Introduction

The ICECAP-A (ICEpop CAPability measure for Adults) is a measure of capability well-being, intended for use in economic evaluation [1]. The measure was developed in the UK, but has also been used in other English speaking countries and across Europe [2].

The measure is distinct from those typically used in health economics, both because of its conceptual alignment to the capability approach of Amartya Sen [3, 4] and the breadth of its evaluative space–well-being in a broad sense, rather than health-related quality of life. The capability approach relies on a distinction between functionings and capabilities (achieved or observed outcomes versus the ‘real’ opportunity that a person has to be or do the things they have reason to value) [3]. ICECAP-A seeks to elicit information on capability by phrasing questions in terms of “I am able to…” and “I can have” [1]. The measure covers five areas of life (or five key capabilities): Stability (feeling settled and secure); Attachment (love, friendship and support); Autonomy (being independent); Achievement (achievement and progress); and Enjoyment (enjoyment and pleasure). There is one question relating to each of the five capabilities.

There are four response options available for each of the five questions [1] and responses are coded (for the purpose of data entry) such that level 1 is the worst response option (for example, “I am unable…” or “I cannot have…”) and level 4 is the best response option (for example, “I can have all/a lot of…”). A value set (or scoring system) exists for ICECAP-A [5], such that a respondent who selects level 1 for all five questions (giving a profile of: 1,1,1,1,1) will have a score of 0 and a respondent who selects the top level for each question (giving a profile of 4,4,4,4,4) will have a score of 1. A score of zero is interpreted as “no capability” and a score of one is interpreted as “full capability” [5].

ICECAP-A is one of four ICECAP measures, which also include the ICECAP-O for older people [6], the ICECAP-SCM (Supportive Care Measure) [7] and the ICECAP-CPM (Close Person Measure), a measure for assessing the impact of end of life care on those close to the patient [8].

Al-Janabi et al [9] note that, in the context of human development (where the capability approach was first developed), functionings have frequently been used as a proxy for capability (often with reliance on interrogation of existing datasets), whereas in health research conducted in high income countries, there is commonly an emphasis on primary data collection, relying on self-reporting. Hence, although ICECAP-A has generally been found to have good construct validity [10], Al-Janabi et al conducted qualitative research to explore whether respondents comprehended the concept of capability and responded appropriately [9]. It was found that a majority of informants participating in think-aloud interviews were able to comprehend questions phrased in terms of capability and felt able to judge their own capability well-being. Indeed, in some cases, informants confirmed that their capability was greater than their level of functioning [9], for example where family circumstances or a relationship constrained their ability to be completely independent. Al-Janabi et al. relate this constraint to Nussbaum’s notion of ‘combined capability’ [9, 11], the level of capability reflecting external conditions.

Whilst the results reported by Al-Janabi et al. are promising, there will be cases where a patient or recipient of social care (sometimes referred to as long-term care) will be too ill and/or too vulnerable to self-report their own capability well-being, and in these cases it may be necessary to elicit proxy responses. For example, Makai et al. relied on staff from nursing homes within Germany [12] and The Netherlands [13] to report the well-being of residents using ICECAP-O. Bailey et al. elicited proxy responses using ICECAP-A from close persons and care staff for patients receiving hospice care [14].

Well-being, in general, is ideally assessed by self-report and there are documented challenges associated with proxy completion [15]. A systematic review by Rand and Caiels, covering the period 2004 to 2014 [15], found that whilst a majority of studies reported proxy respondents rating quality of life lower than self-reported quality of life, the direction of difference was not consistent across all identified research studies, and was associated with: methodology (measurement properties of the instrument); the balance of objective/subjective attributes within the measure and the closeness/relationship between the proxy and the subject (pp1-2).

There is also evidence that the cognitive process (or perspective) adopted by the proxy may influence the difference between proxy and self-ratings [15]; Pickard and Knight [16] set out two perspectives from which the proxy can assess the health of a patient. The first involves the proxy assessing the patient in terms of how they think the patient would respond, and is labelled the ‘proxy-patient’ perspective. The second perspective is ‘proxy-proxy’, where the proxy responds based upon their own perception of the patient’s health.

Most empirical studies investigating proxy completion report the level of agreement between responses from self-completion and proxy completion, often on measures of health-related quality of life (as per the studies identified by Rand and Caiels). Makai et al. instead report correlation between different proxy groups; they found little agreement between proxy responses provided by care staff and family members in the Dutch study, and suggest that these two proxy groups may have different reference points, with care staff referencing one resident’s capability against that of other residents and family referencing the resident’s capability against the resident’s own previous level of capability [13].

