Abstract
The aim of this Campbell systematic review is to examine the effectiveness of speech and language interventions for children with primary speech and language delay/disorder.
Thirty-six studies were found, based on a total of 1,500 children involved in thirty-three randomised controlled trials.
This Review concludes, on the basis of the available international studies in this field, that speech and language therapy generally has a positive effect on children with expressive phonological difficulties. The therapy also helps children who have a problem with their active vocabulary, i.e. children who have difficulty in using words they understand. Children, who have difficulty with active vocabulary alone, but no problem understanding and acquiring language, also benefit from speech and language therapy. Speech and language therapy for children who have difficulty in applying grammar and constructing sentences is, however, only thought to have a positive effect if the child does not have other significant receptive speech and language difficulties.
Synopsis
Primary speech and language delay/disorder is a common developmental difficulty which, if unresolved, can cause difficulties of both learning and socialisation lasting into adolescence and beyond. This review examines the effectiveness of speech and language therapy interventions for children with primary speech and language delay/disorder.
The review concludes that whilst there may be some support for the effectiveness of speech and language therapy for children with expressive phonological and expressive vocabulary difficulties, the evidence concerning the effectiveness of interventions for expressive syntax is mixed, and no evidence is available concerning interventions for children with receptive language difficulties.
Abstract
Background
It is thought that approximately 6% of children have speech and language difficulties of which the majority will not have any other significant developmental difficulties. Whilst most children's difficulties resolve, children whose difficulties persist into primary school may have long-term problems concerning literacy, socialisation, behaviour and school attainment.
Objectives
To examine the effectiveness of speech and language interventions for children with primary speech and language delay/disorder.
Search strategy
The following databases were searched: The Cochrane Controlled Trials Register (Cochrane Library, CENTRAL: 2002/3), CINAHL (1982 – July 2002), EMBASE (1980 – Sept Week 4 2002), ERIC (1965 – 2002), MEDLINE (1966 – Sept Week 3 2002), PsycINFO (1872 – 2002/10 Week 2), The National Research Register (2002/3). In addition to this references were taken from reviews of the literature and reference lists from articles.
Selection criteria
The review considered randomised controlled trials of speech and language therapy interventions for children or adolescents with primary speech and language delay/disorder.
Data collection & analysis
Titles and abstracts were identified and assessed for relevance, before the full text version was obtained of all potentially relevant articles. The data were categorised depending on the nature of the control group and considered in terms of the effects of intervention on expressive and receptive phonology, syntax and vocabulary. The outcomes used in the analysis were dependent on the focus of the study with only the primary effects of therapy being considered in this review.
Main results
The results of twenty-five studies were used in the meta-analysis. The results suggest that speech and language therapy is effective for children with phonological (SMD=0.44, 95%CI: 0.01,0.86) or vocabulary difficulties (SMD=0.89, 95%CI: 0.21,1.56), but that there is less evidence that interventions are effective for children with receptive difficulties (SMD=−0.04, 95%CI: −0.64,0.56). Mixed findings were found concerning the effectiveness of expressive syntax interventions (n=233; SMD=1.02, 95%CI: 0.04–2.01). No significant differences were shown between clinician administered intervention and intervention implemented by trained parents, and studies did not show a difference between the effects of group and individual interventions (SMD=0.01, 95%CI: −0.26,1.17). The use of normal language peers in therapy was shown to have a positive effect on therapy outcome (SMD=2.29, 95%CI: 1.11,3.48).
Reviewers' conclusions
The review shows that overall there is a positive effect of speech and language therapy interventions for children with expressive phonological and expressive vocabulary difficulties. The evidence for expressive syntax difficulties is more mixed, and there is a need for further research to investigate intervention for receptive language difficulties. There is a large degree of heterogeneity in the results, and the sources of this need to be investigated.
Background
Speech and language delay/disorder is a common developmental difficulty in childhood. It may present either as a secondary difficulty (where it can be accounted for by a primary condition such as: autism, hearing impairment, general developmental difficulties, behavioural or emotional difficulties or neurological impairment), or it may be considered primary when it cannot be accounted for by any other condition (Stark 1981, Plante 1998). Whilst prevalence figures for speech and language difficulties as a whole vary from 1–15% (Law 2000) depending on inclusion criteria, it is thought that on average approximately 6% of children may have speech and language difficulties (Boyle 1996), of which a significant proportion will have primary speech and language difficulties. It is recognised that there may be overlap between primary and secondary delay/disorders especially when the features of the primary disability are less pronounced but this distinction remains clinically useful and it is one commonly reported in the literature (Leonard 1998).
Presentation of primary speech and language delay/disorder is heterogeneous. Difficulties may be persistent or transient and present as delayed or disordered speech, expressive or receptive language difficulties or a combination of these. There is little consensus on the aetiology of primary language delay/disorder but a number of studies suggest correlations with multiple risk factors such as chronic otitis media, genetic factors, socio-economic status, difficulties in pregnancy, and oralmotor difficulties (Whitehurst 1991, Tomblin 1997). It may be that these act in a cumulative fashion to increase the severity of the presenting disorder (Aram 1980). From current evidence, it is unclear whether primary speech and language delay/disorder represents varying levels of a single condition or a number of different conditions with diverse aetiologies but similar presenting patterns (Law 1998).
Primary speech and language delay/disorder is of significant concern to those involved with child development and has far reaching implications for the child, parent and carer. Studies indicate that language delay may have adverse effects upon school achievement (Aram 1984, Bishop 1990, Catts 1993, Tallal 1997, Baker 1987) and/or be associated with social, emotional and behavioural problems (Huntley 988, Rice 1991, Rutter 1992, Cohen 2000, Stothard 1998). Children with primary language delay/disorder can also have long-term difficulties which persist to adolescence and beyond (Haynes 1991, Rescorla 1990) with some 30–60% experiencing continuing problems in reading and spelling. Therefore, primary speech and language delay/disorder, has the potential to impact significantly on the individual, family and society in both the short and long term.
Because of the range of difficulties within the diagnosis primary speech and language delay/disorder interventions may take many forms. Law et al (2000) define approaches to intervention using three categories: didactic (direct training of linguistic behaviours), naturalistic (teaching the child to respond effectively to the linguistic demands of the environment) and hybrid (a combination of didactic and naturalistic). These terms provide a useful framework to outline approaches to intervention, although, a therapist may use an eclectic approach and adapt interventions according to the response of the client.
Intervention may take place in many different environments, for example, the home, school or clinic and will vary in duration and intensity dependent on the resources available, perceived needs of the child and policies of individual speech and language therapy services. Intervention may also be delivered indirectly through a third person or directly through the clinician. Direct intervention focuses on the treatment of the child either individually or within a group of children depending on the age and needs of the children requiring therapy and the facilities available. Indirect intervention is often perceived to be a more naturalistic approach where adults in the child's environment facilitate communication. Traditionally this approach is used to create an optimum communicative environment for the child by promoting positive parent-child interaction. Indirect approaches are increasingly being employed within a range of settings, where speech and language therapists train professionals and carers who work with the children and provide programmes or advice on how to maximise the child's communicative environment and enhance communicative attempts.
There are no universal guidelines on what type of intervention to offer children with primary speech and language delay/disorder or on its’ timing, nor is there consistent evidence upon which to base a decision, meaning that the decision is often left to individual therapists and services. Whilst a number of reviews have been carried out in order to summarise the literature, these have largely been narrative and non-systematic and therefore may be prone to bias and inaccuracies (Enderby 1996, Law 1997, Leonard 1998, McLean 1997, Olswang 1998). Two reviews have also included a meta-analysis (Nye 1987, Law 1998). In the former a variety of study designs were included in the analysis which may have affected the accuracy of the results due to potential bias from the inclusion of poorer quality study designs. In the second review only studies that compared speech and language therapy interventions with no treatment were included, and therefore the conclusions could not make distinctions between the relative effects of different treatments in treating the similar participants. This systematic review is based on randomised controlled trials that are methodologically least prone to potential sources of bias, and it also considers evidence from randomised controlled trials that compare traditional and experimental treatments. In doing this it extends the evidence base and has the potential to help practitioners decide between different types of intervention.
The aim of this review is to search systematically for, and combine evidence from, randomised controlled trials (RCTs) relating to the effectiveness of intervention approaches for primary speech and language delay/disorder in children and adolescents.
Objectives
To determine the effectiveness of speech and language intervention for children with a primary diagnosis of speech/language delay/disorder.
Criteria for considering studies for this review
Types of studies
Randomised controlled trials of interventions for primary speech and language delay/disorder.
