Effectiveness and Determinant Variables of Augmentative and Alternative Communication Interventions in Cerebral Palsy Patients with Communication Deficit: a Systematic Review

Accepted: September 20, 2020 Study conducted at Armenian State Pedagogical University after Khachatur Abovyan – ASPU Yerevan, Armenia 1 Education and Research Foundation Yerevan, Armenia 2 Department of Speech and Rehabilitation Therapy, Armenian State Pedagogical University after Khachatur Abovyan – ASPU Yerevan, Armenia 3 Technologies for Management of Health Ltd Yerevan, Armenia 4 School of Medicine, Yerevan Haybusak University – Yerevan, Armenia 5 School of Public Health, American University of Armenia Yerevan, Armenia Financial support: nothing to declare. Conflict of interests: nothing to declare ABSTRACT


INTRODUCTION
issue that should be involved in the assessment process of CP patients as communication disorders may persist lifelong.
To go deeper in the mechanism of how a pediatric patient with CP responds to different categories of AAC interventions, we need to look at a scope of interventions and outcomes associated with them. In a review presented by Lund and Light (25) the patient's age, level of intellectual disability and intervention setting were analyzed as potential moderating factors influencing the efficacy of AAC interventions including picture-based systems like PECS and speech generating devices (SGDs). Both categories of AAC methods (PECS and SGDs) proved to be effective in improving the speech outcomes, but the data analysis revealed higher efficacy of SGDs in pediatric patients with CP without a co-manifestation of intellectual disability (25) . The evidence collected by Lund SK and Light focused on potential variety of predictive factors influencing the AAC intervention outcomes. The authors evaluated the predictive role of cognition and looked into this domain trying to reveal all social-cognitive factors involved. The work scrupulously analyzes all pre-intervention patient characteristics that possessed a potential to predict communication outcomes for patients with CP. The evidence suggests that the identified factors: age, comorbid conditions as well as social-cognitive factors are critical variables affecting the outcomes. The evidence accumulated is only for limited number of AAC methods such as PECS, SGDs and new studies are needed to evaluate the predicting or moderating role of all the influencing factors for a large variety of low or high-tech AAC devices. Identification of the possible predictors, moderators, and mediators that may influence the outcomes of AAC intervention is of major importance, as they should be discussed as potential targets included in AAC assessment questionnaires, making the possible outcome more effectively measureable.

PURPOSE
The objective of the presented review was to evaluate the effectiveness of AAC interventions in patients with CP, pointing on those determinant variables of main intervention outcomes that have to be considered by the specialists in the assessment and intervention process.

RESEARCH STRATEGY
The systematic review was conducted based on the Cochrane Collaboration guidelines (26) and reported using the statement of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) (27) .   (29) . Studies included in review articles and reference lists of the reviews were checked to reveal relevant studies. EndNote bibliographic software (Clarivate Analytics, Philadelphia, PA, U.S.A.) was used to identify and download the references.

SELECTION CRITERIA
Full-text and peer-reviewed studies in English studying the effectiveness of augmentative and alternative communication in patients with cerebral palsy were included. No publication date or publication status restrictions were imposed (including dissertations and theses to address publication bias). Two review authors (T.R.P. and H.H.M.) performed the screening of results independently and selected the articles by title and abstract according to initially elaborated eligibility criteria. Any disagreement between the two authors was resolved by a third author (F.A.S.) of the review. All selected items were independently checked as full text by the same review authors. In case of disagreement the full text was assessed by the third author (F.A.S.) to gain consensus. Studies with pediatric patients (<18 years) diagnosed with CP were included (Table 1). CP diagnosis should include the description of motor dysfunction (spastic, atetoid, ataxic), as well as define all other co-manifestations (sensory deficit, speech disorder, mental disorder, etc.).

TYPES OF INTERVENTIONS
AAC is defined as a multimodal intervention including at least (1) an access mode (e.g., upper limb function, eye gaze, etc.) and AAC method (e.g., unaided AAC (body movements, gestures, facial expressions, signs); aided AAC (objects, object symbols, photographs, picture symbol boards and books; communication devices). The formulation of AAC was based on the recently conducted studies and systematic reviews (30,31) . AAC interventions had to be provided individually or in a group and conducted by a speech therapist (or an aligned specialist) working in specialized centers and departments (special education center, a local primary care facilities, schools or preschool centers, office of a speech therapist, a hospital speech therapy department). For a more detailed analysis of variables contributing to outcomes, data reported by other partners (e.g., peers, family members) were also included in the review. Studies evaluating multidisciplinary interventions provided by different professionals were not included. The control interventions (if applicable) included primary care treatments without an AAC method (e.g., usual speech therapy, language therapy, swallowing and communication therapy), no treatment or the waiting list. Not all the studies included in this review were with a comparator (comparing strategy of intervention), yet we were interested in estimation of variables not only between the interventions, but also within the selected AAC intervention.

Types of outcome measures
The primary outcomes were determined based on the recently conducted reviews on communication therapy interventions (32,33) and included the following: receptive and expressive language skills. The primary outcomes were derived from assessment tools applied in included studies (Vocabulary Acquisition (Mean proportions of errors); MLU; PLS-4; FOCUS©; PPVT-R; TACL-R; GSRT; Analysis of conversational samples).
Secondary Outcomes were also derived from assessment tools applied in the pooled studies (Improvement in child's literacy skills; Reduction in child's frustration; Battelle age equivalence scores; Increase in CACE vocabulary; YC-PEM; CP-Quality of Life Questionnaire; DMQ; GAS; RAACS; Arc's Self-Determination Scale; The Quality of Life Profile for People with Physical and Sensory Disabilities). Outcomes were classified as short term (up to 3 months), medium term (>3 to 12 months), and long term (>12 months). Measurement categories and derived outcomes are presented in Table 2.

DATA ANALYSIS
Data extraction from the selected studies was performed by one of the review authors (T.R.P.) and then screened by a second review author (A.V.A) to increase the number of relevant studies identified for use in this systematic review. Extracted data contained relevant information about the study participants (age, gender, type and duration of intervention); intervention type (unaided or aided AAC methods); type of comparators (if applicable); type of analysis methodology; outcomes at baseline and at follow-up. The review of the extracted data from all selected studies revealed that the studies were heterogeneous to initiate a meta-analysis. Therefore, a narrative analysis was conducted to evaluate the efficacy of AAC methods. A randomeffects model meta-analysis was used to assess determinants of AAC intervention outcomes. The Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) approach was used to assess the overall quality of the evidence and the strength of recommendations that could be applicable across a wide range of interventions and studies. The GEADE provides the best estimate of the effect size for each outcome in the review. The overall quality rating in GRADE is addressed to the body of evidence across outcomes, and selects the lowest quality of evidence from all of the outcomes that are influential to decision making. Evidence was categorized as high, moderate, low or very lowquality evidence (34) . A random-effects model meta-analysis was conducted for each predictor. For this purpose comprehensive Meta-Analysis software (V. 2; Biostat, Englewood, NJ) was used. The significance was acknowledged at p < 0.05. We have used Cohen's (35) conventions for the interpretation of effects sizes as small (Cohen's d = 0.2, Pearson's r = 0.1), medium (Cohen's d = 0.5, Pearson's r = 0.3), or large (Cohen's d = 0.8, Pearson's r = 0.5). The main effects of each meta-analysis were processed using one of two statistics: Pearson's r to show the relationships between continuous predictors and outcomes; and Cohen's d. to show the relationships between nominal predictors and outcomes. The heterogeneity was assessed using a chi-square test of the Q value and estimated the magnitude of heterogeneity using the I 2 value (36) . The value of I 2 was classified as low (< 0.25), moderate (0.50), or high (0.75). Potential moderators of the relationship between predictors and outcomes were analyzed with meta-regression for the group of continuous variables (method-of-moments model) and subgroup analysis for nominal variables. The moderation analyses are applied in cases when moderate or high heterogeneity (I 2 > 0.25) was revealed (a minimum of eight studies are required for the meta-regression) (37) .

Risk for bias assessment
Risk of bias assessment aimed to provide transparency of results and findings presented in the review. It was performed for all included studies to eliminate bias in their findings, reported in results or conclusions. Risk of bias assessment tried to reveal the validity of findings and objectively estimate the intervention effect. The risk for bias assessment was performed independently by two review authors (T.R.P. and H.H.M.), and in case of disagreement a third review author (J.M.A.) was involved in the assessment process. The authors of the review were not blinded for the name of the journal, author names or institutions. Corresponding authors were contacted, if there was a need for additional data. To perform a quality analysis of included studies the EPHPP (Effective Public Health Practice Project) tool was used, as it ensures assessment of multiple studies with different study designs (38,39) . According to guidelines of EPHPP each study was assessed regarding its adequacy with respect to (a) addressing selection bias, (b) research design, (c) limiting confounders, (d) blinding, (e) data collection methods, and (f) accounting for withdrawals and dropouts from the study. In EPHPP assessment, the studies are rated as strong, moderate or weak on each component and the overall quality rating is derived from the combination of all scores. Studies with less than four strong component ratings and no more than one weak rating were considered to have an overall assessment of moderate. The authors individually evaluated studies against the EPHPP criteria and reached 85% on individual rating components and 100% on the overall rating score. Inconsistencies in the ratings were discussed until a 100% agreement was reached.

RESULTS
The online database search revealed 442 records (Figure 1). Three additional records were identified through manual reference checking. All the duplicates were removed and 257 records remained and were checked on title and abstract. The screening identified 19 items to be evaluated as full text for eligibility. After the screening 9 studies met predefined eligibility criteria and were included in the systematic review (40)(41)(42)(43)(44)(45)(46)(47)(48) . From 9 studies included in the systematic review 4 were studies with singlesubject multiple-baseline research designs, 3 were longitudinal cohort studies, 1 case control study and 1 case series. All of the selected items were written in English. Studies were conducted in the USA (40)(41)(42)(43)(44) , Malaysia (45) , Spain (46) , UK (47) and Australia (48) . Among the included studies seven were interventional and two observational studies. All studies provided data at baseline, and post-intervention assessment.

Participants
Participants of the studies were recruited via speech therapist (or an aligned professional) at special education centers, local primary care facilities, schools or preschool centers, offices of a speech therapist, hospital speech therapy departments. The retrospective searches of patient records were also included. The sample sizes of the selected studies ranged from 4 to 215 participants and the overall sample size of included CP patients was 309 (Table 3). Mean levels of expressive and receptive communication skills were assessed by different standardized measurement tools, with moderate to severe scores at baseline (Table 3).

Interventions
The Two studies compared an AAC intervention method with no speech and no AAC therapy (40,47) . One study used another AAC method (Computerized AAC devices for children without disability) as a comparator (41) . One other study (46) used the same AAC strategy in the control group with other category of patients (Table 3).  analyzed. Nine studies, involving 309 participants, contributed to the analysis of the relationship between the selected predictors and expressive and receptive language outcomes (Figure 2). According to the random effects meta-analysis results, there was a statistically significant main effect of the correlation between dysarthria severity and language skills (r = 0.369, p=0.029). Heterogeneity was moderate (I 2 =42.315, Q = 22.326, p < 0.05). Statistically significant correlation existed also between the cognitive ability and language skills (r=0.340, Q=12.717, P=0.042 and I 2 =29.228). However, none of the hypothesized moderators-playing skills of the child, play interaction, object exploration, speech imitation, joint attention explained a significant amount of heterogeneity. Examination of a funnel plot revealed asymmetry (Figure 3), and so we applied a Duval-Tweedie trim-and fill correction. This correction reduced the main effect but did not change the finding (r = 0.126, p < 0.01). Sensitivity analysis indicated that the finding is robust to the removal of any individual sample. Results of the meta-analyses can be categorized as medium significant effects, small significant effects, and nonsignificant effects. Visual impairments, hearing impairments and limb paresis demonstrated a small significant effect on expressive and receptive language outcomes.
Dysarthria severity and cognitive ability showed medium significant effects. The nominal variables such as the age

Speech therapist
The number of speech therapists (or aligned professionals) providing the AAC interventions was different within the selected studies (from an individual therapist to professional group intervention). Speech therapists mostly conducted training program with a duration ranging from 2 weeks to a maximum of 27 weeks (Table 3). Exception were two studies with singlesubject multiple-baseline research design that have evaluated the AAC efficacy after a longer application period (45,46) . Majority of therapists received special training and instructions on the applied AAC method.

Analysis of determinant variables that influence the AAC outcome in CP patients
The basic design of the review includes the assessment of AAC intervention efficacy in pediatric patients with CP discussing the effectiveness of different aided or unaided alternative communication strategies and pointing to factors that may potentially influence the efficacy of intervention. These determinant factors are potentially associated with intervention outcomes and in majority of studies are not properly considered or interpreted. In almost all studies where complex intervention methods are used the component factors involved are not for vocal play (AVP) and age for first words (AFW) showed nonsignificant effects. Our results indicate that dysarthria severity is a better predictor than hearing or visual impairments. The findings show that children with better cognitive ability have a better prognosis for language development. The results support the concept insisting that issues in comprehension are correlated with the most severe language disorders (49,50) . From the list of continuous variables the hearing or visual impairments showed small significant effect, possibly due to different aiding methods used in therapy process. The nonsignificant correlation between AFW (and/or AVP) and language outcomes shows that early word acquisition could be qualitatively different from the types of language/grammatical skills that are difficult for children with CP. The determinant factors can serve not only as predicting variables, but also may represent a group of variables that influence the strength of relationship between the predicting variable and the outcome. The described variables are moderators and in the studies included in this review they are identified as child characteristics: speech imitation, joint attention, play interaction, object exploration. These factors are not directly predicting the outcome but may enhance the influence of predictors (positively or negatively) upon the outcome variable. In addition, there is a need to emphasize the mediator variables that explain the mechanism by which the independent predicting variable affects the dependent outcome variable. Usually the mediators are analyzed after the study completion and are helpful to explain why and how the intervention was effective. Among the mediating factors identified in this review are the influences of the communication partners on different components of intervention. Mediators responsible for the reinforcement of predicting influences on outcomes are the frequency of therapy, communication partner knowledge, perception of AAC benefits, and parent (partner) input at home.

