Background
Dementia describes a class of neurologic illnesses that cause progressive decline in cognitive functioning. Areas that are affected most severely are memory, reasoning, communication skills and the ability to carry out daily activities. In addition, people with dementia frequently suffer from behavioural and noncognitive symptoms such as depression, wandering, aggression, agitation, sleep disturbances, shouting, repeated questioning and psychosis [
1].
Dementia is posing a great threat for the future of current health care expenditures as future scenarios claim that dementia prevalence may have doubled or tripled by 2050 [
2,
3]. With an increase in the number of patients with dementia, many governments will have to change their policies and focus on keeping patients out of nursing homes as long as possible. This is only possible with adequate pharmacological and psycho-social interventions. The evaluations of such programs should incorporate outcome measures such as health-related quality of life (HRQoL). Ideally, these HRQoL measures should be valid, reliable, and precise.
However, measurement of HRQoL in dementia is not without difficulties. First and foremost, the concept lacks a generally accepted definition. Several reviews have been published that describe the number and applicability of HRQoL instruments in dementia [
4‐
7]. They show that more than a dozen dementia-specific instruments are available, each covering different domains and applying different methods of measurement.
Many researchers and clinicians argue that HRQoL is subjective in nature and thus only patient ratings should be considered valid. However, the very problem in dementia is that patients’ cognitive functioning decreases and therefore their ratings might become less valid or even unusable. Therefore, reliable and valid informal caregiver (proxy) ratings could be extremely useful in the field of dementia. Yet, using proxy ratings might have disadvantages.
Numerous studies report on agreement and differences between the dementia patients’ HRQoL ratings and those of patient-by-proxy [
8‐
11]. They show a systematic underreporting of patient HRQoL by caregivers, as compared to patient self-assessment. These findings are in line with proxy reporting from other disease areas [
12‐
19]. These studies identified factors that improved agreement between patients and caregivers. Such factors were higher patient education [
20], a family caregiver (vs. other), lower levels of patient functional disability, higher levels of patient depression and lower caregiver burden [
19], the type of perspective employed and higher patient cognitive functioning [
13].
Dementia is a disease that not only affects patients but also their caregivers. Surprisingly, not much research has been done to identify caregiver characteristics that could influence agreement between patients and caregivers. Other than the type of caregiver, the perspective used [
13], and caregiver burden [
21], no caregiver characteristics have been identified. It has been acknowledged that providing care often results in personal, psychological, social and financial losses [
22‐
26], although there are also care giving uplifts such as feelings of satisfaction, personal growth, enhancement and enrichment [
27,
28]. One could imagine that changes in these domains might have an influence on patient-by-proxy HRQoL ratings.
One study showed that caregiver HRQoL is related to the type of dementia and the coping style of the caregiver [
29]. In addition, providing care for a patient with dementia is associated with a decline in HRQoL in both mental and physical domains [
30]. Furthermore, it is unclear whether caregivers ‘project’ part of their own HRQoL problems onto patient HRQoL. For example, caregivers might underestimate the HRQoL of patients if they experience a diminished HRQoL themselves. This mechanism of caregiver ‘projection’ has been investigated in stroke [
31] and other contexts such as end-of-life decisions [
32,
33]. In the latter context it is acknowledged that caregivers are imperfect decision makers and projection of caregivers’ preferences guide their decisions. If such a mechanism is present in patient-by-proxy HRQoL ratings then this is a type of bias of which clinicians and policy makers should be aware. The aim of this explorative study is to assess whether certain caregiver characteristics contribute to a bias of patient-by-proxy HRQoL assessments.
Discussion
The current study attempted to explore the existence of potentially biasing factors that might influence patient-by-proxy HRQoL assessments in dementia. A significant correlation between patient-by-proxy and caregiver self-assessed EQ-VAS scores was found. This correlation remained statistically significant over time. Moreover, characteristics of caregivers were identified that bias their VAS ratings on patients.
