Introduction
Fragile X syndrome (FXS) is a genetic condition caused by a mutation in the
FMR1 gene resulting in cognitive impairment and behavioural problems [
1]. FXS is the most common inherited form of intellectual disability, affecting approximately one in 4,000 males and one in 8,000 females [
2]. Behavioural characteristics include anxiety, aggression, hyperarousal, attention deficits, hyperactivity, irritability, self-injurious and avoidant behaviour [
3]. Males will typically have intellectual disabilities linked to below average IQ. Language deficits are common, as well as problems with sequential processing, working memory and attention [
4]. Psychiatric problems such as generalised anxiety disorder, social phobia and obsessive compulsive disorder were found to occur in 83 % of individuals with FXS [
5]. Approximately 50 % of males with FXS will also have an autistic spectrum disorder (ASD) [
6]. FXS can exert a substantial burden on caregivers [
7], and many patients are unable to live independently [
8].
Recent research has resulted in a new understanding of the molecular pathways affected by FXS, and a new generation of targeted treatments is currently being tested in clinical trials [
9,
10]. Given the prevalence of social and behavioural problems in FXS, one commonly used measure is the Aberrant Behavior Checklist-Community Edition (ABC-C), a proxy-completed instrument for rating maladaptive and inappropriate behaviours of individuals with intellectual disabilities [
11]. It has been shown to be sensitive in FXS [
12‐
15] and is commonly adopted as a primary outcome measure in clinical trials [
16]. The 58 item ABC-C measures problem behaviour in five domains: hyperactivity, socially unresponsive/lethargic behaviour, stereotypy, inappropriate speech and irritability [
11]. Recently, an adjusted factor structure for individuals with FXS has been reported [
16], which identified a sixth ABC-C domain in FXS, which separates out social avoidance behaviour from socially unresponsive/lethargic behaviour (ABC for FXS).
While caregiver-rated scales can demonstrate the efficacy of an intervention on key characteristics of FXS, only limited data are available regarding the impact of FXS on health-related quality of life (HRQL), particularly for adults. Many decision makers such as the National Institute for Health and Care Excellence (NICE) in the UK prefer to evaluate treatments in terms of impact on survival and HRQL using the quality-adjusted life year (QALY) metric. The estimation of QALYs relies upon HRQL scales that reflect the value (or utility) that people place on health states on a scale from zero (dead) to one (full health). A lack of HRQL data and even suitable HRQL measures in FXS limits ability to estimate QALYs for this condition.
Different methods exist for capturing HRQL data suitable for estimating QALYs, the most common of which, and preferred by reimbursement agencies such as NICE [
17], is use of standardised generic questionnaires where the patient describes their HRQL in a series of questions. Scoring/preference weights are applied to these responses to estimate a utility score. For the purposes of reimbursement review, commonly, it is the societal perspective that is important [
18,
19], so the scoring weights are elicited from the general public in a separate exercise. Examples of such measures include the EQ-5D, SF-6D and Health Utilities Index (HUI) [
20‐
23].
However, generic HRQL measures such as the EQ-5D (covering mobility, self-care, usual activities, pain/discomfort and anxiety/depression) may not accurately capture the impact of certain aspects of FXS, a condition with predominantly behavioural, social and cognitive characteristics. Also, the more subjective aspects of HRQL such as mood, affect, psychological state or pain can be difficult for a proxy to judge, as evidenced by higher rates of missing data on more subjective domains completed by parents of children with an ASD [
24]. Agreement between patient and proxy assessments of HRQL has been found to depend on the concreteness, visibility and importance of aspects of HRQL [
25].
An alternative approach is to develop a utility scoring algorithm, or index, from an existing disease-specific measure or one designed to measure problems associated with certain types of conditions. Disease-specific descriptive systems include items relevant to the patient’s condition that may not be captured by generic measures [
26,
27]. Where a validated disease-/problem-specific measure is commonly used to capture primary outcome data in clinical trials, the ability to estimate utilities from these same data is an added benefit. As the ABC-C in its original form cannot be used to estimate QALYs, the current study was designed to develop and evaluate an ABC-utility index [ABC-UI] to report health state utility scores for children, adolescents and adults with FXS based on patient-level responses to the ABC-C.
