Introduction

Intellectual disability occurs in 1.9 per 100 children and approximately 15% of these children have severe impairment [1]. Worldwide the prevalence is higher in low- and middle-income group countries [2]. Children with intellectual disability have greater exposure to the social determinants of poor health such as economic disadvantage [3, 4]. The effects of intellectual disability are pervasive for both physical and mental health.

An important outcome to evaluate the effectiveness of services for children is quality of life (QOL), which refers to satisfaction with a composite of life experiences and includes domains that are universal (e.g. physical and mental wellbeing) with additional domains for particular populations [5]. As part of an evidence-based platform for patient-centred clinical care, service delivery and formulation of policy, it is critical that outcomes are evaluated in terms of individual experiences rather than professional assessment of what is important to children with intellectual disability [6]. We also need outcomes that incorporate the important features of impairment, activity, participation and the environment in which the child lives, as presented in the International Classification of Functioning, Disability and Health (ICF) [7]. Multifaceted interventions for complex conditions are unlikely to impact one outcome, and composite outcomes such as QOL could be suitable and efficient to measure [8]. However, our capacity to measure the impact of interventions in children with intellectual disability is limited because available QOL measures were not developed with the specific issues of children with intellectual disability in mind and did not involve families in their development [9]. For example, two qualitative studies on children affected by cerebral palsy [10] and autism spectrum disorder (ASD) [11] with comorbid intellectual disability found important themes missing from generic scales.

We recently undertook four qualitative studies to investigate the domains of QOL important to children with intellectual disability. In-depth interviews were conducted with parents of 6–18-year-old children with either Down syndrome (n = 17), Rett syndrome (n = 21; a severe genetic neurodevelopmental disorder mainly affecting females [12]), cerebral palsy (n = 18) or ASD (n = 21) [13,14,15,16]. Together these conditions represent a range of characteristics seen in the broader population of those with intellectual disability including functional, behaviour and socialisation difficulties; medical comorbidities; and different needs for autonomy. QOL domains were consistent across the four groups and included physical health and emotional wellbeing; pleasure in communication, movement and day-to-day routines; and satisfaction derived from social connectedness, leisure activities and the natural environment. Domains such as emotional wellbeing mapped broadly to other QOL measures but included many elements that were unique to our subject group [13,14,15,16]. These family-reported data were consistent conceptually with the ICF [7] and could usefully inform the content of a specific QOL measure for children with intellectual disability.

QOL is a concept evaluated through self-reflection, but this is challenging if cognitive abilities preclude self-report. For children, parents often act as proxies particularly if their child does not have verbal skills [17]. Based on our extensive qualitative dataset, this paper describes the development and validation of the Quality of Life Inventory-Disability (QI-Disability).

Methods

Data sources

Participants were parents of children registered with one of five databases. Families with a child with Down syndrome born from 1980 to 2004 who had previously participated in our research [18] were invited to take part, and additional families were recruited through Developmental Disability WA (a community organisation in the disability sector) and advertising on Facebook. Families with a child with Rett syndrome were recruited from the Australian Rett Syndrome Database, an ongoing population-based register established in 1993 that collects longitudinal data [19]. Families with a child with cerebral palsy and intellectual disability were recruited from the Victorian Cerebral Palsy Register, a population-based register of individuals with cerebral palsy born in Victoria since 1970 [20]. Families with a child with ASD and intellectual disability were recruited from the WA Autism Biological Registry [21] or the WA Autism Register [4].

Development

A working group (JD, AE, NM, HL) extracted statements from 77 interview transcripts [13,14,15,16] to illustrate observable aspects of each QOL domain. The statements were discussed by the group and edited to form questionnaire items. Where possible, items were worded positively to measure wellbeing rather than the converse and to reduce threats to the self-esteem of the parents who were completing [22]. However, negative behaviours were framed to explicitly acknowledge these behaviours as endorsed during consumer consultation. Each item was accompanied by a five-point Likert scale indicating the frequency of each aspect of the child’s wellbeing. The items were reviewed iteratively by the authors and the initial questionnaire draft comprised of 50 items.

