Utility-Based Instruments for People with Dementia: A Systematic Review and Meta-Regression Analysis

Transcription

1 VALUE IN HEALTH 21 (2018) Available online at journal homepage: SYSTEMATIC REVIEWS Utility-Based Instruments for People with Dementia: A Systematic Review and Meta-Regression Analysis Li Li, MHEcon(Adv), BPharm 1, *, Kim-Huong Nguyen, PhD 1,2, Tracy Comans, PhD 1,2,3, Paul Scuffham, PhD 1 1 Centre for Applied Health Economics, Menzies Health Institute Queensland, School of Medicine, Griffith University, Nathan, Queensland, Australia; 2 The NHMRC Cognitive Decline Partnership Centre, Sydney, New South Wales, Australia; 3 Metro North Hospital and Health Service District, Brisbane, Queensland, Australia ABSTRACT Background: Several utility-based instruments have been applied in cost-utility analysis to assess health state values for people with dementia. Nevertheless, concerns and uncertainty regarding their performance for people with dementia have been raised. Objectives: To assess the performance of available utility-based instruments for people with dementia by comparing their psychometric properties and to explore factors that cause variations in the reported health state values generated from those instruments by conducting metaregression analyses. Methods: A literature search was conducted and psychometric properties were synthesized to demonstrate the overall performance of each instrument. When available, health state values and variables such as the type of instrument and cognitive impairment levels were extracted from each article. A meta-regression analysis was undertaken and available covariates were included in the models. Results: A total of 64 studies providing preference-based values were identified and included. The EuroQol five-dimension questionnaire demonstrated the best combination of feasibility, reliability, and validity. Meta-regression analyses suggested that significant differences exist between instruments, type of respondents, and mode of administration and the variations in estimated utility values had influences on incremental quality-adjusted life-year calculation. Conclusions: This review finds that the EuroQol fivedimension questionnaire is the most valid utility-based instrument for people with dementia, but should be replaced by others under certain circumstances. Although no utility estimates were reported in the article, the meta-regression analyses that examined variations in utility estimates produced by different instruments impact on costutility analysis, potentially altering the decision-making process in some circumstances. Keywords: Alzheimer s disease, preference-based measures, systematic review, utility assessment. Copyright & 2018, International Society for Pharmacoeconomics and Outcomes Research (ISPOR). Published by Elsevier Inc. Introduction Dementia has become one of the leading chronic disease contributors to disability and dependency among the elderly worldwide. The total number of people living with dementia was estimated globally to be 47.5 million in 2015 and has been predicted to reach million in 2030 and million in 2050 [1]. The estimated worldwide cost of dementia in 2010 was approximately $604 billion, and has been conservatively predicted to increase by 85% by 2030 [1]. As the global population ages and the prevalence of dementia increases, the proportion of constrained health care resources required for dementia care is also likely to increase. Health care reimbursement authorities around the world have recommended the use of cost-utility analysis (CUA) to determine the costeffectiveness of new health care interventions. Converting the effectiveness data to cost per quality-adjusted life-year (QALY) gain enables the CUA to simultaneously integrate the changes in both quantity (length) of life (mortality) and quality of life (morbidity) [2]. The QALY weights can be obtained directly using elicitation methods or indirectly using questionnaire-type multiattribute utility instruments. A number of multi-attribute utility instruments such as the EuroQol five-dimension questionnaire (EQ-5D) [3 5] and recently developed dementia-specific utility-based instruments such as the DEMQOL-U [6,7] have been applied in CUA to assess preference weights for people with dementia. Nevertheless, significant variations were observed in the literature and concerns have been raised regarding the performance of those instruments [8,9]. For instance, generic instruments such as the EQ-5D are widely used because they facilitate comparisons across disparate diseases and interventions in standard economic evaluations. Nevertheless, the classification systems of generic instruments may lack sensitivity to important differences in health status for * Address correspondence to: Li Li, Centre for Applied Health Economics, Menzies Health Institute Queensland, School of Medicine, Nathan Campus Griffith University, Sir Samuel N78 Room 2.32, Brisbane, Queensland 4111, Australia. li.li4@griffithuni.edu.au $36.00 see front matter Copyright & 2018, International Society for Pharmacoeconomics and Outcomes Research (ISPOR). Published by Elsevier Inc.

