basic reading). The six-measure language and literacy profile available for child siblings (from the basic and the language batteries) showed no initial association with proband overall severity score at age 14, but after adjustment for both gender and performance IQ (of both proband and relative) significantly lowered profiles were found (F(6,40) = 4.36, P = 0.002; individual tests: P = 0.003 for verba IQ, P = 0.009 for spelling, P = 0.001 for basic reading, P = 0.001 for reading comprehension, P = 0.001 for CELF els and P = 0.01 for CELF rls). In these analyses results showed that for both groups of relatives a higher personal performance IQ was associated with higher language scores, but a higher proband performance IQ was associated with lower language scores.

For the proband severity measure based on the assessment of probands at age 7, no associations were found among the scores of adult relatives or child siblings.

Mean score profiles and proband type of language problem

Proband expressive severity and proband receptive severity at age 14 were examined separately. After accounting for performance IQ and gender each showed a significant association with the profiles of adult relatives and child siblings. However, when expressive and receptive severity were considered jointly the associations were not significant.

Similar results were found when considering proband expressive and receptive severity at age 7.

After controlling for performance IQ and gender of both proband and relative, the scores of relatives of probands classified as ER-SLI at age 14 were always lower than the scores of relatives of E-SLI probands, though few differences were individually significant (adult relatives: P = 0.2 for verbal IQ, P = 0.02 for spelling and P = 0.2 for basic reading; child siblings: P = 0.4 for verbal IQ; P = 0.4 for spelling, P = 0.9 for basic reading, P = 0.2 for reading comprehension, P = 0.3 for CELF els and P = 0.03 for CELF rls).

With the classification of the type of language problem of the proband based on measures assessed when the probands were 7 years old, no significant differences in profiles were found for adult relatives or child siblings.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

This study confirms the familial nature of specific language impairment (Stromswold 1998). Rates of language problems were elevated among first-degree relatives and the relationship of proband severity with rates of affectedness of relatives was broadly confirmed. Familial effects were evident among adult relatives for verbal IQ and measures of literacy, but for child siblings the most marked effects were for direct measures of language, measures that were unavailable in standardized form for adults. Controlling for the performance IQ of the relative to examine discrepancy made it evident that these effects were not explained by relatives’ performance IQ. Nonetheless, though adult relatives’ performance IQ scores were not depressed when compared to somewhat dated norms, those of child siblings were lower than the population norms. These patterns did not change when account was taken of probands’ performance IQ.

In contrast to Lahey & Edwards (1995) we found increased affectedness among relatives where the proband had both receptive and expressive problems rather than expressive problems alone. It should be noted that Lahey & Edwards’ sample was small and ranged widely in age from 4;0 to 9;6 years. Thus, this study covered a wide developmental period within a cross-sectional design. In addition, these authors did not examine directly the issue of severity of types of problems. Interestingly, in our study, the difference in results between relatives of probands with ER-SLI versus E-SLI was not explainable by any difference in the severity of expressive problems of these groups of probands, nor did it appear, when probands were characterized quantitatively by receptive and expressive severity, that receptive problems were consistently more familial than expressive problems.

Proband severity at age 7 showed little association with affectedness among any of the groups of relatives, nor did the proband ER-SLI versus E-SLI typology show any specificity. Thus, even though relatives of those with SLI impairment at age 7 are at increased risk of language or literacy impairments themselves, variation of proband severity at this age was not associated with variation shown among relatives. The lack of association, in contrast to that shown when measured at age 14, could reflect a preponderance of transient language impairments lacking any familial aetiology being shown when the probands were younger. However, it may also be because our probands were less variable at that age. Age 7 was the age of recruitment into the sample and all the children were drawn from language units, i.e. specialist language classrooms providing educational support. Although far from homogeneous at that age, variation in progress from age 7 to age 14 is likely to have increased that heterogeneity, providing greater statistical power to detect (non-group based) familial effects. By age 14, some subjects with less persistent/severe impairments appeared to belong to families with lower levels of affectedness. Language development is in part a social process, in which the family provides a significant part of the learning environment. Within this context, the association we found at age 14 may reflect the social processes involved in language learning. Alternatively, it may be that the more complex language elements being tested at age 14 are more heritable. Thus, the differences observed may constitute a developmental trend by which heritable influences, cultural or genetic, increase with age. It might therefore be unwise, at this stage, to attempt to impose too simple a model on these findings.

Finally, our expectation under the multifactorial model for higher levels of affectedness to be found among relatives of female rather than male probands was entirely disproved. In contrast, in our analyses we found that relatives of female probands were less likely to be affected than those of male probands.

