Speech and language therapy to improve the communication skills of children with cerebral palsy

Abstract

Background

The production of speech, language and gesture for communication is often affected by cerebral palsy. Communication difficulties associated with cerebral palsy can be multifactorial, arising from motor, intellectual and sensory impairments. Children with this diagnosis can experience mild to severe difficulties in expressing themselves. They are often referred to speech and language therapy (SLT) services to maximise their communication skills and help them to take as independent a role as possible in interaction activities. Therapy can include introducing augmentative and alternative communication (AAC) systems, such as symbol charts or communication aids with synthetic speech, as well as treating children's natural forms of communication. Various strategies have been used to treat the communication disorders associated with cerebral palsy, but evidence of their effectiveness is limited.

Objectives

To determine the effectiveness of SLT that focuses on the child or their familiar communication partners, as measured by change in interaction patterns.
 To determine if individual types of SLT intervention are more effective than others in changing interaction patterns.

Search methods

Searches were conducted of MEDLINE, CINAHL, EMBASE, PsycINFO, LLBA, ERIC, WEB of SCIENCE, Scopus, NRR, BEI, SIGLE (to January 2011). A previous version of this review included studies up to the end of 2002. References from identified studies were examined and relevant journals and conference reports were handsearched.

Selection criteria

Any experimental study containing an element of a control was included in this review. This includes non‐randomised group studies and single case experimental designs in which two interventions were compared or two communication processes were examined.

Data collection and analysis

All authors searched for and selected studies for inclusion. L Pennington (LP) assessed all papers for inclusion, J Goldbart (JG) and J Marshall (JM) independently assessed separate random samples, each comprising 25% of all identified studies. Two review authors independently abstracted data from each selected study. Disagreements were settled by discussion between the three review authors.

Main results

Sixteen studies were included in the review. Nine studies evaluated treatment given directly to children, seven investigated the effects of training for communication partners. Participants in the studies varied widely in age, type and severity of cerebral palsy, cognitive and linguistic skills. Studies focusing directly on children suggest that this model of therapy delivery has been associated with increases in treated speech and communication skills by individual children. However, methodological flaws and small sample sizes prevent firm conclusions being made about the effectiveness of the therapy. In addition, maintenance of these skills was not investigated thoroughly. The studies targeting communication partners used small exploratory group designs which often contained insufficient detail to allow replication, although more recent studies have improved in this area. Overall, the studies of indirect intervention have very low power and cannot provide evidence of effectiveness of this type of treatment.

Authors' conclusions

Firm evidence of the positive effects of SLT for children with cerebral palsy has not been demonstrated by this review. However, positive trends in communication change were shown. No change in practice is recommended from this updated review. Further research is needed to describe this client group, and its possible clinical subgroups, and the methods of treatment currently used in SLT. Research is also needed to investigate the effectiveness of new and established interventions and their acceptability to families. Rigour in research practice needs to be extended to enable firm associations between therapy and the communication change to be made. There are now sufficient data to develop randomised controlled studies of dysarthria interventions and group parent training programmes. Such research is urgently needed to ensure clinically effective provision for this group of children, who are at severe risk of social and educational exclusion.

Plain language summary

Speech and language therapy for children with cerebral palsy might improve their communication skills, but more research is needed.

Cerebral palsy (CP) is a movement disorder caused by damage to the brain before, during or soon after birth. The ability for people with CP to communicate effectively is often impaired by problems with speech and also gestures usually used in communication. Speech and language therapy aims to help people with CP maximise their communication skills. This can include ways of enhancing natural forms of communication, introducing aids such as symbol charts or devices with synthetic speech, and training communication partners. The review found some weak evidence that speech and language therapy might help children with CP, but more research is needed.

Background

Cerebral palsy describes a "group of persistent disorders of the development of movement and posture, causing activity limitation, that are attributed to non‐progressive disturbances that occurred in the developing fetal or infant brain" (Bax 2005). Subgroups of cerebral palsy have been classified according to the clinical signs of spastic, ataxic and dyskinetic syndromes, plus mixed forms (SCPE 2000).

The prevalence of cerebral palsy is approximately 2.5 per 1000 live births in countries with neonatal intensive care facilities (Colver 2000; Yeargin‐Allsopp 2008; Himmelmann 2010). Prevalence is higher in children born with very low birth weight. However, a decline in prevalence in this group from 60.6 (99%CI 37.8‐91.4) per 1000 live births in 1980 to 39.5 (28.6‐53.0) per 1000 in 1996 has recently been observed in Europe (Platt 2007). Communication difficulties can be associated with any type of cerebral palsy and may relate to limitations in the production of movements for speech, gesture and facial expression; receptive or expressive language; hearing; vision; or a combination of limitations in these functions. Speech impairments are estimated to affect approximately 36% of children with cerebral palsy and communication difficulties are observed in around 42% (Parkes 2010). Prevalence of speech, language and communication impairment increases with severity of motor and intellectual impairment (Kennes 2002; Bax 2006; Parkes 2010; Sigurdardottir 2010). Children may experience communication difficulties from early infancy and, as cerebral palsy is a persistent condition, communication impairments are chronic and children may require long term intervention. In a review of speech and language therapy caseloads in the UK, Enderby 1986 estimated that cerebral palsy was the sixth most common medical cause of speech disorder, and the proportion of referrals of children with this diagnosis remains static (Petheram 2001).

Speech and language therapists (also known as speech therapists, speech‐language pathologists) assess, diagnose and treat the communication disorders associated with cerebral palsy. The aim of treatment is to maximise children's ability to communicate, through speech, gesture and supplementary means such as communication aids, to enable them to become independent communicators. As the problems experienced by children with a diagnosis of cerebral palsy are wide in range there is no single, universally appropriate form of treatment. Intervention can focus directly on spoken output, expressive or receptive language development, or helping children to develop conversation skills such as asking questions and repairing conversation when misunderstandings occur (for example Letto 1994). Work to develop children's language or communication skills could involve children using any method of communication. Intervention can also involve children's familiar conversation partners, such as their families, friends and teaching staff (Culp 1988; Pennington 1996). Such indirect therapy aims to teach people who are in close contact with children with cerebral palsy, to facilitate their communication development by creating opportunities for them to use new skills in conversation. Effective indirect intervention would lead to changes in conversation style for both the familiar conversation partners and the children.

Speech and language therapy may be delivered in a range of settings, including clients' homes, community clinics, hospitals and schools (RCSLT 1999). It is usual for speech and language therapists to liaise with families and teaching staff regarding therapy to ensure that intervention goals are incorporated into daily life, where possible (Calculator 1991). Therapy may be delivered on an individual basis or in groups. Interventions may also vary in duration and intensity.

Speech and language therapy for this group of children is often long term, requiring significant health service resources. The effectiveness of speech and language therapy has been called into question (Enderby 1997). For this client group it is necessary to know if changes that occur in children's communication are a result of SLT intervention or other factors, such as maturation. If it is demonstrated that speech and language therapy is effective, information about the effectiveness of different kinds or components of therapy is needed for children from different clinical subgroups to ensure appropriate use of resources.

In 2001 we set out to conduct an exploratory systematic review of studies of speech and language therapy for children who have communication disorders associated with cerebral palsy. This initial, broad review investigated the forms of SLT currently used to remediate different types and severities of communication disorders associated with cerebral palsy, and their relative effectiveness, with a view to providing directions for future research. As little evidence was found in the original review, this current updated review remains exploratory in nature investigating all areas of speech and language therapy intervention.

Objectives

1 To assess whether direct intervention aimed at improving the communication skills of children with cerebral palsy is more effective than no intervention at all.

2 To assess whether intervention aimed at changing the conversational style of the familiar communication partners of children with cerebral palsy is more effective than no intervention at all in: (i) changing partners' conversational style, and (ii) developing communication skills of children with cerebral palsy.

3 To assess whether individual types of intervention are more effective than others in improving the communication skills of children with cerebral palsy.

4 To assess whether one particular type of intervention is more effective than others in changing the conversation style of the familiar communication partners of children with cerebral palsy.

Methods

Criteria for considering studies for this review

Types of studies

To provide an overview of the evidence for speech and language therapy interventions provided to a diverse client group we included any controlled study of interventions aimed at improving communication skills, reported in any language, in this review. Translations were sought, when necessary. Controlled studies included group and single case experimental designs. Group studies were included if participants were allocated to different interventions or acted as their own control, receiving the novel intervention following usual treatment. Single case experimental designs were included if communication behaviours were allocated to treatment or control and both behaviours were measured at baseline, intervention and follow‐up phases, thereby allowing causal inference. Observational studies which used an AB design replicated across participants were excluded from this review.

Types of participants

Any child or individual under 20 years of age with any communication disorder associated with cerebral palsy, including dysarthria, dyspraxia, ataxia and mixed syndromes; or their communication partners. No exclusions were made on the basis of additional impairments (intellectual or sensory impairments, the presence of epilepsy) or prior receipt of speech and language therapy. This age range was selected as people who have identified special needs are entitled to statutory education provision up to 19 years of age in England, which could specify speech and language therapy.

Types of interventions

Any therapy aimed at improving communication skills whether provided individually or in groups; in the child's home, school or health service settings. Exceptions were therapies provided as part of a holistic approach (for example, in conductive education).

1. Therapies given directly to the child with the aim of developing the child's communication skills.

These are distinguished from the following.
 2. Therapies given to familiar communication partners (families, teachers, teaching assistants, peers) with the aim of changing the communication partners' conversation style to help them facilitate children's communication development.

Types of outcome measures

1. Measures of communication:

a) World Health Organisation International Classifcation of Functioning, Disabiltiy and Health (ICF) body function level outcomes: children's expressive and receptive language skills, speech production;

b) ICF activity level outcomes: conversation and pragmatic skills, intelligibility, communicative competence;

c) ICF environmental level outcomes: partners' communication and interaction strategies.

Measures used may be, for example: rating scales, language tests, coding schemes developed for individual research studies that include validity and reliability data.

2. Family stress and coping (e.g. Questionnaire on Resources and Stress, Carer Strain Index).

3. Children's quality of life.

4. Children's particpation.

5. Satisfaction of patient and family with treatment.

6. Noncompliance with treatment.

Search methods for identification of studies

1. The review is based on the following search strategy:

a. cerebral palsy AND child
 b. speech OR speech disorder OR speech intelligibility OR speech therapy OR speech and language therapy
 c. language OR language disorders OR language development disorders OR sign language OR child language OR language therapy
 d. communication OR communication aids for disabled OR communication disorders OR communication methods, total OR manual communication OR nonverbal communication
 e. #b OR #c OR #d
 f. a AND e

The following electronic databases were searched (up until January 2011): MEDLINE (from 1966); CINAHL (from 1982); EMBASE (from 1980); PsycINFO (from 1967); Web of Science (from 1981); Scopus (from 2002); Language and Linguistic Behaviour Abstracts (from 1973); British Education Index (from 1986); National Research Register (completed and ongoing research); ERIC (from 1966); SIGLE (from 1980).

2. The following journals were handsearched (from their inception or from 1980 onwards): International Journal of Language and Communication Disorders; Augmentative and Alternative Communication; Child Language Teaching and Therapy; Developmental Medicine and Child Neurology; Child: Care, Health and Development and the Ambulatory Child; Journal of Child Psychology and Psychiatry and Allied Disciplines; Topics in Language Disorders; European Journal of Special Needs Education; Journal of Communication Disorders; Journal of Psycholinguistic Research; Journal of Special Education; International Journal of Rehabilitation Research; Folia Phoniatrica et Logopaedica; Applied Psycholinguistics; Journal of Speech, Language and Hearing Research; Asia Pacific Journal of Speech, Language and Hearing; International Journal of Speech‐Language Pathology; American Journal of Speech‐Language Pathology; International Journal of Disability, Development and Education, Speech, Language and Hearing in Schools. The current titles are given for journals experiencing name changes since 1980.