Discussion
This study was designed to develop a utility index for the ABC-C, an established outcome measure commonly used in the assessment of FXS. This utility index allows the estimation of values at the individual patient level within the range of 0.92–0.21, reflecting substantial perceived HRQL burden of problems.
The utility index was developed using relatively standard methodology where a subset of items from a validated psychometric outcome measure are used to describe health states for societal valuation. Without reducing the number of items, the valuation task would be impossibly complex and reduction of the complexity of an existing psychometric instrument is an approach common to the development of other utility indexes [
28,
29]. However, the ABC-C is unusual in the extent to which the measure had to be reduced. The original instrument included 58 items [
11], which meant that 80–90 % of original items had to be discarded to achieve a health state classification system of sufficient simplicity for valuation. This presented a significant challenge for retaining the validity and scope of the ABC-C.
To address this challenge, the item reduction process was guided by a number of different sources of evidence, including advanced statistical methods and expert review. Care was taken to identify items capturing a range of severity in aspects of FXS that affect males and females, children and adults, are related to cognitive and emotional characteristics of FXS and have wider impact for individuals’ HRQL. However, it should be noted that the item selection process was for the purpose of developing a utility index for HRQL and cannot be considered to reflect the entire ABC-C measure or aspects of FXS not captured by the ABC-C. As a result, although the ABC-UI reported here draws on items across the ABC-C domains [
11,
16], it is designed to complement the profile of validated domain scores derived from the ABC-C and cannot be considered a proxy for a total ABC-C score, which is not considered an appropriate or valid summary score to calculate [
39,
40].
In the FXS treatment context, there are several reasons why a preference-weighted adaptation of the ABC-C may better estimate utilities than an existing generic preference-weighted measure such as EQ-5D [
21]. For example, the proxy-rated version of EQ-5D could have been used for assessing health status of adults and the proxy-rated EQ-5D-Y [
41] could have been used with children and young people. However, the ABC-C is specific to the problems people with FXS experience, comprising conceptually very different items to those in the EQ-5D and thus has the potential to better assess the impact of disease and treatment. Secondly, the ABC-C was developed and validated as a proxy-rated measure, whereas the EQ-5D was only adapted for proxy use and the evidence to support its validity as a proxy-rated measure is limited. Indeed, there is evidence that the proxy-rated EQ-5D has poor reliability when assessing more subjective elements of health status (such as anxiety/depression and pain/discomfort) [
42]. The ABC-UI offers reduced measurement burden to future clinical trials in FXS with the possibility of using a single outcome measure, as well as the potential to estimate utility values from existing ABC-C FXS datasets.
One notable aspect of the study design is the use of a combined TTO and lead-time TTO (LT-TTO) valuation method. The conventional TTO approach only allows states to be valued as better than dead, requiring the investigator to use a substantially different valuation task for states worse than dead [
43]. LT-TTO offers a simpler method which is compatible with conventional TTO [
44]. In the present study, the combined TTO and LT-TTO approach appears to have worked well. It was well understood by participants, and there were no obvious patterns of systematic bias in the results. Valuations by some participants in the present study that indicated a belief that certain health states were worse than being dead, were supported by interview field notes confirming that these participants understood the implications of their valuation, e.g. participants were concerned about caregiver burden and potential institutionalisation if behaviour was considered sufficiently severe.
A further methodological issue is the fact that FXS affects both children and adults. The valuation of child health states leads to some points for debate. Should adult participants value states knowing that they describe children or should they be asked to assume they are adult states? Should valuation of such states be restricted to parents because they have greater insight into the needs of children? Should the views or values of children themselves be sought? In this study, adult members of the general public were asked to value health states by imagining they are in the health state, i.e. as if they are an adult patient. This approach was taken for a number of reasons. Firstly, the ABC-C is used with children and adults and is not just a paediatric measure. Secondly, asking people to imagine that they are a child with symptoms of FXS raises concerns about introduction of bias in the data. It is not clear that people would be willing to trade years of a child’s life (even hypothetically) in order to improve their HRQL and it becomes unclear what the participant would be valuing in such an exercise. There is relatively little research that has properly addressed these issues, and it is clear that this is needed [
45].