Types of participants
Children or adolescents with a diagnosis of primary speech and language delay/disorder. No upper age limit was set during the searches, however the oldest participants in studies identified for the review were fifteen. Children whose difficulties arose from stuttering or whose difficulties were described as learned misarticulations e.g. lateral /s/ or labialised /r/ were excluded from this review.
Types of interventions
Any type of intervention designed to improve an area of speech or language functioning concerning either expressive or receptive phonology (production or understanding of speech sounds), expressive or receptive vocabulary (production or understanding of words), or expressive or receptive syntax (production or understanding of sentences and grammar). Three comparisons were investigated: Interventions compared to delayed or no treatment conditions e.g. studies where control children are assigned to a delayed treatment or no treatment condition. Interventions compared to general stimulation conditions e.g. studies where control children are assigned to a control condition designed to mimic the interaction found in therapy without providing the target linguistic input. This may be a cognitive therapy, general play sessions or a speech and language therapy treatment that does not focus on the area of interest in the study. Interventions compared to other speech and language therapy approaches e.g. studies may compare what they consider to be a ‘traditional treatment’ with what they consider to be an experimental treatment. This may be a different approach carried out by the same person e.g. targeting early versus late developing sounds, or the same approach carried out by different people e.g. focused stimulation given by clinicians versus parents.
A general rule was applied to studies comparing treatments: Group therapy was considered to be experimental and individual therapy the control Indirect interventions were considered to be experimental and direct interventions the control Interactive approaches were considered to be experimental and directive approaches the control
If conditions could not be categorised according to these rules, then the conditions were labelled as experimental or traditional according to background information in the literature review.
Types of outcome measures
Types of outcome measure included formal standardised tests, criterion referenced tests, parent report and language samples. Areas measured included aspects of expressive or receptive language functioning in areas of semantics, syntax, and phonology. Non-linguistic outcomes such as behaviour, esteem, and literacy measures were not used in this review.
Outcomes used in the review were dependent on the focus of the intervention e.g. for studies specifying intervention for expressive difficulties effect size was calculated using expressive language measures.
Outcomes for analysis were considered on three levels: At the level of the target for intervention e.g. improvement in targeted speech sounds or syntactic structure. At the level of language functioning e.g. improvement in overall phonological or expressive language maturity. At the level of broader language functioning e.g. improvement in intelligibility, improvement in parent report of language.
Search strategy for identification of studies
Studies were identified from the following sources: The Cochrane Controlled Trials Register (CCTR) (Cochrane Library, central: 2002/3) Cumulative Index of Nursing and Allied Health (1982 – July 2002) EMBASE (1980 – Sept Week 4 2002) ERIC (1965 – 2002) MEDLINE (1966 – Sept Week 3 2002) PsychINFO (1872 – 2002/10 Week 2) The National Research Register (2002/3) C2PSPECTRE (Database to December 2002)
The following search terms were used to identify articles. MESH terms were adapted to suit the fields of each of the databases: randomi∗ clin∗ trial∗ (clin∗ adj trial∗) singl∗ doubl∗ tripl∗ trebl∗ mask∗ blind∗ (5 or 6 or 7 or 8) and (9 or 10) crossover random∗ allocate∗ assign∗ random∗ adj (allocat∗ or assign∗) RANDOM ASSIGNMENT exp CLINICAL TRIALS exp META ANALYSIS 16 or 12 or 11 or 4 or 1 or 17 or 18 or 19 CHILD child∗ or infant∗ or baby or babies or boy∗ or girl∗ or pre-school∗ or preschool∗ or teen∗ or adolescen∗ or schoolchild∗ 21 or 22 COMMUNICATION DISORDERS speech near disorder∗ speech near delay∗ language near disorder∗ language near delay∗ verbal near disorder∗ aprosodi∗ cluttering dysglossia rhinolalia central and auditory and processing and disorder semantic-pragmatic and disorder 25 or 26 or 27 or 28 or 29 or 30 or 31 or 32 or 33 or 34 or 35 speech near screen∗ language near screen∗ speech near patholog∗ language near patholog∗ speech near therap∗ language near therap∗
Further studies were identified through reference lists from articles and reviews of the literature.
Methods of the review
Selection of trials
All references from the search strategy were managed using reference manager. The titles and abstracts were screened against the inclusion criteria by the second author (ZG) in order to identify relevant articles. If it was unclear whether an article met the inclusion criteria then the full text version was obtained and the article more fully assessed. Once all potentially appropriate studies had been obtained, each study was evaluated independently by two reviewers (JL) and (ZG) for inclusion. In the event that there was a question as to the possible inclusion or exclusion of any individual study, a final consensus decision was made between JL and ZG. If the primary reviewers could not come to a consensus regarding the inclusion or exclusion of a study, the full article was submitted to the third reviewer (CN). Reviewers were not blinded to the name(s) of the author(s), institution(s) or publication source at any level of review.
Initially the studies were judged on the basis of their allocation concealment. Two reviewers independently assigned one of three quality codes as described in the Cochrane Collaboration Handbook (Clarke 2003): Indicates adequate concealment of allocation e.g. pre-numbered or coded identical containers administered serially to participants. Indicates uncertainty about whether the allocation was adequately concealed e.g. where authors did not describe the allocation methods. Indicates that the allocation was not adequately concealed e.g. alternate assignment.
For this review only studies considered to be of quality (A) or (B) were included in the review. The studies were then critically appraised by asking the following questions. Were the assessors blind to treatment allocation? Did the authors report a range of baseline characteristics for the participants? If baseline characteristics were reported and tested using significance tests, were significant differences between the groups found? Is there an explanation of why children withdrew? How were the data from these children used? What was the proportion of withdrawals? Was an intention-to-treat analysis used? Were the eligibility criteria adequately defined? Was a power calculation completed to ensure there were adequate participants to see a significant effect?
The studies were labelled using an A/B/C categorisation for each of the above areas: Component mentioned and adequate Component not reported Component mentioned but not adequate
A description of the quality codes for each of the appraised areas is provided in the methodological quality table (Table 01; Table 02).
Data management
Two reviewers independently extracted the data from the articles using a form covering the following: Methods, Participants, Interventions Outcomes.
Uncertainty and disagreement was resolved through discussion and consultation with the third reviewer. If further information was required then the first author of the study was contacted. In total fifteen authors were contacted to ask for further information or clarification, and eleven responses were received.
Due to the heterogeneity of controls used in the studies and the potential impact of this on effect size, the studies were initially categorised on the basis of their control groups. Three main comparisons were made: Speech and Language Therapy Intervention versus No Treatment Controls Speech and Language Therapy Intervention versus General Stimulation Speech and Language Therapy Intervention versus Traditional Therapies
For each of the comparisons outcomes were analysed in seven categories, where measured: Expressive phonology outcomes Receptive phonology outcomes Expressive syntax outcomes Receptive syntax outcomes Expressive vocabulary outcomes Receptive vocabulary outcomes Composite language outcomes
And each considered at three levels: At the level of the target for intervention At the level of overall language or speech maturity At the level of broader functioning e.g. improvement in conversation
Data analysis
Missing Data
Authors were contacted and asked to supply missing data. Where authors could not be contacted or data could not be supplied, missing data and drop-outs were assessed for each individual study. The number of participants in the final analysis, as a proportion of all participants in each study is reported in the tables concerning methodological quality (Table 01; Table 02).
Multiple Treatment Arms
Nine studies combined more than one eligible intervention versus control intervention (Fey 1993; Gibbard 1994b; Lancaster 1991; Law 1999; Shelton 1978; Sommers 1964; Sommers 1966; Sommers 1962; Tufts 1959). With the exception of Shelton (1978), all of these studies compared clinician intervention with parent intervention or additional parental intervention. In the study by Shelton (1978) the authors compared two parent interventions with no intervention. The data from all these studies were analysed by pooling the number of participants, means and standard deviations using a pooled standard deviation formula which weighted the conditions on the basis of the number of participants in each trial arm. Sub-group analyses were then used to separate the data in order to compare clinician and parent intervention.
Measures of Treatment Effect
Binary Outcomes
None of the outcomes used in the analysis were reported as binary outcomes.
Continuous Outcomes
Studies reporting continuous outcomes were summarised using standardised mean differences (SMD) to allow the reviewers to combine studies that measured the same outcome using different methods.
Assessment of Heterogeneity
Statistical heterogeneity was assessed using the Chi-squared test for heterogeneity, and through visual inspection of forest plots. Analysis of heterogeneity investigated the potential impact of differences in child language characteristics, duration of treatment and administrator of treatment. Where excessive heterogeneity was found then the studies were not pooled in the analysis.