2 472 VALUE IN HEALTH 21 (2018) specific conditions and diseases, such as dementia. Consequently, dementia-specific instruments may be more attractive to both researchers and funding authorities. Furthermore, assessing utility values is a complex and challenging issue for people with dementia, who suffer from constraints in judgment, thinking processes, and communicative abilities, resulting in high rates of nonresponders. In this case, it is likely that neither generic nor disease-specific utility-based instruments are able to capture every aspect of quality of life for people with dementia. People with dementia also consistently have higher self-reported utility scores than those reported by a proxy [10 12]. Considering the loss of insight that has been found to be common in dementia, the higher ratings of patients may reflect a lack of insight into their impairments. Hence, proxy ratings may be more reliable and accurate for people with dementia and, thus, may be more reliable in CUA for policy decisions. On the contrary, studies have shown that caregivers may emphasize the negative aspects and overlook all positive aspects of a patient s quality of life, resulting in relatively low utility ratings, whereas the meaningful information that can be transmitted by people with dementia is overlooked, rather than interpreted by others [13]. No consensus has been reached in the literature as to whether the patient-reported or proxy-reported scores are more appropriate to be usedinqalycalculation. Furthermore, research has shown that differences exist within utility values obtained using various preference-based instruments from the same samples. For example, reported mean utility scores were the highest for the EQ-5D, followed by the Quality of Well-Being (QWB) and the Health Utilities Index (HUI), in studies in which these three instruments were administered to the same patients [12,14]. Studies suggested that both preference measurement technique to weight instrument items (time trade-off vs. standard gamble) and the utility combination rule (additive, multiplicative, and regression) may internally lead to differences across different utility-based instruments [15,16]. Nevertheless, wide variations in reported utility values were also found for the same instrument across different samples. For instance, the reported HUI scores for people with moderate dementia varied between 0.53 and 0.87 [14,17]. Such differences may indicate that not only the instrument itself but also the study-level covariates such as patients characteristics and methodology could have impacts on reported utility values. The need for greater understanding of utility weights for people with dementia has been highlighted in the most recent literature [18,19]. It is particularly important to ensure that the values of the measured health states obtained from utility-based instruments are reliable and robust and enable accurate and consistent calculation of QALYs for economic evaluations. To our knowledge, one study has reviewed preference weights for people with dementia obtained from the EQ-5D between 1990 and 2009 [9]. Another study reviewed all relevant utility-based instruments, but only for people with Alzheimer disease (AD) and on the basis of data collected during the period from 2000 to 2011 [8]. Consequently, this study aimed to fill the literature gap by 1) providing a comprehensive comparison of performance (feasibility, precision, reliability, validity, and responsiveness) of all utility-based instruments for people with dementia through a systematic review of published evidence and 2) investigating the factors contributing to the variations in utility values obtained from different utilitybased instruments through meta-regression analyses. Methods Literature Search and Inclusion Criteria This systematic review was conducted in line with the Preferred Reporting Items for Systematic Review and Meta-Analyses guidelines [20]. A database search was performed in April The search included the electronic databases of EuroQol, MED- LINE, CINAHL, Cochrane Library, Embase, PsycINFO, Econlit, and PubMed, using the search terms listed in Appendix Table A1 in Supplemental Materials found at Additional articles were found through citations and an updated database search in January The study selection process was performed by two independent reviewers who initially screened the titles and abstracts for eligibility. The fulltext versions of all included studies were then obtained and reviewed for eligibility using the same inclusion and exclusion criteria. Differences between the results of the two reviewers were resolved by discussion with a third reviewer. Articles were included in this review if they 1) were published in English; 2) involved any type and stage of dementia, including AD, vascular dementia, or dementia with Lewy bodies; and 3) were studies in which health state values or preference weights were presented as outcome measures. Working papers, protocols, editorials or letters, systematic reviews, abstracts, and studies that lacked reported health state values were excluded. Systematic Review and Analysis of Utility-Based Instrument Performance From the studies that satisfied the inclusion criteria, information or data relevant to the instruments psychometric properties were extracted to investigate their performance for people with dementia. An instrument s measurement performance can be assessed by its feasibility, precision, reliability, validity, and responsiveness. Feasibility represents the willingness or ability of participants to complete an instrument [21]. Although difficult to evaluate directly, this is most readily assessed through time taken to complete the instrument, response rates, and number of missing items [9]. Precision is the ability of an instrument to distinguish changes in relation to the reported health states [21]. When more than 20% of respondents have maximum good or bad health scores, this indicates ceiling or floor effects, respectively [21]. Reliability is defined as the ability to provide stable instrument scores when repeated on an unchanged population, which can be assessed by measuring the test-retest correlation over two points in time [22]. In addition, reliability in instruments refers to the comparability of responses across different assessors (such as patient and proxy report), which can be measured by the interrater agreement [23]. Instruments are considered to be strongly reliable and recommended if correlation coefficient estimates are greater than 0.70 [21,24]. Validity is the ability of the dimensions of an instrument to adequately cover the core determinants of health-related utility. Applied to health state preferences, content validity refers to the adequacy of the health state description (classification system) in representing health status [25]. Construct validity is assessed by either convergence validity (correlation between targeted instrument and other quality-of-life measures) or known-group validity (correlation of utility values obtained from groups that are expected to differ in cognitive impairment levels) [21,23]. The strength of correlation is considered to be weak if estimates are less than 0.3, moderate if less than 0.5, and strong if more than 0.50 [23]. Responsiveness is defined as the ability of an instrument to measure changes in health-related utility over time [21]. It is difficult to assess responsiveness because no criterion standard is available for comparison. Nevertheless, effect size can be viewed as the most common indicator for assessing responsiveness for instruments. Instruments are considered to be the most responsive if the effect size score is more than 0.80, moderately responsive if the score is between 0.5 and 0.8, and mildly responsive if the score is only larger than 0.2 [21].