We tested the meaning of the questionnaire items with a sample of parents using cognitive interviewing [23], to provide feedback on the comprehensibility and relevance of the items. Sixteen parents registered with one of the five databases were recruited and their children represented different ages, genders, clinical severity and comorbidities. During a recorded telephone interview, each parent was asked to complete the draft QI-Disability, describe their understanding of each item and share other thoughts (e.g. why they chose the rating category). Parents were recruited until thematic saturation was achieved as observed by repetition of responses. Parent wording and rationale for each of the items were tabulated. The wording of 24 items were clarified to better reflect the intended meaning of the item, three items were merged with other items and six items were removed due to a lack of utility (e.g. did not capture the intended meaning). The questionnaire then comprised of 41 items.

Validation

Between November 2016 and April 2017, QI-Disability was administered to 253 parents/primary caregivers using the REDCap (Research Electronic Data Capture) tool, with a paper format or telephone interview also available. With 41 candidate items, this sample was larger than the generally recommended sample size of five participants per item [24]. Of those families contacted, 98.4% (61/62) families with a child with Down syndrome, 95.6% (66/69) families with a child with Rett syndrome, 77.2% (64/83) families with a child with cerebral palsy and 91.2% (62/68) families with a child with ASD responded. Most (89.7%) respondents were biological mothers of whom 53.6% (n = 135) worked full- or part-time and 17.4% (n = 44) lived in a rural community. The mean age of the children was 12.2 (SD 4.1 years, range 5–18 years). Data describing the distribution of child and family characteristics are presented in Table 1.

Table 1 Frequency distribution (%) for children in the validation study
Full size table

Analyses

Exploratory factor analysis using all available data was performed to identify the factor structure using the iterated principal factor method incorporating promax rotation and pairwise deletion of missing data. A cutoff of 1.0 for the eigenvalue was used to define the domains to be retained and items with a loading < 0.4 on any factor were excluded. Confirmatory factor analysis was then performed to verify the factor structure. To provide a basis for acceptance or rejection of the model, goodness of fit was assessed using the following statistics: CMIN/df value, root mean square error approximation, the Comparative Fit Index and the Tucker–Lewis Index. Cronbach’s alpha, the Composite Reliability and Average Variance Extracted statistics were calculated for each factor to assess convergent validity. The maximum correlation squared value was calculated for each factor to assess divergent validity. Confirmatory factor analysis and goodness-of-fit analyses were also performed with the sample restricted to those who could walk independently or not, and those who could talk and be understood by those who did not know the child well or not.

Differential item functioning (DIF) comparing various sub-groups was performed using the STATA DIF detect command [25]. This employs ordinal regression models with item score as the dependent variable and the relevant domain score and group membership as independent variables. We used the recommended criterion of a > 10% change in the domain score coefficient when group is added to the model to identify uniform DIF [25]. Non-uniform DIF is identified when the group membership × domain score interaction coefficient differs significantly from zero at the P < 0.05 level. In addition to groupings based on ability to walk and to talk, we evaluated DIF by age group at the time of the questionnaire (younger than 12 years vs. 12 years and older).

After reverse coding of relevant items, item scores were transformed to a range of 0–100. Specifically, never was scored as 0, rarely as 25, sometimes as 50, often as 75 and very often as 100. Domain scores were calculated by the sum of item scores divided by the number of items. The total score was calculated by the sum of domain scores divided by the number of domains. Linear regression models were then used to examine the associations between total and domain scores and diagnostic (Down syndrome, Rett syndrome, cerebral palsy, ASD) and age (5–11, 12–18 years) groups. Analysis was restricted to questionnaires with a response to all items for either total or domain scores.

Results

Exploratory and confirmatory factor analyses

Exploratory factor analysis of the 41 items resulted in the extraction of six domains. The factor loadings were < 0.4 for seven items and these items were excluded (Table 2). Subsequent confirmatory factor analysis affirmed the same factor structure, but the factor loadings for two items were < 0.4 and these items were then excluded (Table 3). The remaining items loaded strongly on domains describing “social interaction” (n = 7), “negative emotions” (n = 7), “leisure and the outdoors” (n = 5), “independence” (n = 5), “physical health” (n = 4) and “positive emotions” (n = 4).

Table 2 Factor loadings for individual scale items onto each of the six domains from the exploratory factor analysis (n = 253)
Full size table
Table 3 Factor loading (95% confidence interval) values from confirmatory factor analyses for all children and sub-groups based on capacity to walk or talk
Full size table

Reliability, convergent and divergent validity

The inter-factor correlations were of moderate size with coefficients ranging in magnitude between 0.20 and 0.68 (Supplementary Table 1). The six-factor model showed satisfactory indices of relative fit using the CMIN/df and Root Mean Squared Error of Approximation values, although the Comparative Fit and Tucker–Lewis indices were slightly smaller than the recommended cut-point of 0.9 (Supplementary Table 2). Factor loadings (Table 3), correlation coefficient values and goodness-of-fit statistics (Supplementary Table 2) were similar for each of the mobility and communication sub-groups indicating consistency of responses to the questionnaire across different levels of functioning.

Cronbach’s alpha values ranged from 0.72 for “physical health” to 0.90 for “positive emotions” and composite reliability values ranged from 0.75 for “physical health” to 0.91 for “positive emotions”, each > 0.7 and indicative of satisfactory convergent validity (Supplementary Table 3). The average variance extracted values for the “physical health” and “negative emotions” domains were < 0.5 giving conflicting evidence for the convergent validity of these domains. For each domain, the average variance extracted values were larger than the maximum correlation squared value, providing evidence for satisfactory divergent validity (Supplementary Table 3).

Differential item functioning (DIF)

Uniform DIF was displayed for only one item—“Enjoyed making things with their hands” with higher scores (mean = 3.2) among those who were able to walk compared with those unable to walk (mean = 2.6). There were five instances of non-uniform DIF among the three sets of group comparisons on each of the 32 items (Supplementary Table 4). Taking into account multiple testing, this number of significant results is no greater than would be expected by chance.

Comparison of known groups

Descriptive statistics describing QOL total and domain scores are shown in Supplementary Table 5. The mean (SD) total score for all children was 67.9 (14.3) out of a maximum total score of 100 and mean domain scores ranged from 60.4 (24.0) for “independence” to 74.1 (18.6) for “positive emotions”. Low and high scores were obtained for total and domain scores across the diagnostic groups. Comparisons of total and factor scores between diagnostic, functional and age groups are shown in Table 4. Compared to Rett syndrome, children with Down syndrome had higher total (coefficient 10.55, 95% confidence interval [CI] 5.70, 15.39), “social interaction” (coefficient 7.13, 95% CI 0.61, 13.65), “physical health” (coefficient 9.10, 95% CI 2.42, 15.78), “leisure and the outdoors” (coefficient 10.6, 95% CI 3.36, 17.83) and “independence” (coefficient 29.70, 95% CI 22.88, 36.52) scores. Compared to Rett syndrome, children with ASD had lower scores for “social interaction” (coefficient − 12.81, 95% CI − 19.28, − 6.35) but higher scores for “independence” (coefficient 23.31, 95% CI 16.55, 30.07). Children who could walk independently or talk had slightly higher “physical health” and “independence” scores than if unable to walk or talk. Children able to walk independently had higher “leisure and the outdoors” scores, and children able to speak had higher “social interaction” scores. Scores for the “positive emotions” (coefficient − 6.14, 95% CI − 10.71, − 1.56) and “leisure and the outdoors” (coefficient − 5.36, 95% CI − 10.59, − 0.13) domains were lower for adolescents compared with children (Table 4).

Table 4 Linear regression of the relationships between total and factor scores and predictor variables
Full size table

The final item set is shown in Supplementary Table 6.

Discussion

Our recently identified QOL domains and domain elements as observed in children with intellectual disability formed the foundation for the development of QI-Disability, some that were not well represented in available generic QOL measures. These data indicated the need for a measure developed specifically for children with intellectual disability where options are currently extremely limited. Derived from qualitative data, the items in QI-Disability described caregiver observations of behaviours rather than their impression of what was important for the child’s QOL, and were constructed to describe QOL rather than functioning to ensure measurement was broader than health-related QOL [26]. Prior to pilot testing, families then informed the final selection of items for QI-Disability and evaluated their wording for clarity and appropriateness. These processes contrast with the development of KidsLife, also a proxy-report measure of QOL for children with intellectual disability, where items were based on QOL domains for adults with intellectual disability and the judging of experts used to determine their relevance to children [27]. Caregivers of individuals with intellectual disability completed the Pediatric Quality of Life Inventory [28] but scores are difficult to interpret because items do not represent all relevant QOL domains [13,14,15,16]. Best practice methodologies [29] in the current study explain the intrinsic validity of QI-Disability.