Data Synthesis
Data synthesis was carried out using Review Manager 4.1.1. A random effects model was used in the analysis so as to take into account the variation in the data due to potential study differences.
Sub-group Analysis
We planned and undertook sub-group analyses to assess the impact of the following study characteristics: The role of administrator. Studies using clinicians as the administrators of intervention were examined separately and the results compared to the primary analysis. The role of duration of treatment as an indicator of efficacy studies. Studies offering longer lengths of therapy were examined separately and the results compared to the primary analysis. It was not possible to analyse the results in terms of intensity as only nine studies that were not comparable included intensities of therapy over two hours a week. The role of age. It was not possible to complete a sub-group analysis based on the age of the participants because only six studies (which were not comparable) included children over the age of seven.
We undertook one post-hoc subgroup analysis of expressive syntax and vocabulary outcomes removing a trial that specified that all children taking part in the study had severe receptive language difficulties. This was carried out to examine unexplained heterogeneity in the results.
Sensitivity Analysis
Sensitivity analyses were considered in order to take into account the potential effects of study quality. Randomisation, attrition and blinding were considered as potentially important factors (Juni 2001): Randomisation: Only three studies reported their methods of randomisation so sensitivity analysis was not carried out. Attrition: No study reported more than 15% attrition, but a fourteen studies did not report attrition levels. These studies were removed from the analysis and the results compared to the primary analysis Blinding: Studies that reported the assessors were blind to group status were examined separately and the results compared to the primary analysis
Publication bias
Since there is a risk of bias associated with relatively low sample size, funnel plots were used to evaluate the relationship between effect size and study precision. Such a relationship could be due to publication or related biases or due to systematic differences between small and large studies. If a relationship was identified the clinical diversity of the studies was further examined for possible explanations (Egger 1997).
Description of studies
In total 630 citations were found from the database searches. On the basis of the titles and abstracts 49 were judged to be potentially relevant and the full text articles were obtained. Reasons for excluding papers generally concerned the research design or the nature of the difficulties of the participants. Having read the articles 21 were judged to have met the inclusion criteria. This number was augmented by a further eight articles identified from systematic reviews of the literature, six articles identified by the review authors and one article from the Campbell Collaboration trials database (C2SPECTRE). In total searches identified 36 papers for the review, reporting a total of 33 different trials (see the table of included studies).
The studies generally focused on a single aspect of therapy either phonology, or expressive or receptive language. Seven studies had a less defined focus of therapy and stated that they included more than one aspect of language as an aim for therapy. Two of these studies addressed the Ward Infant Language Screening Test: Aassessment, Acceleration and Remediation (WILSTAAR) early intervention programme (Ward 1994) and provided a composite language measures of receptive and expressive language development (Evans [forthcoming], Sutton 1999), three were English trials run within health care trusts (Barratt 1992, Glogowska, 2000, Law 1999), and the sixth and seventh studies did not clearly detail the interventions in terms of expressive and receptive components (Cole 1986, Head 1975). The study by Cole et al stated that intervention focused on syntax, semantics and pragmatics whilst comparing a directive to an interactive intervention approach, while that of Head et al compared using recreational games as an intervention approach to using perceptual motor activities. Neither study could be easily categorised due to the quality of reporting, and so the aims of therapy had to be determined by visual inspection of participants baseline expressive and receptive language scores.
The nature of the interventions were described in varying amounts of detail. While the area of focus, duration, intensity and administrator were most frequently described, other aspects of the intervention such as the tasks used and the relative proportions of different techniques were often not described. Although this did not affect the quality of the individual studies, as it is most commonly an artifact of the writing-up process, it did affect our ability to draw judgements concerning the comparability of the studies and will affect the practical application of techniques used in the studies to clinical practice.
Sixteen studies investigated either expressive or receptive phonology interventions. Five of these studies compared intervention provided by the clinician to delayed treatment (Almost 1998, Glogowska, 2000, Matheny 1978, Munro 1998, Reid 1996), three studies compared parent intervention and delayed intervention of which two used multiple conditions to compare parent and clinician intervention (Lancaster 1991, Shelton 1978, Tufts 1959), and a further four studies compared clinician intervention with and without additional parental intervention (Fudala 1972, Ruscello 1993, Sommers 1962, Sommers 1964). Four studies compared specific approaches to therapy investigating the differences in outcomes between using group and individual therapy (Sommers 1966), targetting either earlier or later developing speech sounds (Rvachew 2001), using speech sound discrimination tasks (e.g. hearing the differences between sounds) in addition to speech production tasks (e.g. saying the speech sounds) (Rvachew 1994) and using two different forms of physical education programme (Head 1975). The approaches to intervention used in the phonology studies were generally comparable with clinicians describing a hierarchical approach to therapy to encourage production of sounds in isolation and then in increasingly more complex linguistic structures. A small number of studies used a cycles approach to therapy that targeted sounds in cycles dependent on the phonological process. The interventions combined varying quantities of speech sound discrimination tasks in addition to speech production tasks, but for only three studies was sound discrimination the primary focus of the intervention; either without additional speech production tasks (Shelton 1978), or in addition to speech production tasks (Reid 1996, Rvachew 1994). The client groups described in the phonology studies were relatively homogenous with the majority of participants ranging in age from three to six years of age. In the earlier studies participants tended to be older and five studies reported participants aged over seven (Fudala 1972; Head 1975; Sommers 1964; Sommers 1966; Sommers 1962) The severity of the difficulties was generally at least one standard deviation below the mean on standardised assessment measures, although in several of the earliest studies participants were included with as few as three mis-articulated phonemes. There was a tendency in the phonology studies to exclude participants with other linguistic difficulties such as expressive delay or receptive delay. Most commonly this excluded any child with a receptive delay below one standard deviation from the mean.
Nineteen studies investigated expressive language interventions. The focus of expressive studies was more likely to be concerned with the provision of intervention through parent programmes such as the Hanen parent programme (Manolson 1992). Three studies compared parent implemented therapy to a control group who received delayed (Girolametto 1996a, Girolametto 1996b, Gibbard 1994a) and a further three studies had multiple conditions and compared parent intervention with clinician intervention and delayed intervention either implemented by parents or clinicians or general stimulation (Fey 1993, Gibbard 1994b, Law 1999). In addition, three studies investigated the effectiveness of clinician delivered therapy compared to delayed therapy (Glogowska, 2000; Robertson 1999; Schwartz 1985). Regardless of the administrator of therapy, the approaches used in these studies tended to be based on language stimulation and used more interactive (e.g. child led) and naturalistic methods to encourage word mapping or the development or more complex syntactic structures. Three studies addressed specific approaches to intervention such as the generalisation of ‘is’ interrogative (Mulac 1977), mimicry and modelling (Courtwright 1979) and play sessions with normal and language impaired play peers (Robertson 1997), and three studies investigated classroom-based interventions. One study compared individual therapy to classroom-based group intervention (Wilcox 1991), one study compared a directive (e.g. clinician led) approach to intervention to an interactive (e.g. child led) approach to intervention within a classroom environment (Cole 1986), and the final study compared weekly intervention to intensive intervention within a nursery setting (Barratt 1992). The participants in expressive language studies were normally of pre-school age, although in one study participants were between six and twelve years old (Head 1975). The children involved in these studies tended to have more severe difficulties with a number of studies specifying two standard deviations below the mean on tests of expressive language. The participants also had a range of other linguistic difficulties; commonly this was in phonology, but this could also be in receptive language and three studies specified that all children had receptive language difficulties (Robertson 1997, Law 1999, Wilcox 1991). Two studies investigated the effectiveness of WILSTAAR intervention looking at language development in children under a year old (Evans [forthcoming], Sutton 1999). These studies aimed to train parents to provide intervention to stimulate language development. To be included in these studies children had to score below one standard deviation in the combined receptive expressive emergent language (REEL) scale (Bzoch 1970).
Five studies investigated interventions for receptive language difficulties. One study specifically investigated comprehension intervention with older children through an approach called Visualising and Verbalising, where children were taught to visualise words and sentences to help them understand language (Dixon 2001). Two studies provided a more general focus to intervention but stated that there were receptive language aims (Glogowska, 2000; Law 1999). These two studies used much younger children than the first study and also included a focus on expressive intervention. The children in the receptive language studies tended to have more severe difficulties than in other studies. The older children in the study by Dixon had receptive language difficulties more than 2 SD below the mean, and the younger children in the studies by Glogowska and Law had receptive difficulties more than 1.25 SD below the mean or 1.5 SD below the mean respectively. In all three of the studies it was stated that children had additional expressive language problems. Two studies were included in this category because the participants were shown to have receptive difficulties, although the studies did not categorise therapy aims in terms of expressive and receptive language specifying instead; semantics, syntax and pragmatics (Cole 1986), and form, content and use (Barratt 1992). Children in the study by Cole had to score 1.5 SD below the mean on either expressive or receptive tests of language, and those in the study by Barratt had to score 1SD below the mean on receptive and/or expressive language tests.