The present study has the following limitations which should be considered. It is unclear how the ABC-UI performs compared with other utility measures. This may well be important for decision-making in a reimbursement setting. One important issue is that the ABC-C is primarily a measure of behavioural problems rather than a measure focused on capturing HRQL. The QALY concept explicitly reflects survival and HRQL, and therefore, it could be argued that the ABC-UI has limitations when used to estimate QALYs. It could equally be argued that because of the nature of FXS, people’s HRQL is in large part determined by their behavioural and functional problems. However, in the valuation exercise, the respondents were provided with little or no information regarding aspects of HRQL not covered in the descriptive system, such as physical functioning. This could potentially be interpreted in different ways by respondents. This is perhaps a general limitation in the use of condition-specific measures to estimate utilities. Condition-specific measures offer the advantage of including potentially specific and sensitive items to assess the burden of a disease, but at the same time, may miss important elements of HRQL that are not affected in that disease. While it should be noted that identifying items with important HRQL impact for people with FXS was a key aim of the clinical expert review and item selection/health state development process for the ABC-UI, this does not change the fact that the ABC-UI does not specifically include aspects of HRQL commonly included in generic instruments such as physical functioning (e.g. mobility), emotional status (e.g. anxiety/depression), self-care and usual activities.
A related concern is the possibility that the focus of the ABC-C on behavioural problems rather than generic HRQL domains may have given rise to health states containing impacts which members of the general public may find hard to understand or value. In the current study, introductory text was developed with clinical expert review and input to provide background information that would help the participants’ understanding of the health states, but without naming the FXS condition. This was piloted alongside example health states, with specific probing in pilot interviews for problematic terms and difficulties general public participants may have imagining the health states described. Another potential limitation of this method is the reliance on proxy assessment for the completion of the ABC-C. However, given the nature of FXS, this remains the only realistic option for data collection.
In addition to behavioural problems, FXS is also characterised by anxiety and attention problems [
3,
5], which the ABC-C is not specifically designed to capture. While this may limit the ability of the ABC-UI to reflect HRQL impact associated with less behaviourally expressed impacts, measures of anxiety and attention have been found to be highly associated with behavioural problems captured by the ABC-C in FXS, which suggest these characteristics co-occur with and may drive problematic behaviour observed [
30]. Specific associations between ABC-C items and measures of non-behavioural FXS characteristics, including anxiety and attention problems, were also considered in developing the ABC-UI. Item selection was informed by the ABC for FXS social avoidance subscale [
16]. However, it is possible that the behavioural focus of ABC-C socially unresponsive/social avoidance items may not characterise the specific social difficulties experienced by FXS patients, who typically seek out but struggle to cope with social situations rather than avoid or are disinterested in social interaction.
It should be noted that this work was conducted in the UK and that additional validation work may need to be conducted in other geographical areas. The UK general public sample was a convenience sample designed to approximate the general population. However, the sample differed slightly from UK general population norms. Regression models run with additional socio-demographic variables resulted in a number of the lower levels of health state dimensions becoming non-significant and showed significant effects of gender, employment and pain/discomfort (on EQ-5D). This suggests that the small differences in the sample from the general population may have had some influence on the results. However, the sizes of the parameter coefficients for the health state dimensions were very similar in the models that also included socio-demographic variables to those reported in Table
4.
Other limitations include the decision during health state development to merge the two most severe ABC-C health states. This was made on the grounds that these were infrequently endorsed in FXS, an observation that was confirmed by expert clinical input. However, it is possible that the resulting ABC-UI may lack sensitivity or demonstrate floor effects when applied to ABC-C scores from more severe FXS cases. The study also used an orthogonal design and the resulting ABC-UI assumes a linear additive functional form. This is not ideal given likely interactions among participants’ preferences across the seven health state dimensions. However, this is a common limitation of most preference-based measures, generic or condition-specific. Where attempts have been made to address non-additivity, these have not allowed for interactions between specific dimensions, but have instead assumed a constant impact that is either additive or multiplicative. To model specific interactions between dimensions requires a far larger sample size.
In conclusion, the ABC-UI appears to be able to report a wide range of utility values from patient-level FXS ABC-C data. This allows estimation of FXS HRQL impact for economic evaluation using an established instrument commonly adopted as a primary outcome measure in FXS clinical trials. However, development of a utility index from a lengthy proxy-rated measure of behaviour raises both conceptual and methodological challenges, along with questions over what the index captures and whether this is an appropriate basis from which to estimate QALYs.