Background
Neurodevelopmental disorders (NDDs) are a group of conditions that affect a child's cognitive, behavioural, and emotional development which include intellectual disability (ID), autism spectrum disorder (ASD), and attention deficit hyperactivity disorder (ADHD) (World Health Organization, 2019). According to Global Burden of Disease (GBD) 2016 estimates, developmental disabilities, including NDDs, affect 52.9 million children under 5 years of age. Of these, 95% live in low and middle-income countries (LMICs). Nearly 15 million of these children live in sub-Saharan Africa (Olusanya et al., Reference Olusanya, Davis, Wertlieb, Boo, Nair, Halpern, Kuper, Breinbauer, de Vries, Gladstone, Halfon, Kancherla, Mulaudzi, Kakooza-Mwesige, Ogbo, Olusanya, Williams, Wright, Manguerra, Smith, Echko, Ikeda, Liu, Millear, Ballesteros, Nichols, Erskine, Santomauro, Rankin, Smith, Whiteford, Olsen and Kassebaum2018). Ethiopia is one of the top 10 nations globally, with an estimated 1.3 million children living with developmental disabilities (Olusanya et al., Reference Olusanya, Davis, Wertlieb, Boo, Nair, Halpern, Kuper, Breinbauer, de Vries, Gladstone, Halfon, Kancherla, Mulaudzi, Kakooza-Mwesige, Ogbo, Olusanya, Williams, Wright, Manguerra, Smith, Echko, Ikeda, Liu, Millear, Ballesteros, Nichols, Erskine, Santomauro, Rankin, Smith, Whiteford, Olsen and Kassebaum2018). Children affected by NDDs and their caregivers experience severe stigma and social exclusion; this is especially true for those with ID and/or ASD (Tekola et al., Reference Tekola, Baheretibeb, Roth, Tilahun, Fekadu, Hanlon and Hoekstra2016, Reference Tekola, Kinfe, Girma, Hanlon and Hoekstra2020; Tilahun et al., Reference Tilahun, Hanlon, Fekadu, Tekola, Baheretibeb and Hoekstra2016).
Despite the high prevalence and burden of NDDs, there is a wide gap between needs and service provision in sub-Saharan Africa. This applies to both fund allocation and the availability of trained personnel (World Health Organization, 2013; Strand et al., Reference Strand, Chisholm, Fekadu and Johansson2016; Chisholm et al., Reference Chisholm, Docrat, Abdulmalik, Alem, Gureje, Gurung, Hanlon, Jordans, Kangere, Kigozi, Mugisha, Muke, Olayiwola, Shidhaye, Thornicroft and Lund2019). Ethiopia, with a population of nearly 110 million, is served by only one specialist child psychiatrist. There are 0.08 general psychiatrists and 0.04 psychologists per 100000 people, but these cadres of workers have no specialist expertise in NDDs (World Health Organization, 2018).
This gap in resources and services for children with NDDs extends to research (Patel et al., Reference Patel, Kieling, Maulik and Divan2013; Tomlinson et al., Reference Tomlinson, Yasamy, Emerson, Officer, Richler and Saxena2014). Only a negligible fraction of research on child development and mental health is conducted in LMICs (Kieling et al., Reference Kieling, Baker-Henningham, Belfer, Conti, Ertem, Omigbodun, Rohde, Srinath, Ulkuer and Rahman2011; Durkin et al., Reference Durkin, Elsabbagh, Barbaro, Gladstone, Happe, Hoekstra, Lee, Rattazzi, Stapel-Wax, Stone, Tager-Flusberg, Thurm, Tomlinson and Shih2015; Nielsen et al., Reference Nielsen, Haun, Kärtner and Legare2017; Hoekstra et al., Reference Hoekstra, Girma, Tekola and Yenus2018). This research gap results in an incomplete and biased body of knowledge (Durkin et al., Reference Durkin, Elsabbagh, Barbaro, Gladstone, Happe, Hoekstra, Lee, Rattazzi, Stapel-Wax, Stone, Tager-Flusberg, Thurm, Tomlinson and Shih2015; Hoekstra et al., Reference Hoekstra, Girma, Tekola and Yenus2018). When evidence-based tools are lacking, diagnosis and intervention initiation tend to occur later or not occur at all, potentially impairing the prognosis and increasing the risk of comorbidities in affected children (Ruparelia et al., Reference Ruparelia, Abubakar, Badoe, Bakare, Visser, Chugani, Chugani, Donald, Wilmshurst, Shih, Skuse and Newton2016; Guralnick, Reference Guralnick2017).
