Differences in utility elicitation methods in cardiovascular disease: a systematic review

Abstract

Aims: Utility values inform estimates of the cost-effectiveness of treatment for cardiovascular disease (CVD), but values can vary depending on the method used. The aim of this systematic literature review (SLR) was to explore how methods of elicitation impact utility values for CVD.

Materials and methods: This review identified English-language articles in Embase, MEDLINE, and the gray literature published between September 1992 and August 2015 using keywords for “utilities” and “stroke”, “heart failure”, “myocardial infarction”, or “angina”. Variability in utility values based on the method of elicitation, tariff, or type of respondent was then reported.

Results: This review screened 4,341 citations; 290 of these articles qualified for inclusion in the SLR because they reported utility values for one or more of the cardiovascular conditions of interest listed above. Of these 290, the 41 articles that provided head-to-head comparisons of utility methods for CVD were reviewed. In this sub-set, it was found that methodological differences contributed to variation in utility values. Direct methods often yielded higher scores than did indirect methods. Within direct methods, there were no clear trends in head-to-head studies (standard gamble [SG] vs time trade-off); but general population respondents often provided lower scores than did patients with the disease when evaluating the same health states with SG methods. When comparing indirect methods, the EQ-5D typically yielded higher values than the SF-6D, but also showed more sensitivity to differences in health states.

Conclusions: When selecting CVD utility values for an economic model, consideration of the utility elicitation method is important, as this review demonstrates that methodology of choice impacts utility values in CVD.

Keywords:

• Cardiovascular disease
• stroke
• myocardial infarction
• heart failure
• angina
• utilities
• standard gamble
• time trade-off
• EQ-5D

Introduction

Cardiovascular disease (CVD) is the leading cause of mortality in both the United States (US) and the European Union (EU)^1–3. The annual total cost of CVD, including direct healthcare costs, lost productivity, and informal caregiver burden, was ∼ $320 billion (2011 US dollars) in the US and more than €196 billion (2009 euros) in the EU^4,5. Such evidence of the clinical and economic burden associated with CVD emphasizes the need for new interventions that are both effective and affordable.

Assessing whether new candidate treatments for CVD meet health economic standards involves using cost-utility analyses to estimate the cost per quality-adjusted life-year associated with the interventions. This method relies on quality adjustment of any life-year gains through the use of utilities—values that represent the preferences of individuals for health states and that can be thought of as measures summarizing health-related quality-of-life^6–9. To this end, the accuracy of cost-utility analyses and their estimates depends on the reliability and appropriateness of the chosen utility values. This raises questions about the various ways health states are elicited (direct, indirect) and the comparability of their results.

Common direct methods of eliciting health states include standard gamble (SG, which produces utilities) and time trade-off (TTO, which produces values). During these interviews, individuals engage in choice-based tasks to identify their preferences for either their own current health or scenarios (also called vignettes) that describe various health states^10,11.

Indirect methods of eliciting utility values include the EuroQol–Five Dimensions (EQ-5D; which requires respondents to rate their health and has three or five levels of response), the Short Form–Six Dimensions (SF-6D), and the Health Utilities Index (HUI; two versions, HUI-2 or HUI-3; both are scored using country-specific tariffs)^12,13,14,15. These tools measure quality-of-life on a number of domains such as physical functioning, pain, and emotional well-being; the domains for each of these common measures are described in Table 1. For these metrics, a utility value is calculated by applying a scoring algorithm called a tariff; this tariff weights each item in a survey and is derived from preferences in the general population based on direct methods (such as SG or TTO).

Table 1. Domains measured in common indirect utility tools.

EQ-5D	HUI-2	HUI-3	SF-6D
Mobility Self-care Usual activities Pain/discomfort Anxiety/depression	Mobility Sensation Emotion Cognitive Self-care Pain Fertility	Vision Hearing Speech Ambulation Dexterity Emotion Cognition Pain	Physical functioning Role participation Social functioning Bodily pain Mental health Vitality

Given that studies can vary in the methods used to elicit utilities (direct, indirect, or both) as well as the populations studied, it is not surprising that the values for CVD health states can vary widely from study to study, thus complicating the choice of which estimates to use in cost-utility analyses. Because CVD has an impact on quality-of-life and, thus, utilities in patients who survive cardiovascular (CV) events (myocardial infarction [MI], stroke, etc.), utility values may be impacted by both the condition and the time since the CV event¹⁶. While reviews of health utilities that use different methods have been conducted independently, there are few reviews that provide an overall picture of utility values across CVD conditions¹⁷. Thus, to provide insight into this topic and characterize utility values across CVD conditions, a systematic literature review (SLR) was conducted to identify and assess published utilities for stroke, heart failure (HF), MI, and angina. An examination of head-to-head studies allowed for the exploration of methodological factors that may contribute to the variation between estimates for CV health states.