A variety of measures was used to investigate the outcomes of treatment. Authors investigating the effectiveness of phonology interventions were most likely to use standardised measures of overall phonological development. A smaller number of authors also used narrower measures such as production of target processes (Munro 1998, Ruscello 1993, Rvachew 2001), and broader measures such as percentage of consonants correct in conversation (Almost 1998). Studies investigating expressive language interventions tended to use broader outcome measures, drawing on information from parental report and language samples. These were used to measure the increase in number of words or utterances, and the increase in complexity of utterances (See for example: Fey 1993, Gibbard 1994a, Law 1999). Standardised measures of expressive syntax and vocabulary were also used, but were less likely to be administered than in the phonology studies.
The outcomes measured in the studies focusing on receptive language were more diverse due to the differing ages of the client groups and the broader range of intervention targets within the studies. However, there was a tendency for these studies to use standardised outcomes measures. Second-order effects were also investigated and a number of studies reported phonology outcomes following expressive language intervention in separate papers (Fey 1994, Girolametto 1997). Four studies also included non-linguistic second-order effects that may have benefited from the speech and language therapy. These included measures of behaviour or socialisation (Girolametto 1996a; Glogowska, 2000; Law 1999; Robertson 1999), and the effects of intervention on parents in terms of esteem (Law 1999) and stress (Robertson 1999).
Methodological quality of included studies
Each of the studies was assessed for methodological quality and assigned a quality code for a number of potentially important areas: blinding of assessors, reporting of participants’ baseline characteristics and assessment of similarity, explanation of withdrawals, discounting of missing values, degree of participant attrition, use of intention to treat analysis, completion of a power calculation and description of eligibility criteria (Table 01; Table 02). In addition the methods of selecting participants and the size of the sample were also noted. The results reported here are for the thirty-three different studies that were used in the review, although methodological quality ratings were also given to follow-up studies and separate papers reporting different outcomes for the same study.
In eighteen instances it was not clear how the participants in the trials had been recruited and selected. In the remaining studies one had used a random sample of participants referred by clinicians (Sommers 1966), and six had enrolled all eligible participants over a specified time span (Almost 1998; Evans [forthcoming]; Glogowska, 2000; Law 1999; Munro 1998; Sutton 1999). Five studies had actively enrolled participants through advertisements (Fey 1993; Girolametto 1996b; Robertson 1999), and letters home from school (Sommers 1962; Tufts 1959), and in another three cases participants agreed to complete the full number of intervention sessions before being included in the study (Dixon 2001; Fudala 1972; Ruscello 1993). The use of self-selecting samples means that in some instances results may need to be interpreted cautiously; these may create larger effect sizes than investigations using a more mixed clinical population.
The sample sizes ranged from eight to 240, although the majority of studies included between ten and 30 participants (18 studies). This size is relatively small and therefore the degree to which some of the studies are able show significant effects needs to be considered. Four studies completed power calculations (Almost 1998; Barratt 1992; Law 1999; Glogowska, 2000). However for two of these studies recruitment problems meant that the study could not achieve the planned level of power.
The papers generally provided clear eligibility criteria, typically excluding children with general developmental delays, neurological impairments and hearing loss. The majority of studies described the main language characteristics of the participants in terms of phonology, and expressive and receptive language skills. In eight studies details were only provided of the linguistic area examined in the study, or were unclear (Courtwright 1979; Fudala 1972; Head 1975; Mulac 1977; Sommers 1964; Sommers 1966; Sommers 1962; Tufts 1959). This is significant because other factors may have had an effect on the response to therapy. There was variability in the reporting of other potentially confounding variables such as socio-economic status and behaviour. Ten studies reported some details of socio-economic status, this was provided typically as social classes (Gibbard 1994a; Gibbard 1994b; Law 1999; Sommers 1964; Sommers 1966), or parent education (Girolametto 1996b; Glogowska, 2000; Robertson 1997; Robertson 1999; Tufts 1959). In addition to this three studies described their participants as being from middle class, or lower class backgrounds without including data (Evans [forthcoming]; Fudala 1972; Sommers 1962). Seventeen studies reported baseline language scores, and twelve of these studies included some other potentially significant variables such as behaviour, esteem, or parent education. Of the papers that statistically analysed baseline scores, three papers found significant differences at baseline in non-linguistic measures (Fey 1993; Girolametto 1996a; Law 1999).
All of the papers described their participants as being randomly assigned although only three papers described their methods (Almost 1998; Law 1999; Glogowska, 2000). In addition, two papers stated that randomisation had been completed by an independent person, but gave no explanation of the method (Lancaster 1991; Munro 1998) and three papers stated that random numbers tables had been used but provided no other details (Barratt 1992; Sommers 1962; Sommers 1964). In the papers describing their methods, random number tables were used to create sealed envelopes that were opened in the presence of the parents once consent had been gained.
In fourteen studies the assessors were blind to the group status of the participants, or had all their assessments marked by blind transcribers. In the other studies two stated that assessors were blind to group allocation at pre-test, and seventeen did not report whether assessors were blind. In a number of cases a sample of papers were independently marked by blind assessors to give a reliability measure.
Twelve studies accounted for their withdrawals and six stated that no withdrawals occurred. There was only one study that showed withdrawals that were not accounted for (Sommers 1966). Withdrawals were mainly discounted from analysis, although in two studies the withdrawals or non-finishers were included in the analysis by using the last score obtained (Almost 1998; Lancaster 1991). The amount of attrition was generally small, no study mentioned attrition rates higher than 15%, and only eight studies had attrition rates between 10% and 15%. In fourteen studies attrition (e.g. proportion of participants not completing the intervention) and withdrawal (e.g. an explanation of why participants did not complete intervention) were not mentioned and therefore these figures need to be interpreted cautiously; non-reporting may have covered up significant levels of attrition. Two studies reported intention to treat analysis where participant data were analysed according to the group to which they were assigned initially rather than whether they finished the intervention or requested therapy (Almost 1998; Glogowska, 2000). A third study was categorised as an intention-to-treat analysis as participants that had not yet managed to complete the intervention package (e.g. through holidays or illness) were assessed at the end of the study and included in the final analysis (Lancaster 1991).
Results
Seven studies were not included in the meta-analysis because further information regarding results could not be obtained (Barratt 1992; Fudala 1972; Mulac 1977; Reid 1996, Ruscello 1993; Rvachew 1994; Sutton 1999). Of these studies, four were phonological interventions, one was an expressive language intervention focusing on ‘is’ interrogative, one used WILSTAAR intervention, and one was a general language programme carried out within an English healthcare trust. A final study reported outcomes as ratings data and could not be entered into Review Manager (Tufts 1959). A narrative summary of these studies is provided below.
Two of the five phonology studies that could not be used in the meta-analyses compared clinician-delivered intervention to clinician- delivered intervention alongside additional parent-implemented intervention. The first study (Ruscello 1993) used Speech Viewer software and found that involving parents in intervention led to no significant differences in outcomes. The second study (Fudala 1972) found that the children of parents observing the clinician completing the intervention had better speech outcomes than those children whose parents did not observe. Interestingly, no differences were found between the outcomes for children whose parents observed the sessions once a week and those who observed once a month. Another phonology study (Tufts 1959) compared three conditions: parent implemented intervention, clinician implemented intervention and no intervention. The study reported significant differences between the control group and the two intervention groups, but no significant differences between the two intervention groups. Two other phonology studies compared different lengths of intervention (Reid 1996) or specific intervention tasks (Rvachew 1994). The first study (Reid 1996) compared three conditions: no treatment, ten weeks of Metaphon training (levels one and two), and six weeks of Metaphon training (level one only). The study showed significant differences for those children who had attended the ten weeks of speech intervention when compared to either six weeks of intervention and no treatment. The final phonology study (Rvachew 1994) focused on sound discrimination comparing the effects of three conditions: discrimination between a rhyming word pair (e.g., ‘sheet’ and ‘meat’), a non-rhyming word pair (e.g. ‘cat’ and ‘Pete’) and a misarticulated word pair where one word of the pair rhymed but did not start with an English consonant. The results suggest that those children listening to either rhyming word pairs or misarticulated word pairs progressed further in sound production tasks than did the children listening to non-rhyming word pairs.