A biasing factor on VAS ratings that was identified was ‘life as a whole’. This factor has at face value a great contribution to projection. Caregivers incorporate part of their overall assessments of their own lives into the assessments of the patients’ lives. Should this finding be replicated then such a bias can be overcome by measuring caregiver HRQoL alongside patients. Researchers might then adjust the ratings of caregivers on patients by using a correction algorithm.
Another biasing factor on VAS ratings that was identified was ‘money’. This factor contributes strongly to the overall rating. As this estimate was negative this implies that the better the caregivers’ financial situation are, the worse their ratings on patient HRQoL will be and vice versa. This is a new finding and to the authors’ knowledge has not been previously reported elsewhere. These results seem counterintuitive, as previous research has demonstrated that having more or enough money would improve (HR)QoL [
45]. A possible explanation for this relationship could be that financial status functions as a mediating variable for socio-economic status (SES). Caregivers with a higher SES might perceive the impact dementia has on the patient and themselves to be bigger than lower SES proxies. This might indicate that caregivers with higher SES perceive more shame and experience the difference in HRQoL between themselves and the patient to be larger. More research is needed to further explore this finding.
The least strong characteristic was caregiver age. Age contributed little to the overall rating, but older caregivers gave higher ratings. This finding is in line with general findings that aspects of QOL such as happiness are rated higher as age increases [
46,
47]. Moreover, because spousal caregivers are generally older than child caregivers, it is highly likely that spousal caregivers rate patients higher than child caregivers do. If such is the case, then this might have serious implications for outcomes research. For example, if a new study is to be initiated to evaluate a new intervention which uses spousal caregivers to assess the HRQoL of people with dementia, such a study might overestimate the effects on HRQoL when compared to a study that solely uses child caregivers. Further research is required to corroborate this finding. It is acknowledged that age is one of the most common confounders in observational research. Nevertheless, clinicians and policy makers should be made aware of its potentially biasing effects on HRQoL ratings.
There were two biasing factors on utility values, caregiver age and ‘ability to do things for fun’. The implications for caregiver age are similar to the VAS ratings that were discussed previously. Interestingly, the ability of caregivers to do things for fun biases their assessments of the EQ-5D items on patients. Since the β-coefficient is positive this means that caregivers who are better able to do things for fun give better ratings of patient functioning. One possible explanation for this phenomenon might be that caregivers who undertake many fun activities with a patient also think that the patient experiences a similar level of fun. It might thus be possible that the amount of fun caregivers experience bias their assessment of patient functioning.
These newly identified factors differ from those previously identified in other research areas. For example, patient depression has previously been identified as a factor leading to an increase in patient and proxy differences for elderly patients visiting the emergency room [
38]. In addition, burden and psychological distress in caregivers was a significant predictor of patient and proxy differences in psychosocial scores in veterans [
48]. A different study [
49] that focused more on functional status through (instrumental) activities of daily living identified the following factors that contribute to more disagreement between patients and proxies: female proxies, proxies who lived with the patient, proxies who were not first-order relatives of the patient, and proxies who assisted patients with (instrumental) activities of daily living. The newly identified factors thus provide fruitful grounds for new research on systematic differences between patient and proxy assessments.
It should be noted that the explained variance of both linear mixed models was low. However, with the current study design, this is a desirable outcome. If the models would explain all of the variance then this would imply that patient-by-proxy assessments would only be based on proxy characteristics and not on patient characteristics. In this study the explained variance in the models was less than 10% which suggests that the bias that is present in patient-by-proxy HRQoL assessments is small compared to the influence of actual patient characteristics.