3. Published conference proceedings of the following organisations were checked: European Academy of Childhood Disability (1996 to 2010), International Society for Alternative and Augmentative Communication (1996 to 2010), American Speech and Hearing Association (1999 to 2002), Royal College of Speech and Language Therapists (1998 to 2009).

4. Reference lists of all studies selected for possible inclusion were checked for other possibly eligible studies.

5. Authors of included trials were contacted for unpublished studies. Calls for assistance were made via national professional associations.

Data collection and analysis

One review author (LP) assessed the studies identified by the search strategies for inclusion according to specified criteria. The other two authors independently assessed separate random samples each comprising 25% of all identified studies plus any studies whose inclusion status was ambiguous. Agreement on inclusion was calculated using the Kappa statistic. The opinion of the third review author was sought if there was any disagreement regarding the inclusion of a trial.

Two review authors reviewed each identified study, abstracted data using forms developed for the review and graded the study's methodological quality. Where necessary, authors were contacted at their last known address to provide missing data for included trials.

As per the Cochrane Handbook, attention was paid to whether studies demonstrated protection from the following types of bias:

• selection bias, i.e. true random sequencing, true concealment up to the time of allocation, comparison of known confounding variables between groups, comparison of developmentally similar processes in single case experimental designs;

• performance bias, i.e. differences in types of treatment (co‐interventions) between the two groups;

• exclusion bias, i.e. withdrawal after entry to the trial;

• detection bias, i.e. 'unmasked' assessment of outcome.

Decisions regarding potential biases were recorded in the risk of bias table for each study.

The methodological quality of single case experimental designs was also rated on the description of the participant and intervention, whether baseline performance was adequately established, the duration of treatment and follow up and the frequency of measurement across the phases of the experiment.

Individual criteria were rated as 'met', 'partially met', 'unmet' or 'unclear'. Disagreements were resolved with a third review author. Agreement on methodology assessment was calculated using the Kappa statistic (K).

Information from studies meeting criteria for inclusion was entered into RevMan. Most studies that were included used single case experimental designs. Four group trials were identified but only one included randomisation and the participants were heterogeneous. Data were therefore not combined for the review and data were not analysed using RevMan.

Results

Description of studies

Searches yielded 911 abstracts. Of these, 771 clearly did not fit the inclusion criteria for the review. Full texts of 140 papers were considered for potential inclusion. Authors agreed on 78 of the 81 papers randomly selected for reliability check, K = 0.873. Disagreements were resolved with a third review author for the other papers. Seventeen papers (reporting 16 studies) were included for full review. The main reasons for exclusion were that included participants did not have cerebral palsy or those with cerebral palsy could not be disaggregated from other participants, or the study did not include any experimental control. Most reports were written in English. Papers in other languages were read by translators who discussed the content with review authors; none were found to fit the inclusion criteria for the review.

Therapy focusing on children

Nine of the included studies evaluated therapy that focused directly on children, who varied widely in age, type and severity of cerebral palsy and additional impairments. These studies aimed to facilitate the development of pre‐intentional communication skills (behaviours such as mutual gaze, anticipation of behaviours in familiar routines that can be interpreted as communication by others, but which are not performed with the intention of conveying a message), pragmatic or communicative functions used in conversation, such as asking questions, providing information or repairing misunderstandings, speech production, expressive language structures or receptive vocabulary. The studies focusing on dysarthria are also discussed in a separate review (Pennington 2009).

Pre‐intentional communication

Richman 1977 used operant teaching strategies to train a nine year old girl with severe cognitive impairment, who lived in an institution, to produce three pre‐intentional communication skills: maintaining eye contact and head control and increasing vocal imitations (ICF body functions). Forty hours of therapy were given over 20 weeks. Ten minute intervals were sampled for the presence of the three behaviours.

Communicative functions

Five studies focused on the production of nonverbal messages, teaching children to use individual communicative functions (ICF activity level outcomes). Hunt 1986, Pinder 1995 and Sigafoos 1995 taught children to use requests for objects or actions. Hunt 1986 included one seven year old girl with cerebral palsy who had severe cognitive impairment and multiple disabilities in a multiple probe multiple baseline across participants design. Other participants did not have cerebral palsy. The subject was taught to request four objects or events by eye pointing to line drawings symbolising the object or action. Operant teaching methods were used, including interrupted chain training. Treatment was given twice daily, with 55 sessions in total. Requests were probed across the treatment sessions. Pinder 1995 taught four infants with cerebral palsy to produce either requests for objects or requests for more of an activity using micro teaching techniques (creating a communication environment, modelling the target skill, expectant delay, prompting and reinforcement). The children were aged 11 to 13 months, had severe cerebral palsy with no independent sitting and less than 50 on the Mental Development Index. Therapy was given twice a week for up to 12 weeks. Taught and untaught requests were probed in the teaching situation and across a second familiar communication situation. Sigafoos 1995 reported the training of a six year old boy with severe cerebral palsy and moderate cognitive impairment to request three items using micro‐teaching strategies, requests were probed throughout treatment. Three sessions were given per week, with 19 sessions in total.

Davis 1998 taught two children to produce responses to statements made by others in conversation. One of the participants was a 15 year old boy with severe cerebral palsy who usually communicated by yes/no responses only but who had access to a voice output communication device with pre‐stored phrases and spelling for novel words. Communication partners provided structured opportunities for the boy to respond to statements in conversation with further information that maintained the interaction. These elicitations were added to the conversation of three partners in succession. Responses to statements were recorded across the treatment sessions with the three partners. Therapy was given two to three times per week, 36 sessions in total.

Hurlbut 1982 trained three teenage boys with severe cerebral palsy and cognitive impairments to label objects using Blissymbols or iconic line drawings using micro‐teaching strategies. The duration and frequency of therapy sessions was not stated. The proportion of Blissymbols and iconic symbols used to label taught and untaught items was calculated before and throughout training.

Speech production

One study focused on speech production (Pennington 2009b), training children to control the loudness of their speech and maintain their respiratory effort for speech. Sixteen children aged 12 to 18 years, 15 of whom had cerebral palsy and one who had Worster Drought Syndrome, were taught to control their speech rate and loudness. Individual sessions, lasting 35 to 40 minutes took place three times per week over 6 weeks. Speech intelligibility in single words and connected speech to familiar and unfamiliar listeners was measured at six weeks before, one week before, one week after and six weeks after intervention (ICF activity level outcomes).

Receptive vocabulary

Dada 2009 taught three children with cerebral palsy aged 8 to 12 years to understand 24 spoken words (ICF body function level outcome), using an aided language stimulation programme which involved pairing a spoken word with a graphic symbol. Eight vocabulary items were taught in one activity carried out five times in one week (each activity lasting 15 to 25 minutes). In the second week a different activity with eight new vocabulary items was used and the activity and vocabulary items were changed again in the third week, so that 24 vocabulary items were taught in total. Teaching comprised repeated pairing of spoken word and visual symbols representing the target vocabulary. The proportion of items correctly selected when named was measured three times a week prior to and across the three week intervention period.

Expressive language

Campbell 1982 used operant training techniques to teach a 10 year old boy with severe cerebral palsy and moderate language delay to produce "is/are" in three linguistic structures (ICF body function level outcome). Two 15 minute therapy sessions were given each day, with 155 sessions in total. Frequency of correct "is/are" production in each of the three target structures was measured during each training session.

Therapy focusing on parents or other conversation partners

Eight papers investigated the success of training communication partners to facilitate the communication development of children with cerebral palsy, measuring ICF environmental level outcomes (Basil 1992; Hanzlik 1989; McCollum 1984; McConachie 1997; Olswang 2006; Pennington 1996a; Pennington 2009a; Tait 2004). Pennington 1996a reported the same information, but in different format, as McConachie 1997 and will be excluded from further discussion.

Participants

Children

Children whose parents and educators received training in the seven studies appear heterogeneous. However, insufficient information was given to provide a clear picture of their overall level of functioning. They ranged from eight months of age to 17 years, had cerebral palsy classed as mild to severe and cognitive skills ranging from within normal limits to severely impaired. Hanzlik 1989 included 20 infants aged 8 to 32 months, who had cerebral palsy of different types and severity ranging from mild to severe. Mental age was at least one standard deviation below the mean, range two to 18 months. None of the infants were able to ambulate either independently or with aids. Some, although it was difficult to tell how many, fell into the category containing speech impairment. However, some of those children may not have been expected to communicate intentionally given their chronological and mental age. Levels of communication development were not specified. McCollum 1984 included one child with severe cerebral palsy, of unknown type, aged 18 months. He was reported to vocalise but to exhibit few social behaviours. No other information was given regarding his developmental level. Basil 1992 studied four Spanish children aged seven to eight years who had cerebral palsy of unstated type. They had no independent mobility and upper limb function was severely affected. One child scored 4.5 years on a test of mental development, the others did not reach baseline. These children communicated by vocalisation, eye gaze, facial expression and produced one symbol messages on their communication boards, which contained 52 to 188 symbols. McConachie 1997 included nine children aged seven to 17 years who had cerebral palsy of differing types. No information was given on the severity of their motor impairments, cognitive or sensory skills. All had symbol communication systems (six used Blissymbolics, three Rebus Symbols), with access to 175 to 1000 plus symbols. Two children also had voice output communication aids. No information was given about how the children used their communication systems or their communicative level. Olswang 2006 studied two children with cerebral palsy aged 14 and 20 months, who were unable to sit independently, had severe cognitive impairments and vision and hearing that were within normal limits. Both children were preverbal communicators, who looked at objects, but did not look back to the parents to request items. Pennington 2009a included 11 children, 10 of whom had cerebral palsy of differing types. The children were aged 19 to 36 months at the start of the study. Gross motor function was rated using the gross Motor Function Classifaction Scale (Palisano 1997); most children had difficulty using their hands but could bring two hands together to act on a toy. All had severe dysarthria and were unintelligible to their parents out of context. Most children had severe receptive language delay (Preschool Language Scales mean percentile rank = 6, SD = 9) and severe intellectual impairment. Tait 2004 studied six children aged 16 to 47 months who had spastic type cerebral palsy which affected all four limbs. Two children had cortical visual impairments, one child had a mild visual impairment, one child had a hearing impairment and two children had epilepsy. Four of the children had receptive language scores that were within six months of their chronological age, one child had receptive language skills within nine months of their chronological age and one child had a profound language delay and scored at a two month developmental level.

Adult conversation partners

With the exception of McConachie 1997, who trained teachers and education assistants, parents were the subjects of the research. Overall, very little information is provided on the people who were trained, their communication style before intervention, previous training and relationship with the participant children. None of the studies included information on parental stress and coping, which has been found to affect communication (Dunst 1988). Basil 1992 trained three mothers and one father. They were compared with teachers who received no training. No information was given on prior training or other characteristics of either group other than the pre‐intervention interaction measures which showed different communication styles between the two groups. The mothers who participated in Hanzlik 1989 had completed varying levels of education, from partial high school to college graduation. Half of the families in each group had other children. The employment of parents ranged from major professionals to semi‐skilled workers. However, it was not clear if any of the mothers were employed outside the home or how social status was classified. The mother in McCollum 1984 was a single parent with a lower‐middle income, no other information was given. McConachie 1997 included nine teachers and 10 assistants in the experimental group who received training and eight teachers and six assistants who received no intervention. No other information was given on the adult participants, who volunteered to take part and who were assigned by their managers to the two groups. Authors stated that the participants and controls were matched on gender and extent of interaction with the participant children, however no supporting evidence was presented in the paper. Olswang 2006 trained two mothers. One mother had older children, was Caucasian and had been born and brought up in the USA. The other was a second generation Ukranian immigrant living in the USA with her mother (who spoke only Ukranian) and had no other children. Both mothers had graduated from high school. No other details were provided. Pennington 2009a trained 11 mothers. Two were single parents; five had completed high school education; three had received received some further education and three had completed university degrees. Five mothers worked outside the home, four on a part‐time basis. Families lived in urban and rural areas and levels of deprivation varied across the sample. All mothers were White British or White Australian. Tait 2004 trained mothers of children in their study, but no information on the mothers was provided.