Three studies that could not be used in the meta-analyses focused on language interventions. The first study (Sutton 1999) compared WILSTAAR intervention to no intervention and did not show any significant differences in language outcomes. The second study (Barratt 1992) compared intensive intervention (four days a week for three weeks once every three months) to weekly intervention (once a week for six months) delivered in a nursery setting. The study showed that although children in the intensive group did not receive any more hours of intervention than those in the weekly group, there was more improvement in expressive language scores from those children receiving intensive intervention. The same result was not seen for language comprehension measures. The third expressive language study (Mulac 1977) compared a language intervention programme focusing on ‘is interrogative’ to the same programme with additional tasks to facilitate generalisation, and to a control programme described a speech sound intervention. The study found that although both experimental groups improved in an ‘is interrogative’ measure, only the group receiving the additional tasks generalised their use of ‘is interrogative’ to outside of the clinic setting.
Expressive Phonology Outcomes
Two studies measured outcomes at the level of the production of the target sound (Munro 1998; Rvachew 2001). One of these studies compared intervention to delayed treatment (Munro 1998), and the second compared treatment of early developing phonemes with later developing phonemes (Rvachew 2001). The former did not show a statistically siginificant difference either for the production of the target sounds (n=11; SMD=0.98, 95%CI: −0.35, 2.31) or for the variability in the production of target sounds (n=11; SMD=0.91, 95%CI: −0.41, 2.23). The latter showed a statistically significant effect that favoured the treatment of early developing phonemes (n=48; SMD=−1.00, 95%CI: −1.60, −0.40).
Eleven studies measured the effects of treatment through standardised measures of overall phonological development. Seven of these provided a comparison of speech and language therapy with no therapy (Almost 1998; Glogowska, 2000; Lancaster 1991; Matheny 1978; Munro 1998; Rvachew 2001; Shelton 1978), and six studies compared different therapies (Lancaster 1991; Rvachew 2001; Shelton 1978; Sommers 1964; Sommers 1966; Sommers 1962). The overall effect estimate was statistically significant favouring the use of speech and language therapy when compared to no treatment (n=264; SMD=0.44, 95%CI: 0.01, 0.86). This estimate increased when parent administered treatments were removed (n=214; SMD=0.67, 95%CI: 0.19, 1.16), and when interventions lasting less than eight weeks were removed (n=213; SMD=0.74, 95%CI: 0.14, 1.33). Studies comparing treatments were varied and could not be combined. One study (Lancaster 1991) showed no significant difference between intervention administered by trained parents and clinicians (n=10; SMD=−0.90, 95%CI: −2.25, 0.44), although two studies (Sommers 1962; Sommers 1964) showed a statistically significant effect favouring the use of trained parents in addition to clinician intervention (n=120; SMD=1.20, 95%CI: 0.17, 2.23). One study (Sommers 1966) did not show a difference between group and individual therapy (n=216; SMD=0.01, 95%CI: −0.26, 1.17), and a final study (Shelton 1978) did not show a difference between parent delivered auditory discrimination therapy and no treatment (n=30; SMD=−0.21, 95%CI: −0.83, 0.41).
Three studies measured broader outcomes for phonology intervention, of which two used the percentage of consonants correct in conversation (Almost 1998; Rvachew 2001), and the third focused on the number of correct target consonants when the child was re-telling a story (Munro 1998). The effect estimates for the two studies comparing phonological therapy with no treatment showed a statistically significant effect favouring phonological therapy when measured as the percentage of consonants correct in conversation (n=26; SMD=1.91, 95%CI: 0.96, 2.86), but not for the re-telling of a story with target consonants (n=11; SMD=1.29, 95%CI: −0.11, 2.69). The third study did not show a significant difference between targeting later and earlier developing sounds in intervention when measured as percentage of consonants correct in conversation (n=48; SMD=0.12, 95%CI: −0.45, 0.68).
The confidence intervals in most of these analyses were over one standard deviation wide, showing a wide range of response to intervention. This reflects the small number of studies and the small number of participants in each of the analyses. One exception to this was the overall measure of treatment compared to no treatment and measured using standardised assessment measures. This had a much larger number of participants (264) and the confidence interval was 0.01–0.86 reflecting less variation in response than other outcome measures.
Receptive Phonology Outcomes
One study focused on receptive phonology intervention as a means of improving productive phonology and measured outcomes using a test for auditory association (Shelton 1978). The effect estimate showed that parental listening and reading-talking intervention was not significantly effective for receptive phonology compared to no intervention (n=45; SMD=0.53, 95%CI: −0.10, 1.16), and that neither intervention showed a more favourable outcome against the other (n=30; SMD=0.00, 95%CI: −0.72, 0.72), nor made any improvement on productive phonology (n=45; SMD=−0.21, 95%CI: −0.83, 0.41).
Expressive Syntax Outcomes
Three studies measured outcomes at the level of the target of therapy. One study compared an intervention approach involving mimicry of sentences to an approach to intervention involving modelling of sentences (Courtwright 1979) and did not show a significant difference between the two approaches (n=24; SMD=0.57, 95%CI: −0.25, 1.39). A second study (Schwartz 1985) also found no statistically significant differences when they compared focused stimulation to general stimulation (n=10; SMD=0.74, 95%CI: −0.87, 2.34). The third study investigated the number of play related speech acts used by children when they were paired with either a normal language play peer or a language impaired play peer (Robertson 1997). This study showed a significant effect favouring the use of normal language peers within play sessions (n=30; SMD=3.20, 95%CI: 1.79, 4.62).
Seven studies measured the outcomes of expressive interventions using tests of overall expressive syntax. Five studies compared expressive interventions to no treatment or to cognitive therapy (Fey 1993, Gibbard 1994a, Gibbard 1994b, Matheny 1978, Law 1999, Glogowska, 2000). The effect estimate from these studies was not statistically significant when compared to no treatment (n=271; SMD=0.70, 95%CI: −0.14, 1.55), although a significant effect was shown when speech and language intervention was compared to cognitive therapies (n=25; SMD=0.93, 95%CI: 0.05, 1.82). The effect size of the no treatment comparison decreased when only data from clinician studies were included (n=214; SMD=0.28, 95%CI: −0.19, 0.75), and increased when studies using clinician administered therapy with duration less than eight weeks were excluded (n=187; SMD=0.43, 95%CI: −0.06, 0.93). However, in neither case did differences reach statistical significance. A post-hoc subgroup analysis was performed on the basis of child characteristics when it became apparent that children were often excluded from trials if they had significant receptive language difficulties. Removing the study where all children had a severe receptive language difficulties showed a significant effect favouring speech and language therapy (n=233; SMD=1.02, 95%CI: 0.04, 2.01). Three studies directly compared parent delivered interventions with clinician interventions (Fey 1993, Law 1999; Gibbard 1994b). These studies did not show a statistically significant difference between the two approaches (n=66; SMD=−0.04, 95%CI: −0.56, 0.48).
Six expressive intervention studies included measures of parental report of phrase complexity, and total number of utterances and mean length of utterance from language samples (Cole 1986, Gibbard 1994a; Gibbard 1994b; Girolametto 1996b; Law 1999; Robertson 1999). Studies measuring the total number of utterances in a language sample produced an effect estimate that was not statistically significant when compared to no intervention (n=99; SMD=0.68, 95%CI: −0.45, 1.82), or to cognitive therapy (n=25; SMD=0.88, 95%CI: 0.00, 1.76). The same result was seen when the effect size was measured as mean length of utterance; no treatment (n=95; SMD=0.74, 95%CI: −0.33, 1.81), cognitive therapy (n=25; SMD=1.36, 95%CI: 0.42, 2.29), and measured as parental report of phrase complexity; no treatment 1.02 (n=99; 95%CI: −0.17, 2.22), cognitive therapy 0.78 (n=25; 95%CI: −0.09, 1.65). When a subgroup analysis was performed excluding the study where all children had receptive language difficulties, the effect estimates significantly favoured speech and language therapy for total number of utterances (n=61; SMD=1.20, 95%CI: 0.33, 2.07), mean length of utterance (n=57; SMD=1.28, 95%CI: 0.66, 1.89), and parent report of phrase complexity (n=61; SMD=1.54, 95%CI: 0.42,2.65). Two studies (Gibbard 1994b; Law 1999) compared interventions with administration by parents and clinicians. These studies did not show a significant difference for total utterances (n=45; SMD=0.15, 95%CI: −0.45, 0.74), mean length of utterance (n=45; SMD=0.28, 95%CI: −1.41, 1.96), and parental report of sentence complexity (n=45; SMD=0.01, 95%CI: −0.63, 0.66).