The standard multifactorial threshold liability model makes strong assumptions, allowing a gender difference in the threshold or mean of the liability distribution but not the variance, and heritability that does not change with severity. Viding et al. (2004), using 4-year-old twins found increasing group heritability with severity, a gender difference in mean (0.189 SD) but not variance in the distribution language impairment scores and in addition no evidence for gender-specific genes. Participant age, study design and measurement differences make direct comparison difficult (our own study had insufficient power to properly assess severity invariant familiality), but their findings would probably suggest an expectation of a gender of proband effect in our context. Its absence is therefore striking. Our data are thus suggestive of a different mechanism or level of risk factor exposure in females vs. males in SLI. It is also interesting to note that in other developmental disorders, such as autism and mental retardation, the hypothesis that the sex that is the least frequently affected by these disorders (female) is also the relatively more severely affected, has not received strong support (Eme 1992). However, unlike autism where affected relatives show the discrepant gender ratio of that observed in the general population, our study showed no gender of relative differences in rates. Thus, it appears that future research is needed to investigate more fully the role of gender in the familiality in SLI.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments
  • American Psychiatric Association (1994) Diagnostic and Statistical Manual of Mental Disorders IV. American Psychiatric Association, Washington D.C., USA.
  • Beitchman, J. H., Hood, J. & Inglis, M. A. (1992) Familial transmission of speech and language impairment. Can J Psychiatr 37, 151156.
  • Bishop, D. V. M. (1989) Test for Reception of Grammar, 2nd Edn. Age and Cognitive Performance Research Centre, University of Manchester, Manchester.
  • Bishop, D. V. M. (1997) Uncommon Understanding: Development and Disorders of Language Comprehension in Children. Psychology Press, Hove.
  • Bishop, D. V. M. & Snowling, M. J. (2004) Developmental dyslexia and specific language impairment: same or different? Psychol Bull 130, 858886.
  • Byrne, B. M., Willerman, L. & Ashmore, L. L. (1974) Severe and moderate language impairment: evidence of distinctive aetiologies. Behav Genet 4, 331345.
  • Choudhury, N. & Benasich, A. A. (2003) A family aggregation study: the influence of family history and other risk factors on language development. J Speech Lang Hear Res 46, 261272.
  • Conti-Ramsden, G. & Botting, N. (1999) Classification of children with specific language impairment: longitudinal considerations. J Speech Lang Hear Res 42, 11951204.
  • Conti-Ramsden, G., Botting, N., Simkin, Z. & Knox, E. (2001) Follow-up of children attending infant language units: outcomes at 11 years of age. Int J Lang Commun Disord 36, 207219.
  • Conti-Ramsden, G., Crutchley, A. & Botting, N. (1997) The extent to which psychometric tests differentiate subgroups of children with specific language impairment. J Speech Lang Hear Res 40, 765777.
  • Cytel Software Corporation (2004) LogXact: Software for Exact Logistic Regression. Cytel Software Corporation, Cambridge.
  • Elliot, C. D. (1983) British Ability Scales. NFER-Nelson, Windsor.
  • Eme, R. F. (1992) Selective female affliction in the developmental disorders of childhood: a literature review. J Clin Child Psychol 21, 354364.
  • Falconer, D.S. (1965) The inheritance of liability to certain diseases, estimated from the incidence among relatives. Ann Hum Gen 29, 5176.
  • Flax, J. F., Realpe-Bonilla, T., Hirsch, L., Brzustowicz, L. M., Barltlett, C. W., & Tallal, P. (2003) Specific language impairment in families: evidence for co-occurrence with reading impairments. J Speech Lang Hear Res 46, 530543.
  • Huber, P. J. (1967) The behaviour of maximum likelihood estimators under non-standard conditions. Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1, 221233. University of California Press, Berkeley.
  • Johnston, R., Stark, R., Mellits, E., & Tallal, P. (1981) Neurological status of language-impaired and normal children. Ann Neurol 10, 159163.
  • Kirk, S. A., McCarthy, J. J. & Kirk, W. D. (1968) Illinois Test of Psycholinguistic Abilities. University of Illinois, Urbana, IL.