Factor analyses streamlined the item set and consolidated the qualitative themes into six domains. In broad terms, the domains have conceptual validity because they represent aspects of physical and mental wellbeing, social and recreational functioning illustrated in other child QOL measures [9] and are consistent with the ICF structure [7]. More specifically, items describing QOL in relation to physical health, and positive and negative emotions were extracted from qualitative data representing those domains. Otherwise, items from different qualitative themes were grouped to form the domains “social interaction”, “leisure and the outdoors” and “independence”. However, these groupings also made conceptual sense. For example, items describing communication experiences in social settings loaded together to represent the child’s social interactions. Items describing the pleasures of movement and balance loaded with items describing a range of leisure activities and spending time in the natural environment, providing a comprehensive picture of aspects of participation. The factor “independence” comprised of items necessary for day-to-day communications, routines and everyday tasks in daily living.

The diagnoses of the children together represented the range of health and functioning issues that are observed in children with intellectual disability and a wide range of scores were calculated across each of the domains and within each diagnostic group. In our sample, children with Rett syndrome or severe cerebral palsy were more likely to experience comorbidities such as epilepsy and scoliosis [30, 31], whereas most children with Down syndrome or ASD could walk independently and feed themselves [32, 33]. In this diverse group, some goodness-of-fit analyses were slightly lower than recommended but taken together, the model appears satisfactory. Statistics indicating convergent and divergent validity were also satisfactory, except the average variance extracted values were slightly lower than the recommended cut-point for two of the six domains. There was only one instance of uniform differential item functioning when evaluating each of the 32 item responses by three different sub-groupings of our sample. When replicating factor analyses and validity testing across different levels of communicative and mobility functioning, the validation held. These findings suggest that QI-Disability will be useful across diverse groups of children with intellectual disability and within different groups who experience different impairments and severity.

Variation in QI-Disability factor scores for the diagnostic and age groups was consistent with known between-group heterogeneity and conceptually in alignment with the difficulties experienced. For example, children with Down syndrome had significantly higher total scores than those with Rett syndrome. With regard to specific domains, individuals with Down syndrome where disability is milder scored higher for the “independence” factor [32] than individuals with Rett syndrome who are dependent for most activities of daily living [34]. Alternatively and consistent with other literature [35], scores for “social interaction” were higher for children with Down syndrome who often have a more sociable nature in contrast to children with ASD who experience social difficulties.

Adolescents scored lower for “positive emotions” compared to the younger children, possibly reflecting changes experienced by adolescents in the general population [36] or emotional disorders as reported in adolescents with intellectual disability in a national survey in the United Kingdom [37]. Interestingly, “leisure and the outdoors” scores were also lower for adolescents, perhaps consistent with lower “positive emotions” scores or with encountering barriers to participation such as issues of access, limited opportunities or attitudes of others of discrimination or exclusion. These data suggest that important differences are identifiable, and some point to opportunities for interventions to increase QOL in adolescents with intellectual disability.

Pilot testing of QI-Disability involved a sample derived from population-based databases representing a range of child and family characteristics. Rett syndrome is caused by a pathogenic mutation in the MECP2 gene located on Xq28 and almost exclusively affects females [38] and so our sample was entirely female. The gender distribution in the other diagnostic groups was broadly consistent with the literature with some differences. For example, autism is more prevalent in males [39] as reflected in our sample. Epidemiological studies of children with cerebral palsy and comorbid intellectual disability [40] and those with Down syndrome [41] indicate a slightly higher prevalence in males, whereas our samples include slightly more females. Our sample enabled factor analysis and related validation but it will be important to ensure representativeness in future studies when investigating the determinants of QOL. Within each diagnostic group there was a range of strengths and difficulties as seen in clinical care. There were high recruitment fractions with little missing data. We acknowledge that QI-Disability collects proxy-reported data and that there may be differences between parent and child reports [17]. Whilst self-report is preferable where feasible, there is still substantial reliance on parent/proxy reports in the paediatric literature and practice, and particularly in the field of intellectual disability. For intellectual disability, it is necessary to develop a proxy-report measure of QOL that would enable population-based investigations and include the substantial proportion of children unable to self-report. Importantly, the development of QI-Disability is a vital step in the preparation for the development of a child-report measure based on child-reported domains of QOL and appropriate for children with communication difficulties who can self-report.

Conclusion

More validation studies will be an area for future research [42], but the developmental processes, theoretical underpinnings and psychometric testing provide evidence that QI-Disability can be used as an outcome measure to support evaluation in children with intellectual disability. With complex needs, multifaceted outcome measures such as QI-Disability are necessary to assess practice and enable new lines of inquiry on the determinants of QOL and novel interventions.