The confidence intervals for these analyses show a much wider range of responses than those for phonology interventions. Even when a larger number of participants are included in the analyses the confidence intervals remain broad e.g. a comparison of treatment versus no treatment and measured using standardised assessments included 271 participants and produced confidence intervals of −0.14–1.55. This is twice the breadth of the measure obtained from expressive phonology interventions.
Expressive Vocabulary Outcomes
Four studies measured outcomes at the target of therapy. Two of these studies measured the number of target words used during a probe exercise (Girolametto 1996a; Girolametto 1996b), one study measured the number of target words used productively within the clinic setting (Wilcox 1991), and the final study measured the number of words used in play scripts (Robertson 1997). The first two studies compared indirect parent administered therapy with no treatment, and showed a significant effect supporting the use of indirect therapy (n=41; SMD=0.93, 95%CI: 0.27, 1.58). The third study compared the use of individual sessions and group sessions within the class context and did not show a significant difference between individual and class based intervention (n=20; SMD=0.35, 95%CI: −0.53, 1.24). The final study compared the use of normal language and language impaired play peers for SLI children and showed a significant effect supporting the use of normal language peers (n=20; SMD=2.29, 95%CI: 1.11, 3.48).
Three studies used standardised measures to measure the effect of expressive interventions on vocabulary (Gibbard 1994a; Gibbard 1994b; Law 1999). The studies did not show a significant difference when compared to no therapy (n=74; SMD=0.98, 95%CI: −0.59, 2.56), or cognitive therapy (n=25; SMD=0.76, 95%CI: −0.11, 1.63). Subgroup analysis produced a significant difference favouring speech and language therapy when children only had expressive language difficulties (n=36; SMD=1.79, 95%CI: 1.01, 2.58). Data from trials directly comparing parent and clinician administered interventions did not show a significant difference between parent and clinician administered interventions (n=45; SMD=0.20, 95%CI: −0.40, 0.79).
Six expressive intervention studies measured outcomes using broader measures of language functioning, either through parent report of vocabulary size or the total number of words in a language sample (Gibbard 1994a; Gibbard 1994b; Girolametto 1996a; Girolametto 1996b; Law 1999; Robertson 1999). The effect estimate of total number of different words in a language sample significantly favoured speech and language therapy when compared to no intervention (n=82; SMD=1.08, 95%CI: 0.61, 1.55) but not when compared to cognitive therapy (n=25; SMD=0.62, 95%CI: −0.24, 1.49). A similar effect estimate was seen when considering parental report of vocabulary as an outcome measure (n=136; SMD=0.89, 95%CI: 0.21, 1.56). Studies that directly compared parent administered and clinician administered treatments did not show significant differences when outcomes were measured using parent report of vocabulary size (n=45; SMD=−0.16, 95%CI: −0.76, 0.44), or the number of words in language samples (n=17; SMD=−0.50, 95%CI: −1.48, 0.47).
Receptive Syntax Outcomes
Four studies used standardised measures of receptive syntax to measure the outcomes of interventions for receptive language (Cole 1986; Dixon 2001; Glogowska, 2000; Law 1999). The latter two studies compared speech and language therapy intervention to no treatment, these studies did not show a statistically significant difference (n=193; SMD=−0.04, 95%CI: −0.64, 0.56). When subgroup analyses were performed removing parent administered interventions (n=182; SMD=0.01, 95%CI: −0.53, 0.55), and trials carried out over a shorter duration (n=155; SMD=0.19, 95%CI: −0.12, 0.51), no changes in effect were seen. Three studies compared speech and language therapy interventions, though none showed statistically significant differences between the comparators. Law 1999 compared parent and clinician administered interventions (n=28; SMD= −0.11, 95%CI: −0.87, 0.65), Dixon 2001compared traditional speech and language therapy with a technique called Visualising and Verbalising (n=4; SMD=−0.07, 95%CI: −2.08, 1.93), and Cole 1986 compared directive and interactive treatment programmes (n=44; SMD= − 0.33, 95%CI: −0.93, 0.27).
Receptive Vocabulary Outcomes
One study used a standardised measure of receptive vocabulary to measure the outcomes of an intervention for receptive and expressive language (Cole 1986). This study compared directive with interactive speech and language therapy, and did not show a significant difference between the two approaches (n=44; SMD=−0.22, 95%CI: −0.82, 0.38).
Composite Language Outcomes
One study with a focus on the development of expressive and receptive language used composite language measures to measure the outcomes of therapy in terms of expressive and receptive language (Evans [forthcoming]). This study compared WILSTAAR intervention to no treatment and did not show a significant difference between the two approaches (n=55; SMD=0.22, 95%CI: −0.32, 0.76).
Quality
Sensitivity analyses were performed on data from studies comparing intervention to no treatment that measured the overall development of phonology, expressive or receptive abilities. Two sensitivity analyses were performed investigating the effect of non-reporting of attrition and blinding of assessors on effect size. When studies that did not report attrition were removed from the analyses the results did not change either the effect size of expressive phonology (n=211; SMD=0.40, CI: −0.08, 0.89), or expressive syntax (n=217; SMD=0.67, CI: −0.33, 1.66). In both analyses the confidence intervals became broader as a result of the removal of the study. Removing studies where authors did not state that the assessors were blinded to group allocation had a larger effect on the effect sizes. Positive results for phonology increased when lower quality studies were removed (n=188; SMD=0.66, CI: −0.07, 1.40), but expressive measures decreased in both expressive vocabulary (n=38; SMD=0.19, CI: −0.54, 0.91) and expressive syntax (n=219; SMD=0.14, CI: −0.47, 0.75).
Discussion
The objective of this review was to consider the effectiveness of speech and language therapy for children with primary speech and language delay/disorder. The review located 33 different trials of which 25 included sufficient data to be used in the meta-analyses. All of the studies located were written in English, although no language limits were set during the searches. The data were analysed with particular consideration given to the primary effects of intervention at the level of the target of therapy, overall development and broader levels of linguistic functioning. Subgroup analyses were carried out to help explain the results and the variation in response to therapy. Due to the focused nature of the analyses and the heterogeneity of the comparisons some of the effect sizes are based on small numbers of participants and the results of single studies are drawn on. Some of these results therefore need to be considered tentatively.
The findings of this review provide some support for the effectiveness of speech and language therapy for children with phonological and expressive vocabulary difficulties. It is possible that therapy may be effective for children with expressive syntax difficulties if they do not have concomitant difficulties in receptive language. This result is based on sub-group analyses and therefore no causal relationships can be drawn; however this finding corresponds with what is known about the long-term prognosis for children with speech and language difficulties (Law 1998). For children with receptive language difficulties there is still less evidence that interventions are effective. This needs careful consideration and further evaluation as natural history studies have shown that children with receptive language problems are least likely to resolve without intervention and most likely to have longer term difficulties (Law 1998).
The effect sizes in a number of the meta-analyses were relatively large favouring speech and language interventions, however the confidence intervals were generally broad due to the heterogeneity of the studies included in the analyses and the wide variation in response. The majority of the confidence intervals crossed zero, and are therefore non-significant results. Children with phonological difficulties tended to vary in their response to treatment within studies (e.g. Munro 1998; Reid 1996), but between studies showed similar patterns of variation, as shown in their narrower confidence intervals. This is in contrast to children with expressive language difficulties who showed a much wider range of response between studies, as shown in much wider confidence intervals. There is a need for more clearly defined interventions and analyses that are stratified on the basis of participant characteristics, so that the sources of the variation can be more fully understood.
Within the categories of analysis e.g. expressive and receptive phonology, syntax and vocabulary, a wide range of outcomes were measured, and not all outcomes within any one area were significant. It is hard to make judgements on which outcomes are the most important; while narrow outcomes measuring the target of therapy may best measure what the clinician is working on and the short term aims for the child, standardised assessments allow for comparison with same age peers, and broader measures best show generalisation of skills to everyday situations. In addition important outcomes that were not measured in this review may have improved such as the communication skills of communicative partners e.g. parents or teachers. For some children the most important outcome may not be improvement in their abilities, but an improvement in those around them to support their communicative environment. For this reason it is also hard to make judgements about the size of effect required to benefit the child. A better understanding of natural history and the long term outcomes for children with primary speech and language delay/disorder would help to clarify whether the goals of intervention for particular client groups should focus on changing the child's speech and language abilities or in improving the communication skills of other people within the child's environment.