In general, measurement of HRQoL in dementia is difficult. In other disease areas patient self-assessment is usually regarded as a gold standard against which proxy assessment is compared. HRQoL is a very subjective concept and thus patients have ‘privileged access’. In dementia however, researchers have questioned the assumption that people with dementia should be regarded as the gold standard since cognitive impairments might lead to less valid self-assessments. For example, Lawton [
50] noted “most cognitively impaired patients do not introspect, or at least do not report reliably on interior phenomena”. However, the authors feel that patients’ self-assessment is the best measure of HRQoL, as long as the patients can deliver this measure [
51]. If patients cannot give their HRQoL assessment anymore, one has to rely on proxy measures. However, since cognitive functions are primarily affected by dementia, caregivers may be less capable of assessing the internal state of the dementia patients they care for and therefore they might provide less valid HRQoL assessments compared to other disease areas.
Patient-by-proxy assessments can be used for two distinct purposes. The first is substitution of patient self-assessment. In this situation the patient self-assessment is considered a gold standard to compare patient-by-proxy assessment with. However, when patient-by-proxy assessments are used in addition to patient self-assessment, for example to provide extra information for clinical decision making, then patient-self-assessment should not be considered a gold standard. In this context, proxy reporting might even be more valid than patient-self assessment as the disease progresses. Nevertheless, the biases that were identified might occur in both substitution judgments and informing clinical decision making.
One major limitation of the current study is that the proxy perspective investigating how HRQoL is
according to the patient was not measured. We therefore cannot conclude whether or not caregivers actually know how patients would assess themselves. The addition of this perspective could provide additional information on caregiver bias, since it might be different when multiple perspectives are used [
52]. In addition, the relatively short follow up time and the relatively homogenous sample make it difficult to generalize the current findings to a broader context.
Conclusion
Proxy assessment is an important aspect of the evaluation of people with dementia, yet a bias is present. Many caregivers might report on different aspects than patients, and thus the patient self-assessment and patient-by-proxy assessment perspectives might be complementary instead of being regarded interchangeable. However, this makes it more difficult to deal with the proxy bias that is present as shown in this study. Proxy bias might have serious implications for clinical and policy decisions. Dementia is a progressive disease, in which patients at some point become unable to express their HRQoL in a meaningful and valid way. Therefore, after such a point in time, one has to consider alternatives such as proxies or behavioural observations. Nonetheless, if the bias found in this study is generalizable to broader contexts, clinicians and policy makers should be made aware of its influence on proxy HRQoL assessments. For future studies we recommend measurement of HRQoL of patients and proxies, with identical instruments, and multiple perspectives. Should future studies discover more complete causal models of reported HRQoL values, it might then be possible to constrain proxy biases to a minimum.
Ethical approval
The study was approved by the Medical Ethics Committee of the Radboud University Medical Centre.
Acknowledgements
We thank all participants for their contribution. We also thank the research assistants for carrying out all the measurements and the staff of the memory clinics for their participation. In addition we would like to thank A. Simon Pickard and Hilde Verbeek; their review comments helped us improve the quality of the manuscript.
Funding
This work was supported by ZonMw (Netherlands Organization for Health Research and Development; project number 945077039) and by the Radboud University Medical Centre.
Data sharing
The raw dataset is available from the corresponding author at s.arons@ebh.umcn.nl. Participants’ consent was not obtained, but the data presented are anonymised and risk of identification is low.
Competing interests
Marcel Olde Rikkert is consultant for Numico, ECHO Pharmaceuticals, and was consultant for Novartis, Janssen-Cilag, and Schering-Plough. Marcel Olde Rikkert’s fees for presentations and honoraria were only paid to the institution. Marcel Olde Rikkert has no financial interests that conflict with the integrity for this paper. All other co-authors had no competing interests, financial or otherwise.
Authors’ contributions
MOR and PK made substantial contributions to the AD-Euro study design which the current manuscript draws its data from. CSD made substantial contributions to data collection. AA was responsible for the analyses and interpretation and drafting the initial manuscript. MOR, PK, CSD, GJvdW, and AA subsequently contributed to critically revising the manuscript. All authors read and approved the final manuscript.