3 VALUE IN HEALTH 21 (2018) Meta-Regression Analysis of the Utility Values Examining how the properties of different instruments influence the utility values derived for people with dementia (the study s second aim) will help us understand the potential impacts of using a particular instrument in calculating the most important outcome measure in the CUA the QALY. To this end, we conducted a meta-regression analysis. This approach has the advantage of combining all published data on utility values and potentially yields more accurate utility estimates for mild, moderate, and severe levels of dementia, as well as providing insight into the factors that influence health state values. In our meta-regression analyses, the dependent variable was mean utility values and the explanatory variables included utility-based instruments, cognitive impairment levels, mean age, and other relevant variables, which were all extracted from studies included in the systematic review. The respondents variable was included as one of the controlled variables because patient self-reporting scores have been consistently reported in the literature to be higher than proxy-reporting scores [26]. The administrations variable was chosen because the preferencebased weights might be more precise if collected via face-to-face interview, compared with a survey or phone interview method [27]. Mean age, sample size, disease types, and severity levels were taken into consideration in the regression because they were believed to correlate with different health state values [26,28]. Last, the year of publication was considered in this study to explore the possibility of improved methodology (both surveys and statistical analysis) in recent years [29]. The base model (model 1), estimated by the least squares method with robust standard error, is specified as follows: Mean utilities¼ αþβ 1 Instrumentsþβ 2 Respondents þ β 3 Administrations þ β 4 Mean age þ β 5 Sample size þβ 6 Severity levels þ β 7 Disease types þ β 8 Year of publication þ ε Findings in the literature indicated that using utility weights derived from populations other than that being studied (for instance, a European utility weight on an American population) could bias the utilities measured and adversely affect the results of economic evaluations [30]. The results from a recent investigation by Fang et al. [31], however, indicated that no important differences were found in obtaining utility values by using other countries population utility weights. Hence, value sets for each instrument and study regions for individual studies were added in model 2 to investigate whether differences between the population of interest and the population from which the utility weight was derived will contribute to the variations in measured health state values. In many studies, patients cognitive impairment levels were not measured as a continuous variable but by categories (e.g., mild, moderate, and severe). To explore the impact on utility values of using different measures for patients cognitive impairment levels, the continuous variable of severity levels in model 1 was replaced by categorical severity levels variable in model 3. Similarly, the severity levels variable was replaced by the categorical severity levels variable to form model 4 in our analyses. The number of studies per outcome of interest required for the regression analysis to be plausible in the multivariate setting is about 40 and an absolute minimum of 10 depending on the level of heterogeneity across studies [32]. According to this guideline, the meta-regression analyses include only variables with observations from at least 10 different studies. All statistical analyses were performed using STATA 14.1 (StataCorp LP, College Station, TX). Results The database search identified 64 articles including 3 identified from the reference lists and citations of retrieved articles (Fig. 1). Six utility-based instruments (Table 1) were identified from studies that have been used to assess health state values for people with dementia. Descriptive characteristics of the 64 reviewed articles are presented in Appendix Table A2 in Supplemental Materials found at Evaluation of the Utility-Based Instruments Performance Feasibility Completion time, response rate, and missing items of each instrument are presented as the average values across studies in Table 2. Overall, the EQ-5D demonstrated the best feasibility among six utility-based instruments, with the lowest completion time and missing items but higher completion rate for both patient self-reporting and proxy-reporting [14,33]. The completion rates were 100% for the caregivers using the HUI2 as well as patients with mild AD using the HUI3 [13,16]. Missing items were only 2.8% for the caregiver version of the DEMQOL-U [6]. Precision The systematic review indicated that a ceiling effect on the EQ-5D was reported across various studies [6,12,14,31,34]. No ceiling effect was found for the HUI, but the authors suggested a potential floor effect [14]. A ceiling effect was found for the DEMQOL-U and both ceiling and floor effects were discovered for the DEMQOL-Proxy-U. Other instruments such as the QWB and the Assessment of Quality of Life (AQoL) reported no ceiling or floor effects, and so were considered to have better precision compared with instruments with significant ceiling or floor effects. Reliability The EQ-5D [14,33,35 37], HUI3 [13], and QWB [14] were reported to have moderate to strong test-retest reliability compared with the 15D, AQoL, and DEMQOL-U. This finding indicates that stable utility values can be achieved using those three instruments on an unchanged population, particularly for patients with mild cognitive impairment. Nevertheless, the inter-rater agreement, which represents the association between patient- and proxyreported values, was found to have a moderate level for all utility-based instruments. Validity First, the content validity was assessed for three instruments and it was concluded that there was a mismatch of reported content between patients and caregivers reports of what was critical to their quality of life and the content of the three instruments. The QWB included the largest number of items that were identified by patients and caregivers as important in assessing their quality of life compared with the HUI and the EQ-5D [38]. Studies showed moderate to strong convergence validity for the EQ-5D [39 43], the AQoL [44], and the QWB [12,14]. The findings for the HUI were not consistent across studies. For example, no convergence validity for the HUI was reported in three studies, whereas a moderate convergence validity was reported in two other studies [14,35,37,45]. Most of the articles reported significant moderate to strong known-group validity of the EQ-5D [6,46 49]. Again, conflicting evidence was found across studies for the known-group validity of the HUI. For instance, no known-group validity was