The development and evaluation of contextually appropriate interventions and harmonised and contextually valid outcome measures are essential to effectively address the service gap (Kieling et al., Reference Kieling, Baker-Henningham, Belfer, Conti, Ertem, Omigbodun, Rohde, Srinath, Ulkuer and Rahman2011; Tomlinson et al., Reference Tomlinson, Yasamy, Emerson, Officer, Richler and Saxena2014). There is currently no consensus on which outcome measures should be used in the evaluation of interventions targeting NDDs (Kohli-Lynch, Tann and Ellis, Reference Kohli-Lynch, Tann and Ellis2019). There is an urgent need for tools that are both accessible and appropriate for use in low-resource settings. Such tools are recommended to be: (i) of high quality, (ii) open-source and open-access, (iii) culturally appropriate, (iv) intuitive, (v) brief, (vi) acceptable, and (vii) easy to administer (Prince, Reference Prince M, Patel, Prince, Minas and Cohen2013; Durkin et al., Reference Durkin, Elsabbagh, Barbaro, Gladstone, Happe, Hoekstra, Lee, Rattazzi, Stapel-Wax, Stone, Tager-Flusberg, Thurm, Tomlinson and Shih2015; Ruparelia et al., Reference Ruparelia, Abubakar, Badoe, Bakare, Visser, Chugani, Chugani, Donald, Wilmshurst, Shih, Skuse and Newton2016; de Vries, Reference de Vries2016; Carruthers et al., Reference Carruthers, Kinnaird, Rudra, Smith, Allison, Auyeung, Chakrabarti, Wakabayashi, Baron-Cohen, Bakolis and Hoekstra2018; Bakolis et al., Reference Bakolis, Thornicroft, Vitoratou, Rüsch, Bonetto, Lasalvia and Evans-Lacko2019; Kohli-Lynch et al., Reference Kohli-Lynch, Tann and Ellis2019). These criteria rule out many of the existing outcome measures, which are often prohibitively expensive or rely on administration by highly qualified specialists (Durkin et al., Reference Durkin, Elsabbagh, Barbaro, Gladstone, Happe, Hoekstra, Lee, Rattazzi, Stapel-Wax, Stone, Tager-Flusberg, Thurm, Tomlinson and Shih2015).
In this study, we validate two scales of the Communication Profile-Adapted (CP-A) as a brief, culturally appropriate, caregiver-reported outcome measure for Ethiopia. The CP-A assesses caregivers' perceptions of their child's abilities and activities for communication, and participation in family and community events (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). It was developed to assess communication in children with complex communication needs and is thus suitable for use in children with NDDs. It has yet to be validated and assessed in its psychometric properties. The CP-A meets important criteria for use in low-resource settings. It has a solid theoretical background, based on the International Classification of Functioning, Disability and Health Framework (World Health Organization, 2001; Hartley and Wirz, Reference Hartley and Wirz2002; Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). It also meets the requirement of cultural relevance for sub-Saharan Africa as it was developed in Uganda and Kenya (Baker and Hartley, Reference Baker and Hartley1998, Reference Baker and Hartley1999; Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). Because it is an open-access tool, translations and adaptations can be readily made. It is also easy to administer and does not use technical terminology (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014).
We selected those scales of the CP-A that most closely reflect the key aspects to be targeted in interventional investigations for children with NDDs and hold promise as quantitative scales. These scales focus on the child's communicative mode and function. Both were adapted to the local context and assessed for validity and reliability. We hypothesised that children with NDDs would score significantly lower on the CP-A than controls and that CP-A scores would be negatively correlated with the severity of clinically diagnosed NDDs.
Methods
Setting
This study was carried out in Addis Ababa, at Yekatit 12 and St Pauls Millennium Medical College government hospitals between August 2018 and May 2019. The ethics protocol was approved by the College of Health Sciences Institutional Review Board at Addis Ababa University (062/16/Psy) and King's College London (HR-16/17–3489). In addition to validation of the CP-A, the data collection comprised further questionnaires, the validation of which are reported in Borissov et al. (Reference Borissov, Bakolis, Tekola, Kinfe, Ceccarelli, Girma Bayouh, Abdurahman, Zerihun, Hanlon and Hoekstra2021).