Methods

Search strategy

This systematic review was designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards and conducted in MEDLINE and Embase using search algorithms to identify relevant English-language publications presenting utilities related to CV conditions. Searches identified publications using the main keywords of “stroke”, “myocardial infarction”, “angina”, and “heart failure”. Complete details are provided in the Supplementary Appendix. To identify studies of health state utilities, additional keywords included “utility”, “time trade-off”, “standard gamble”, “HUI”, “SF-6D”, and “EQ-5D”. Keywords had to appear in the title and/or abstract of publications since September 1992, and the search covered publications through the most recent update to this review on August 3, 2015.

Study review, selection, and data extraction

Abstracts were manually reviewed against pre-determined inclusion and exclusion criteria (Table 2) to select primary studies and systematic reviews reporting utilities for CV events. We excluded studies that implemented visual analog scale methods, as visual analog scales are meant to introduce respondents to the health states and to get an approximation of the utility score, but are not utilities themselves. Torrance et al.¹⁸ (2001) specify that visual analog scores are prone to bias and should not be used alone. To supplement the search, gray literature (i.e. materials such as meeting abstracts that are published, but are not peer reviewed) on MEDLINE or Embase, as well as at scientific organizations or health technology assessment body websites, was examined for the final 4 years of the review period (i.e. 2011–2015). The bibliographies of SLR articles were examined to identify additional articles noted as relevant by prior reviewers. All abstracts were examined manually by a single reviewer against the inclusion criteria. All articles accepted during abstract screening were reviewed in full text by two independent investigators against the inclusion and exclusion criteria. Any discrepancies in the decisions were reviewed and resolved by a third investigator. Data (study design, sample size, utility measure, tariff, country, interventions, CV condition, respondents, age, gender, disease severity, and time since event) were extracted into pre-determined table shells and validated by a second investigator. We compared indirect vs direct methods, indirect vs other indirect methods, as well as direct vs other direct methods.

Table 2. Inclusion/exclusion criteria.

	Inclusion	Exclusion
Health states	≥18 years with any of the following events:  • Stroke (including TIA)  • MI  • Angina (stable and unstable)  • HF	≤18 years and/or without CV conditions of interest
Outcomes	Any health state utility value	No health state utility value reported
Study design	Study generated the utility value (i.e. “primary” utility values)	Study reported previously published utility values (e.g. economic models or literature reviews)

Abbreviations. CV, cardiovascular; HF, heart failure; MI, myocardial infarction; TIA, transient ischemic attack.

Definition of minimally important difference

Minimally important differences for utilities vary by measure and are not well established. However, thresholds of 0.05–0.10 (TTO) have been suggested to represent clinically important differences in direct utility elicitation^19–24. For the indirect methods EQ-5D, HUI, and SF-6D, minimally important differences have generally been reported in a range of 0.01–0.08^25,26. Some authors consider half of the standard deviation as the minimally important difference^27,28.

Results

In total, 4,341 citations were screened, and 290 articles (283 from the literature databases and seven documents from the gray literature) reported utility values for one or more of the CV conditions of interest and were included in the SLR (Table 3; Figure 1). We observed a large range of values across each CV diagnosis (stroke = –0.23–0.99; HF = 0.199–0.995; MI = 0.11–1.0; angina = 0.05–1.0; acute coronary syndrome [ACS] = 0.63–0.847). The studies showed a clear trend of utility values being lowest for stroke, followed by HF, angina, and MI, although the variation within each health state was substantial. Some of this variation was explained by disease severity, disease sub-types, timing of elicitation, comorbidities, and interventions. However, even across studies evaluating similar health states, utility values varied. Consequently, the relationship between key methodological factors—including how the utilities were elicited, with regards to method, type of respondents, or tariff—and the variation in scores was evaluated. Therefore, we evaluated the sub-set of 41 studies that directly compared the method of elicitation, tariff, or type of respondent; we found that 23 studies compared methods of elicitation, seven compared tariffs used to calculate EQ-5D index values, and seven compared types of respondents. Of the 41 studies, 27 used the EQ-5D; all of the 27 either specified using the three-level version (no problems, some/moderate problems, severe/extreme problems) for each domain or did not specify (and therefore we can assume the three-level version was used). No study used the newer five-level version. The selection of these articles is described in Figure 1, and the sub-set of 41 studies is listed in the Supplemental Appendix. The results below reflect findings from the sub-set of 41 articles reporting head-to-head comparisons of utilities.

Figure 1. Systematic review study attrition (PRISMA) diagram. Abbreviations. CV, cardiovascular; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Table 3. Characteristics of the 290 included articles.

	Stroke (n = 155)	HF (n = 100)	Angina^a (n = 50)	MI^a (n = 74)
Method of utility elicitation^b
EQ-5D	102 (66%)	70 (70%)	31 (62%)	45 (61%)
HUI	13 (8%)	10 (10%)	1 (2%)	6 (8%)
SF-6D	10 (6%)	6 (6%)	8 (16%)	6 (8%)
SG	14 (9%)	5 (5%)	4 (8%)	4 (5%)
TTO	16 (10%)	18 (18%)	9 (18%)	10 (14%)
Type of respondent
Patients	148 (95%)	91 (91%)	44 (88%)	68 (92%)
General population	8 (5%)	7 (7%)	5 (10%)	6 (8.1%)
Carers	7 (5%)	2 (2%)	0 (0%)	0 (0%)
Geographic setting
US	24 (15%)	30 (30%)	10 (20%)	16 (22%)
UK	33 (21%)	21 (21%)	9 (18%)	11 (15%)
Canada	20 (13%)	7 (7%)	6 (12%)	7 (9%)
Germany	5 (3%)	3 (3%)	4 (8%)	7 (9%)
Sweden	9 (6%)	8 (8%)	2 (4%)	3 (4%)
Australia	5 (3%)	3 (3%)	1 (2%)	4 (5%)
Spain	4 (3%)	8 (8%)	0 (0%)	0 (0%)
Netherlands	7 (5%)	3 (3%)	0 (0%)	0 (0%)
Norway	6 (4%)	0 (0%)	0 (0%)	2 (3%)
South Korea	7 (5%)	0 (0%)	4 (8%)	5 (7%)
Other countries	37 (24%)	17 (17%)	13 (26%)	18 (24%)

Abbreviations. EQ-5D, EuroQol–Five Dimensions; HF, heart failure; HUI, Health Utilities Index; MI, myocardial infarction; SF-6D, Short Form–Six Dimensions; SG, standard gamble; TTO, time trade-off.

A comprehensive listing of all studies included here can be found in the Supplemental Appendix.

Some totals may add to more than 100% in a column due to some studies evaluating more than one method, type of respondent, or country.

^a Acute coronary syndrome values are included in both the angina and MI columns.

^b Utility value reported on a 10-point scale.

Studies that compared direct and indirect methods of elicitation

Direct methods of utility elicitation (i.e. SG and TTO) yielded larger scores than indirect methods, particularly the EQ-5D compared with the UK tariff. The three studies comparing TTO methods to the EQ-5D, all of which were conducted in HF patients, found that the TTO yielded similar or slightly larger scores than the EQ-5D^29–31. The EQ-5D compared with the UK tariff yielded values somewhat smaller than the TTO (by 0.07–0.09)^29,31. In contrast, one of these studies found a greater score with the EQ-5D, using the US tariff, than the TTO (by 0.03)³¹. The SG yielded even greater values than the EQ-5D (by +0.17) in a study of stroke patients, although the authors did not report what tariff was used³².

Studies that compared indirect methods

When comparing indirect methods, the EQ-5D typically yielded higher values than the SF-6D, but also showed more sensitivity to differences in health states (Table 4). Eight studies in the review compared EQ-5D and SF-6D values for the same group of respondents. Six of these studies, evaluating stroke, HF, MI, and angina, reported SF-6D scores lower than the corresponding EQ-5D scores by a small difference, typically ∼0.08 points^31,33–37. In the other two studies, both evaluating stroke patients, the SF-6D yielded values between 0.03 and 0.2 points higher than the EQ-5D^38,39. Across studies comparing HUI methods (used in studies of stroke, HF, and MI), HUI-2 scores were typically very similar to EQ-5D scores^35,39,40, whereas HUI-3 scores were consistently much lower than EQ-5D scores by differences typically ∼0.15 points, and as much as 0.32^35,39–41. Studies described above as well as additional literature surveyed are reported in Table 4 (Wu 2014⁴², Hebert 2008⁴³, Kaplan 2011⁴⁴).

Table 4. Average utility values in studies comparing methods of elicitation.

Reference (year)	Health state	Sample size	Time point	Utility elicitation method
Studies comparing indirect methods				EQ-5D	SF-6D	HUI-2	HUI-3
Feeny et al.^35 (2012)	HF	36	Baseline	0.77 (0.18)	0.63 (0.11)	0.76 (0.22)	0.62 (0.32)
			6 months	0.76 (0.18)	0.66 (0.13)	0.76 (0.24)	0.65 (0.33)
Darlington et al.^33 (2007)	Stroke	80	2 months	0.69 (0.25)	0.69 (0.12)	NR	NR
			1 year	0.72 (0.22)	0.69 (0.12)	NR	NR
De Smedt et al.^34 (2014)	AMI	NR	NR	0.78 (0.66–1.00)^a	0.66 (0.62–0.80)^a	NR	NR
	Stroke	NR		0.69 (0.52–0.85)^a	0.66 (0.57–0.74)^a	NR	NR
Han et al.^36 (2012); Wu et al.^37 (2012); Wu^42 (2014)	Stable angina	411	NR	0.75 (0.19) (0.73–0.77)^b	0.68 (0.12) (0.67–0.69)^b	NR	NR
Pickard et al.^39 (2005); Pickard et al.^41 (2004)	Ischemic stroke	98	Baseline	0.31 (0.38)	0.55 (0.09)	0.51 (0.20)	0.19 (0.30)
			6 months	0.62 (0.33)	0.68 (0.15)	0.64 (0.23)	0.44 (0.35)
Stevanovic et al.^31 (2014)	HF	NR	NR	0.6385 (US) 0.7593 (UK)	0.6382	NR	NR
Barton et al.^38 (2008)	Stroke	62	NR	0.612 (0.318)	0.644 (0.162)	NR	NR
Haacke et al.^40 (2006)	Hemorrhagic stroke	5	2 years	0.74 (0.39)	NR	0.67 (0.28)	0.55 (0.50)
	Infarction and/or TIA	54		0.68 (0.33)	NR	0.61 (0.24)	0.36 (0.38)
	TIA	18		0.90 (0.16)	NR	0.86 (0.10)	0.79 (0.23)
Hebert^43 (2008)	HF patients in a nurse-led program	146	1 year	0.708	NR	NR	0.6619
	HF patients receiving usual care	140		0.6651	NR	NR	0.6122
Kaplan^44 (2011)	HF	160	From baseline to 1 month	+0.005 (0.14)^c	+0.022 (0.087)^c	+0.009 (0.16)^c	+0.012 (0.20)^c
			From 1 month to 6 months	–0.000 (0.16)^c	+0.014 (0.086)^c	+0.003 (0.15)^c	+0.020 (0.20)^c
Studies comparing indirect and direct methods				EQ-5D	SF-6D	TTO	SG
Stevanovic et al.^31 (2014)	HF	NR	NR	0.6385 (UK) 0.7593 (US)	0.6382	0.7287	NR
Kraai et al.^29 (2013)	HF	100	NR	0.68 (0.26)	NR	0.77 (0.26)	NR
Wannasiri and Kapol^32 (2011)	Stroke	284	NR	0.55	NR	NR	0.72
Sánchez et al.^30 (2010)	HF treated with peritoneal dialysis	17	6 months	0.6727	NR	0.675 (0.078)	NR
	HF treated conservatively	NR	NR	0.4305	NR	0.456 (0.213)	NR
Studies comparing direct methods				TTO		SG
Hallan et al.^10 (1999)	Minor stroke	158	NR	0.88 (0.64–0.98)^d		0.91 (0.69–0.99)^d
	Major stroke			0.51 (0.12–0.79)^d		0.61 (0.10–0.91)^d
Lalonde et al.^45 (1999)	Angina	30	6 or more months	80.3		92.4
Yi et al.^46 (2004)	HF	81	NR	0.8 (0.20)		0.91 (0.10)
Lewis et al.^47 (2001)	HF	99	3 or more months	0.65		0.64
Gage et al.^48 (1996)	Moderate stroke	70	NR	0.39		0.26
Nease et al.^49 (1995)	Chronic stable angina (CCS1)	38	>1 year in 80% of respondents	1.0^a		0.965^a
	Chronic stable angina (CCS2)	107		0.997^a		0.97^a
	Chronic stable angina (CCS3)	66		0.929^a		0.875^a

Abbreviations. AMI, acute myocardial infarction; CCS, Canadian Cardiovascular Society; CI, confidence interval; EQ-5D, EuroQol–Five Dimensions; HF, heart failure; HUI, Health Utilities Index; IQR, interquartile range; NR, not reported; SD, standard deviation; SF-6D, Short Form–Six Dimensions; SG, standard gamble; TTO, time trade-off; TIA, transient ischemic attack; UK, United Kingdom; US, United States.

Mean (SD) unless otherwise specified.

^a Median or median (IQR).

^b Mean (95% CI).

^c Mean difference (SD).

^d Median (10th–90th percentile).

Among the studies of indirect methods, it appears that the EQ-5D shows the greatest sensitivity to changes over time. Pickard et al.^39,41 reported on a prospective cohort of patients in Canada hospitalized for acute ischemic stroke; they found that health status improved over a 6-month follow-up, and that the EQ-5D and HUI-3 showed the greatest improvements (+0.31 and +0.25, respectively), whereas SF-6D and HUI-2 yielded modest improvements (both +0.13). Similarly, a study in the Netherlands reported that improvements in stroke patients from 2 months to 1 year were reflected by the EQ-5D (+0.03), but not by the SF-6D (no change)³³.

Studies that compared direct methods

There were no clear trends in the six studies that compared direct methods (SG and TTO). Three studies that evaluated stroke, angina, and HF used TTO methods, which yielded higher values than SG methods when evaluating the same health states; in these studies, most health states showed a 0.1–0.14-point difference in scores depending on the method^10,45,46. One study in HF showed nearly identical TTO and SG values for HF⁴⁷, and the TTO yielded lower values than the SG in a study of angina and another of stroke, with differences as high as 0.13^48,49. Utility values in studies comparing respondents for hypothetical health states measured with direct methods of elicitation are reported in Table 5.

Table 5. Median (IQR) utility values in studies comparing respondents for hypothetical health states measured with direct methods of elicitation.

Reference (year)	Method	Health state	Respondent type
			Stroke survivors (n = 41)	Healthy population (n = 66)	Non-stroke patients (n = 51)
Hallan et al.^10 (1999)	SG	Minor stroke for ages 20–44	NR	0.91	0.95
		Minor stroke for ages 45–64	0.95	0.90	0.95
		Minor stroke for ages 65–84	0.95	0.87	0.82
		Major stroke for ages 20–44	NR	0.52	0.74
		Major stroke for ages 45–64	0.85	0.54	0.64
		Major stroke for ages 65–84	0.99	0.59	0.51
			Atrial fibrillation patients with history of stroke (n = 20)	Atrial fibrillation patients without history of stroke (n = 50)
Gage et al.^48 (1996)	TTO	Minor stroke	0.95	0.94
		Moderate stroke	0.00	0.12
		Major stroke	0.00	0.00
			Stroke survivors (n = 75)	General population (n = 75)
Slot and Berge^11 (2009)	SG	Mild stroke	0.93 (0.90–0.98)	0.91 (0.88–0.95)
		Moderately severe stroke	0.78 (0.70–0.90)	0.68 (0.58–0.80)
		Severe stroke	0.18 (0.08–0.24)	0.11 (0.04–0.15)
			Stroke survivors (n = 11)	Health professionals (n = 20)	Non-stroke patients (n = 22)
Murphy et al.^54 (2001)	SG	Mild stroke	0.93 (0.50–0.98)	0.88 (0.75–0.98)	0.78 (0.48–0.98)
		Moderate stroke	0.73 (0.03–0.98)	0.68 (0.41–0.78)	0.30 (0.01–0.63)
		Severe stroke	0.40 (0.48–0.98)	0.14 (<0–0.45)	<0 (<0–0.48)

Abbreviations. IQR, interquartile range; NR, not reported; SG, standard gamble; TTO, time trade-off.

Region and tariff

Across countries, few differences in utility values were observed; however, in the UK, EQ-5D values were notably lower than those in other countries, particularly for stroke and angina; this trend was most notably seen in a comparison with the US. EQ-5D utility values in studies comparing tariffs are reported in Table 6. Four studies in the review directly compared EQ-5D scores calculated with a UK tariff vs a US tariff; all were conducted in multi-national populations and found that the US tariff resulted in higher scores for HF, angina, and MI, although differences ranged from 0.03 to 0.12^31,47,50,51. A hospital-based cohort study in Taiwan, however, reported EQ-5D utility values that were 0.1 higher with the UK tariff vs the US tariff⁵². The UK tariff also resulted in lower EQ-5D index scores compared with Swedish and Malaysian tariffs⁵³. Studies mentioned above are reported in Table 6.

Table 6. Average EQ-5D utility values in studies comparing tariffs.

				Tariff
Reference (year) country	Health state	Sample size	Time point	US (Shaw^73 2005)	UK (Dolan^74 1997)
Lee et al.^52 (2010) Taiwan	Ischemic stroke (lacunar infarction)	170	NR	0.8 (0.2)	0.9 (0.2)
	Ischemic stroke (non-lacunar infarction)	263		0.7 (0.3)	0.8 (0.3)
	Intracerebral hemorrhage	53		0.6 (0.3)	0.7 (0.3)
Lewis et al.^47 (2014) Multiple	MI with additional CV event	597	NR	0.76 (0.22)	0.70 (0.29)
	MI without additional CV event	1,785	NR	0.83 (0.17)	0.80 (0.23)
Eurich et al.^50 (2006) Multiple	HF	298	NR	0.74 (0.17)	0.66 (0.26)
	HF +2 NYHA classes	2	Baseline	0.78 (0.11)	0.75 (0.19)
			6 weeks	0.82 (0.06)	0.79 (0.14)
	HF +1 NYHA class	50	Baseline	0.76 (0.16)	0.68 (0.25)
			6 weeks	0.77 (0.16)	0.70 (0.24)
	HF with no change in NYHA class	206	Baseline	0.74 (0.18)	0.66 (0.27)
			6 weeks	0.77 (0.16)	0.71 (0.22)
	HF –1 NYHA class	40	Baseline	0.73 (0.18)	0.65 (0.27)
			6 weeks	0.74 (0.17)	0.65 (0.25)
				US (Wijeysundera^75 2011)^a	UK (Goldsmith^76 2010)^a
Kohn et al.^51 (2014) Multiple	Chronic stable angina (all patients)	548	6 weeks	0.75 (0.69–0.80)^b	0.68 (0.62–0.77)^b
	No angina symptoms	28		0.89 (0.84–0.92)^b	0.87 (0.77–0.91)^b
	Monthly angina symptoms	188		0.80 (0.75–0.85)^b	0.76 (0.70–0.81)^b
	Weekly angina symptoms	283		0.72 (0.68–0.76)^b	0.65 (0.61–0.70)^b
	Daily angina symptoms	49		0.65 (0.61–0.69)^b	0.54 (0.52–0.61)^b
				US (unspecified)	UK (unspecified)
Stevanovic et al.^31 (2014) Multiple	HF	NR	NR	0.6385	0.7593
				Sweden (Burstrom^77 2014)	UK (Dolan^74 1997)
Berg et al.^53 (2015) Sweden	HF	5,334	Baseline	0.828 (0.135)	0.696 (0.302)
		3,495	12 months	0.846 (0.126)	0.732 (0.275)
				Malaysia (Yusof^78 2012)	UK (Dolan^74 1997)
Azmi^79 (2015) Malaysia	ACS	104	Baseline	0.75	0.63
		53	12 months	0.82	0.72

Abbreviations. ACS, acute coronary syndrome; CV, cardiovascular; EQ-5D EuroQol–Five Dimensions; HF, heart failure; IQR, interquartile range; MI, myocardial infarction; NR, not reported; NYHA, New York Heart Association; SD, standard deviation; UK, United Kingdom; US, United States.

Mean (SD) unless otherwise specified.

^a Kohn et al.⁵¹ mapped EQ-5D scores from the Seattle Angina Questionnaire.

^b Median (IQR).

General population respondents

Hypothetical vignettes of disease states can be used to elicit utility values from general population respondents instead of self-assessments of patients’ health. Use of such a technique across studies in this review elicited utility values for stroke, HF, angina, and MI that were similar to those elicited from patients, albeit within a somewhat narrower range for each of the conditions (Table 6). Four studies, all evaluating stroke, directly compared utility values from general population respondents and patients^10,11,48,54 (Table 5). The three studies using SG methods consistently showed that higher utilities were derived from patients compared with general population respondents^10,11,54. The first of these, a Norwegian study that compared stroke survivors and age-matched control subjects who had at least one risk factor for stroke, reported that controls rated health state descriptions such that median SG utility values were 0.91 for mild stroke, 0.68 for moderately severe stroke, and 0.11 for severe stroke. Compared with these controls, stroke survivors rated the same health state descriptions as having significantly higher utility values (p < .01 for all comparisons; Table 5). Specifically, they reported median SG utility values of 0.93 for mild stroke, 0.78 for moderately severe stroke, and 0.18 for severe stroke¹¹. A second study in Norway recruited healthy individuals, non-stroke patients, and stroke survivors, and similarly found that, for nearly all age stratifications, the healthy individuals and non-stroke patients estimated lower utilities than stroke survivors for both major and minor stroke¹⁰. Also, a study by Murphy et al.⁵⁴ reported that patients who had experienced a severe stroke assigned higher median SG utility values to hypothetical stroke scenarios than age- and sex-matched controls who had not experienced a stroke. In particular, these patients assigned higher SG values to mild stroke (0.93 vs 0.78, p = not reported [NR]), moderate stroke (0.73 vs 0.30, p = .06), and severe stroke (0.40 vs 0.0, p = NR) compared with the controls⁵⁴. This trend, however, was not seen in the fourth study that directly compared utility values from general population respondents and patients using a TTO method. Specifically, this study found that patients with atrial fibrillation (AF) and a history of stroke yielded utility estimates for mild and major stroke nearly identical to those of patients with AF and no history of major stroke. However, for moderate stroke, patients with AF who had experienced a stroke estimated a somewhat lower utility (0.00) than patients with AF who had not experienced a stroke (0.12)⁴⁸.

Caregiver respondents

Another type of respondent used to evaluate utility values was caregivers of individuals with CVD. Three studies we surveyed compared utility values for patients elicited by caregivers to patients’ own utility values, in order to establish the validity of using caregivers as proxies (Table 7). When family members or carers evaluated the health states of patients, the EQ-5D index scores so elicited were slightly lower than the patients’ own EQ-5D ratings in two studies of stroke^41,55, as well as in a third study in patients who had experienced a severely disabling stroke, wherein hospital staff evaluated the same hypothetical health states for mild, moderate, and severe stroke via an SG⁵⁶. Differences ranged from 0 to 0.1, and were statistically significant in some cases. Notably, the study also found that the hospital staff scores were higher than those derived from the controls matched by age and sex to the patients evaluated⁵⁴. Like general population respondents, health professionals’ estimates of utility for CVD appear lower than those for patients or their family member proxies, especially immediately after an event. A UK study of informal carers (e.g. close relatives) and formal carers (e.g. nurses) for patients with acute HF showed that informal carers reported higher EQ-5D utility values at baseline (0.675), but perception of utility value increased only slightly over time (0.726), whereas formal caregivers reported low utility values at baseline (0.199) and their perception of utility value showed a higher increase over time (0.817)⁵⁷.

Table 7. Mean (SD) utility values in studies comparing patient and caregiver proxy assessments of patient’s health status via indirect methods of elicitation.

Reference (year)	Health state	Time point	Method	Patients	Caregivers
Pickard et al.^41 (2004)	Ischemic stroke	Baseline	EQ-5D	0.31 (0.38)	0.24 (0.39)
			HUI-3	0.21 (0.30)	0.21 (0.36)
		1 month	EQ-5D	0.55 (0.36)	0.51 (0.36)
			HUI-3	0.42 (0.36)	0.36 (0.35)
		3 months	EQ-5D	0.61 (0.30)	0.56 (0.33)
			HUI-3	0.45 (0.34)	0.42 (0.36)
		6 months	EQ-5D	0.62 (0.34)	0.59 (0.35)
			HUI-3	0.44 (0.37)	0.44 (0.39)
Franceschini et al.^55 (2003)	Stroke	NR	EQ-5D	57.4 (22.1)	54.3 (22.2)
Hung et al.^56 (2007)	Stroke	NR	HUI-3	+0.008^a

Abbreviations. EQ-5D, EuroQol–Five Dimensions; HUI, Health Utilities Index; NR, not reported; SD, standard deviation.