4 474 VALUE IN HEALTH 21 (2018) Fig. 1 PRISMA flow diagram summarizing the study selection process. AQoL, Assessment of Quality of Life; CEA, costeffectiveness analysis; CUA, cost-utility analysis; EQ-5D, EuroQol five-dimension questionnaire (three-level); HUI, Health Utilities Index; PRISMA, Preferred Reporting Items for Systematic Review and Meta-Analyses; QWB, Quality of Well-Being. reported in the study by Neumann et al. [17] but a strong (40.7) validity was reported for the HUI in other studies [45,50]. Responsiveness Only the EQ-5D has been reported as having an effect size higher than 0.5, indicating at least medium responsiveness of the EQ-5D compared with other instruments. Meta-Regression Analyses Results The studies and observations selected for the meta-regression analyses are outlined in Figure 2. In total, 46 studies were included in the analysis. The most commonly applied utilitybased instrument was the EQ-5D, followed by the HUI2/3. Most of the EQ-5D values were based on a UK value set (n ¼ 35), whereas the Danish value set was least presented (n ¼ 4). Definitions, summary statistics, and expected impact on the mean utility values of those extracted variables are presented in Table 3. Results of the meta-regression analyses are presented in Table 4. Overall, the instruments, respondents, and severity levels variables were found to increase mean utility values across four different models (positive coefficients). The categorical severity levels variable expectedly declined utility values (negative coefficients). The coefficients of the administrations variable were positive and significant in models 3 and 4 but not in models 1 and 2, whereas the coefficients of mean age were positive and significant in models 1 and 2 but not in models 3 and 4. Disease types, sample size, and year of publication (both continuous and categorical) had no impact on mean utility value estimation across models. The direction of impacts confirmed our expectations specified in Table 3, except for the coefficient for mean age. As can be seen from Table 4, our meta-regression analyses results indicate that the self-reported health state values for patients with dementia are generally around 0.14 points higher than values reported by their caregivers. Patients with severe dementia predictably rate their utility values approximately 0.12 points lower compared with people with moderate dementia and double the magnitude compared with people with mild dementia. Most importantly, utility-based instruments were found to be a statistically significant predictor of mean health state values. The estimated utility values obtained from the EQ-5D are higher than those obtained from the HUI. More specifically, it was 0.13 points higher than in model 1 and 0.17 points than in model 3. To illustrate the impact that variability in utility estimates would have on the CUA, we created hypothetical scenarios that

5 VALUE IN HEALTH 21 (2018) Table 1 Overview of utility-based instruments identified in the study. Instrument Country of origin Description of domains Population for preference weights Preferences elicitation method No. of health states Scoring algorithm Scale range No. of studies 15D Finland Mobility, vision, hearing, breathing, sleeping, eating, speech, elimination, usual activities, mental function, discomfort and symptoms, depression, distress, vitality, and sexual activity AQoL Australia Illness, independent living, social relationships, physical senses, and psychological well-being DEMQOL-U/ DEMQOL- Proxy-U United Kingdom QWB United States Positive emotion, memory, relationships, negative emotion and loneliness/positive emotion, memory, appearance, and negative emotion Mobility, physical activity, and social activity HUI2 Canada Sensation, mobility, emotion, cognition, self-care, pain, and fertility HUI3 Canada Vision, hearing, speech, ambulation, dexterity, emotion, cognition, and pain EQ-5D United Mobility, self-care, usual activities, Kingdom pain/discomfort, and anxiety/ depression Public VAS 31 billion Additive 0.11 to Public TTO 16.8 million Multiplicative 0.04 to Public TTO 1,024/256 * Regression/ additive 0.00 to Public VAS 1,170 Additive 0.00 to Parents VAS/SG 24,000 Multiplicative 0.03 to Public VAS/SG 972,000 Multiplicative 0.36 to Public TTO/VAS 234/3,125 Regression/ additive 0.59 to AQoL, Assessment of Quality of Life; EQ-5D, EuroQol five-dimension questionnaire (three-level); HUI, Health Utilities Index; QWB, Quality of Well-Being; SG, standard gamble; TTO, time tradeoff; VAS, visual analogue scale. Number of health states generated from the DEMQOL-U and the DEMQOL-U-Proxy, respectively. The three-level EQ-5D included a three-level scale in the health state descriptive system of having no problems, having some or moderate problems, being unable to do/having extreme problems. The five-level EQ-5D included a five-level scale of having no problems, having slight problems, having moderate problems, having severe problems, and being unable to do/ having extreme problems. The EQ-5D contained various value sets (previously known as tariffs ), representing valuations that have been elicited for all possible health states defined by the EQ-5D descriptive system. Currently, 16 country-specific value sets are available for the three-level EQ-5D and six for the five-level EQ-5D. Valuation studies for the five-level EQ-5D are still under development in a number of countries. In most countries, value sets are elicited by either the VAS or the TTO. Number of health states generated from the three-level and the five-level EQ-5D, respectively.