The results of this review generally complement those other reviews (e.g. Nye 1987; Law 1998), although the effect sizes in some instances are smaller. This can be explained in two ways; firstly, the effect sizes for each of the linguistic outcomes only includes studies where authors said that they had an explicit language focus in that area. A number of studies reported larger effects of intervention for receptive language (e.g. Gibbard 1994a; Shelton 1978), but these studies did not have explicit receptive language aims, and the children had not been identified as having receptive language difficulties. Therefore the outcomes were not included in this analysis. Secondly, a larger number of trials in this review were completed within British health care trusts (7 out of 25) and tended to produce lower effect sizes than those completed in America, possibly due in general to the smaller number of hours of contact time.
No statistically significant differences were found between using trained parents and clinicians as the administrators of intervention. Results of a number of the studies showed that response to treatment is more varied when using parent administrators, and suggest that some parents may be more suited to parent treatment than others (Fey 1997; Gibbard 1994a). Further research is required to assess parent characteristics that may lead to better outcomes. There is an indication that interventions carried out over longer periods of time may be more effective than those carried out over shorter periods. This result is based on a subgroup analysis rather than the direct results of any randomised studies and therefore needs to be interpreted cautiously due to potential confounding variables. There needs to be further randomised studies comparing interventions over different lengths of time in order to test this hypothesis. A small number of studies have investigated the impact of different lengths of treatment; Fey et al (1997) showed much smaller gains in a second five month phase of therapy following an initial five months of therapy, and a previous meta-analysis reported a larger effect size for treatments lasting four to twelve weeks (Nye 1987). In considering the impact of differing lengths of treatment it is also important to consider the role of intensity on treatment outcome; the studies included in this review could not be reliably grouped to consider the impact of intensity on treatment outcomes. However, one study that could not be used in the analysis found that children who received intensive therapy had better outcomes than those who received weekly therapy, even though they received no more contact time (Barratt 1992). Further research is also required to investigate the impact of intensity on outcomes. Studies comparing the use of individual and group therapies showed no difference in the effectiveness of therapy (Sommers 1966; Wilcox 1991). However a study that employed the use of normal language peers in play sessions with language impaired children found a statistically significant difference in outcomes (Robertson 1997). These findings support the use of collaborative approaches in schools where therapists have the potential to draw on normal language peers.
Studies that address specific aspects of phonology intervention support clinician treatment of earlier rather than later developing phonemes (Rvachew 2001), but do not support the use of parent delivered auditory discrimination therapy as a means of improving productive phonology (Shelton 1978). While the majority of the studies used in the review included aspects of auditory discrimination, few of them focused on the effectiveness of auditory discrimination or phonological awareness therapy. Two papers that could not be used in the meta-analyses provide positive indicators for the use of metaphon therapy (Reid 1996), and auditory discrimination in addition to articulation therapy (Rvachew 1994). Given the widespread use of these approaches within therapy, this area of intervention may benefit from randomised controlled trials to assess effectiveness.
Studies comparing different approaches to expressive language interventions did not show any statistically significant differences between interactive and directive approaches (Courtwright 1979; Cole 1986). There is some research evidence that suggests that higher functioning children and children with learning disabilities may benefit from interactive learning, whilst children who have more severe difficulties but normal intelligence measures may benefit from a directive approach (Yoder 1991). Further research is required considering different learning styles and child characteristics to ensure that children receive an intervention most suited to them. This review found no significant effect of WILSTAAR early intervention (Evans [forthcoming]). This suggests that further evaluation of this technique is required as these results are smaller and less conclusive than those obtained from a large “first generation” non-randomised control trial (e.g. Ward 1999). Finally this review found no evidence to support the use of Visualising and Verbalising as an approach to receptive intervention when compared to traditional therapy involving inference making, identifying main ideas, and following sequences (Dixon 2001).
Studies often provided a range of outcome measures, measured at the target of therapy, overall development and broader measures of linguistic functioning. Results of the meta-analysis generally complemented each other at each of the levels of analysis. In general larger effect estimates were obtained from either measures of the therapy target or broader measures of either language sample data or parent report. However, in most cases these measures were also associated with broader confidence intervals.
The quality of the studies used in the review was variable. A number of studies did not report potentially important factors such as attrition, method of randomisation and baseline characteristics. While sensitivity analyses did not show that this consistently altered the effect sizes, it is possible that non-reporting and poor methodology may have had an effect on the results gained in the trials which may have exaggerated the meta-analysis effect estimates (Juni 2001).
The lack of information in some older studies concerning participant characteristics also made it hard to analyse the impact of the language difficulty on the effectiveness of therapy. This was made more difficult because of the variability in reporting standard scores and the variation in the range of assessments used. A number of papers gave baseline characteristics as raw scores making it difficult to identify the severity of participants’ difficulties, and making comparisons between papers difficult. In addition, while it is recognised that the causes of primary speech and language delay/disorder are likely to be multi-factorial, the reporting of non-linguistic characteristics was also variable. This meant it was difficult to consider other potentially important factors that may have influenced therapy outcomes e.g. behaviour, esteem, socio-economic status and developmental history. The final concern reflects the fact that the sources of the samples were not always well described, and in some cases were actively recruited through advertisements or letters. The use of self-selecting samples may mean that effect sizes are higher than could be expected in a true clinical population. The high rate of compliance in a number of the studies may also contribute to this effect and is likely to be partially as a result of the way participants were selected.
Funnel plots were investigated to assess the possibility of small sample bias and publication bias. Overall the studies had relatively large amounts of variance and were widely scattered on both sides of the pooled estimate. This produced symmetry in the plots, but meant that some lost their characteristic funnel shape. This can be explained by the inclusion of small studies that included both significant and non-significant results.
Limitations
The majority of the interventions reported in the present review can be described as being delivered with a clinical model, that is they are separate from other contexts in which the children learn -namely early years provision and school. Furthermore they are circumscribed in terms of intensity and duration. The implicit understanding is that these relatively short “innoculation” interventions are effectively adjusting the child's developmental trajectory. While a number of the interventions report significant results it is not clear that children's speech and language development is “normal” post intervention. Indeed the follow-up literature would suggest that many of these children go on to have persistent problems. Furthermore it is not clear that the additional input that many children receive equates to this clinical model. Many of these children receive a comprehensive package of interventions over an extended period of time of which these periods of speech and language therapy are only a part.
Measures of second-order effects were sometimes reported in the included studies, but were not included in this review. For example, interventions for expressive language sometimes included an analysis of the effect on phonology, and a number of studies show positive second order effects of expressive language intervention on phonology measures (Girolametto 1997; Matheny 1978; Robertson 1999). A number of other studies also included broader non-linguistic outcome measurements including behaviour gains, esteem and stress. These studies provide positive indicators that speech and language therapy interventions can have a positive effect on socialisation (Robertson 1999), and behaviour (Girolametto 1996a; Law 1999), as well as parental stress (Robertson 1999) and self-esteem (Law 1999). It is a limitation of this review that the analysis does not consider second-order effects of therapy, as there is a need to address the wider potential of speech and language therapy interventions. The authors will seek to address this deficit in future updates of the review.
Studies rarely included a measure of longer term follow-up, and therefore it is hard to know how the effects of speech and language therapy maintain over time. Five studies followed up children after intervention had finished at three weeks (Robertson 1997), eight weeks (Sommers 1964), four months (Almost 1998), five months (Fey 1997), and six months post intervention (Law 1999). Sommers (1964) showed that children continued to make gains in phonology measures following intervention, although Almost (1998) reported that at four months post intervention gains in phonology were maintained but not improving. In terms of expressive language skills, Robertson (1999) reported that expressive language gains were maintained at three weeks post intervention, as did Fey (1997) at five months, although Law et al (1999) report that the overall effects of speech and language therapy had largely worn off at six months post intervention. Of interest is the fact that parent interventions which aim to change the nature of the interaction between parent and child did not show further gains following therapy (Fey 1997, Law 1999). Further research is needed in this area to examine the long-term impact of speech and language therapy.
In a large number of the studies the focus of the comparison was either on the overall effectiveness of speech and language therapy compared to no treatment, or on the effectiveness of using parent administrators in therapy. A small number of studies compared other aspects of intervention including group and individual therapy, and directive and interactive approaches. There is a need to identify and investigate other potentially important factors that lead to positive outcomes both in terms of intervention approaches used and the physical characteristics of the intervention process. While two randomised controlled trials will hopefully add to the knowledge concerning the intervention approaches and participant characteristics that lead to better clinical outcomes (Boyle, Ongoing; Broomfield, FC), there is a need to consider other forms of research that may help identify a broader range of potential indicators that could be used to focus more rigorously controlled studies.