^a Mean difference.

Discussion

When examining studies that conducted direct comparisons between CV utilities, we observed a wide variation across utility scores. This was mostly due to differences in the populations or health states being measured. In particular, variables affecting the range of values observed included disease severity, as well as time since disease onset or study enrollment for MI, angina, and hemorrhagic stroke. Another contributor to the variation in utilities was methodological differences that various studies used, including differences in type of direct or indirect method of elicitation, and tariffs (for indirect methods only). This review builds on previously published works, such as the systematic review by Smith et al.⁵⁸, to show the contribution of methodological differences to utility scores in CVD. It is critical to understand the differences in utility methodologies, because the utility score directly influences the calculated cost-effectiveness of new therapies in cost-utility models, which drive payer decision-making.

Vignette health states

Vignette health states allow for comparison across CV conditions, with values elicited from the same groups of respondents. For example, Matza et al.^16,59 elicited utility values for acute and chronic stroke (0.33 and 0.52, respectively), HF (0.57 and 0.60), and ACS (0.67 and 0.82) vignettes from a group of general population respondents in the UK, and found that values were the lowest for stroke, followed by HF and ACS—trends that were consistent with the published values among patients. There is some evidence that estimates derived from general population respondents may be somewhat lower than those from patients. Since all of these studies evaluated only stroke, it is unclear whether or not these trends can be generalized to other measures or other CV conditions.

Indirect utility measures

The EQ-5D was the most common method used in studies we surveyed, and resulted in a large range of values. In head-to-head studies, the EQ-5D yielded higher values than other indirect methods, including the SF-6D and HUI-3; but, upon comparison, EQ-5D and HUI-2 scores were found to be similar. This is likely due to the similarity between domains of the EQ-5D and HUI-2, such as mobility, self-care, emotion, and pain, although the HUI-2 also includes cognition and fertility. In contrast, the other indirect methods include several additional domains such as dexterity, vision, hearing, and speech in the HUI-3 and social functioning, vitality, and role participation in the SF-6D. However, it is unclear whether these domains explain the higher sensitivity of the EQ-5D over the other indirect utility methods in capturing changes over time.

It is also possible that the scales of utility values may impact mean scores. While all utility values are anchored on a scale of 0–1, where 0 represents a health state equal to death and 1 represents a health state equal to full health, some scores allow for negative values representing health states worse than death. For example, the lower bounds of the HUI-2 and HUI-3 are –0.03 and –0.36, respectively^12,13; the EQ-5D may have varying lower bounds depending on the tariff used. Publications evaluating studies with low lower bounds (including patients with utility values worse than death) could have lower mean scores than studies using methods with higher lower bounds (or those that do not accommodate negative utility values).

These trends for the EQ-5D were often modified by the tariff used, as the UK tariff consistently resulted in lower scores than the US and other tariffs. Notably, the SF-6D yielded a very narrow range of values, indicating a low level of sensitivity to differences among CVD health states, particularly for milder states, in both ceiling and floor effects. In contrast, the direct methods (SG and TTO) often yielded higher scores than the indirect methods, particularly when compared with EQ-5D index scores calculated using the UK tariff. The lower scores typically seen with the SF-6D may reflect a reduced ability of this measure to capture better health states. For example, in one study, SF-6D scores for a history of stroke were similar to EQ-5D scores (0.66 vs 0.69), but were significantly lower for a history of acute MI (0.66 vs 0.78, p < .001)³⁴. Additionally, in a study of stroke patients in the Netherlands, SF-6D and EQ-5D values were the same (0.69) at 2 months after stroke, but only the EQ-5D captured an improvement at 1 year (0.72)³³.

Similar trends have been observed in published studies evaluating other diseases. Peasgood and Brazier⁶⁰ conducted a review of meta-analyses of utility values, and found substantial differences across the direct and indirect methods used. In fact, they proposed that, because the heterogeneity among utility methods was so great, meta-analysis techniques may not be appropriate. Four of the included meta-analyses compared direct methods to the EQ-5D and evaluated chronic liver disease, chronic kidney disease, hip fracture, or diabetes. All of these studies reported that direct methods yielded statistically significantly higher values (0.068–0.36) than the EQ-5D scores^61–64. However, there are differences among direct methods as well; a fifth study we surveyed reported significantly lower SG scores (–0.13) compared with TTO scores for colorectal cancer⁶⁵.

Indirect utility measures also vary in sensitivity, which may further hinder comparisons of utility values across measures. For example, the EQ-5D index score has been shown to have a ceiling effect (i.e. lower discriminatory power at the highest levels of quality-of-life), and the SF-6D has been shown to have a floor effect (i.e. lower discriminatory power at the lowest levels of quality of life)^66–68.