Participants
Participants were 300 caregivers with long-term responsibilities for children aged 2–9 years, attending either the general paediatric or child mental health clinic of the two hospitals. The paediatric clinics consecutively recruited children with physical health conditions for the clinical control group. The mental health clinics consecutively recruited children with either NDDs alone or with a comorbid mental health condition for the case group, in accordance with the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) (American Psychiatric Association, 2013). Given that no standardised diagnostic tests are available in Ethiopia, health practitioners relied on their clinical judgment based on interviews with caregivers, and observations and interactions with children. The clinical diagnoses in the mental health clinic were provided by general psychiatrists without specialist expertise in child psychiatry, as this specialty training is not available in Ethiopia.
Eligible families were approached by the attending clinician at either clinic and given a flyer with information about the study. Data collection and consent taking was done by clinic nurses who worked independently from the clinicians supporting the families. A subgroup of 40 caregivers was invited for a retest of an average of 19.6 days (s.d. = 3.8) after initial test data collection. All participants provided written informed consent.
Measures
CP-A
The CP-A is composed of 51 questions, divided into 3 main sections named ‘body function and structure’, ‘activities for communication’, and ‘participation’. The portion ‘activities for communication’ includes six scales. Two of those, namely communicative mode (CP-A-mode) and communicative function (CP-A-function) were selected for validation in this study, see Tables 2 and 3 for item content. We refer to the combined items of these two scales as the CP-A-total.
Items' responses refer to a 0–4 Likert scale where [0] stands for ‘never’, [1] ‘rarely’, [2] ‘sometimes’, [3] ‘usually’, and [4] ‘always’. Participants were shown a visual ladder representation of this scale with 0 being the lowest and 4 the highest rung (online Supplementary material Fig. S1).
Demographic information
Clinician-assigned severity levels were available for a subgroup of participants with a diagnosis of ID (N = 24) and ASD (N = 50). General psychiatrists working in the two hospitals rated the child's condition, based on the DSM-5 severity ratings, as [1] ‘mild’, [2] ‘moderate’, or [3] ‘severe’. A structured questionnaire was used to collect caregiver-reported demographic information.
Procedure
CP-A adaptation, translation, and pre-testing
The CP-A was translated to Amharic, one of Ethiopia's official languages, and adapted to the local context. The instrument was considered by a consensus committee comprising native Amharic speakers fluent in English with expertise in the field. Following forward and backward translation, the committee met to formulate a version for pre-testing. This draft was pre-tested with 20 participants through cognitive interviews to understand how respondents interpreted instructions and items (Willis, Reference Willis1999). Feedback was recorded by data collectors and subsequently discussed in the committee to establish a final draft.
Within the CP-A-mode, the content of question 1j was changed to ‘Amharic’ to reflect the main language spoken in Addis Ababa. Item l, Behaviour, was followed by examples of challenging behaviours in brackets (e.g. crying, shouting) as cognitive interviews indicated that the direct translation of the term ‘behaviour’ was unclear to respondents. The original CP-A response scale comprised 8 rungs of ladders; this was shortened to 4 rungs (online Supplementary material Fig. S1) after cognitive interviewing indicated that participants could not meaningfully differentiate between the small increments of the original ladder. Within the CP-A-function, the original version of the scale did not just require a response using the ladder but also asked the extent to which the item is a problem for the respondent. This second part was removed to improve consistency as well as administration ease and time. In its original format, the CP-A asked both questions on how the child communicates with the respondent, and how the respondent communicates with the child. Caregivers participating in the cognitive interviews had difficulty distinguishing between these questions. In response, the section on how the caregiver communicates with the child was removed, focusing solely on the mode of communication of the child.
Data collection and entry
Data collection took place in Yekatit 12 and St Paul's Millennium Medical College general paediatric and child mental health clinics. To allow for the participation of non-literate caregivers, instruments were administered through face-to-face interviews by nurses who had received training in data collection procedures and questionnaire administration. Participating caregivers were reimbursed for their travel costs. Data were double entered using Epidata, version 3.1 (Christiansen and Lauritsen, Reference Christiansen and Lauritsen2003), to reduce the risk of data errors.
Data analyses
Demographics and item checks
The data were analysed using STATA, version 16 (Stata Corp, 2019). Group differences in demographic variables between cases and controls were explored using unpaired t test, Mann–Whitney test, Chi-Square, or Fisher's exact test depending on variables' characteristics. Missing values (N = 3) in the CP-A were replaced by median imputation (Zhang, Reference Zhang2016).