Homogeneity
There was a large degree of variation in the studies even when similar intervention techniques were desribed. This was most apparent in the expressive language interventions. In part the variation in effect size can be explained by differences due to having different administrators of treatment. Once parent interventions are removed from the analysis the studies become more homogenous. However even having taken this into consideration differences in response still occurred, which may have been due to factors such as; socio-economic status, comprehension levels, and duration and intensity of therapy. There is a need for trials to more closely define their intervention procedures and to stratify analyses on the basis of participant characteristics so that some of the variation can be more easily explained.
Reviewers’ conclusions
Implications for practice
This review concludes that there may be some support for the effectiveness of the use of speech and language therapy interventions for children with expressive phonological and expressive vocabulary difficulties. The evidence suggests that there is a differential effect of intervention and that therapy for expressive syntax difficulties may be effective when children do not also have severe receptive language difficulties. The nature of these mixed findings means that further research would be beneficial to the field. For children with receptive language disorders the evidence base is limited, but evidence from studies in this review suggests that the effect of speech and language therapy interventions on language outcomes for this group is much smaller than for other client groups. Studies directly comparing administrators of intervention did not show a statistically significant difference between the use of trained parents and clinicians. Studies comparing group and individual interventions did not show a statistically significant difference. This conclusion is drawn from a limited number of studies that mainly focused on phonological interventions. Its application to expressive or receptive language interventions may be limited. Subgroup analysis indicated that interventions lasting longer than eight weeks may be more effective than those lasting less than eight weeks. This result needs to be interpreted tentatively due to the limited range of studies in the analyses and differences between the individual studies. This should be viewed as a possible indicator of better clinical outcome, with further research required. The evidence supports the use of normal language peers in intervention as language models for children with language impairment.
Implications for research
Many of the basic questions about natural history and the appropriateness of outcomes would at this stage best be answered by an examination of appropriate cohort data in order to develop a fuller picture of the factors that can influence intervention. However this review provides a partial set of indicators that would benefit from further experimental research. There needs to be further research investigating the effects of intervention for children with receptive language disorders and how the outcomes of therapy may be optimised. This is particularly important in light of research that shows these children are least likely to resolve and are more likely to have long term sequelae. It is important that research is carried out on the relative value of these more systemic, contextualised interventions within the educational setting, using educational outcome measures as well as more traditional speech/language outcomes. Before this can happen protocols need to be developed of the different “care pathways” available to individual children. Interventions need to be clearly defined to help explain variation in response to treatment both within studies and between studies. The use of stratified analysis on the basis of participant characteristics needs to be considered so as to identify potential effect modifiers. Research is needed exploring the optimum starting point for intervention. It is tempting to think that early intervention is the obvious solution because it will reduce the need for subsequent support and thus reduce the call on resources. However, there is also a case to be made for targeting intervention to the point where the child is ready to change. Research is needed that investigates the impact of severity on therapeutic outcome. It is currently not known whether those children who have the most severe difficulties are likely to show the most change because they start at a lower level and have most potential for change, or if they may show the least change because they start at a lower level and their difficulties prevent them from benefiting from intervention. A variation in response is shown to parent administered interventions amongst children with expressive language difficulties. Further research is required to consider the parent characteristics that lead to better therapy outcomes, and the characteristics of therapy that best help to engage parents. There needs to be consideration of the learning styles of children with different language difficulties and the influence of this on the child's responsiveness to therapy. The long term impact of therapy needs to be investigated to look at the maintenance over time of the skills that have been acquired in therapy. It is unclear to what extent “downstream” effects are primarily linguistic or whether they include other aspects of behaviour. It may be appropriate to include preference arms in randomised control trials, giving parents the choice of which treatment they would like to receive. This would ascertain the potential effect on therapeutic outcome of having parents who were receiving what they perceive to be the best possible treatment for their child. Non-linguistic factors need to be included in intervention reports both as baseline characteristics and outcome measures. This is because therapy may be influenced by non-linguistic factors and because the therapeutic process may influence additional non-linguistic factors. Linguistic baseline characteristics need to be presented as age equivalent or standardised scores to allow comparison between studies. Due to the variety of assessments available, further research on the comparability of assessments would also facilitate the interpretation of results. The interpretation of the studies would be facilitated by the use of the CONSORT statement to structure report writing (Moher 2001).
Characteristics of included studies
APPR: Assessment of Phonological Processes, BPVS: British Picture Vocabulary Scale, CA: Chronological Age, CAPP: ?, CDI: Communicative Developmental Index, CELF: Clinical Evaluation of Language Fundamentals, DSS: Developmental Sentence Score, EAT: Edinburgh Articulation Test, HANEN: ?, ITPA: Illinois Test of Psycholinguistic Properties, MLU: Mean Length of Utterance, PLS: Pre-school Language scale PAT: Picture Articulation Test, PCLT: PPKP: PPVT: Peabody Picture Vocabulary Test, RAPT: Renfrew Action Picture Test, REEL: Receptive Emergent Language Scale, SES: Socio-Economic Status, SLI: Specific Language Impairment, SS: Standard Score, WILSTAAR: Ward infant Language Screening Test: Assessment, Acceleration, Remediation.
Characteristics of excluded studies
Characteristics of ongoing studies
Footnotes
Table of comparisons
Speech and language intervention vesus delayed or no treatment Expressive phonology outcomes Production of target sound Variability in production of target sound Measures of overall phonological development Percentage of consonants correct in conversation Re-telling a story with target sound Receptive phonology outcomes Auditory association test Expressive syntax outcomes Measures of overall expressive syntax development Total number of utterances in a language sample Mean length of utterance from language sample Parent report of phrase complexity Receptive syntax outcomes measures of overall receptive syntax development Expressive vocabulary outcomes Number of different target words learnt Measures of overall expressive vocabulary development Different words in language sample Parent report of vocabulary Receptive vocabulary outcomes Composite language measures REEL Language quotients Subgroup analysis (clinician only data) measures of overall expressive phonology development measures of overall expressive syntax development measures of overall expressive vocabulary development measures of overall receptive syntax development Subgroup analysis (trials of longer than eight weeks) measures of overall phonological development measures of overall expressive syntax development measures of overall expressive vocabulary development measures of overall receptive syntax development Subgroup Analysis (excluding data from children with receptive and expressive difficulties) Measures of overall expressive syntax development Total number of utterances in language sample Mean langth of utterance based on language sample Parent report of phrase complexity Measures of overall vocabulary development Parent report of vocabulary size Sensitivity analysis (excluding studies not reporting attrition) Measures of overall expressive phonology development Measures of overall receptive phonology development Measures of overall expressive syntax development Measures of overall receptive syntax development Measures of overall expressive vocabulary development Measures of overall receptive vocabulary development Composite language measures Sensitivity analysis (excluding studies not reporting blinding) Measures of overall expressive phonology Measures of overall receptive phonology Measures of overall expressive syntax Measures of overall receptive syntax Measures of overall expressive vocabulary Measures of overall receptive vocabulary Composite language measures Speech and language intervention versus general stimulation programmes Expressive phonology outcomes Receptive phonology outcomes Expressive syntax outcomes Multi-word utterances using target stimuli Measures of overall expressive syntax development Number of play related speech acts during play Total number of utterances in language sample Mother description of phrase complexity Mean length of utterance based on language sample Receptive syntax outcomes Expressive vocabulary outcomes Number of words in play scripts Measures of overall expressive vocabulary development Number of words in language sample Receptive vocabulary outcomes Speech and language interventions versus traditional speech and language programmes Expressive phonology outcomes Production of target sound Measures of overall expressive phonology development Percentage of consonants correct in conversation Receptive phonology outcomes Test of Auditory Association Expressive syntax outcomes Correct utterance of 20 unusual sentences Measures of overall expressive syntax development Total utterances from a language sample Mean length of utterance from a language sample Parent report of complexity of phrases Receptive syntax outcomes Measures of overall receptove syntax development Expressive vocabulary outcomes Overall use of target words in clinic interaction Measures of overall expressive vocabulary development Different words in language sample Parent report of vocabulary size Receptive vocabulary outcomes Measures of overall vocabulary understanding Composite language outcomes Subgroup analysis (clinician versus parent) Measures of overall expressive phonology development Measures of overall expressive syntax development Measures of overall expressive vocabulary development Measures of overall receptive syntax development Subgroup analysis (excluding data from children with expressive and receptive difficulties) Measures of overall syntax development Total number of utterances in a language sample Mean length of utterance derived from a language sample Parent report of phrase complexity Measures of overall vocabulary development Parental report of vocabulary size