Our findings also suggest a potential ceiling effect, in addition to the documented floor effect with the SF-6D. However, the EQ-5D scale has been reported to be more sensitive than the SF-6D scale in monitoring values for health-related quality-of-life, particularly at the lower end of the scale for patients with chronic obstructive airway disease, osteoarthritis, irritable bowel syndrome, lower back pain, leg ulcers, and for post-menopausal women and healthy elderly individuals (aged 75+ years)⁶⁶; this trend was also observed in the current SLR in relation to CVD. In addition, the HUI focuses on physical and emotional health and does not include questions on social functioning or satisfaction⁶⁹.

Tariffs

A number of published studies in other disease areas also showed that the UK tariff produced lower EQ-5D index scores than tariffs used in other countries. For example, a UK tariff yielded lower EQ-5D scores than US or Danish tariffs in a Swedish rheumatoid arthritis population^70–72 and US tariffs in a Thai diabetes population⁷¹. The differences we observed between the SF-6D and EQ-5D may have been driven by the tariff algorithm used to calculate the EQ-5D index utility score. In a meta-regression of utility data from HF patients, lower utility values were observed for the SF-6D and EQ-5D with a UK tariff (0.6382 and 0.6385, respectively) compared to the EQ-5D with a US tariff (0.7593)³¹. However, the lack of tariff reporting made it difficult to see whether this trend was consistent across the studies.

Selection of utilities

Because of the variation reported here and in other studies, it may be important to select utility values for economic models that precisely represent the health states meant to be evaluated with respect to such factors as clinical diagnosis, comorbidities, interventions received, age, geography, and duration of disease. Furthermore, it would be prudent to select a utility method that is appropriate for both the health state being measured and the intended use of the utility value. Additionally, indirect methods should evaluate domains (e.g. pain, or physical, social, and emotional functioning) that are relevant to the health attributes of the condition being evaluated. Furthermore, the tariff that is used to weight the value of each of those domains should be compatible with the population being evaluated. Lastly, because of the variation seen across utility methods, cost-utility models using utility values derived from the published literature should employ sensitivity analyses using alternate utility values, derived from other appropriate methodologies and populations.

Study limitations

The under-reporting of relevant study and population characteristics is a substantial limitation to selecting utility values for economic models from the published literature. In this review, some studies provided few details regarding patients’ medical conditions (e.g. stable vs unstable angina, ischemic vs hemorrhagic stroke), demographics, and comorbidities. This limits the assessment of the effect of variations in the patient populations. Additionally, less than half of the studies in the SLR reported how much time had passed between the onset of the CV condition and the utility assessment, making it difficult to determine whether the utility value reflected the acute or chronic impact of the CV condition described. This is particularly relevant to MI. Since many of the respondents contributing to an MI utility value may have had an MI several years prior to completing an assessment, the resulting utility is likely to under-estimate the impact of an acute MI on an individual’s quality-of-life.

Our study included publications between September 1992 and August 2015. While our study is comprehensive, future studies examining data through 2017 and beyond will be important for confirming our findings. In addition, different methodological factors (not captured here) may have contributed to different utility values. For example, differences between utility values in two articles could have been a result of both the direct methods technique and the type of respondent the study surveyed. We were unable to disentangle this effect in our review, and further empirical research is needed to better understand this effect. Lastly, while we searched the gray literature and congress abstracts in addition to MEDLINE and Embase, other large electronic databases that we did not target may include studies that we did not capture in our survey. Future studies surveying these sources may provide additional data to help us understand the variation in utility values.

Conclusions

This review found that health utility values for stroke, HF, MI, and angina are readily available in the published literature; however, such data should only be used after considering the limitations of each individual study, as well as the heterogeneity among studies. Because of the variation in utility values across different utility-elicitation methodologies, utility values for different health states or timeframes for a cost-effectiveness model should be taken from publications based on the same study, or at least studies using the same methodologies. Care should be taken when choosing utility values to use in a particular model, and, wherever possible, they should match the target population’s CV condition, geographical location, and time since stroke or MI. However, because published utility studies often lack these details, it may not be possible to determine whether a particular published value is appropriate for a target population. Ultimately, in such circumstances, it may be necessary to conduct de novo utility studies to obtain values relevant to the specifications of health states in economic models and the reimbursement setting. Regardless of whether values are selected from the literature or generated de novo, consideration of the utility-elicitation methodology is important, as our work here demonstrates that utility methodology impacts utility values in CVD.

Transparency

Declaration of funding

This study was funded by Amgen Inc.

Declaration of financial/other relationships

MB is an employee of Evidera. AS was an employee of Evidera at the time of this study and is currently a consultant for Pacira Pharmaceuticals and Takeda Pharmaceuticals. PPT is a consultant to Akcea, Amarin, Amgen, Kowa, Merck, Regeneron-Sanofi, and Gemphire and on the speakers bureaus for Amarin, Amgen, Kowa, Merck, Novo Nordisk, and Regeneron-Sanofi. SRG is an employee of Amgen Inc. and owns Amgen stock/stock options. L-IC was an employee of Amgen Inc. and holds Amgen stock/stock options. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

The authors acknowledge Annalise M. Nawrocki, PhD, of Amgen Inc. for medical writing and editorial assistance.