  • Kovas, Y., Hayiou-Thomas, M. E., Oliver, B., Bishop, D. V. M., Dale, P. S. & Plomin, R. (2005). Genetic influences in different aspects of language development: the etiology of language skills in 4.5-year-old twins. Child Dev 76, 632651.
  • Lahey, M. & Edwards, J. (1995) Specific language impairment: preliminary investigation of factors associated with family history and with patterns of language performance. J Speech Hear Res 38, 643657.
  • Lai, C. S. L., Fisher, S. E., Hurst, J. A., Vargha-Khadem, F. Monaco, A. P. (2001) A forkhead-domain gene is mutated in a severe speech and language disorder. Nature 413, 519523.
  • Leonard, L. B. (1998) Children with Specific Language Impairment. MIT Press, Cambridge, MA.
  • Liang, K.-Y., & Zeger, S. L. (1986) Longitudinal data analysis using generalized linear models. Biometrika 73, 1322.
  • Little, R. A., & Rubin, D. B. (1987) Statistical Analysis with Missing Data. John Wiley and Sons, New York.
  • Mehta, C. R. & Patel, N. R. (1995) Exact logistic regression: theory and examples. Stat Med, 14, 21432160.
  • Melnick, M., Bixler, D., Fogh-Andersen, P & Coneally, P.M. (1980) Cleft lip +/− cleft palate: an overview of the literature and an analysis of Danish cases born between 1941 and 1968. Am J Med Genet 6, 8397.
  • Neuhaus, J. M., Kalbfleisch, J. D. & Hauck, W. (1991) A comparison of cluster specific and population averaged approaches for analysing correlated binary data. Int Stat Rev 59, 2535.
  • Neils, J. & Aram, D. (1986) Family history of children with developmental language disorders. Percep Mot Skills 63, 655658.
  • Renfrew, C. (1991) The Renfrew Language Scales: Bus Story Test. Winslow Press Ltd, Bicester, UK.
  • Rice, M. L., Haney, K. R. & Wexler, K. (1998) Family histories of children with SLI who show extended optional infinitives. J Speech Lang Hear Res 41, 419432.
  • Royall, R. M. (1986) Model robust confidence intervals using maximum likelihood estimators. Int Stat Rev 54, 221226.
  • Rust, J., Golombok, S., & Trickey, G. (1993) Wechsler Objective Reading Dimensions. The Psychological Corporation, San Antonio, TX.
  • Semel, E., Wiig, E., & Secord, W. (1987) Clinical Evaluation of Language Fundamentals – Revised. The Psychological Corporation, San Antonio, TX.
  • Snowling, M., Bishop, D. V. M. & Stothard, S. E. (2000) Is preschool language impairment a risk factor for dyslexia in adolescence? J Child Psychol Psychiatr 41, 407418.
  • StataCorp (2003) Stata Statistical Software: release 8.0. Stata Corporation, College Station, TX.
  • Stromswold, K. (1998) Genetics of spoken language disorders. Hum Biol 70, 297324.
  • Tallal, P., Hirsch, L. S., Realpe-Bonilla, T., Miller, S., Brzustowicz, L. M., Bartlett, C. & Flax, J. F. (2001) Familial aggregation in specific language impairment. J Speech Lang Hear Res 44, 11721182.
  • Tallal, P., Ross, R. & Curtiss, S. (1989) Unexpected sex-ratios in families of language/learning-impaired children. Neuropsychologia 27, 987998.
  • Tomblin, B. J. (1989) Familial concentration of developmental language impairment. J Speech Hear Dis 54, 287295.
  • Tomblin, J. B., Records, N. L., Buckwalter, P., Zhang, X., Smith, E. & O’Brien, M. (1997) Prevalence of specific language impairment in kindergarten children. J Speech Lang Hear Res 40, 12451260.
  • Tomblin, J. B., Records, N. L. & Zhang, X. (1996) A system for the diagnosis of specific language impairment in kindergarten children. J Speech Hear Res 39, 12841294.
  • Viding, E., Spinath, F.M., Price, T.S., Bishop, D.V.M., Dale, P.S. & Plomin, R. (2004) Genetic and environmental influence on language impairment in 4-year-old same sex and opposite-sex twins. J Child Psychol Psychiatr 45, 315325.
  • Wechsler, D. (1986) Wechsler Adult Intelligence Scale – Revised, UK. The Psychological Corporation.
  • Wechsler, D. (1992) Wechsler Intelligence Scale for Children – Third Edition – Revised, UK. The Psychological Corporation, San Antonio, TX.
  • Whitehurst, G. J., Arnold, D. S., Smith, M., Fischel, J. E., Lonigan, C. J. & Valdez-Menchaca, M.C. (1991) Family history in developmental expressive language delay. J Speech Hear Res 34, 11501157.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Acknowledgments

The authors gratefully acknowledge the support of the Wellcome Trust (Grant 060774) and the Nuffield Foundation (Grants AT251[OD] and DIR/28). Thanks also go to Helen Betteridge, Emma Knox and Catherine Pratt for their help with data collection. The authors would also like to thank the schools and families who helped us with this research.