The study by Bailey et al. involved patients recruited from a UK hospice (along with close persons and care staff) completing ICECAP-A, ICECAP-SCM and EQ-5D within think-aloud interviews [14]. The paper reports error rates relating to comprehension, retrieval, judgement and response. Qualitative analysis revealed that healthcare workers found it easier to complete EQ-5D (as a measure of health functioning) on behalf of patients than they did to proxy using the ICECAP measures (with their focus on broader well-being); there were also fewer errors when healthcare workers completed EQ-5D, in comparison to the ICECAP measures [14].

In summary, proxy respondents are commonly observed to report the subject’s well-being as worse than the subject’s own self-assessment; different proxy respondents may adopt a different reference point or find it easier to report on some aspects of a subject’s well-being than on others, depending upon their relationship with the subject; and the perspective adopted by a proxy respondent (proxy-proxy or proxy-patient) may also influence their proxy responses.

This study is the first to elicit qualitative and quantitative data from multiple proxy respondents for the same four, hypothetical individuals. The aim of the study was not to explore differences between proxy and self-reported responses, but instead to explore variation across multiple proxy responses for the same ‘subject’ individual and attempt to explain any significant variation in proxy responses. Objectives were to: (i) assess the extent of divergence in proxy responses for the same ‘subject’ individual; (ii) investigate whether there was any evidence of proxy responses varying systematically across proxy raters of different socio-demographic groups; and to understand the cognitive process, reported difficulties, struggles and considerations of proxy raters.

Methods

A series of eight one-day deliberative workshops (involving a total of 62 members of the public) were held across the West Midlands region of the UK, between August and November 2016. The primary objective of the workshops was to elicit a sufficient level of capability well-being (as defined by ICECAP-A), for use in decisions about the allocation of scarce public health and social care resources [17]. The intuition behind eliciting a sufficient level of well-being is that society will not attach additional value to improvements in well-being resulting from social policies beyond what is deemed to be a ‘good enough level’. Practically, this means prioritising improvements in well-being for the worst off in society above improvements for those already in a “comfortable” or “adequate” state of well-being [17].

Deliberative processes involve presenting options and information to participants and encouraging open discussion, in order to elicit informed and considered views. To prompt participants to consider a range of social care and public health services and how these services may affect the lives of those dependent upon them, participants were presented with hypothetical vignettes. Vignettes depicted the needs and circumstances of hypothetical individuals and the services/sources of support which may be available to them. To develop familiarity with the ICECAP-A, workshop participants were asked both to self-complete ICECAP-A and to provide proxy responses on behalf of the ‘subject’ individuals described in the vignettes. Consideration of the vignettes and the completion of ICECAP-A on behalf of the individuals they depict provided the data that is analysed and interpreted in this paper.

Results

Sample

3,685 invitation letters were sent to voters within the selected electoral wards and 62 members of the public participated, across the eight workshops. Attendance at each workshop varied between four and 10. Table 1 reports sample characteristics. Older people and females were over-represented in the sample.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographics (Citizens’ workshops).

https://doi.org/10.1371/journal.pone.0236584.t001

61 participants (24 male and 37 female) completed ICECAP-A based on vignettes A, B and D. 53 participants completed ICECAP-A on behalf of person C (visual impairment)–this number excludes participants from the pilot workshop who saw a different vignette ‘C’.

The mean (self-reported) ICECAP-A score for participants (n = 62) was 0.8802 (SD: 0.1444, minimum: 0.2695, maximum: 1.0).

Proxy ICECAP-A scores

Tables 2 and 3 show that mean proxy scores differ across the four vignettes, with the highest ICECAP-A score being associated with person B (planned pregnancy) and the lowest score being associated with person A (young adult feeling depressed). No statistically significant difference was found between the mean scores of male and female participants, either for own scores or for proxy scores across the four vignettes (Table 2). With respect to age, there were no statistically significant differences between means as determined by one-way ANOVA (Table 3). No correlation was observed between own ICECAP-A scores and proxy scores (for any of the four vignettes) (see Supporting Information).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Proxy ICECAP-A scores by sex.

https://doi.org/10.1371/journal.pone.0236584.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Proxy ICECAP-A scores by age.

https://doi.org/10.1371/journal.pone.0236584.t003

Discussion

Much of the literature on proxy completion has focused on the assessment of health related functioning and on the correlation between patient and proxy responses. The ICECAP measures assess well-being in a broader sense and rely on the potentially complex concept of capability (a distinction between functioning and the ability to function). There is some evidence of differences between proxy assessments of well-being by care workers and proxy assessments of well-being by family members for care home residents, when assessed using ICECAP-O [13] and some evidence that health professionals find proxy completion of EQ-5D easier than proxy completion of ICECAP-A in the context of care at the end of life [14].

This study has adopted the novel approach of eliciting multiple proxy responses for a defined set of hypothetical vignettes. The study found no evidence that participants’ own demographic or well-being characteristics systematically influenced proxy responses, but did find that there were large ranges of values for the same individual/vignette. Variation across proxy respondents for the same vignette appears to have been driven by four factors: the perspective adopted (factual versus empathetic); the proxy’s ability to empathise with the subject; the introduction of external information and considerations; the proxy’s interpretation of the measure attributes.