6 476 VALUE IN HEALTH 21 (2018) Table 2 Summary assessing the utility-based instruments against the psychometric properties. Psychometric properties Criteria Instruments EQ-5D HUI2/3 QWB DEMQOL-U DEMQOL- Proxy-U AQoL 15D Feasibility Average patientrated NA NA NA completion time (min) Average proxyrated NA 22.5 NA NA completion time (min) Average patientrated 89.5% 77.1% 55.4% 87.9% NA NA NA completion rate Average proxyrated 97% 100% 78.6% NA 74.7% NA NA completion rate Average missing 0.6% 19.2% 24% 15% 2.8% NA NA items Precision Shows ceiling Yes No No Yes Yes No NA effect * Shows floor No Yes No No Yes No NA effect * Reliability Test-retest Moderate Moderate Strong Weak Weak Weak Weak reliability Inter-rater Weak Weak Weak Weak Weak Weak Weak agreement Validity Number of NA NA NA NA relevant attributes included Convergence Strong Inconclusive Moderate Weak Weak Moderate Weak validity Known-group Moderate Inconclusive NA Weak Weak Moderate NA validity according to MMSE Responsiveness Responsiveness Medium Low Low Low Low Low NA AQoL, Assessment of Quality of Life; EQ-5D, EuroQol five-dimension questionnaire (three-level); HUI, Health Utilities Index; MMSE, Mini- Mental State Examination; NA, not available; QWB, Quality of Well-Being. Ceiling or floor effects were indicated when 20% of respondents achieved maximum good or bad health scores, respectively. Reliability was considered to be weak when the correlation coefficient was o0.4, moderate when between 0.4 and 0.7, and strong when There were 56 attributes identified from both patients and caregivers that were important in measuring the quality of life. Validity was considered to be moderate when the coefficient was between 0.3 and 0.5 and strong when larger than 0.5. Instruments were considered to be the most responsive if the effect size score was more than 0.80 and moderately responsive if the score was more than 0.5. compared a new intervention (drug B) with an existing best alternative (drug A) (see Appendix Table A3 in Supplemental Materials found at Drug B was assumed to effectively delay the dementia progression from mild to moderate from 2 years to 5 years. We then compared the changes in incremental QALYs resulting from differences in the magnitude of utility estimates in different scenarios. First, the predicted incremental QALYs were and for the EQ-5D and HUI, respectively, after adjusting other factors. Second, the predicted incremental QALYs were and for the patient self- and proxy-reported EQ-5D scores, respectively, after controlling for other factors. Third, the predicted incremental QALYs were 2.76 and 1.71 for using the EQ- 5D administered by face-to-face interview and other methods (such as phone interview), respectively, after holding other factors constant. On the basis of the aforementioned calculations, the incremental cost-effectiveness ratio (ICER), which is expressed as the ratio of incremental costs (cost of drug B minus cost of drug A) over incremental QALY (QALY of drug B minus QALY of drug A), is expected to be the highest in the CUA when using the EQ-5D self-completed scores through face-to-face interview and the lowest when using the HUI proxy-rated scores via other administration modes (see Appendix Figure A1 in Supplemental Materials found at

7 VALUE IN HEALTH 21 (2018) Fig. 2 Systematic review and meta-regression analysis procedure flowchart. CDR, Clinical Dementia Rating; MMSE, Mini- Mental State Examination. *Participants cognitive impairment levels were measured by the MMSE or CDR and defined as mild (MMSE 4 21 or CDR ¼ 1), moderate (10 o MMSE o 20 or CDR ¼ 2), and severe (MMSE o 10 or CDR 3) cognitive impairment. Instruments and value sets cannot appear in the same regression models because of perfect collinearity. Discussion In this study, we conducted a literature review to compare the performance of utility-based instruments for people with dementia and a meta-regression analysis to explore factors that contributed to the variation in utility values derived from different instruments. This is the first and most comprehensive systematic review that examines the psychometric properties of available utility-based instruments for people with dementia (including AD). The meta-regression analyses quantified the variation between preference weights obtained from different utility-based instruments. The EQ-5D demonstrated better feasibility, reliability, and validity than did other utility-based instruments in the review, but the usefulness for people with dementia may be restricted by its ceiling effect. The review showed that the EQ-5D was the most commonly used utility-based instrument for researchers to derive preference weights for people with dementia. This may be due to its relatively simple three-level scale descriptive system compared with other instruments such as the HUI. Nevertheless, the relatively simple descriptive system may contribute to its ceiling effect. The meta-regression analyses results showed that the EQ-5D produces higher estimated utility values than do other instruments, which is a reflection of its ceiling effect. A new version of the EQ-5D with a five-level scale was developed to overcome the ceiling effect limitations. This five-level EQ-5D has been validated in reducing the ceiling effect while improving discriminatory power and establishing convergence and knowngroup validity [51]. Nevertheless, no similar investigation has been conducted for patients with dementia and, therefore, its performance for people with dementia remains unclear. The HUI and QWB may perform better than the EQ-5D in terms of content validity and precision. The QWB covered a large number of attributes that people with dementia and their caregivers identified to be important in measuring their quality of life. For instance, the QWB addresses all attributes of physical wellbeing considered important by both people with dementia and caregivers. The EQ-5D asks only for the level of pain/discomfort that the individual is experiencing rather than directly addressing the attributes of physical illnesses, ailments, and frailty [38]. In addition, both the QWB and the HUI perform well for people with mild cognitive impairment with no ceiling or floor effect evidence, and the completion rate for the HUI was reported as 100% in the literature [14]. This finding indicates that the selection of the appropriate utility-based instrument for people