Intervention

The training given all related to facilitating communication development. McCollum 1984, Hanzlik 1989, Olswang 2006 and Tait 2004 concentrated on pre‐verbal communication. McCollum 1984 provided direct teaching of target skills specific to the parent and child receiving therapy. In total 10 weekly home visits were made, in which target behaviours were watched on video‐tape and practiced and treated and untreated communication behaviours measured. Hanzlik 1989 gave a generic model of training to each mother, focusing either on interaction and the use of adaptive seating for the experimental group, or neurodevelopmental therapy for the control group. Training in this study was given at home in one session that lasted one hour. Olswang 2006 and Pennington 2009a used a specified training protocol in which parents were all taught to create communication opportunities, to wait for their child to communicate, to recognise their child's communication and to shape this communication into more sophisticated signals. Olswang 2006 devised training specifically for their study. Individual training took place twice per week for three weeks. Each session lasted approximately 45 minutes. The parents in Pennington 2009a received It Takes Two to Talk®, the Hanen parent program, which was delivered to groups of parents in seven or eight sessions of two and half hours over 12 to 13 weeks. In both the intervention by Olswang and It Takes Two to Talk parents also received home visits for individual coaching. Tait 2004 developed an individualised training programme for each parent‐child pair, on how to respond contingently to their child's pre‐linguistic communication and how to prompt the use of more sophisticated communication signals. A written summary of the teaching strategies was also provided for parents. Parents then implemented the teaching strategies in 6 to 12 individual sessions of 30 minutes, which comprised three types of interactions: mealtime, play with a toy and social interaction (for example peek‐a‐boo, pretend play). Parents received feedback on their use of target strategies after each practice session. Basil 1992 and McConachie 1997 both undertook group teaching to facilitate interaction with individual AAC users. Basil 1992 trained a group of parents in one session then followed this training up with three home visits to each family to individualise intervention and help parents practice techniques. McConachie 1997 trained teachers and assistants in their own school in five 90 minute workshop sessions which concentrated on one child. Both Basil 1992 and McConachie 1997 used short talks, brainstorming and videotapes in their group teaching.

Outcome measures

Each study used outcome measures developed specifically for the research project, which related to the specific aims of the therapy. Only one (Hunt 1986) had information on validation. Inter‐rater reliability of use of the coding schemes was given in each paper.

Risk of bias in included studies

See Table 1and Table 2 for ratings of the methodological quality of included studies.

1. Methodological quality of group studies.

Study	Assignment	Allocation concealed	Eligibility criteria	Groups similar	Sample size	Blinding	Protocol compliance	Missing values	Loss to follow‐up
Basil 1992	I	I	U	I	I	U	U	G	U
Hanzlik 1989	I	I	A	A	U	I	P	G	G
McConachie 1997	I	I	U	U	U	I	A	G	P
Pennington 2009a	N/A	I	A	N/A	U	I	P	G	G
Pennington 2009b	N/A	A	P	N/A	P	A	A	G	G
KEY	A = Adequate	P = Partial	I = Inadequate	U = Unclear	N/A=not applicable			KEY G = Good P = Poor U = Unclear	KEY G = Good P = Poor U = Unclear

2. Methodological quality of single case studies.

Study	Subject description	Therapy description	Blinding	Control similar	Assignment	Baseline	Intervention	Follow‐up	Measurement
Campbell 1982	I	P	I	A	U	A	A	I	A
Dada 2009	A	A	P	A	U	A	A	P	A
Davis 1998	P	A	P	A	U	A	P	I	P
Hunt 1986	P	P	I	A	U	I	A	P	P
Hurlbut 1982	P	P	I	A	U	A	A	I	P
McCollum 1984	I	P	I	A	U	I	I	I	I
Olswang 2006	P	A	P	I	U	P	I	I	A
Pinder 1995	P	P	P	A	U	A	P	I	A
Richman 1977	I	A	I	I	U	A	I	I	A
Sigafoos 1995	I	A	I	A	U	I	I	I	I
Tait 2004	P	P	I	U	U	P	I	P	A

It is rarely possible or advisable to blind patients and clinicians to the type and aims of intervention in trials of speech and language therapy, but this does leave them open to performance and attrition bias.

Group studies

(Basil 1992; Hanzlik 1989; McConachie 1997; Pennington 2009a; Pennington 2009b)

1. Randomisation and concealment of allocation

Basil 1992, McConachie 1997, Pennington 2009a and Pennington 2009b did not randomly assign participants to treatment or control groups. Basil 1992 gave training to parents and compared their communication with that of teachers who received no training. Teachers and assistants who participated in the McConachie 1997 study were assigned to treatment and control group by their school managers on the basis of school timetable, as staff were released to participate in training workshops. Allocation was not concealed as the person(s) who allocated participants also decided on their eligibility. Both of the studies, therefore, had significant weakness in their allocation strategies and selection bias was likely. In both Pennington 2009a and Pennington 2009b an interrupted time series design was used in which participants acted as their own controls. In both studies participants were recruited through the researchers' speech and language therapy colleagues, who were given a list of inclusion and exclusion criteria. Referring therapists selected participants from their caseloads and thus selection bias is likely. Hanzlik 1989 recruited parents through colleagues. Parents who were willing to take part in the study were allocated to group as they were recruited by the investigator by her taking a piece of folded paper out of a bag; 20 papers were created, 10 consigned parents to control and 10 to experimental group.

2. Similarity of participants at baseline

Information on recruitment to the studies was not provided for Basil 1992 and McConachie 1997, nor were inclusion and exclusion criteria cited. For Basil 1992 participants and controls differed in their relationship to the children; parents received training, teachers were controls and received no training. The two groups clearly differed in their pre‐intervention patterns of interaction. No information was provided on other possible confounding variables such as previous training in communicating with children who use augmentative and alternative communication (AAC), beliefs about interaction, age, education, socio‐economic status, and extent of knowledge and experience of AAC. This study was rated as inadequate on participant similarity. Participants and controls in McConachie 1997 were matched on gender, occupation and extent of contact with the target children by managers. Pre‐intervention communication ratings and information on possible confounders such as those listed above were not given. Therefore, it was not possible to detect how similar the two groups were before training. Hanzlik 1989 provided sufficient information on participants to assess the similarity of the groups and to replicate the research with similar samples. She cited inclusion criteria that related to children's locomotor, cognitive and sensory skills and excluded mothers who had received previous training in either of the intervention strategies used in the study. The gender, type and severity of cerebral palsy, extent of locomotor skills, chronological and mental age was given for the children in each group in terms of frequencies, means and SDs, with groups seeming to be equally matched. Mothers were similar across groups in education and half of those in each group had other children. The range of socio‐economic status (SES) of the households of the two groups was slightly wider for the control group, and the numbers of participant families in each SES group is not given. Pre‐intervention scores (means and SDs) for interaction behaviours were given for the mothers and infants in both groups and appear similar. Pennington 2009a provided sufficient detail on participants to allow replication. Selection criteria were chosen to reflect the population of children with cerebral palsy referred to speech and language therapy clinics. Criteria related to presence, type and extent of motor disorder; vision; hearing; communication and lack of previous parental communication training. The gender, type and severity of cerebral palsy, extent of locomotor skills, upper limb function, speech production, chronological age, receptive language, nonverbal understanding and expressive vocabulary is given for all children. The number of mothers in the group with university degrees was higher than in the general population. Pennington 2009b provided exclusion and inclusion criteria. Information was provided on group scores for type and distribution of motor disorder, severity of dysarthria, age and gender and ability to follow simple instructions. No further information was given on children's cognitive and language functioning.

3. Participant numbers

Most of the studies provided information on how the number of participants was chosen. Pennington 2009b stated that sample size was determined by feasibility, with restrictions imposed by the data collection schedule, therapy duration and by the school day and term times. Number of participants ranged from eight (Basil 1992) to 20 (Hanzlik 1989). With such small numbers of participants it is unlikely that the sample can reflect the population of people who regularly converse with children who have cerebral palsy. The studies also have very low chances of detecting a true effect of training.

4. Blinding

Appropriately for therapies involving training and participant co‐operation, none of the studies included the blinding of the participants or of the clinicians providing therapy. However, with the exception of Pennington 2009a and Pennington 2009b the outcomes of the interventions were inappropriately assessed by the clinicians providing the therapy, which increased the risk of detection bias. In Pennington 2009a the first coder collected the data and was aware of time at which data were collected when analysing behaviour. A second (naive) rater coded a random 20% of set of data from each participant, thereby reducing but not eliminating detection bias. In Pennington 2009b listeners heard speech recordings blind to the data collection point. In the studies in which the therapists who undertook intervention also measured outcomes bias was reduced, but not eliminated, by the inclusion of a reliability check of coding with a blind assessor. Basil 1992 checked a nonrandom sample of 12.5% sessions from before, during and after therapy, with agreement 90%, 92%, 98%. Hanzlik 1989 reported K = 0.75 to 1.0 agreement from data from each of the children, across 14 categories but did not state the amount of data on which this was calculated. In McConachie 1997 half of the data were coded by the second author, half by a blind assessor, with agreement calculated as 76% (71% to 79%) on 15% of the total data. As only small proportions of data were included in the reliability checks, each of the studies was still open to detection bias.

5. Description of the intervention

From the information given in the studies it would not be possible to replicate the intervention provided by Basil 1992 or Pennington 2009b. It was also unclear how similar the intervention was between participants within the groups. McConachie 1997 provided fuller description of the intervention and the training programme used has been published (Pennington 1993), allowing replication. Hanzlik 1989 provided additional information on the treatment protocols, which would allow partial replication. Pennington 2009a followed a well known, published therapy protocol for which therapists also receive training and certification.

6. Analysis

Data were analysed in the category to which participants were originally allocated. No cross‐over was reported or could be detected in any of the three group comparison studies. Basil 1992 and Hanzlik 1989 analysed data from the very small numbers of participants as groups, using parametric tests, which were unsuitable for such a small sample size. McConachie 1997, Pennington 2009a and Pennington 2009b used appropriate statistical tests. Hanzlik 1989 measured 14 variables from the samples of interaction, and Basil 1992 measured 10, increasing the likelihood of obtaining a statistically significant result by chance. However, this was not taken into account in the authors' conclusions. Losses to follow up occurred only in McConachie 1997, where a high attrition rate was observed, especially for the control group. The attrition was unexplained and left the study open to attrition bias.

Single case studies

1. Participant description

For replication of single case studies and moving from hypothesis generation to hypothesis testing participants need to be described in detail. All of the studies included in the review gave the participant child's chronological age and most gave a rating of their severity of cerebral palsy (mild, moderate, severe). Some gave children's type of cerebral palsy and rated the severity of any additional cognitive impairments. Few gave information on sensory impairments (Olswang 2006; Pinder 1995; Tait 2004) and epilepsy or details of children's receptive language development. Most cited children's present modes of communication and gave a very brief overview of their use of their communication skills in interaction. Only Dada 2009 was judged to include sufficient detail to select, with certainty, other participants with a similar type of cerebral palsy, level of locomotor skills, cognitive and communication development. Davis 1998, Hunt 1986, Hurlbut 1982, Olswang 2006, Pinder 1995 and Tait 2004 were judged to give a partial account of children's level of functioning. The descriptions given by Campbell 1982, McCollum 1984, Richman 1977 and Sigafoos 1995 were judged to be inadequate for replicating the study.

2. Equality of skills assigned to treatment and control

Intervention is deemed to have an effect if outcome measures change at the point of, or after, intervention for each child in succession but no change is observed prior to intervention. Allocation of skills to control and treatment and similarity of skills in treatment and control were judged to be not applicable for this study. When different skills are assigned to treatment and control, in order to avoid selection bias and the effect of maturation, skills need to be of similar developmental level and prognostic indication and assigned at random to treatment or control with later treatment in multiple baseline designs. Richman 1977 compared communication skills with a motor skill. Tait 2004 compared looking at an object with looking at a graphic symbol, for which the visual‐cognitive processing is quite different. Pinder 1995 and Sigafoos 1995 selected target skills that were very similar, and which may have been expected to generalise for the included participants. Therefore, an increase in control skill as well as treated skill would be expected. The other studies investigated skills of similar prognostic indication and were rated adequate in skill selection. However, none of the studies stated if skills were assigned to treatment (or a place in a sequence of treatments for multiple baseline across processes designs) or control randomly, which could introduce selection bias. All studies were rated as unclear on this criterion.

3. Description of the intervention

For single case studies to be replicated, interventions, which are often innovative in these designs, need to be described in detail. Dada 2009, Davis 1998, Olswang 2006, Richman 1977 and Sigafoos 1995 were judged to describe the intervention in sufficient detail for it to be replicated. Dada 2009, Davis 1998, Hunt 1986; Olswang 2006 and Tait 2004 reported checks of treatment integrity, which show fidelity of treatment across participants and that treatment was undertaken according to the protocol. Campbell 1982, Hunt 1986, Hurlbut 1982, McCollum 1984, Pinder 1995 and Tait 2004 were judged to give only part of the information needed to replicate intervention. Information was usually provided on the frequency and duration of treatment but was lacking on the exact methods of eliciting skills from individual children. For example, incomplete information was provided on which communication situations were used to elicit particular skills, the point in an activity at which communication opportunities were provided and the methods used to teach a communication strategy to a mother.

4. Blinding

None of the studies included blinding participants or clinicians to the aims or type of therapy. In all studies except Olswang 2006 data on outcome measures were collected by the investigators, but included checks on the reliability of coding by a second observer, which could reduce detection bias. Olswang 2006 used students who were trained to criterion on communication coding, but it was not clear if coding was undertaken blind to time of data recording. All studies except McCollum 1984 used data collected from the participants during the study. Amount of data checked ranged from 17% to 50%; only that used by Pinder 1995 was selected randomly. Most studies calculated agreement using percentage (agreement‐disagreements/total number of behaviours coded), which does not adjust for chance agreement. Agreement ranged from 75% to 100%. Pinder 1995 calculated agreement using Kappa (K), achieving more than K = 0.60 for each participant. Taking into account the amount of data checked, the selection method used and the agreement achieved, Dada 2009, Davis 1998, Olswang 2006 and Pinder 1995 were judged to partially meet the blinding criterion. Campbell 1982, Hunt 1986, Hurlbut 1982, McCollum 1984, Richman 1977, Sigafoos 1995 and Tait 2004 were judged inadequate and to be at considerable risk of detection bias.

5. Duration of phases and measurement

To show that intervention leads to change in single case experiments, frequent measurements should be taken in baseline, intervention and follow up or maintenance phases, and phases should be of similar duration. Without the use of randomisation tests (Edgington 1995), baseline should be adequately established with a plateau across at least three measurements or with a downward trend. If treatment is successful a clear upward trend should be observed during the intervention phase. In studies aiming for the acquisition of new skills the behaviour should continue at similar levels to the intervention phase in follow up or maintenance with no intervention.

Campbell 1982, Dada 2009, Davis 1998, Hurlbut 1982, Pinder 1995 and Richman 1977 showed baselines that were adequate, with demonstration of stable behaviours. Hunt 1986, McCollum 1984 and Sigafoos 1995 did not demonstrate stable behaviour at baseline and were rated inadequate. Some participants in Olswang 2006 and Tait 2004 did not show a stable baseline and these studies are rated as partial. The interventions in Campbell 1982, Dada 2009, Hunt 1986 and Hurlbut 1982 showed a clear upward trend in target behaviour. Similar changes were partially demonstrated by Davis 1998, Pinder 1995 and Tait 2004 with higher scores than baseline but variability. In Olswang 2006 and Sigafoos 1995 scores were higher in intervention but were variable within treatment, which should have been continued to investigate possible trends. No clear trends were demonstrated by McCollum 1984 or Richman 1977 with lots of variation in the scores. The follow‐up phases of all studies were rated as partial or inadequate due to their absence, short duration or change in target behaviours.

Measurements of all target skills were taken continuously across phases by Campbell 1982, Dada 2009, Olswang 2006, Pinder 1995, Richman 1977 and Tait 2004. Data across sessions were aggregated by Davis 1998, taken infrequently for control behaviours by Hunt 1986 and presented as means by Hurlbut 1982; partially meeting the criterion relating to measurement. McCollum 1984 and Sigafoos 1995 included one measurement only for follow up and Sigafoos 1995 measured control processes at baseline and follow up only.

6. Confounding variables

None of the studies discussed confounding variables and all were rated unclear on this criterion. It is possible that for Sigafoos 1995 and Pinder 1995 the control skill was too similar to the treated skill and would be expected to generalise without treatment for the participants.

7. Analysis

Statistical tests have been developed for single case experimental designs (Edgington 1995). However they have not been widely used and none of the studies included in the review employed statistical analysis. Analysis involved visual inspection of the graphed data and subjective interpretation.

8. Replication

Dada 2009, Hurlbut 1982, Olswang 2006, Pinder 1995 and Tait 2004 included replication across participants, who appeared similar in prognostic indication. Other studies included in the review did not systematically replicate their interventions to other children with cerebral palsy. Some included children with other medical diagnoses.

Effects of interventions

Studies of interventions focused on children

The studies focusing on children aimed to facilitate different aspects of communication development. Each aspect targeted is discussed separately.

Pre‐intentional communication

Richman 1977aimed to increase a child's amount of eye contact, time she kept her head in an upright position and her imitative vocalisations. These behaviours were compared with control of drooling, which received no intervention. Wide variation was seen in each of the behaviours across baseline. Increases in each behaviour were observed during their individual intervention phases. Behaviours reduced during reversal and then increased again once the treatment was recommenced. However, during the second treatment phase behaviours did not reach the levels of the initial treatment phase. Follow up at one month after intervention had ceased showed similar levels to the second treatment phase for head control and imitative vocalisation. Increased scores were observed for the three behaviours at 12 month follow up.

Communicative functions

Hunt 1986, Pinder 1995 and Sigafoos 1995 all trained children to produce requests. Hunt 1986 taught one girl to make requests for objects or actions in a multiple baseline design. Baseline was stable, showing infrequent use of any of the requests. The first request showed a steady increase and reached criterion (three successive correctly produced requests) in 16 sessions, the second in the sequence was produced without direct teaching. The third request in the sequence also increased steadily in the intervention phase reaching criterion in 13 sessions. The final request also generalised without direct teaching. Pinder 1995 taught four children to request either an object or 'more' by looking at the adult and the object, the untaught request acted as a control. Requests were taught in play with toys and also assessed in snack time as a generalisation situation. Baselines were stable for three of the children, with requests made to less than 20% of probes. For one child, who had earlier been taught to make the same requests by actively reaching towards an object, increases in the target behaviour appear to have been made towards the end of baseline. For each of the four children increases in the production of both the taught and untaught requests were observed during intervention across both the treatment and generalisation situations. For two children increases were noted with the onset of intervention. For the other two increases in the behaviours were observed after three to four sessions of therapy. Levels of requests were maintained for four weeks after therapy had been withdrawn. Sigafoos 1995 aimed to teach a boy to use three requests for objects in a multiple baseline design. During baseline percentage correct production of the three requests (not separated) ranged from 0% to 35%. For the first request production increased to 35% to 60% with verbal prompting and increased to 80% to 100% when expectant delay was used and verbal prompts were faded. However, although requests increased from the first to the second phase of intervention they showed a downward trend in the latter part of the second phase. The other target requests were tested after intervention for the first and were correct for 65% and 30% of 17 trials, showing some generalisation. The trial was then stopped due to the school year ending.

Davis 1998 trained a boy to produce responses to statements in conversation partners in a multiple baseline design across three communication partners, by pairing obligatory requests (questions) with a nonobligatory request (statement). Prior to intervention responses to statements were rare, being produced following 0% to 20% of statements made by each of the three partners in conversation (means = 1.8%, 2.5% and 4.0%). During intervention responses immediately increased, following an average of 41.7% and 52% of statements by the first two partners. Increases were only observed with the individual partners once the treatment had started. However, there was considerable variation in frequency of responses during intervention, ranging from 0% to 60% and 20% to 80% with each partner. Intervention was not carried out with the third partner due to the child's family moving away from the area in which the research was conducted. Responses to statements with this partner remained at baseline level throughout the study.

Hurlbut 1982 trained three children to use Bliss and iconic symbols to name objects. For each child trials to criterion were faster for iconic symbols than Bliss. Each child also produced iconic symbols more frequently than Blissymbols in maintenance and generalisation probes, and named more untrained objects using iconic symbols than Bliss.

Expressive language

Campbell 1982 taught one child to use "is/are" in three linguistic structures in a multiple baseline design. In baseline is/are were produced correctly in 0% to 10% of wh questions, 0% to 10% of yes/no reversal questions and 0% to 35% of statements. For the first two structures baselines were stable, whereas statements seemed to show an upward trend in correct production. During intervention the percentage of correct productions rose steeply for all three targeted structures. Levels were also maintained at a much higher rate than baseline for these structures, but showed considerable variation during the maintenance phase. Generalisation to use in spontaneous speech showed increases from baseline for yes/no questions, but much lower levels than observed with intervention. Wide variation was noted for the generalisation of is/are in statements, with no clear pattern observed during baseline, treatment or maintenance phases.

Receptive vocabulary

Dada 2009 taught four children to select 24 graphic symbols when named. During baseline two children selected two out of the 24 items named. During intervention the percentage of correct identification rose steeply for all target items. During follow up children continued to select items from the first two sets of vocabulary items. However, follow up was not long enough to show retention of the third set of taught words.

Speech production

Pennington 2009b delivered dysarthria therapy focusing on respiratory and phonatory control, and control of speech rate and phrase length. No change in percentage of words in single word utterances and connected speech that were understandable to familiar and unfamiliar adults was observed at six weeks and one week prior to treatment. Following treatment the estimated in increase in intelligibility to familiar listeners was 14.7% (95% CI 9.8 to 19.5) for single words and 12.1% (95% CI 4.3% to 20.0%) for connected speech. For unfamiliar listeners the immediate post‐therapy estimated increase was 15.0% (95% CI 11.73% to 18.17%) for single words and 15.9% (95% CI 11.8% to 20.0%) for words in connected speech. No differences were observed between post‐intervention scores and follow‐up scores taken at one and six weeks after intervention completion for either single words or connected speech when heard by either familiar or unfamiliar listeners.

Therapy focusing on communication partners

Parents

Basil 1992, Hanzlik 1989, McCollum 1984, Olswang 2006, Pennington 2009a, Tait 2004 all trained parents, with the intention of changing their interaction style and thus facilitate children's communication. Basil 1992 found no difference between the percentage of turns taken in conversation, or the proportion of responses to children's utterances by trained parents or untrained teachers before and after intervention. Parents asked fewer open questions than teachers prior to therapy, but increased these after intervention whilst teachers' use of open questions remained stable ( F (3, 1) = 8.35, P = 0.063). After one hour of instruction parents in Hanzlik 1989 changed behaviour that related to 'doing', but not that which involved verbal interaction. Mothers who received instruction on changing physical and verbal interaction used less physical guidance (F(1, 18) = 6.34, P = 0.02), more face to face contacts (F(1, 18) = 28.49, P = 0.00005) and less physical contact (F(1, 18) = 10.11, P = 0.005) than mothers in the control group who received neurodevelopmental therapy. No differences were observed in mothers' verbal directiveness, praise, questions or verbal interaction before and after instruction for either group. McCollum 1984 trained a mother to bring her face close to her child's. The behaviour increased from baseline and was maintained after intervention had finished. The mother's imitation of her child's vocalisation increased during intervention but showed a lot of variation, and a possible downward trend towards the end of treatment. The skills appeared to generalise to an untreated play situation, but were not maintained once treatment had stopped. Contrary to expectation one mother in the study by Olswang 2006 used higher rates of target behaviours at the start of each phase of the study and rates reduced during each of the three phases: baseline, intervention and withdrawal of treatment. Her child showed variable rates of engagement in interaction in each of the three phases, but a gradual increase from baseline to treatment withdrawal. The other mother showed a fairly stable baseline, increases in rates of target behaviours during the treatment phase and a reduction in rates of target behaviours when treatment was withdrawn. Her child's engagement was correspondingly stable during baseline, increased during intervention and reduced during the withdrawal phase. In Pennington 2009a no differences were observed in mothers' communication at four and one month prior to intervention. Mothers started fewer conversational exchanges in the month after therapy (t(10) = ‐2.730, P = .011, d = ‐.823) and responded more to children's communication (t(10) = 3.891, P = .002, d = 1.173) than in the month prior to training. In the month after intervention mothers also used fewer directives (t(10) = ‐2.630, P = .013, d = ‐.793). The changes observed in mothers' communication were maintained without further therapy at follow up, four months after intervention completion. The complexity of mothers' language directed to their children did not change during the study, nor did their responses to the Parental Sense of Competence questionnaire.

Teachers and educational assistants

Teachers and educational assistants who received training in McConachie 1997 used more strategies to facilitate children's communication four months after training (X² (4) = 15.84 P ≤ 0.01). Post hoc analysis suggested that these differences were already observable for teachers at one month post‐training, but not for assistants (X² (4) 11.82, P < 0.01). Teachers and assistants who did not receive training showed no changes in their communication patterns.

Secondary outcomes for children

Basil 1992, Hanzlik 1989 and McCollum 1984 also looked at changes in children's communication that were associated with training given to parents. In Basil 1992 prior to parent training children failed to respond to parents' interaction more often than to teachers', but increased their responses to parents after intervention (F (3, 1) = 17.94, P = 0.024). Similarly, children communicated less often using their symbol communication boards with their parents than with their teachers, but their use with parents increased after training (F (3, 1) = 16.93, P = 0.026). Hanzlik 1989 observed an increase in voluntary responsiveness (F (1, 18) = 11.53, P < 0.003) and less physically directed compliance (F (1 to 18) = 4.44, P < 0.05) but no differences in the amount of independent play for the infants whose mothers had received interaction training. The child in the study by McCollum 1984 showed an increase in vocalisation concurrent with his mother's training and increase in the frequency with which she brought her face close to her child's. The child of the mother in the study by Olswang 2006 who used higher rates of target behaviours at the start of each phase of the study but whose rates reduced during each phase showed variable rates of engagement in interaction in each of the three phases, but a gradual increase from baseline to treatment withdrawal. The child of the mother who had a fairly stable baseline, with increases in rates of target behaviours during the treatment phase and a reduction in rates of target behaviours when treatment was withdrawn, showed engagement which was correspondingly stable during baseline, increased during intervention and reduced during the withdrawal phase. The communication patterns of children in Pennington 2009a showed no differences between recordings taken at four and one month prior to parent training. Children initiated more conversations (t(10) = 3.150, P = .005, d = .950) and used more of their turns in conversation to control the interaction and their mother's behaviour (t(10) = 2.987, P = .007, d =.901) in the month after their parents had received training than in the month prior to training. Changes observed following therapy were maintained at four month follow‐up, during which time no further intervention was given to parents or children. Tait 2004 observed children's use of more sophisticated communication signals to make requests, make choices, protest and to show response to their name before, during and after their parents were given training in communication teaching techniques. Four children increased their use of more sophisticated communication signals in three communication activities during treatment and maintained the use of these more sophisticated signals during follow up. Two children produced more sophisticated signals for one out of three messages targeted. One of these two children may have maintained the new behaviour but one reverted to prelinguistic behaviours during follow up.

Discussion

Principal findings

• This exploratory review found 17 papers (reporting 16 studies) that investigated the effects of different methods of speech and language therapy (SLT) for children with cerebral palsy, who differed in age, type and severity of motor disorder, presence and severity of intellectual impairment, or their communication partners. Nine of these studies evaluated therapy that focused on children. Seven studies concentrated on adult conversational partners (one study contained data subsumed into another trial and the larger study only is discussed). Therapy for children targeted pre‐intentional communication skills, the use of individual communicative functions, expressive language, receptive vocabulary and speech production. Training for conversational partners included parents and education workers, teaching them to facilitate the communication of individual target children, usually augmentative communication system users.

• Although the results observed suggest possible trends in communication change, the methodological quality of the studies included in the review is generally poor and this review provides insufficient evidence to support the general effectiveness of SLT for either children with cerebral palsy or their communication partners.

• Since the original review in 2003 group studies have been published and show the potential impacts of therapy on children's speech production and training for parents on parent‐child interaction patterns. These studies provide the data necessary to develop rigorous controlled investigations of the effectiveness of therapy.

• The participants of the studies included in the review are heterogeneous and are often poorly described. Consensus on the description of participants and the choice of outcome measures in research reports is needed to establish potential clinical subgroups. Children and conversational partners within subgroups may resemble those with other primary disorders, for example children with severe cognitive impairment.

• Consensus is needed on the aims and methods of standard therapies targeting different areas of communication used with clinical subgroups. Once consensus is gained, investigations of the effectiveness of standard therapies can be developed. Consensus among communities of practice could be gained through focus groups followed by a survey of SLTs working in the clinical field.

• New therapy techniques should be applied in single case experimental designs, which should be rigorously designed and reported. These need to be replicated with similar participants, from a defined clinical subgroup, and evaluated in exploratory group trials. Should they show positive findings, the intervention should be tested in pragmatic trials.

• Participants in trials of SLT interventions should be followed up to evaluate the long term impact of therapy on communication activity, participation and quality of life.

• Valid, reliable generic outcome measures are needed to assess communication activity outcomes and allow cross trial comparisons.

• The acceptability of interventions for families has not been evaluated and needs further study.

Therapy focusing on children

Nine studies were found that investigated the effects of therapy given directly to children. All but one (Pennington 2009b) used single case experimental designs to show the impact of treatment for individual children. Children included in the studies ranged in age from infancy to late teens, had moderate to severe motor impairments, mild to severe speech, language and communication disorders and intellectual impairment ranging from mild to severe. Although each of the studies has methodological flaws, the provision of therapy does seem to be associated with increases in the production of pre‐intentional communication behaviours (Richman 1977), requests for objects or actions (Hunt 1986; Pinder 1995), responses to others' communication (Davis 1998) use of expressive language structures (Campbell 1982) and understanding of spoken words (Dada 2009) for the children studied. For three teenage children with severe cognitive impairment it appeared that iconic communication symbols were easier to acquire than Bliss symbols (Hurlbut 1982). For the studies using single case methodology we can only conclude that the intervention employed in the studies may have been effective in helping the individual children involved to develop communication skills. Given the methodology employed we cannot extend the findings to other children with cerebral palsy. Replication of the studies with other participants and exploratory group studies are needed to generalise findings to possible clinical subgroups, and move from hypothesis generation to hypothesis testing. The group study by Pennington 2009b, along with others described in the systematic review by Pennington, Miller and Robson (Pennington 2009), suggests that speech and language therapy targeting control of respiration and phonation for speech may be associated with gains in intelligibility to both familiar and unfamiliar listeners. Pennington 2009b provides the data necessary to develop rigorous controlled trials of the effectiveness of this type of intervention.

Training for conversation partners

The studies that focused on communication partners provided training in facilitating the communication skills of individual children for parents, teachers and education assistants. Four studies were group trials, three studies used single case experimental design. Four of the studies (Basil 1992; McCollum 1984; McConachie 1997; Olswang 2006) have serious methodological flaws and cannot demonstrate the effects of therapy for the participants who participated. The study by Hanzlik 1989, involved parents of infants receiving a one hour individual training session focusing on the use physical and verbal interaction techniques. Results suggest that following the short period of intervention mothers changed their interaction style using more face to face communication and less physical contact with their infants. Overall, interaction was rated as more positive following training, but use of verbal interaction strategies did not appear to change. Follow up was not included in the study, therefore it is not possible to determine if change was maintained for the participants or if children's communication development was facilitated. Replication of this study with follow up is needed to investigate the effectiveness of the training programme used. Pennington 2009a observed the effects of the group training programme It Takes Two to Talk® for a group of parents and their preschool children aged 19 to 36 months. Although this programme was not specifically designed for parents of children with motor impairments positive changes were observed in the communication patterns of mothers and their children after training. Changes were maintained four months after intervention without any further therapy. Furthermore, in a separate interview study (Pennington 2010) parents reported their experiences of the programme as largely positive, although some advised that more reference to augmentative and alternative communication (AAC) should be included when using the programme with families of children with severe motor disorders. Even with this suggestion parents appeared to view the training as acceptable and effective in helping to learn more about their child's communication needs and fostering their child's communication development. The study provides the information needed to develop a pragmatic randomised control trial of the effectiveness of the training programme for parents and their young children with motor disorders. Further, more rigorous investigations are needed of the training given in the other studies, as they aimed to teach the same communication strategies, which are widely acknowledged by clinicians to affect the communication of children with speech disorders and cerebral palsy.

Methodological quality of investigations

Participants

Children with cerebral palsy who receive SLT range in age from infancy to late teens and vary widely in their functional levels of movement, learning, communication, vision and hearing. When reporting new interventions it is necessary to describe for whom they may be suitable. However, the descriptions of children and adults who participated in the studies included in this review were generally poor. It would therefore not be possible to replicate most studies or to decide whether children on clinical caseloads were similar to those in the original study and may benefit from the intervention. Descriptions of subjects should include all features that may confound studies. This includes children's chronological age, type and severity of cerebral palsy, gross and fine motor functioning, cognitive developmental level, presence, type and severity of epilepsy, sensory skills, receptive and expressive language development, educational placement and previous therapy. Valid systems now exist to classify the motor skills of children with cerebral palsy (Eliasson 2006; Palisano 2007) and these should be routinely applied in research. Communication skills should be described in detail, and should include measures of speech intelligibility, methods of communication used and communicative functions produced in conversation. Schemes are currently being developed to classify the communication of children with cerebral palsy (Barty 2009; Hidecker 2009) and it is hoped that in the near future they will be validated for use in research For training of conversation partners details of their relationship to target children, gender, educational level, previous training and present communication style should be given. With such descriptions it may be possible to identify clinical subgroups of children with cerebral palsy who display similar skills and who react to interventions in similar manners. However, as cerebral palsy is associated with a wide range of disorders it is possible that some children will not fit into such groups and the evaluation of interventions for them will comprise N of one trials.

Intervention

The interventions investigated in the studies included in this review were generally well described and their primary features could be replicated. For therapy focusing directly on children, techniques included operant and micro‐teaching. Training for conversation partners included short talks, brainstorming, video examples, practice and feedback. Full description with examples of interaction during intervention would facilitate replication. However, some differences would still be likely to occur due to the fluid nature of conversation and effects of different communication environments and circumstances. Some studies (Dada 2009; Davis 1998; Hunt 1986; Olswang 2006; Tait 2004) reported checks of treatment integrity, which should be included in study design to show constancy of treatment across participants and that treatment was undertaken according to the protocol.

Blinding

Due to the nature of participation in therapy and training it is not possible to blind participants and clinicians to therapy, which leaves trials of SLT, including those in this review, open to attrition bias. People may agree to participate, but withdraw when allocated to the intervention they least support. Attempts were made in each of the studies to reduce detection bias by including checks of data coding by a second rater. To improve the rigour of studies, outcomes should be assessed by persons other than those giving the therapy, who are blind to the allocation of treatment and control.

Sample size

The group studies in the review were exploratory in nature. Some (Pennington 2009a; Pennington 2009b) could be used by future researchers to calculate sample sizes reliably to test the effects of similar interventions.

Single case experimental design

It is important to show, in these hypothesis generating studies, that intervention addresses a target behaviour and that changes in behaviour are not due to maturation. This demands the establishment of an adequate baseline, with sufficient data collection points throughout the baseline, intervention and follow‐up phases, and the comparison of a treated skill with an untreated behaviour that is similar in prognostic indication. Some studies failed to show that behaviours were stable before therapy, and it is therefore possible that behaviours attributed to intervention may have developed without it. Randomisation tests may have addressed the lack of a stable baseline, but these were not used. Other studies included control behaviours that were untreated, or treated later, which were too similar to the treated behaviour and also changed, probably as a result of the intervention. One of the studies (Richman 1977) used a motor skill which would not have been expected to show the same pattern of development as the treated communication skills. None of the studies included adequate follow up to show the maintenance of behaviour change, which is vital if we are seeking to show the acquisition of communication skills.

Outcome measures

Communication

The aim of SLT is to improve communication. As such, outcome measures should relate directly to aspects of communication behaviour. Depending on the particular difficulties children experience, therapy could aim to improve a child's speech production, understanding, expressive language, voice, range of communicative functions or use of an augmentative or alternative communication system. Training for parents and other communication partners involves changing their communication patterns to give children opportunities to develop and use new communication skills. The studies involved in this review targeted different aspects of communication and used different outcome measures. Even studies that looked at similar skills, for example those targeting requests for objects and actions, used different measures to evaluate outcome. This makes replication of studies harder than if generic tools were used, as clinicians and researchers need to be trained to use the measures reliably. The use of the same outcome measures across studies would also help in the collection of a bank of information about the communication of children with cerebral palsy and their conversation partners, in the formation of clinical subgroups and in the assessment of the clinical significance of reported interventions. In addition to describing the change in the individual skills targeted it would be useful if authors examined the rate of change in other areas of communication, using well known outcome measures. This would provide rates of change for individuals and groups that may or may not be associated with intervention and which may be used to aid clinical practice and to inform future research.

Quality of life and participation

Recent research has observed an negative correlation between communication skill and both relationships with parents and participation of children with cerebral palsy (Dickinson 2007; Fauconnier 2009). With the exception of Pennington 2009a, which had an accompanying qualitative study (Pennington 2010), the studies included in this review did not examine the wider impact of therapy. In addition to investigating the change in children's communication, and that of their conversation partners, it is important to examine if children and their families find interventions acceptable and worthwhile, and if interventions are associated with children's increased quality of life and participation in social, education and family life. Such additional information could be gained by the use of published measures of participation, family stress and functioning and through interviews with both parents and children. Qualitative studies are now more prevalent (Clarke 2001; Goldbart 2004; Lund 2007a; Lund 2007b; Marshall 2008) and some have shown unexpected results. For example in Clarke 2001 young AAC users supported a model in which children are withdrawn from classrooms to learn new communication skills, contrary to current clinical practice in which skills are taught in normal class activities. We cannot assume that parents and families involved in the studies of this review view the intervention they received positively as therapy was of short duration and with minimal follow up, making attrition due to unsuitability of treatment less likely.

Authors' conclusions

Implications for practice.

Considering the range of aspects of communication targeted, methods used and participants involved in the studies included in this review, and the methodological weaknesses of the studies, it is not possible to conclude at the present time that speech and language therapy focusing on children with cerebral palsy or their communication partners is more effective than no intervention at all. However, no evidence has been found of any harmful effects of SLT for children with cerebral palsy and their families, and therapy has not been shown to be ineffective. Changes in therapy provision are not warranted given current evidence.

Implications for research.

This exploratory review highlights the paucity of rigorous research on the effectiveness of speech and language therapy (SLT) that aims to improve the communication skills of children with cerebral palsy. Further research is needed to define possible clinical subgroups of children with cerebral palsy and their communication partners and to investigate the most effective methods of intervention for these subgroups. To this end a bank of research evidence is needed, including the following.

• Detailed description of research participants including their age, type and severity of cerebral palsy, gross and fine motor function (e.g. Gross Motor Function Classification System, Manual Ability Classification System), cognitive level, presence and type of epilepsy, sensory skills, receptive and expressive language skills, method of communication, range of communication skills and speech intelligibility. Where possible researchers should use the same validated measures across reports. Communication partners should also be described thoroughly, including information on their relationship to the child, age, gender, educational history, employment, previous training in communication, attitudes towards augmentation and alternative communication (AAC), present communicative style.

• Development of valid and reliable generic measures of speech function and communication activity for children with motor impairments.

• Definition of the methods currently used to treat different areas of communication development for (subgroups of) children with cerebral palsy and their conversational partners, gained through focus groups and surveys.

• Randomised controlled trials of the effectiveness of dysarthria therapy focusing on respiratory and phonatory control and communication training for parents.

• Rigorous series of single case experiments to test new interventions with clients from a potential subgroup, and for clients who do not fit inclusion criteria for identified subgroups.

• Exploratory trials of new interventions with groups of children/conversational partners to investigate the feasibility of using the new therapy in typical clinical situations and of extending the therapy to a group of clients who vary more than those involved in a single case series. If positive results are achieved these studies would lead to pragmatic trials comparing new and standard therapies for subgroups of children and conversation partners.

• Follow up of participants for at least three months after therapy to investigate the maintenance of skills development.

• The inclusion of participation and quality of life measures to evaluate wider impacts of interventions.

• Qualitative research studies to investigate children's and families' perceptions of intervention techniques and the need for these interventions.

As this review has shown, SLT for children with cerebral palsy is a complex intervention. Children have complex communication disorders, associated with their varied underlying impairments, and each disorder may require a different type of treatment. In addition, children will experience different social relationships and interact with many different people in many different environments, each of which will influence communication and its treatment. It is probable that because of the heterogeneity of the children, their conversational partners and their communicative environments, and the interaction between these variables, that a broad evaluation of the effectiveness of SLT for children with cerebral palsy may not be possible. Instead, evaluations should concentrate on the effectiveness of interventions given to ameliorate disorders affecting different areas and stages of speech, language and communication development for groups of clients with particular sets of skills and needs and to facilitate children's and families' participation in chosen life situations.

What's new

Date	Event	Description
1 March 2016	Amended	This review is being partially updated and will be replaced by a new protocol, and in due course, a new review. Please see 'Published notes' section for more information

History

Protocol first published: Issue 1, 2002
 Review first published: Issue 2, 2004

Date	Event	Description
6 February 2011	New search has been performed	Studies 2003‐2011 included in updated review. Conclusions changed.
31 January 2011	New search has been performed	Substantive update
27 March 2003	Amended	Review updated. No new studies.

Notes

This review is being partially updated and will be replaced by a new protocol, and in due course, a new review. The new review will focus on parent‐mediated interventions only.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Basil 1992.

Methods	Controlled before and after study, comparing communication strategies of trained (parents) versus untrained (teachers) communication partners. Assessor seemed to be experimenter. Reliability checked on non‐random sample.
Participants	Three mothers and 1 father of 4 Spanish children. Children: 3 F, 1M, aged 7.4‐8.8 years, severe CP affecting all four limbs, severe intellectual impairment, language age 3‐5 years, communicated using vocalisation, facial expression, eye gaze and Picture Communication Symbols (52‐188 symbols available). No details provided on subjects (parents) or controls (teachers).
Interventions	One group session on using communication boards, child's methods of selecting symbols, reducing own speech rate, prompting AAC use, asking open questions and increasing responses to child's communication, followed by 3 home visits of unspecified duration or frequency. Controls received no training.
Outcomes	Frequency of adult's initiations, responses, nonresponses, open questions, closed questions; and child's initiations, responses, nonresponses, utterances conveyed by their communication board, utterances conveyed by other modes (not AAC) were measured in 3 sessions before and 3 sessions after treatment by one non blinded assessor, and in 12.5% of sessions, with a blinded assessor. Agreement of treatment with protocol not reported.
Notes	Number of controls (teachers) unclear
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	No random allocation. Parents trained, teachers not trained.
Allocation concealment (selection bias)	Unclear risk	Not used
Blinding of outcome assessment (detection bias) All outcomes	High risk	Coding from transcribed interaction. Blinding of second rater only, who assessed only 12.5% of data. High agreement between raters (>90%).
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Campbell 1982.

Methods	Single case experimental design: within subject multiple baseline across 2 behaviours, plus one control untreated behaviour.
Participants	One boy aged 10 years with CP affecting lower limbs, and moderate language delay.
Interventions	Correct production of "is/are" in three syntactic structures ("wh" questions, "yes/no" reversal questions and statements) was reinforced using behaviour modification techniques. Two 15 minute sessions were given each school day, with 155 sessions in total.
Outcomes	Frequency of correct "is/are" production in the three target syntactic structures was recorded online by an unblinded observer in each training session, and by a second assessor in 17% of sessions.
Notes	Second single case using same design also reported in same paper. Second child did not have cerebral palsy and information not reported in this review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	High risk	Online, live data collection. Reliability between two independent raters on 17% of sessions ranged from 68‐90%
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Unclear risk	All expected outcomes reported

Dada 2009.

Methods	Single case experimental design replicated across participants: within subject multiple baseline across 3 activities.
Participants	Three children with fewer than 15 spoken words, aged 8‐12 years.
Interventions	Aided language stimulation. One set of 8 vocabulary items taught in a week, same activity repeated each day for five days. Activity 15‐25 minutes in duration. Three weeks intervention. Total of 24 vocabulary items taught.
Outcomes	Number of objects correctly selected when named.
Notes	Intervention targeted receptive vocabulary. Intervention provided in English. English was not children's first language.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	High risk	Coding from videotaped recordings. Not clear if the outcome assessor was blind to time of recording. 20% of data checked by second rater. High agreement between raters (>90%).
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Davis 1998.

Methods	Single case experimental design: multiple baseline design across 3 communication partners. Intervention was not implemented with one partner, who acted as control.
Participants	American boy aged 15 years, with spastic quadriplegia with athetosis, who communicated using vocalisation, gesture and one word phrases via voice output communication aid containing 500+ stored messages. Other development not reported. Communication partners: 2 female graduate students employed as home tutors of maths, reading and communication, and a male personal care attendant. No further details on the communication partners given.
Interventions	Communication partners trained to use nonobligatory requests in conversation to promote response. Treatment 2‐3 times per week at home. 36 sessions in total.
Outcomes	Percentage responses to blocks of 5 elicitation sequences was recorded by unblinded assessor. Reliability of treatment according to protocol and data coding were checked on 25% of sessions with a second, unblinded assessor.
Notes	Two children took part in the study. The second child did not have cerebral palsy and data from that subject is not included in this review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Online, live coding of interaction. Second, independent rater coded 25% of sessions. Inter‐rater agreement > 94%
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Hanzlik 1989.

Methods	Randomised controlled trial, comparing intervention with neurodevelopmental therapy.
Participants	Parents of 20 US children with non‐ambulatory cerebral palsy who had received no previous training on interaction or neurodevelopmental therapy. Children: aged up to 32 months, with no major sensory handicaps and mental age at least one standard deviation below mean for age. in experimental and control groups. Unstratified random allocation, with no attempt to match groups on possible confounding variables.
Interventions	Experimental group: One hour home visit from therapist with instruction on turn taking in interaction, increasing responsiveness, increasing face to face contact, reducing directiveness, and therapeutic holding techniques. Mothers practised techniques with the therapist and received an advice leaflet. Control group: One hour home visit from therapist to demonstrate neurodevelopmental therapy technique, practice with feedback from therapist and handout.
Outcomes	Parent‐child interaction during free play was video‐taped for 15 minutes during the visit in which intervention was given and again two weeks later. The proportion of 15 second samples in which target interaction strategies were observed was compared before and after training. Assessments were made by the therapist and by a blind assessor.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Methods: drawing lots
Allocation concealment (selection bias)	High risk	Lots to be drawn not adequately concealed
Blinding of outcome assessment (detection bias) All outcomes	Low risk	All interaction was coded by two researchers from videotape, one blind to the aims of the study. Inter‐rater agreement was high k<0.75
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Hunt 1986.

Methods	Single case experimental design. Multiple baseline across four request situations.
Participants	North American girl aged 7 years with severe intellectual impairment and multiple disabilities. No further details provided on underlying impairments. Communicated by vocalisation, 1 gesture, 2 manual signs, and by touching the listener. Could not use pictures for communication. Limited success matching representation to real object.
Interventions	Interrupted chain training of 4 requests. Treatment given twice daily in familiar routines, with 55 sessions in total.
Outcomes	Probes were made daily of the request currently under investigation. Content, form and function of communicative behaviour was assessed by therapist. Reliability of assessment was checked with by independent observer in 20% sessions.
Notes	Three children took part in the study. Only one had cerebral palsy. The other children's results will not be included in this review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	High risk	Online, live coding of interaction. Second, independent rater coded 20% of sessions. Inter‐rater agreement > 92%
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Hurlbut 1982.

Methods	Single case experimental design. Alternating treatments design across 3 subjects. Compared trials to acquisition and response generalisation for Blissymbols and iconic symbols.
Participants	Three US males, aged 14, 16, 18 years with severe spastic quadriplegia, moderate athetosis and severe choreoathetosis and severe speech impairment. No other further information supplied on cognitive and sensory skills. Communicated by idiosyncratic gestures, yes/no responses and 1‐3 Blissymbols (1).
Interventions	Participants trained to use 5 Blissymbols and 5 iconic symbols to criterion (10 correct responses) in response to "What's this?". Teaching strategies included modelling, verbal prompting, physical and verbal prompting and reinforcement. Duration and frequency of therapy sessions not specified.
Outcomes	Percentage correct naming of 10 trained and 10 untrained items using Bliss and iconic language was measured before and after intervention. Trials to acquisition for both systems was also calculated. Data were measured by an unblinded assessor, and by an independent observer on approximately half of the sessions.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Online, live coding of interaction. Second, independent rater coded 50% of baseline, 50% of intervention phase and 33‐50% of sessions in which spontaneous use of behaviours was coded. Mean inter‐rater agreement 98%
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

McCollum 1984.

Methods	Single case experimental design. Multiple baseline across 2 communication targets in two conditions treated (play without toys) and untreated (play with toy).
Participants	US mother and her son, aged 18 months, severe cerebral palsy. No further information given on developmental levels. Lower middle income, single parent family.
Interventions	Six weekly home visits of unspecified duration, training mother to move her face closer to child and to imitate child's vocalisation in play without toys, by modelling and providing feedback on practice.
Outcomes	Percentage of 10 second intervals containing vocalisations by the child and/or target behaviours by the mother were measured across 3 baseline, 6 treatment and 1 follow‐up sessions in play without toy (experimental condition) and play with toy (control). Reliability of coding was established prior to the experiment.
Notes	Two other children also included in study. They did not have cerebral palsy and their results are not reported in this review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	High risk	Interaction coded from videotapes. Assessor not blind to outcome or phase of study. Second rater coded 11% of data. Inter‐rater agreement >80%
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No missing data
Selective reporting (reporting bias)	Unclear risk	All expected outcomes reported

McConachie 1997.

Methods	Controlled before and after study in which school staff were assigned to training or control (no treatment) trial.
Participants	35 UK education staff who worked with 9 children (5 M, 4 F) aged 7‐17 years who had cerebral palsy of mixed, dystonic or spastic type, and who used symbol systems containing 175‐1000+ symbols to communicate. 2 children also used voice output communication aids. Experimental group: 9 teachers, 10 teaching assistants. Controls: 8 teachers, 6 assistants. Staff were allocated to group by school management and were matched on gender, occupation and extent of contact with child.
Interventions	Training followed "My Turn to Speak", which comprises five 90 minute sessions over 10‐12 weeks, concentrating on child's positioning on function, methods of accessing AAC systems, communication and communication breakdown. Teaching strategies included short talks, video demonstrations, written tasks, brainstorming, role play and intervention planning. Controls received no training related to AAC during the study.
Outcomes	Five minute video clips of each adult interacting with their target child were made in the month prior to training, one month after its completion and four months later. Facilitation of AAC users' communication was rated on 11 categories using a 3 point scale by one of the trainers and by a blind assessor coding 50% of the clips each, with reliability checked on approximately 20% of the clips.
Notes	High attrition rate, not explained.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Allocation assigned by school managers
Allocation concealment (selection bias)	High risk	Not used
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Two raters each coded 50% of data. One rater only blind to group allocation and data collection point. Agreement on 20% of data >80%
Incomplete outcome data (attrition bias) All outcomes	High risk	50% of controls lost at Time 2, >50% of participants and controls lost at Time 3
Selective reporting (reporting bias)	Unclear risk	All expected outcomes reported

Olswang 2006.

Methods	Single case experimental design replicated across participants
Participants	Two girls with quadriplegia 14 and 20 months of age, unable to sit independently, severe cognitive impairments, vision and hearing within normal limits. Preverbal communicators, who looked at objects, but did not look back to the parents to request items
Interventions	Responsive interaction training. Parents trained to respond contingently to child's communication and to shape more sophisticated signals. Two training sessions each week for three weeks.Sessions = 45 minutes.
Outcomes	Rate (number of times per minute) parents created communication opportunities, waited for their child to communicate, recognised their chid's communication and shaped the child's communication. Rate at which children looked actively at their parent or an object;looked back and forth between parent and object; produced a gesture; failed to respond; protested.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Outcome assessed by independent raters. Second raters independently coded one third of the data. Inter‐rater agreement k=0.78.
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Pennington 1996a.

Methods	Data included in McConachie 1997 study
Participants
Interventions
Outcomes
Notes

Pennington 2009a.

Methods	Interrupted time series
Participants	Eleven children with nonprogressive motor disorders. 10 children with cerebral palsy aged 19‐36 months; GMFCS median = 3, IQR = 2‐5; severe dysasrthria, unintelligible to familiar adults out of context.
Interventions	It Takes Two to Talk® the Hanen parent program.
Outcomes	Proportions of turns taken in conversation by mothers and children that were initiations or responses and proportions of turns that were directive. Complexity and rate of mothers' spoken language and number of different vocabulary items used. Parental Sense of Competence Scale (Gimbaud and Waterson, 1979).
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Participants acted as own controls
Allocation concealment (selection bias)	Unclear risk	Participants acted as own controls
Blinding of outcome assessment (detection bias) All outcomes	High risk	Interaction coded from videotapes. Primary rater not blind to data collection point. 20% of all recordings coded by independent second rater, k=0.75.
Incomplete outcome data (attrition bias) All outcomes	Low risk	One family lost to follow‐up
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Pennington 2009b.

Methods	Interrupted time series
Participants	15 children with cerebral palsy, 1 child with Worster Drought, aged 12‐18 years (M=14, SD = 2). 9 spastic type cerebral palsy, 2 dyskinetic, four mixed (spastic and dyskinetic). Dysarthria rate mild ‐ severe by referring therapists. All children able to comprehend simple instructions.
Interventions	Individual therapy focused on stabilising respiratory and phonatory effort and control, speech rate and phrase length/syllables per breath.
Outcomes	Percentage of words intelligible in single words and connected speech to familiar and unfamiliar listeners.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Participants acted as own controls
Allocation concealment (selection bias)	Unclear risk	Participants acted as own controls
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Listeners blind to time of recording
Incomplete outcome data (attrition bias) All outcomes	Low risk	One child's data missing at Time 1
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Pinder 1995.

Methods	4 single case experiments.
Participants	Four US children, (2 M, 2 F), aged 11.5‐13.5 months with mixed athetoid or spastic diplegia type cerebral palsy, who had difficulty grasping and releasing objects and did not sit independently. All with IQ < 50 Bailey Mental Development Index, vision correctable with glasses and hearing within normal limits.
Interventions	Twice weekly sessions of 50‐60 minutes for up to 12 weeks in which children were taught to request objects or request more by gaze and /or reaching and grasping. Teaching strategies included modelling, expectant delay and reinforcement.
Outcomes	Requests for more and requests for objects were probed once per week in play with toys (experimental condition) and at snack time (control condition). Unblinded assessor recorded response to elicitations and modes used to make response. Reliability checked with a second observer using randomly selected 20‐25% of data for each child.
Notes	Generalisation across acts expected. Design not able show effects of treatment.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	High risk	Coding of interaction from videotapes. Primary rater not blind to data collection point. Second rater, independently coded 22% of all data, k>0.69
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing data
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Richman 1977.

Methods	Single case experimental design. Multiple baseline with reversal and reinstatement of treatment across three behaviours.
Participants	US girl aged 9 years, severe spastic quadriplegia and severe cognitive impairment. No further developmental information supplied.
Interventions	Operant teaching strategies were used to encourage the maintenance of eye contact and head control and the production of vocal imitations in 10 minute therapy sessions given four days per week for 40 weeks.
Outcomes	Percentage of time eye contact and head control were maintained during each training session. Vocal imitation was requested 30 times in each session, percentage response recorded. Data collected during each session by the therapist. Reliability checked with a number of trained observers on 12.5% session.
Notes	Child absent for 3 sessions over treatment period.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Online, live coding. Second, independent observer coded 25% of samples, inter‐rater agreement >80% (mean = 92%)
Incomplete outcome data (attrition bias) All outcomes	Low risk	3/80 sessions missed
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Sigafoos 1995.

Methods	Single case experimental design.
Participants	Australian boy aged 6 years with severe cerebral palsy of unspecified type, who had moderate cognitive impairment, very poor upper limb control and required assistance for all activities of daily living. Participant was reported to understand various spoken commands and communicated using eye gaze.
Interventions	Trained to request objects by eye gaze in 19 sessions over 8 weeks. Teaching strategies included: creating communicative environment, expectant delay, verbal prompting, increasing expectant delay. reinforcement of response by use of object requested.
Outcomes	Therapist assessed percentage of trials in which object requested. Reliability of coding established with independent observer using approximately 50% of sessions.
Notes	Requests for objects generalised across the three objects. All used in same activity, probably inter‐related in communication.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Online, live coding. Second, independent observer coded approximately 50% of samples, inter‐rater agreement >83%
Incomplete outcome data (attrition bias) All outcomes	High risk	Child absent from school for replication phase
Selective reporting (reporting bias)	Low risk	All expected outcomes reported

Tait 2004.

Methods	Single case experimental design: multiple baseline design replicated across subjects
Participants	Six children with spastic quadriplegia aged 16‐47 months (M= 29 months); 1 child had hearing impairment; 2 children had cortical visual impairments; 1 child had mild visual impairment; 2 children had epilepsy. 5 children's receptive language was 6‐9 months behind chronological age; 1 child had profound language delay equivalent to 2 month developmental level.
Interventions	Functional Communication Training. Parents taught to how to train their child to use more sophisticated communication signals in place of prelinguistic communication. 18‐21 practice sessions for parents to use strategies taught. Each session approximately 30 minutes, comprising three activities to elicit specific communication behaviours. Feedback and coaching given to parents after each session.
Outcomes	Number of communication opportunities provided by parents; number of prelinguistic communication signals produced by child; number of times signals replaced with more sophisticated communication behaviours.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not used ‐ single case experimental design
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Videotaped interactions coded. Second, independent observer coded approximately one third of samples, inter‐rater agreement ranged from 79‐99%, mean =93%.
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Baseline for all children for two of three activities tested
Selective reporting (reporting bias)	Unclear risk	All expected outcomes reported

(1). Blissymbols: symbol system with written words printed beneath symbol.
 (2). Allocation concealment: A = allocation could not be predicted, B = method of allocation not made clear, C = allocation could be predicted or circumvented, D = no random allocation of subject or process

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Abrahamsen 1989	Subjects with different aetiologes included. Not possible to study children with cerebral palsy as subgroup.
Alant 1996	Before and after study. No control group or control intervention.
Almeida 2005	Before and after single case study. No experimental control.
Amari 1999	Subject had cererbal palsy but speech production intact. Treated for selective mutism.
Batarowicz 2006	Description of intervention and case studies. No experiment.
Bedrosian 1997	Conceptual review. No subjects or experimentation.
Bedrosian 1999	Conceptual review. No subjects or experimentation.
Binger 2008	Narrative review
Bishop 1994	Conceptual review. No subjects or experimentation.
Blackstone 1994	Discussion paper. No subjects or experimentation.
Boose 1999	Observational group study. No control group or control intervention.
Bruno 1989	Descriptive case study. No control processes or control intervention.
Bruno 1998	Before and after study. No control group or control intervention.
Buzolich 1991	Multiple baseline across subjects design. No control process.
Buzolich 1994	Before and after study. No control group or control intervention.
Carter 1998	Multiple baseline across subjects design, insufficient iterations across similar subjects. Baselines not adequately established. Intervention not replicable.
Chan 2002	Description of communication. No intervention.
Cohen 2000	Observational study. No control group or control intervention.
Darrah 2004	Generalised intervention approach, not specifically speech therapy
DiCarlo 2000	Subjects did not have cerebral palsy.
Dowden 1995	Conceptual review. No subjects or experimentation.
Durand 1993	Multiple baseline across subjects design. No control processes.
Enderby 1981	Description of an AAC system and users' views. No experimentation.
Erickson 1997	Descriptive case study design. No experimentation.
Fox 2005	Multiple baseline across subjects design. No control processes.
Galliers 1987	Descriptive case study. No experimentation. Aetiology of subject not given.
Gibbon 2003	No experimental control
Glennen 1985	Single case study with insufficient sessions at baseline and follow‐up to observe behaviour. "Control" process only tested at baseline and follow‐up.
Goossens 1985	Two descriptive case studies. No experimentation.
Goossens 1989	Descriptive case study. No experimentation.
Hall 1997	Multiple baseline across subjects design. No control processes.
Harris 1982	Description of communication. No experimentation.
Harris 1996	Subject did not have cerebral palsy.
Heim 1990	Description of communication. No experimentation.
Hetzroni 2000	Subjects did not have cerebral palsy.
Hooper 1987	Descriptive case study. No experimentation.
Horn 1996	Method of accessing AAC system rather than use for communication investigated.
Hsieh 1999a	Method of accessing AAC system rather than use for communication investigated.
Hsieh 1999b	Descriptive case study. No experimentation.
Hulme 1989	Investigated effects different types of seating. Not a communication intervention.
Hunt 1996	Not specific SLT intervention.
Hunt 2002	Not specific SLT intervention.
Hustad 2002a	Includes adults with cerebral palsy
Hustad 2002b	Includes adults with cerebral palsy
Hustad 2003a	Includes adults with cerebral palsy
Hustad 2003b	Includes adults with cerebral palsy
Hustad 2008a	Includes adults with cerebral palsy
Hustad 2008b	Includes adults with cerebral palsy
Iacono 1993	Subjects did not have cerebral palsy.
Jeffries 1987	Not a communication intervention.
Jouannaud 1972	Description of tongue movements in children with and without cerebral palsy. No intervention.
Kaiser 1993	Subjects did not have cerebral palsy.
Kent‐Walsh 2010	Mulitple probe design replicated across participants. Only one participant with cerebral palsy
Kertoy 2007	Not an intervention study
Ketelaar 1998	Critical review of intervention for children with cerebral palsy. No quantitative analysis.
King 1997	Social skills training, not specifically communication intervention.
King 1998	School based therapy services evaluated; speech and language therapy services cannot be disaggregated. Before and after study. No control group or control intervention.
Kozleski 1991	Single case design. Three treatment phases with one probe after each phase. Treatments not tested across all phases, therefore no control.
Kratzer 1993	Multiple baseline design across two subjects. Multiple interventions given consecutively to improve one skill. Treatments not measured across all phases, therefore no control.
Lagerman 1982	Descriptive case study. No experimentation.
Lancioni 2001	Multiple baseline across subjects design. No control processes. Intervention to increase switch activation, not communication intervention.
Light 1999	One child with cerebal palsy included in multiple baseline across subject design. When the subject examined individually design becomes ABABA with no control process.
Lovett 1994	Subjects with different aetiologies included. Not possible to study children with cerebral palsy as subgroup.
Mathisen 2009	Case study. No experimental control
McCarthy 2001	Observational study. No control group or control intervention.
McEwen 1989	Investigates success of diferent methods of access. Not a communication intervention.
McNairn 2000a	Discussion paper. No experimentation.
McNairn 2000b	Discussion paper. No experimentation.
McNairn 2000c	Discussion paper. No experimentation.
Mechling 2006	Participants did not have cerebral palsy
Millar 2006	Review paper, no experimentation
Mitchell 1995	Conceptual review. No experimentation.
Okimoto 2000	Investigates playfulness. Measures of communication cannot be disaggregated.
Olswang 1995	Observational, qualitative study.
Oxley 2000	Conceptual review. No experimentation.
Pahn 1972	Description of intervention. No experimentation.
Palmer 1997	Critical review of physiotherapy for children with cerebral palsy. Not speech and language therapy.
Park 1979	Description of an AAC system. No experimentation.
Parker 2006	No experimental control
Patel 2006	Case report. No experimental control.
Paul 1997	Conceptual review. No experimentation.
Pennington 2003	Earlier version of part of this review
Pennington 2005	Earlier version of part of this review
Pennington 2006	No experimental control
Pennington 2007	Qualitative study of speech and language therapists' views of It Tales Two to Talk© parent training
Pennington 2009	Systematic review of interventions.
Petersen 2010	Mulitple probe across participants. Only one participant with cerebral palsy.
Pirila 2007	Observational study. No intervention.
Pueyo 2008	Observational study. No intervention.
Puyuelo 2005	No experimental control
Ratcliff 1996	Before and after case study. No control processes or control intervention.
Ray 2001	No experimental control
Reichle 1999	Subjects did not have cerebral palsy.
Reinhartsen 1997	Description of intervention. No experimentation.
Romski 1988	Not a multiple baseline design across phases or subjects design. Description of instruction across phases with baseline, teaching and probe sessions.
Romski 1994a	Subjects with different aetiologies inlcuded. Not possible to study children with cerebral palsy as subgroup.
Romski 1994b	Subjects with different aetiologies inlcuded. Not possible to study children with cerebral palsy as subgroup.
Romski 1995	Subjects with different aetiologies inlcuded. Not possible to study children with cerebral palsy as subgroup. No control group.
Romski 1997	Conceptual review of research. No experimentation.
Rowland 2000	Subjects did not have cerebral palsy.
Salmien 2004	No experimental conrtol
Sanger 2007	No speech therapy intervention
Sevcik 1995	Subjects with different aetiologies inlcuded. Not possible to study children with cerebral palsy as subgroup.
Sigafoos 1999	Aetiology of subjects not given.
Sigafoos 2004	Review paper, no experimentation
Signorino 1997	Description of rehabilitation for deaf‐blind children, including those with cerebral palsy. No experimentation.
Soro 1993	Three single case studies. No control processes or control intervention.
Soto 2008	Mulitple probe across participants design. No control process.
Soto 2009	Single case AB design, no control process
Spiegel 1993	Single case study containing description of intervention with baseline and intervention scores for trained stimuli. Assessment of untrained stimuli before and after intervention.
Sternberg 1983	Subjects with different aetiologies included. Not possible to study children with cererbal palsy as a subgroup.
Swinehart‐Jones 2009	Literacy study
Thomas‐Stonell 2009	No experimental control. Chidlren with cerebral palsy cannot be separated from group.
Treviranus 1987	Observational study. No experimentation. Intervention to access devices, not communication intervention.
Truxler 2007	Intervention targeted phonological awareness, not communication
Udwin 1987a	Critical review. No experimentation.
Udwin 1987b	Observational study. No experimentation.
Udwin 1990	Observational study. No experimentation.
Udwin 1991	Observational study. No experimentation.
Watkins 1988	Observational study. No experimentation.
Woods 1997	Subject has mild hemiplegia arising from cerebral palsy, speech disorder incidental.
Worth 2001	Observational study. No experimentation.
Wu 2004	Before and after study. No experimental control.

Contributions of authors

All review authors devised the protocol and search strategy. L Pennington selected studies for inclusion, with reliability checks conducted by J Goldbart and J Marshall. All review authors were involved in extracting data from included studies and writing the review. L Pennington was the primary author.

Sources of support

Internal sources

• No sources of support supplied

External sources

• Royal College of Speech and Language Therapists, UK.

• National Institute for Health Research, UK.

  Salary funding for Lindsay Pennington during update of review. This report is independent research arising from a Career Development Fellowship supported by the National Institute for Health Research. The views expressed in this publication are those of the authors and not necessarily those of the NHS, the National Institute for Health Research or the Department of Health.

Declarations of interest

None

Edited (no change to conclusions)