Participants differed in their attitudes to mental health problems, with some reporting that they were unable to empathise or sympathise in particular with the individual described in vignette A (depression). The range of scores for scenario D indicate that proxy respondents struggled when the individual described in the vignette had a cognitive impairment (in this case, dementia). Considering that ICECAP-A scores are anchored between zero and one, proxy scores for vignette D ranged massively (from zero to 0.9308).

In the case of cognitive impairment, there may be significant differences in scores depending upon whether the proxy respondent adopts a factual or an empathetic approach. The two approaches identified here (factual and empathetic) closely relate to and can be retrospectively matched to Pickard and Knight’s two perspectives: proxy-proxy and proxy-patient [16]. We haven’t attempted to explore potential differences in ICECAP-A scores from proxy respondents adopting the two different approaches (factual versus empathetic) because it cannot be guaranteed that participants adopted the same approach consistently across either all four vignettes or across all five questions from the ICECAP-A. It would also not be possible to identify the approach that was adopted by all of the participants–some articulated their approach more explicitly than others.

A limitation of the study is that there was no professional or personal relationship between proxy respondents and subjects, although this fact did not result in participants being noticeably cautious with their proxy responses–participants did not, for example, avoid extreme response levels. The lack of sympathy or empathy for the individuals described in the vignettes (observed in a small number of cases) would perhaps be less likely to be observed in cases where there was a personal relationship between the proxy and the subject. It may be that participants in this study were more prone to introduce additional information into their assessment (as evidenced through the qualitative analysis) because of the lack of a relationship with the subject and perhaps, linked to this, the fact that the vignettes were deliberately fairly brief. The introduction of considerations or reference points which were external to the subject of interest by participants in our study is a phenomenon which Makai et al suspected having occurred in their study; they suggested that care staff in the nursing home context could have been drawing upon experiences of other residents as a point of reference when providing proxy responses [13]. The naming of health conditions or types of disability (for example, using the term ‘dementia’) may have encouraged the introduction by participants of external knowledge and points of reference, but it is reasonable to expect that when providing proxy responses for an actual close person, the proxy rater would be aware of their health constraints and diagnoses. Previous quantitative research has found that labelling health states influences health state valuation [19, 20]; clearly the context and methodologies differ, but the qualitative evidence from this study of proxy raters introducing external considerations relating to a particular illness/diagnosis may be relevant to those undertaking future research involving health state valuation and choosing whether to label health states.

Given the broad scope of the concepts covered by ICECAP-A it is likely that, even where there is a close personal or professional relationship with the subject, proxy responses will be determined to some extent by the proxy’s own interpretation of the attributes. In particular, there was discussion and some disagreement amongst participants in this study over the interpretation of autonomy (independence) and achievement. Participants differed, for example, in the extent to which they considered finances (including home ownership) as an element of independence, and in the extent to which they considered family as a constraint to personal independence. There was some evidence that participants had different expectations of what constituted achievement, given the life-stage of the individual described in the vignette.

Because vignettes were developed depicting hypothetical individuals, it was not possible to compare proxy responses with responses from self-completion, although doing so was not the aim of this particular study.

A limitation of the study was the sample size. Although no other study, to our knowledge, has elicited or been able to compare as many as 61 different proxy responses for the same individual, the sample size was only sufficient to enable somewhat tentative quantitative analysis.

Conclusion

This study adds to the existing body of evidence suggesting that proxy completion is problematic. Given the scope for different proxy respondents to differ in relation to both perspective (empathetic, linking to Pickard and Knight’s ‘proxy-patient’ perspective versus factual, linking to the proxy-proxy perspective) and interpretation of concepts covered by the questions, at the very least the same questions should be asked of the same proxy respondent at each point of data collection. Relying on a single proxy respondent will ensure that variation in responses/scores across different time points is driven by a proxy respondent’s perception of variation in the subject’s well-being, and not by differences in how proxy respondents approach the task of providing proxy responses.

It may be worth explicitly instructing proxy respondents to adopt either a proxy-patient (empathetic) or a proxy-proxy (factual) perspective, as per Pickard and Knight’s recommendation, although it is not clear to what extent proxy respondents will take note of or adhere to this instruction (either at all or consistently across different questions). It may also be worth selecting proxy respondents who are homogenous in terms of the nature of their relationship with the subject, for example, all proxy respondents will be health/care professionals or all will be family/friends. Homogeneity in proxy respondents may promote homogeneity in the type of external information that is potentially introduced by proxies as their point of reference.

Supporting information

Acknowledgments

We are grateful to the study participants for generously giving their time and for sharing their views and experiences. We are grateful to the project advisory group for their support and guidance.