8 478 VALUE IN HEALTH 21 (2018) Table 3 Statistics of variables included in meta-regression analysis. Variables Variable definition No. of observations based on severity levels measured by the MMSE (77), observations (%) Number of observations based on categorized severity levels (83), observations (%) Expected impact on utility values MMSE (44) CDR (39) Instruments This variable represents the preferencebased EQ-5D instrument used in reviewed 47 (61) 35 (80) 13 (33) Positive HUI studies. It is divided into three types: the 14 (18) 3 (7) 24 (62) Reference Others EQ-5D, the HUI, and others (including 16 (23) 6 (13) 2 (5) NA the AQoL, QWB, and 15D because of insufficient sample size). Respondents This dummy captures ways to report health state values. Patient selfreporting ¼ 1 if patient self-report 37 (48) 20 (45) 19 (49) Positive Proxyreporting ¼ 0 otherwise 40 (52) 24 (55) 20 (51) Reference Administrations This dummy variable captures different modes of administration used in reviewed articles. Face-to-face ¼ 1 if face-to face interview 62 (81) 40 (91) 33 (85) Positive interview Others ¼ 0 if survey or phone interview 15 (19) 4 (9) 6 (25) Reference Disease type This dummy variable represents the disease type of sample. AD ¼ 1 if participants are diagnosed with AD 46 (60) 33 (75) 32 (82) NA Other ¼ 0 if participants are diagnosed with other 31 (40) 11 (25) 7 (18) Reference dementia types of dementia Mean age This variable represents the average age of sample in each reviewed study. 77 (100) 44 (100) 39 (100) Negative Year of publication * Sample size Severity levels Categorical severity levels This variable indicates the year of publication of individual study. This variable indicates the sample size collected in each reviewed study. This continuous variable indicates cognitive impairment levels of people with dementia or AD measured by the MMSE. This categorical variable indicates cognitive impairment levels of people with dementia or AD measured by either the MMSE or the CDR. It is categorized into three groups: mild, moderate, and severe. 77 (100) 44 (100) 39 (100) NA 77 (100) 44 (100) 39 (100) NA 77 (100) 44 (100) 39 (100) Negative Mild NA 19 (43) 15 (38) Reference Moderate 15 (34) 9 (24) Negative Severe 10 (23) 15 (38) Negative AD, Alzheimer s disease; AQoL, Assessment of Quality of Life; CDR, the Clinical Dementia Rating; EQ-5D, EuroQol five-dimension questionnaire (three-level); HUI, Health Utilities Index; MMSE, Mini-Mental State Examination; NA, not available; QWB, Quality of Well-Being. Year of publication is measured as a continuous variable and is divided into three periods: before 2000, , and after with dementia should consider not only the simplicity but also other aspects such as the objectives of the study, target population, and their level of cognitive impairment. The performance of the AQoL and the 15D in measuring health states for people with dementia is uncertain because of insufficient information available in the literature. For example, only two studies for the AQoL and three studies for the 15D were included in our review and meta-regression analyses. In addition, their psychometric properties were not fully assessed to be compared with other instruments. The DEMQOL-U, the only dementia-specific preference-weight measure included in this study, did not show better psychometric properties than the EQ- 5D, QWB, and HUI. This result, however, may be constrained by the availability and inclusion of only two relevant studies. Hence, further investigation regarding the validity of the DEMQOL-U for people with dementia is needed. Several results from the meta-regression analyses were in agreement with discussion on the variations in utility values in

9 VALUE IN HEALTH 21 (2018) Table 4 Meta-regression analyses results. Variable Coefficient (SE) Model 1 (N ¼ 77 * ) Model 2 (N ¼ 77 * ) Model 3 (N ¼ 75 ) Model 4 (N ¼ 75 ) Constant 3.42 (8.69) (12.41) (18.89) (24.44) Instruments (HUI as the reference) EQ-5D 0.13 (0.07) NA 0.17 (0.05) NA Others 0.17 # (0.08) NA NA NA Respondents (proxy report as reference) Patients 0.14 # (0.05) 0.13 # (0.06) 0.15 (0.05) 0.14 # (0.06) Administrations (others as reference) Face-to-face (0.05) (0.07) 0.35 (0.07) 0.34 (0.08) Disease types (AD as reference) Other dementia (0.05) (0.06) (0.05) 0.01 (0.07) Mean age 0.03 # (0.01) (0.01) (0.008) (0.01) Year of publication (0.004) (0.006) (0.01) (0.01) Sample size ( ) ( ) (0.0001) (0.0002) Severity levels (0.009) (0.01) NA NA Categorical severity levels Moderate NA NA 0.12 # (0.05) 0.13 # (0.05) Severe NA NA 0.25 (0.07) 0.25 (0.07) R F statistics AD, Alzheimer s disease; CDR, the Clinical Dementia Rating; EQ-5D, EuroQol five-dimension questionnaire (three-level); HUI, Health Utilities Index; MMSE, Mini-Mental State Examination; NA, not applicable; R 2, coefficient of determination; SE, standard error. A total of 33 studies providing 77 observations in models 1 and 2. Includes the value set variable instead of the instruments variable, but no result presented because of insignificance. A total of 13 studies providing 75 observations in models 3 and 4. Denotes statistical significance (P o 0.1). Denotes statistical significance (P o 0.01). The Others variable was excluded in model 3 because the total number of studies in this variable was o10. # Denotes statistical significance (P o 0.05). literature. The findings may have important implications for researchers when choosing a suitable instrument for people with dementia so as to achieve reliable QALYs and ICERs in economic evaluation. For instance, the significant differences between selfand proxy-reported scores may suggest that a utility-based instrument with both self- and proxy-completed versions may be necessary because people with dementia and their caregivers have different opinions in assessing various aspects of their quality of life. Failing to prove that the face-to-face interview is a superior administration method may suggest that other costeffective administration methods, such as an online panel or computer-assisted interview techniques, may be considered valid in future studies [26,52]. In addition, our findings were in agreement with Fang et al. [31] and may indicate that different value sets and adjustments may not be necessary when using instruments to calculate health utility scores for people with dementia. Notably, contradiction was indicated for the mean age and type of disease across models. This finding may be clinically important, and further investigations of a larger population are therefore needed to provide an explanation for our findings. Most importantly, we used hypothetical scenarios to demonstrate the impact that differences in estimated utility values have on the QALY and ICER calculations, and the findings may provide useful information for decision makers to better understand CUA outcomes. Our results indicated that using the EQ-5D as the utility-based instrument for people with dementia yields a higher incremental QALY and hence lower ICER. In