Risk for bias assessment
The number of studies included in the review was small and the risk for overestimation of AAC intervention efficacy was very high. Possible high risk is explained by variety of factors: research favoring positive results, different study design and large variety of outcomes. Strength evaluation for included studies and applied design was based on the Grading of Recommendations, Assessment, Development and Evaluation guidelines. The scores of GRADE (Table 4) represent different required components: level of evidence, quality of evidence, relevance of evidence; statistics precision and size of effect (51) . Our review does not contain any randomized control trials that are known to have less risks of bias. On the other hand selection of study design does not in itself increase or decrease the research quality. Effective Public Health Practice Project (EPHPP) was used to perform quality assessment of included multimodal studies: The Table 5 shows results of EPHPP assessment for individual studies and components, as well as the overall rating of quality assessment.

Outcomes
All studies included in the review reported improvement in the receptive and expressive language skills of patients with  CP. These primary outcomes were derived from corresponding assessment tools (Vocabulary Acquisition (Mean proportions of errors); MLU; PLS-4; FOCUS©; PPVT-R; TACL-R; GSRT; Analysis of conversational samples) applied in studies included in the review. Seven studies out of nine showed quantified evidence of improvement in communication skills. Secondary outcomes were also derived from assessment tools applied in the pooled studies (Improvement in child's literacy skills; Reduction in child's frustration; Battelle age equivalence scores; Increase in CACE vocabulary; YC-PEM; CP-Quality of Life Questionnaire; DMQ; GAS; RAACS; Arc's Self-Determination Scale; The Quality of Life Profile for People with Physical and Sensory Disabilities). They were distributed with certain level of overlap in included studies, though the design of the studies was very heterogeneous. All secondary outcomes were scores of standardized measurements (for participation motivation, quality of life, activities of daily living, goal attaining, responsiveness, improvement in child's literacy, level of frustration, developmental delay, cognitive deficits, self-determination) and in all studies deploying the specific tools significant improvement was registered. In five studies this improvement was statistically significant which exposes the instruments that are more reliable and flexible for use with various AAC strategies ( Table 3).

CONCLUSIONS
In general, this systematic review produced moderatequality evidence indicating that AAC interventions have proven medium and long-term efficacy in improving communication of CP patients. In studies where AAC intervention is compared to speech therapy (or no therapy), high-quality evidence was revealed that AAC is improving communication at medium and long-term interventions (40,47) . This systematic review does not include studies with RCT design, which usually produces the highest level of evidence and lowest level of the risk for bias, compared to other interventional or observational designs. Hence, certain level of bias was included in all studies where the patients, therapists (professionals), and outcome assessors were not blinded. This systematic review included studies with diverse strategies of patient selection, including primary sources such as neurologists, speech therapist, special educator, as well as secondary care specialists. Possible categories of applied assessment tools, many of which are standardized measurement methods, have been integrated and presented in review results. The review tried to focus on all possible outcome measures, both primary and secondary that have been used by the authors of included studies. However, heterogeneity existed among the included studies and part of the included studies were with methodological and practical limitations. It is essential also to select therapists or other professionals based on their AAC skills or to properly train and support them in providing AAC services. Trained therapist might be capable of identifying patients with AAC need (for temporary therapeutic or permanent use). The practical implication of the determinant variables is essential for intervention and assessment planning. Any assessment tool containing more of the determinant variables may ensure the accuracy of the AAC need evaluation or AAC intervention efficacy. In general, the majority of determinant factors identified in this review were rather large categories (language skills, cognition, language comprehension, etc.) and there is a need for new studies of high methodological and practical quality focusing on detailed, clusterized analysis of determinant variables to clarify the evidence.