Exploratory Factor Analysis (EFA)
The factor structure of the (i) CP-A-mode, (ii) CP-A-function, and (iii) CP-A-total were examined with the use of item exploratory factor analysis (EFA) (Bartholomew et al., Reference Bartholomew, Steele, Moustaki and Galbraith2008). For each, we evaluated the polychoric matrix of their respective items. Exploratory factor analysis was applied to the matrix of item correlation coefficients, to identify possible underlying dimensions. We used two criteria to aid the choice of the number of factors and provide empirical support for the selection: the scree plot and the criterion of eigenvalues above 1. χ 2 was used as a goodness-of-fit test to evaluate the adequacy of the number of extracted factors (Pett et al., Reference Pett, Lackey and Sullivan2003). To assist data interpretation promax oblique rotation was used (Bartholomew et al., Reference Bartholomew, Steele, Moustaki and Galbraith2008; Baldwin, Reference Baldwin2019). Factors were labelled referring to theoretical notions and interpretability.
Validity
Our approach of studying the validity and reliability was guided by the consensus-based standards for the selection of health measurement instruments (COSMIN) guidelines (Mokkink et al., Reference Mokkink, Terwee, Patrick, Alonso, Stratford, Knol, Bouter and de Vet2010, Reference Mokkink, Prinsen, Patrick, Alonso, Bouter, de Vet and Terwee2019).
Within-scale validity
We assessed within-scale validity (the extent to which the subscales of an instrument measure the same concept; Brohan et al., Reference Brohan, Clement, Rose, Sartorius, Slade and Thornicroft2013)) by examining the correlation between summative scores of the identified factors as well as those of the CP-A-mode and CP-A-function.
Known-group validity
Known-group validity (i.e. the ability to distinguish among distinct groups; Streiner et al., Reference Streiner, Norman and Cairney2015) was tested by assessing group differences between the NDDs and clinical control group across identified factors. Unadjusted and adjusted mean differences between cases and controls were tested using multivariable linear regression, with group membership and previously identified demographic characteristics as independent variables.
Convergent validity
Convergent validity (how related different measures assessing associated constructs are; Streiner et al., Reference Streiner, Norman and Cairney2015), was assessed by estimating the correlation between the total scores of the identified factors and clinician-rated clinical severity measures available for a subset of children with ID or ASD. The clinical severity scores for each disorder were merged into a single score to maximise sample size; in the case of double diagnoses, the highest severity score was retained.
Reliability
Internal consistency (the extent to which items of the same scale measure the same construct) was evaluated through Cronbach's alpha (α) (Revicki, Reference Revicki and Michalos2014). Test-retest reliability (the agreement between scores on the same scale across timepoints) was assessed through weighted Kappa and Interclass Correlation Coefficient (ICC) (Streiner, Norman and Cairney, Reference Streiner, Norman and Cairney2015). The ICC was calculated for continuous summary scores, weighted Kappa for single categorical items.
Results
Demographics and item checks
Our sample included N = 300 caregivers, comprising N = 139 cases and N = 139 clinical controls. Children in the case group presented with ASD, ID, ADHD, language delay, global developmental delay, and/or Down's syndrome, see Table 1 (severity ratings details are provided in online Supplementary material Table S1). Those in the clinical control group were affected by a range of physical health problems (e.g. cardiovascular, respiratory, neurological conditions; details in online Supplementary Table S2). The remaining participants (rest: N = 22) included children who did not meet the criteria to be in either the case or clinical control group (e.g. they were diagnosed with epilepsy or a mental health condition, but did not have an NDD) and those for whom diagnostic information was incomplete (full details in online Supplementary Table S3).
Table 1. Conditions reported in the case group including comorbidities.
Note. ASD, autism spectrum disorder; ADHD, attention deficit hyperactivity disorder; ID, intellectual disability; LD, language disorder; DS, Down's syndrome; GDD, global developmental delay. N, number of observations; %, frequency
Table 2 presents the demographic characteristics of the caregivers and children. The majority of caregivers were females, mainly mothers of the child, and housewives. Most caregivers were married, living in urban areas, and of Orthodox Christian religion. Most did not receive any education above the primary school level. The mean age of caregivers at the time of the interview was of 34 years (s.d. = 7.1).
Table 2. Demographic information
Note: $Rest group refers to 22 participants included in the EFA analyses that did not meet inclusion criteria for either the case or control group. Significant differences between cases and controls are displayed as *p < 0.05, **p < 0.01, ***p < 0.001; N, number of observations; %, frequency; s.d., standard deviation.
Statistically significant differences between cases and controls were observed for the caregivers' gender (p < 0.001), with a higher proportion of females among cases, and residence (p < 0.001), with more cases living in urban areas. Furthermore, we report significant differences in religion, (p = 0.004), observing a higher proportion of Muslims among cases compared to controls, relationship to the child (p < 0.001), where caregivers in the case group were more often mothers, and occupation (p = 0.02), with a higher proportion of housewives among cases.
62% of children were male, with an average age of 4.9 years (s.d. = 2.0). Among children, comparison across cases and controls showed significant differences in age (p < 0.001), with cases being older than controls, and gender (p < 0.001), with a higher proportion of males in the case group in line with the observation that NDDs are more common in boys (Loomes et al., Reference Loomes, Hull and Mandy2017; Sayal et al., Reference Sayal, Prasad, Daley, Ford and Coghill2018). The retest sample consisted of N = 40 caregivers (N = 19 cases, N = 20 controls, N = 1 rest).
Items i ‘Sign language’, j ‘Speaking English’, and m ‘Other’ from the CP-A-mode, had a median and iqr of 0 across test and retest, suggesting no or limited variability. These items were therefore removed. The remaining N = 10 items for CP-A-mode and N = 23 items for CP-A-function were further analysed.
Exploratory factor analysis
The Scree-test and eigenvalues (online Supplementary material Fig. S2) suggested a 2-factor solution for the CP-A-mode (χ 2 = 1501.84, p < 0.001). The correlations between items and promax-rotated common factors are displayed in Table 3. Factor loadings and structure matrix indicated that items a ‘Facial expression’, c ‘Gestures’, d ‘Body movements’, e ‘Looking or use of eye gaze’, f ‘Pointing’, and h ‘Showing you objects’ loaded more strongly on factor 1. Items b ‘Making noises (vocals)’, g ‘Showing you pictures’, j ‘Speaking Amharic’ and l ‘Behaviour’, instead loaded on factor 2. Factor 1 generally corresponds to physical communication, while factor 2 to verbal communication. Item g ‘Showing you pictures’ presented moderate loadings for both factors (slightly higher for the verbal communication) that were maintained throughout different rotation methods. Item l had a negative loading on verbal communication, indicating that caregivers of children with poor verbal communication tended to highlight the use of behaviour as the main mode of their child's communication.
Table 3. Factor loadings of the items of the CP-A-mode and CP-A-function
Note: Factor loadings extracted through principal axis factoring. Shaded cells indicate factor allocation.
For communicative function, the Scree-test and eigenvalues (online Supplementary material Fig. S3) suggested a 1-factor solution. The model had a good fit to the data (χ 2 = 8784.87; p < 0.001), suggesting that the function items assessed a single construct. All items loaded positively on the factor, see Table 3.
Similarly, for the CP-A-total the Scree-test and eigenvalues (online Supplementary material Fig. S4) indicated a 1-factor solution. The model had a good fit to the data (χ 2 = 2.1e + 04; p < 0.001), suggesting that all items of the analysed CP-A scales can be meaningfully subsumed under a single construct reflecting overall communication activities. All items, besides Item l, loaded positively on the factor (online Supplementary material Table S4).
Validity
Within-scale validity
A summary of the psychometric properties of the CP-A is provided in Table 4. The Spearman's rank correlation coefficient (ρ) between the CP-A-mode's verbal communication and physical communication scores was significant and moderately positive (ρ = 0.41, p < 0.001). Similarly, the correlation between the CP-A-mode and CP-A-function scores was significant and moderately positive (ρ = 0.59, p < 0.001), further justifying the adoption of the CP-A-total as a unidimensional summary scale (Lamping et al., Reference Lamping, Schroter, Marquis, Marrel, Duprat-Lomon and Sagnier2002; Dancey and Reidy, Reference Dancey and Reidy2017).
Table 4. Psychometric properties of the CP-A's identified factors
Note: Significant differences between cases and controls are displayed as *p < 0.05, **p < 0.01, ***p < 0.001; N, number of participants.
a Unadjusted mean difference in summative scores (controls minus cases).
b 6 items.
c 4 items.
d 23 items.
e 33 items.
Known-group validity
The multiple regression analyses indicate that clinical controls scored significantly higher than cases on CP-A-mode's physical communication (Δmean = 2.89, p < 0.001) and verbal communication (Δmean = 2.09, p < 0.001). This holds true also for the CP-A-function (Δmean = 28.94, p < 0.001) and CP-A-total (Δmean 33.93, p < 0.001), Table 5. The significant group differences persisted when adjusting for covariates (caregiver's age, religion, occupation, residence, and relationship to the child; children's age and gender). Cohen's d estimates for CP-A-mode's factors were medium when unadjusted and small when accounting for covariates. Estimates were large for CP-A-function and CP-A-total across both scenarios (Cohen, Reference Cohen1988).
Table 5. Adjusted and unadjusted mean differences in summative scores between cases and controls
Note: Significant differences between cases and controls are displayed as *p < 0.05, **p < 0.01, ***p < 0.001; Δ mean, mean difference (controls minus cases); N, number of items. Mean differences unadjusted or adjusted for covariates.
Convergent validity
There was a significant correlation between clinical severity and CP-A-mode's physical communication (ρ = −0.24, p = 0.04) as well as verbal communication (ρ = −0.27, p = 0.02). Similar results were obtained for the CP-A-total (ρ = −0.25, p = 0–04). For the CP-A-function, the point estimate of the correlation was in the same direction and of similar magnitude (ρ = −0.23) but was not significant (p = 0.59) (Dancey and Reidy, Reference Dancey and Reidy2017).
Reliability
The internal consistency for the CP-A-mode factors was acceptable (physical communication: 4 items, α = 0.80, 95% confidence interval (CI) ⩾ 0.78; verbal communication: 6 items, α = 0.72, 95% CI ⩾ 0.69). Excellent internal consistencies were obtained for the CP-A-function (23 items, α = 0.96, 95% CI ⩾ 0.96) and the CP-A-total (33 items, α = 0.97, 95% CI ⩾ 0.96). Assessing internal consistency separately for cases and controls suggested that levels remained acceptable for the CP-A-mode's physical communication (cases: α = 0.78, 95% CI ⩾ 0.74; controls: α = 0.78, 95% CI ⩾ 0.74), but were low for verbal communication (cases: α = 0.63, 95% CI ⩾ 0.57; controls: α = 0.64, 95% CI ⩾ 0.57). The internal consistency remained excellent for CP-A-function (cases: α = 0.94, 95% CI ⩾ 0.93; controls: α = 0.96, 95% CI ⩾ 0.95) and CP-A-total (cases: α = 0.94, 95% CI ⩾ 0.95; controls: α = 0.94, 95% CI ⩾ 0.93) (George and Mallery, Reference George and Mallery2010; Revicki, Reference Revicki and Michalos2014).
Weighted Kappa coefficients ranged between 0.60 and 0.86 for physical communication and between 0.65 and 0.76 for verbal communication (p < 0.001), demonstrating moderate to a substantial agreement among items. For CP-A-function and CP-A-total, the agreement was moderate to near perfect for all items (min = 0.60, max = 0.95; p < 0.001) (McHugh, Reference McHugh2012; Portney, Reference Portney2020). Please refer to online Supplementary material Tables S5, S6 for details.
The ICC indicated good test-retest reliability for the CP-A-mode's physical communication (ICC = 0.81, 95% CI 0.67–0.89; p < 0.001) and verbal communication (ICC = 0.83; 95% CI 0.70–0.91; p < 0.001). Excellent test-retest reliability was observed for the CP-A-function (ICC = 0.96; 95% CI 0.94–0.98; p < 0.001) and CP-A-total (ICC = 0.97; 95% CI 0.95–0.98; p < 0.001) (Koo and Li, Reference Koo and Li2016; Portney, Reference Portney2020).
Discussion
This paper reports the first validation study of the CP-A, a caregiver-reported tool for the assessment of children's communication. Our aim was to address the need for a psychometrically sound, brief, and culturally appropriate outcome measure for use in Ethiopia. We investigated two sections of the CP-A: communicative mode and function. Within communicative mode we identified two factors, verbal and physical communication; the communicative function scale was unidimensional. EFA results indicated that all items, from both scales, can be meaningfully summarised into one single factor. This suggested the adoption of a summary score (CP-A-total), supported by findings of strong correlations between the identified factors. As hypothesised, children with NDDs (cases) scored lower than clinical controls. Moreover, scale scores were negatively correlated to clinical severity ratings of NDDS, indicating that children with more severe NDDs used fewer modes of communication and applied fewer functions of communication. We observed acceptable to excellent internal consistency as well as test-retest reliability. Overall, these results demonstrated the validity and reliability of the CP-A-mode, function, and communication-total scales.
Within the CP-A-mode, for items i ‘Sign language’, j ‘Speaking English’, and m ‘Other’, most responses were equal to 0 (i.e. ‘never’), suggesting these items have little relevance in assessing communicative mode within our sample. For item i this is likely due to the fact that none of our participants was reported to have hearing loss. Furthermore, formal sign language has received very limited implementation in Ethiopia so far (Wakuma, Reference Wakuma2015), and caregivers could have selected item c ‘Gestures’ to indicate informal signs as modes of communication. The lack of variability in item j can be attributed to the fact that English is not widely spoken across the population (Central Intelligence Agency, 2020). Most caregivers answered ‘never’ to item m ‘Other’, suggesting that all previously administered questions had exhaustively described the communicative modes adopted by their children. For these reasons, items i, j, and m of the communicative mode were dropped from subsequent analyses. The CP-A-function scale was retained in its entirety.
For communicative mode, we found support for a 2-factor structure. All factor 1 items fit with the profile for physical communication. For factor 2, the construct of verbal communication is defined by the items with the highest loadings, b ‘Making noises (vocals)’ and j ‘Speaking Amharic’. Item g ‘Showing you pictures’ showed more moderate loading, with substantial cross-loading on factor 1. Item l ‘Behaviour’ had a strong negative correlation with factor 2, suggesting that caregivers of children who did not express themselves verbally were more likely to report their child's behaviour as a form of communication. This finding is in line with theoretical notions that see verbal acts and behaviours as equivalent in function (Carr and Durand, Reference Carr and Durand1985). When verbal communication is severely impaired, behavioural expression may become challenging to the person and others (Royal College of Psychiatrists and Banks, 2007). The occurrence of challenging behaviours (e.g. self-injury, stereotypy) is reported across NDDs and cultural contexts (McClintock et al., Reference McClintock, Hall and Oliver2003; Adeniyi and Omigbodun, Reference Adeniyi and Omigbodun2016; O'Nions et al., Reference O'Nions, Happé, Evers, Boonen and Noens2018). The Amharic version of the item, contrary to the English original, was followed by examples of challenging behaviours in brackets ‘(e.g. crying and shouting)’. These examples may have contributed to caregivers' interpretation of the item as primarily concerning challenging behaviours rather than behaviour overall.
Clinical controls scored higher than cases on the CP-A-mode, function, and communication-total scales, even after adjusting for covariates. This supports both our initial hypothesis and that of the developers of the measure: these scales were designed to reflect higher perceived competence in communication through higher summative scores (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). This is in line with studies investigating other tools assessing communication in higher-income countries (HICs), with lower ratings consistently indicating more profound impairments (Geurts et al., Reference Geurts, Verté, Oosterlaan, Roeyers, Hartman, Mulder, van Berckelaer-Onnes and Sergeant2004; Norbury et al., Reference Norbury, Nash, Baird and Bishop2004).
Significance
This investigation represents the first exploration of the validity and psychometric properties of the CP-A. Compared to other measures, it is more suitable for application in low-resource settings. Unlike caregiver-reported tools developed in Western HICs that assess similar constructs, the CP-A is free and open access, and this avoids the significant costs and adaptation negotiations associated with copyright-restricted instruments (Durkin et al., Reference Durkin, Elsabbagh, Barbaro, Gladstone, Happe, Hoekstra, Lee, Rattazzi, Stapel-Wax, Stone, Tager-Flusberg, Thurm, Tomlinson and Shih2015). Moreover, instruments developed in Western HICs often require extensive adaptations to be suitable in non-Western lower-income contexts (Marlow et al., Reference Marlow, Servili and Tomlinson2019; de Leeuw et al., Reference de Leeuw, Happe and Hoekstra2020) As the CP-A is one of the very rare measures developed in an LMIC (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014; Goldfeld and Yousafzai, Reference Goldfeld and Yousafzai2018), it does not encounter these issues. Its design and content are more likely to be relevant and appropriate for the African context. Nevertheless, adaptations of limited entities (e.g. referring to the languages spoken locally) are required to fit the specific context of an application. These must be paired with further explorations of the psychometric properties in diverse settings across the continent,
The only previously published research using the CP-A as an outcome measure investigated the impact of a caregiver-driven intervention for children with complex communication needs in rural Kenya (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). While the sample size of this study was small (N = 10) and did not include a control group, results suggested sensitivity to change, as scores in the activities for communication sections, including communicative mode and function, were significantly higher post-intervention compared to pre-intervention (Bunning et al., Reference Bunning, Gona, Newton and Hartley2014). The integration of their findings with that of our investigation demonstrates the potential of the CP-A for use in interventional studies. Adopting an accessible and appropriate measure like the CP-A across investigations on NDDs would increase the comparability of results, aiding the evaluation and implementation of effective interventions in low-resource settings.
Limitations
Limitations should be considered when evaluating our results. Significant differences were reported for some demographic variables, especially in terms of age, where our clinical control group was younger than the case group. Nevertheless, it could be argued that the developmental age of the two groups is more comparable in this situation, given that the control group (with younger children that are naturally in the earlier phases of developing their communication abilities) still scores higher than the case group. The clinical severity assessment was carried out by general psychiatrists, with no specialist expertise in the assessment of child NDDs. Nevertheless, diagnosis and assessment of children with NDDs part of postgraduate training in psychiatry in Ethiopia and the psychiatrists involved in the study were experienced in making these diagnostic assessments. Psychiatrists used their clinical judgement rather than standardised tools to assess the severity of impairment of the children since there are no validated standardised clinical severity assessment scales available in Ethiopia to support the assessment of severity. Furthermore, the limited educational or supportive service provision available for children with NDDs in this setting means that reports from other professionals are not available to inform severity assessments. Thus severity is based on a single report from the caregiver during the clinical encounter and observations of the psychiatrist of the child in a clinical setting. Severity scores were collected for a small sample only, covering a limited range of complex communication needs. This study was conducted as part of a larger project focusing on NDDs. Most cases included in our study had ASD or ID, rather than a wider group of complex communication needs and developmental disabilities for which the CP-A was also developed (e.g. sensory impairments). Lastly, participants were help-seeking families recruited in Addis Ababa. Our sample had an overrepresentation of urban families and may not be fully representative of the Ethiopian population.
Future research
Further research could test our factor structure for the CP-A-mode and CP-A-function through confirmatory factor analysis. Recruitment of cases should be extended to a wider range of complex communication needs. In such studies, the relevance of items reflecting characteristics of developmental disabilities with no representation in our sample (i.e. i ‘Sign language’), should be re-evaluated. Future research should further examine whether the CP-A-mode, CP-A-function, and CP-A-total are sensitive to change induced by interventions. Lastly, future studies may also wish to consider other sections of the CP-A not included in the current evaluation.
Conclusion
This work is the first investigation to explore the validity of CP-A, an open-access measure developed in and for the African context. The communication mode and function and their combined scales met the validity and reliability criteria as a measure for the assessment of caregiver-perceived activities for communication. We tested this among children with NDDs and concomitant complex communication needs in Ethiopia. We recommend the further validation of this scale. The CP-A has potential for application in intervention studies on NDDs across sub-Saharan Africa as a brief, quantitative, and culturally appropriate outcome measure.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/gmh.2021.44
Acknowledgements
The authors are grateful to all caregivers for their participation.
Financial support
This work was supported by joint funding from the Medical Research Council (MRC) (United Kingdom), Department for International Development (DFID), Wellcome Trust and National Institute for Health Research (NIHR) (#MR/P020844/1). BT, CH and RAH receive support from the National Institute of Health Research (NIHR200842) and CH through the NIHR Global Health Research Unit on Health System Strengthening in Sub-Saharan Africa, King's College London (GHRU 16/136/54) using UK aid from the UK Government. The views expressed in this publication are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. CH receives support from AMARI as part of the DELTAS Africa Initiative [DEL-15-01]. IB is supported by the NIHR Biomedical Research Centre (BRC) at South London and Maudsley NHS Foundation Trust and by the NIHR Applied Research Collaboration South London (NIHR ARC South London) at King's College Hospital NHS Foundation Trust.”
Conflict of interest
Conflicts of Interest: None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