addition, different respondents (people with dementia or caregivers) and modes of administration (face-to-face interview or others) will also result in different ICERs using the same instrument. The proposed health intervention will not be considered as cost-effective by the policymakers if the ICER is lower than the decision threshold level (such as $25,000/QALY gain). In other words, the decisionmaking process may be biased if QALYs are calculated from the proxy-rated HUI scores via phone interview for people with dementia because those factors will generate lower QALYs and hence higher ICERs in CUAs. As with all systematic reviews and meta-regression analyses, this study was limited by the quality and quantity of the available publications. First, a considerable number of health state values were included in this study even though a limited number of published studies with health state values were identified. Furthermore, most of the data were collected and extracted from studies using the EQ-5D as the utility-based instrument, which may have resulted in measurement errors and bias. Our analyses were unable to specifically compare the values derived from the EQ-5D or the HUI with other available instruments such as the 15D, AQoL, QWB, DEMQOL-U, and DEMQOL-Proxy-U because of insufficient sample sizes. In addition, no investigation of possible publication bias was conducted, which may have affected the reliability of the analyses and represents a potential problem with this study. A linear regression model with robust standard error estimates was applied for the meta-regression analyses. The overall fit was considered as good, with more than half the variance (R 2 ¼ 0.50) explained by the association between the observed and fitted health state values across all regression models. Nevertheless, it placed no restrictions on the predicted values that normally range from 0 to 1. Moreover, the contradictory results for the effect of mean age on mean utility scores suggested that

10 480 VALUE IN HEALTH 21 (2018) there may have been a model specification problem, indicating that a linear model may not have been the most suitable model for our data. Therefore, other forms of the model, such as the inclusion of interaction terms or longitudinal linear models, deserve further investigation. Conclusions The literature review concluded that the EQ-5D remains the most useful utility-based instrument for people with dementia. It should, however, be noted that this conclusion is based on a literature review in which most of the included studies used the EQ-5D to derive health state values for people with dementia. Meanwhile, other instruments demonstrated equivalent performance, or even better, in specific cases. The meta-regression analyses results and hypothetical scenarios showed that variations in estimated utility values can result in higher or lower QALY calculation and hence lead to different decision-making processes for health interventions. These findings can be viewed as a warning for researchers to choose the most reliable utilitybased instruments to provide reliable QALY and ICER measures in health economic evaluations for policymakers. Assessing quality of life is a complex and challenging issue for people with dementia, who experience constraints in judgment, thinking processes, and communicative abilities. This study revealed that currently available utility-based instruments may not perform well for people with dementia. Future studies could develop a disease-specific utility-based instrument for dementia that is able to capture every aspect of quality of life for people with dementia and display better measurement properties. Acknowledgements Source of financial support: This study received no financial support. T. Comans and K.-H. Nguyen are funded via the Cognitive Decline Partnership Centre. Supplemental Materials Supplemental material accompanying this article can be found in the online version as a hyperlink at jval or, if a hard copy of article, at journal.com/issues (select volume, issue, and article). REFERENCES [1] Prince MJ, Wimo A, Guerchet MM, et al. World Alzheimer Report 2015 The Global Impact of Dementia. London: Alzheimer s Disease International, [2] Weinstein MC, Torrance G, McGuire A. QALYs: the basics. Value Health 2009;12:S5 9. [3] EuroQol Group. EuroQol a new facility for the measurement of health-related quality of life. Health Policy 1990;16: [4] Balestroni G, Bertolotti G. EuroQol-5D (EQ-5D): an instrument for measuring quality of life. Monaldi Arch Chest Dis 2012;78: [5] Coucill W, Bryan S, Bentham P, et al. EQ-5D in patients with dementia: an investigation of inter-rater agreement. Med Care 2001;39: [6] Mulhern B, Rowen D, Brazier J, et al. Development of DEMQOL-U and DEMQOL-PROXY-U: generation of preference-based indices from DEMQOL and DEMQOL-PROXY for use in economic evaluation. Health Technol Assess 2013;17: [7] Rowen D, Mulhern B, Banerjee S, et al. Comparison of general population, patient, and carer utility values for dementia health states. Med Decis Making 2015;35: X [8] Shearer J, Green C, Ritchie CW, Zajicek JP. Health state values for use in the economic evaluation of treatments for Alzheimer s disease. Drugs Aging 2012;29: [9] Hounsome N, Orrell M, Edwards RT. EQ-5D as a quality of life measure in people with dementia and their carers: evidence and key issues. Value Health 2011;14: [10] Aguirre E, Kang S, Hoare Z, et al. How does the EQ-5D perform when measuring quality of life in dementia against two other dementiaspecific outcome measures? Qual Life Res 2016;25: org/ /s [11] Lacey L, Bobula J, Rüdell K, et al. Quality of life and utility measurement in a large clinical trial sample of patients with mild to moderate Alzheimer s disease: determinants and level of changes observed. Value Health 2015;18: [12] Naglie G, Hogan DB, Krahn M, et al. Predictors of patient self-ratings of quality of life in Alzheimer s disease: cross-sectional results from the Canadian Alzheimer s Disease Quality of Life (CADQOL) study. Am J Geriatr Psychiatry 2011;19: b013e a67. [13] Acton GJ, Mayhew PA, Hopkins BA, Yauk S. Communicating with individuals with dementia the impaired person s perspective. J Gerontol Nurs 1999;25:6 9. [14] Naglie G, Tomlinson G, Tansey C, et al. Utility-based quality of life measures in Alzheimer s disease. Qual Life Res 2006;15: dx.doi.org/ /s [15] Hawthorne G, Richardson J. Measuring the value of program outcomes: a review of multiattribute utility measures. Expert Rev Pharmacoecon Outcomes Res 2001;1: [16] Sansoni J. Final Report: Dementia Outcomes Measurement Suite Project. Wollongong, Australia: Centre for Health Service Development, University of Wollongong, [17] Neumann PJ, Kuntz KM, Leon J, et al. Health utilities in Alzheimer s disease: a cross-sectional study of patients and caregivers. Med Care 1999;37: [18] Brazier J, Deverill M, Green C. A review of the use of health status measures in economic evaluation. J Health Serv Res Policy 1999;4: [19] Mellenbergh GJ. A review of quality of life instruments used in dementia. Qual Life Res 2005;14: s [20] Moher D, Liberati A, Tetzlaff J, Altman DG; Prisma Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med 2009;6:e [21] Fitzpatrick R, Davey C, Buxton MJ, Jones DR. Evaluating patient-based outcome measures for use in clinical trials. Health Technol Assess 1998;2:1 74. [22] Longworth L, Yang Y, Young T, et al. Use of generic and conditionspecific measures of health-related quality of life in NICE decisionmaking: a systematic review, statistical modelling and survey. Health Technol Assess 2014;18: [23] Haywood KL, Garratt AM, Fitzpatrick R. Quality of life in older people: a structured review of generic self-assessed health instruments. Qual Life Res 2005;14: [24] Streiner D, Norman GR. Health Measurement Scales: A Practical Guide to Their Development and Use (5th ed.). New York, NY: Oxford University Press, [25] Krahn M, Naglie G. Content validity for dementia of three generic preference based health related quality of life instruments. Qual Life Res 2002;11: [26] Stevanovic J, Pechlivanoglou P, Kampinga MA, et al. Multivariate metaanalysis of preference-based quality of life values in coronary heart disease. PLoS One 2016;11:e pone [27] Norman R, King MT, Clarke D, et al. Does mode of administration matter? Comparison of online and face-to-face administration of a time trade-off task. Qual Life Res 2010;19: [28] Sullivan PW, Ghushchyan V. Preference-based EQ-5D index scores for chronic conditions in the United States. Med Decis Making 2006;26: [29] Wyld M, Morton RL, Hayen A, et al. A systematic review and metaanalysis of utility-based quality of life in chronic kidney disease treatments. PLoS Med 2012;9:e journal.pmed [30] Oremus M, Tarride J-E, Clayton N, Raina P; Canadian Willingness-to- Pay Study Group. Health utility scores in Alzheimer s disease: differences based on calculation with American and Canadian preference weights. Value Health 2014;17: [31] Fang M, Oremus M, Tarride J-E, Raina P. A comparison of health utility scores calculated using United Kingdom and Canadian preference

11 VALUE IN HEALTH 21 (2018) weights in persons with Alzheimer s disease and their caregivers. Health Qual Life Outcomes 2016;14:105. [32] Thompson SG, Higgins JPT. How should meta-regression analyses be undertaken and interpreted? Stat Med 2002;21: org/ /sim [33] Ankri J, Beaufils B, Novella JL, et al. Use of the EQ-5D among patients suffering from dementia. J Clin Epidemiol 2003;56: [34] Orgeta V, Edwards RT, Hounsome B, et al. The use of the EQ-5D as a measure of health-related quality of life in people with dementia and their carers. Qual Life Res 2014;24: s [35] Karlawish JH, Zbrozek A, Kinosian B, et al. Preference-based quality of life in patients with Alzheimer s disease. Alzheimers Dement 2008;4: [36] Diaz-Redondo A, Rodriguez-Blazquez C, Ayala A, et al. Spanish Research Group on Quality of Life and Aging. EQ-5D rated by proxy in institutionalized older adults with dementia: psychometric pros and cons: EQ-5D by proxy in dementia. Geriatr Gerontol Int 2014;14: [37] Karlawish JH, Zbrozek A, Kinosian B, et al. Caregivers assessments of preference-based quality of life in Alzheimer s disease. Alzheimers Dement 2008;4: [38] Silberfeld M, Rueda S, Krahn M, Naglie G. Content validity for dementia of three generic preference based health related quality of life instruments. Qual Life Res 2002;11: A: [39] Thorgrimsen L, Selwood A, Spector A, et al. Whose quality of life is it anyway? The validity and reliability of the Quality of Life- Alzheimer s Disease (QoL-AD) scale. Alzheimer Dis Assoc Disord 2003;17: [40] Selwood A, Thorgrimsen L, Orrell M. Quality of life in dementia a one-year follow-up study. Int J Geriatr Psychiatry 2005;20: [41] Boström F, Jönsson L, Minthon L, Londos E. Patients with dementia with Lewy bodies have more impaired quality of life than patients with Alzheimer disease. Alzheimer Dis Assoc Disord 2007;21: [42] Bhattacharya S, Vogel A, Hansen M-LH, et al. Generic and diseasespecific measures of quality of life in patients with mild Alzheimer s disease. Dement Geriatr Cogn Disord 2010;30: / [43] Naglie G, Hogan DB, Krahn M, et al. Predictors of family caregiver ratings of patient quality of life in Alzheimer disease: cross-sectional results from the Canadian Alzheimer s Disease Quality of Life study. Am J Geriatr Psychiatry 2011;19: b013e a7f. [44] Nikmat AW, Hawthorne G, Al-Mashoor SH. The comparison of quality of life among people with mild dementia in nursing home and home care a preliminary report. Dementia 2015;14: / [45] Kavirajan H, Hays RD, Vassar S, Vickrey BG. Responsiveness and construct validity of the health utilities index in patients with dementia. Med Care 2009;47: [46] Wolfs CAG, Dirksen CD, Kessels A, et al. Performance of the EQ-5D and the EQ-5DþC in elderly patients with cognitive impairments. Health Qual Life Outcomes 2007;5: [47] León-Salas B, Ayala A, Blaya-Nováková V, et al. Quality of life across three groups of older adults differing in cognitive status and place of residence: quality of life of older adults. Geriatr Gerontol Int 2015;15: [48] Heßmann P, Seeberg G, Reese JP, et al. Health-related quality of life in patients with Alzheimer s disease in different German health care settings. J Alzheimers Dis 2016;51: [49] Jönsson L, Andreasen N, Kilander L, et al. Patient- and proxy-reported utility in Alzheimer disease using the EuroQoL. Alzheimer Dis Assoc Disord 2006;20: c9. [50] Neumann PJ, Sandberg EA, Araki SS, et al. A comparison of HUI2 and HUI3 utility scores in Alzheimer s disease. Med Decis Making 2000;20: [51] Janssen M, Pickard AS, Golicki D, et al. Measurement properties of the EQ-5D-5L compared to the EQ-5D-3L across eight patient groups: a multi-country study. Qual Life Res 2013;22: [52] Gwaltney CJ, Shields AL, Shiffman S. Equivalence of electronic and paper-and-pencil administration of patient-reported outcome measures: a meta-analytic review. Value Health 2008;11: