EQ-5D Based Utility Values for Adults with Chronic Obstructive Pulmonary Disease: A Systematic Review, Meta-Analysis, and Meta-Regression

Abstract

Chronic obstructive pulmonary disease (COPD) is a common lung disease that negatively affects health-related quality of life (QoL). Utility values, which measure QoL by weighting health states with societal preferences, are required for the cost-utility models that drive economic evaluations and policy decisions. Moayeri et al. published a systematic review and meta-analysis of utilities (EQ-5D) in COPD in June 2016. The current study investigated changes in mean utilities in more recent studies thereafter, exploring heterogeneity in utilities across diverse clinical and study characteristics. Systematic searches of databases, such as MEDLINE and Embase were undertaken from 1 July 2015 until 20 May 2024. A random-effects meta-analysis of utilities (EQ-5D) was performed which addressed inter-study heterogeneity and subgroup analyses. The pooled general mean (95% CI) utility value was 0.761 (0.726–0.795) from 43 studies, whereas Moayeri et al. reported 0.673 (0.653–0.693) from 32 studies. This improvement in mean utilities could be due to increased awareness, early detection, and better medical interventions over the past decade, but demonstrates that a general utility value should be approached with caution given significant heterogeneity. Four meta-regressions were performed on each subgroup: region, method of elicitation, reported comorbidities, and disease stage; of which, method of elicitation, disease stage, and region were found to be significant moderators of utilities. It is, therefore, important to use meta-analysed utilities for cost-utility analyses that reflect the context and patient population of the model. Moreover, these results provide additional evidence for the precision and sensitivity of EQ-5D-5L over EQ-5D-3L.

Keywords:

• Chronic obstructive pulmonary disease
• quality of life
• EQ-5D
• utility
• heterogeneity
• meta-analysis

Introduction

Chronic obstructive pulmonary disease (COPD) is a common lung disease that causes airflow obstruction and breathing difficulties [1]. Long-term exposure to air pollution has been identified as a risk factor for COPD [2], along with tobacco smoking, occupational exposure, asthma and genetics [3]. The prevalence and impact of COPD have been associated with poor socioeconomic status in the UK and in many other countries [4]. The progressive course of the disease negatively affects a sufferer’s health-related quality of life (QoL) [5], usually referred to as utility values which are measured on a scale between 0 (death) and 1 (perfect health) that represent the strength of preference weights for different health outcomes.

Cost-utility analyses are used by the National Institute for Health and Care Excellence (NICE) to evaluate the cost-effectiveness of interventions and make evidence-based recommendations [6]. Utilities are used as a weighting for health outcomes in cost-utility analyses which use quality-adjusted life-years (QALYs), a composite measure of quality and length of life. However, challenges may arise when utilities are not available, lack sufficient quality, or demonstrate heterogeneity. This challenge is particularly relevant in the context of COPD, a group of lung conditions that lead to breathing difficulties, such as emphysema, chronic bronchitis, and chronic obstructive airway disease. Identifying the most appropriate utility values for determining the cost-effectiveness of interventions for COPD-related issues can be problematic.

Estimation of utilities is a difficult area of research. Both generic (e.g. EQ-5D) and disease-specific QoL measurements (e.g. St. George Respiratory Questionnaire) are used to estimate utilities for patients with COPD. However, utilities produced by generic measures are preferred in cost-utility analyses as they allow comparison across diseases [7], and when they are not available, some disease-specific measures can be mapped onto utilities produced by generic instruments [8]. NICE prefers using utilities based on the generic EQ-5D instrument in health economic analyses to provide consistency across guidance and appraisals [9, 10]. The EQ-5D has two versions: EQ-5D-3L and EQ-5D-5L. Respondents assess their impairment in five health domains: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. The 3L version has three response levels per domain, while the 5L version offers five levels for increased sensitivity. The NICE methods manual currently favours EQ-5D-3L until a quality 5L value set is available [11].

While there is a substantial body of literature measuring utility values of people with COPD, there is significant variability in utility values across studies. This issue of utility variability within COPD has been previously acknowledged [7] and poses a challenge to ensuring consistency in cost-utility analyses. When a large number of utility values are available, a systematic approach is needed to identify the most relevant value considering factors, such as the method, robustness, and publication date of the evidence, as well as the specific research question and population of interest. It is important for researchers conducting cost-utility analyses to base their choice of utilities on the most reliable and up-to-date sources. When possible, researchers should strive to include all available high-quality evidence that is relevant to the specific research question and population of interest in their cost-utility models [12].

Moayeri et al. conducted a systematic literature review and meta-analysis of utilities (EQ-5D) in the area of COPD and found significant heterogeneity between studies (I² = 97.7%) [7]. Building on this, the primary aim of the current study was to investigate potential changes in mean utility values in more recent studies. It also sought to explore potential variations in utility values across diverse clinical and study characteristics with a focus on heterogeneity and variability. Additionally, the study aimed to pool utility estimates from homogeneous studies and conduct a meta-regression to explore whether any differences between studies can be explained by specific pre-defined moderators, which were based on factors known to influence health utility values, such as disease progression, country setting and choice of instrument [13].

Methods

Study design

This review complies with the PRISMA checklist [14]. A review protocol was produced (available from the corresponding author on request) and set out the process to provide an updated version of the systematic review and meta-analysis conducted by Moayeri et al. [7]. As such, the search strategy and inclusion and exclusion criteria from Moayeri et al. [7] were replicated in this study.

Search strategy

A systematic search adapted from the databases used by Moayeri et al. [7] was conducted on 20 May 2024. Those databases included MEDLINE, Embase, Web of Science, CINAHL, ProQuest (which includes PsycINFO and other 61 databases), the Cochrane Library Database (which includes NHS Economic Evaluation Database, Health Technology Assessment Database, Cochrane Database of Systematic Reviews and other three databases), International Society for Pharmacoeconomics and Outcomes Research (ISPOR) and Google Scholar, as well as dissertation and web sites of key academic institutions, such as NICE (National Institute of Health and Care Excellence), CCOHTA (Canadian Cooperating Office for Health Technology Assessment), SBU (The Swedish Council on Technology Assessment in Health Care), Health Economic Evaluations Database (HEED, ceased publishing in 2014) and the Cost Effectiveness Analysis Registry at Tufts-New England Medical Centre. Adapted databases (included in Figure 1) were searched from 1 July 2015 until 20 May 2024. The search terms included a combination of MeSH and free text terms based on Moayeri et al. [7].

Inclusion and exclusion criteria

Studies were included only if they used EQ-5D to estimate utility values for patient level research in COPD.

In addition, studies were identified as being eligible for inclusion if they were:

• studies on health utility that were published from July 2015 until 20 May 2024;

• studies in which their sample population was specifically categorised as COPD as defined by standard criteria for COPD diagnosis and spirometric confirmation (should clearly be addressed in methodology of included studies);

• English language studies and non-English language studies with English abstracts;

• abstracts (e.g., seminar abstracts) and reports if adequate data for analysis were provided;

• studies with more than 10 participants.

Studies were excluded according to the following criteria:

• simulation-based studies

• editorials/opinion pieces, letters, systematic reviews and meta-analyses;

• studies that reported utilities from proxies, not individual participant data (e.g., reported by family member or a health professional);

• studies that obtained utility estimates from the literature, if there was not enough information on the derivation of utility;

• studies that did not distinguish COPD from other types of obstructive pulmonary disease, such as asthma or cystic fibrosis;

• studies using utility values mapped from other reported QoL studies;

• studies that reported utility values from non-stable and exacerbation state COPD patients.

In addition, studies were examined in which results were measured from populations duplicated in other studies and excluded those studies that were less comprehensive and less recent than other studies measuring the same population.

Study selection and data extraction

Two authors (NS: 100% and YZ: 100%) reviewed titles and abstracts of the identified studies. Where it was unclear whether a study should be included it was carried forward to the next stage. During this stage, NS and YZ independently reviewed the full text of the relevant studies, with a 100% review rate each. Studies that met the predefined inclusion criteria were then carried forward to the data extraction process. Any disagreements about study eligibility were discussed with another author (SM) and resolved by consensus.

Data extraction from all included studies was performed independently by NS (100%) and YZ (100%), and any disagreements were resolved by consensus among NS, YZ, and SM. Data were extracted in Microsoft Excel, and the following data were recorded: principal author, year of publication, number of patients, country of origin, clinical characteristics and demographic of patients, age (mean and standard deviation), sex, comorbidities, study design, data collection method, health state utility value measure and utility estimate (mean and standard deviation, or where unavailable upper and lower confidence intervals), FEV1/VC (mean and standard deviation) and FEV% predicted (mean and standard deviation). The method adopted by Moayeri et al. [7] was replicated in this study where baseline characteristics were used to avoid any potential influence of the intervention on utility estimates.

Data synthesis

Extracted data was grouped according to [1]: disease stage [2], population (including age, sex, and region) [3], comorbidities, and [4] method of utility elicitation (i.e. EQ-5D-5L vs. EQ-5D-3L). A first look at the data indicated that disease stage, country, comorbidities, and method of elicitation may be sources of heterogeneity. Given that the country category yielded subgroups with too few data points for a meaningful analysis, the continent was used as the region category. Separate meta-analyses were conducted thereafter, one for each subgroup.

Point estimates and 95% confidence intervals (CIs) for utility values were calculated and displayed in forest plots. Moayeri et al.’s [7] restriction of their meta-analysis to EQ-5D to avoid heterogeneity was replicated in this study; and additional heterogeneity introduced by elicitation methodology was explored in subgroup analyses of EQ-5D-5L and EQ-5D-3L.

Meta-analyses of subgroups were conducted with the meta package in R v4.3.0. To account for the expected between-study heterogeneity, a random-effects model was used to estimate the distribution of true mean utility values. The I² statistic = 100% × (Q − df)/Q and 95% CI was used to indicate a relative measure of between-study variation as well as sampling error. Aligning with Moayeri et al. [7], I² values of 30–60, 50–90, and 75–100% were classified as moderate, substantial, and considerable heterogeneity, respectively. If standard deviations of utility values were not reported, they were calculated from 95% confidence intervals. Where studies did not report enough data for this calculation, they were excluded.

The t-test and analysis of variance (ANOVA) were applied to compare estimated utility means between subgroups. Moayeri et al. [7] did not perform any meta-regression, possibly because their meta-analysis did not possess a sufficient sample size following the recommendation to include at least ten studies per covariate [15]. However, given that for this study, the subgroups region, disease stage, method of elicitation, and comorbidities each had sufficient data points, we determined that conducting a meta-regression for each subgroup would be a valuable tool to evaluate the factors contributing to variability among utility values. A model was constructed for each subgroup separately such that the region model assessed the impact of regional differences on the outcome; the disease stage model assessed the impact of disease stage on the outcome; and so on. The meta-regression was used to interpret the impact of the inclusion of covariates on residual heterogeneity (defined as the variance in mean utility value) for each subgroup. These analyses were performed with the rma function of the metafor package in R v4.3.0.

Results

Study characteristics

A total of 5912 records were retrieved through database searching, of which 3044 were included after duplicates were removed. After title and abstract screening, 2945 records were removed and 99 remained to be screened on full text, after which 34 records were excluded.

Figure 1. PRISMA flow chart of the study identification and inclusion process.

Sixty-five studies reported EQ-5D based utility values for adults with COPD, either in stages or as an overarching condition, and were included in the review. However, 23 of these studies were excluded from the meta-analysis because they did not report mean utility values (and further did not provide enough information to infer mean utility values), did not sample from more than ten participants, or treated COPD as a comorbidity rather than the primary condition. A further two Korean studies (and all their observations) were excluded given that utility values were extracted from overlapping datasets (the Korean National Health and Nutrition Examination Survey) which would otherwise have skewed the meta-analysis. Some studies reported multiple data points e.g. studies that included utility values for different levels of COPD severity. Thus, a total of forty-two studies were included which yielded 104 data points that were used in the meta-analysis.

Moayeri et al. [7] excluded studies that were deemed to have extreme utility values (>0.96); and therefore, the same approach was taken in the current study. However, no studies were found with extreme utility values, which were not previously excluded for other reasons.

Study characteristics are detailed in Table 1. Age and sex data were reported for 37 and 39 studies, respectively. Of these study participants, 64.94% were male and the average age per study was 66.6 years. Of the studies that reported FEV1% predicted, the study average was 54.5%, which indicated moderate airflow obstruction according to the Global Initiative for Chronic Lung Disease (GOLD) [16].

Table 1. Study characteristics.

Author, year	Country	Clinical characteristics	Age: mean (SD)	Male%	Study design	QoL measurement	Data collection method	FEV% predicted (mean)
Ahn et al. 2020 [17]	Korea	Pulmonology outpatient department	69.8 (8)	93.5	Prospective cohort study	EQ-5D	Interview	Not reported
Alcazar-Navarrete et al. 2020 [18]	Spain	Controlled stable COPD	64.9 (7.7)	88	Multicentre, prospective observational study	EQ-5D-5L	Not clear	51.6
Altawalbeh et al. 2022 [19]	Jordan	Stable COPD and emphysema	Not reported	85.7	Cross-sectional multicentre cohort study	EQ-5D-5L	Questionnaire	Not reported
Bae et al. 2020 [20]	Korea	Stable COPD	65.5	85	Multicentre, non-interventional, cross-sectional study	EQ-5D-3L	Survey	25.4
Baltasar-Fernandez et al. 2023 [21]	Spain	Clinically stable COPD	79 (6.4)	Not reported	RCT	EQ-5D-5L	Questionnaire	58.7
Boland et al. 2016 [22]	Netherlands	Stable COPD in primary care	68 (8.7)	45.9	Longitudinal patient-level data from the RECODE study, the largest 2-year cluster randomised trial	EQ-5D-3L	Questionnaire	65.1
Boven et al. 2016 [23]	Netherlands	Primary care	68.8 (10.1)	52.2	Real-world study with a pre-test/post-test design	EQ-5D-3L	Questionnaire	Not reported
Burns et al. 2016 [24]	UK	Stable COPD patients who completed pulmonary rehabilitation program	69.3 (8)	66.7	Cost-effectiveness study alongside RCT	EQ-5D	RCT	Not reported
Chen et al. 2016 [25]	Taiwan	Patients invited to be interviewed	69 (8.2)	93	Cross-sectional survey	EQ-5D	Interview	23.99
Choi et al. 2018 [26]	Korea	Patients visiting Chonbuk National University Hospital	70.95 (8.5)	87.2	Validation of K-LCADL	EQ-5D (Korean version)	Interview	57.12
Ding et al. 2017 [27]	France, Germany, Italy, Spain, the UK, China and the USA	Stable COPD (including patients with emphysema and chronic bronchitis and concomitant asthma	64.6 (8.7)	65.1	Cross-sectional study	EQ-5D-3L	Self-completion record	62.9
Fishwick et al. 2014 [28]	England	Participants with stable COPD in epidemiological study	69.4 (8.4)	62.2	Population-based study of respiratory disease	EQ-5D	Questionnaire	Not reported
Garcia-Gordillo et al. 2017 [29]	Spain	Stable COPD	Not reported	48.67	Cross-sectional study	EQ-5D-5L	Spanish National Health Survey	Not reported
Hansen et al. 2021 [30]	Denmark	Stable COPD outpatients	66.6	56	Inter-day test–retest reproducibility study	EQ-5D-3L	Questionnaire	32
Hoogendoorn et al. 2021 [31]	UK	Recruited from general practices	71 (8.5)	58	Cohort study	EQ-5D-5L	Questionnaire	81
Hoyle et al. 2016 [32]	UK	Stable COPD RCT participants	63.8	68.8	Mapping study	EQ-5D-3L	Questionnaire	Not reported
Husain et al. 2021 [33]	Pakistan	Stable COPD	43.28	24.9	Cross-sectional study	EQ-5D-3L	Questionnaire	Not reported
Jiménez-Reguera et al. 2020 [34]	Spain	Stable COPD	68.1 (12.1)	59.1	RCT	EQ-5D-3L	Questionnaire	43.1
Kokturk et al. 2018 [35]	Turkey, Saudi Arabia	COPD	Not reported	Not reported	Observational, cross-sectional, multicentre study	EQ-5D-3L	Questionnaire	Not reported
Koo et al. 2017 [36]	Korea	Stable COPD	64.7	73.4	Cross-sectional study	EQ-5D	Korean National Health and Nutrition Examination Survey	77.3
Kwon and Kim 2016 [37]	Korea	Stable COPD	60.37 (9.9)	72.36	Cross-sectional survey	EQ-5D	Korean National Health and Nutrition Examination Survey 2007–2012.	Not reported
Kwon 2018 [38]	Korea	Stable COPD	Not reported	Not reported	Cross-sectional study	EQ-5D-3L	Korea National Health and Nutrition Examination Survey	Not reported
Lee and Kwon 2017 [39]	Korea	Stable COPD	63.51 (7.8)	83.6	Prevalence study	EQ-5D (Korean version)	Interview	76.53
Lee et al. 2022 [40]	Korea	Men with stable COPD	66.55 (10.96)	100	Cross-sectional study	EQ-5D-3L (Korean version)	Questionnaire	66.03
Lim et al. 2019 [41]	Korea	Stable COPD patients from tertiary teaching hospitals	69.16 (0.34)	86.3	Cross-sectional study	EQ-5D-3L	Questionnaire	55.85
Merino et al. 2019 [42]	Spain	Stable COPD	71.8 (8.4)	76.2	Cross-sectional observational study	EQ-5D-5L	Patient visit	66.7
Miravitlles et al. 2015 [43]	Spain	Patients from primary care (63%) and from pneumology departments (37%)	66.6 (9.18)	85.4	Transversal observational study	EQ-5D	Questionnaire	54
Nguyen et al. 2019 [44]	Vietnam	Stable COPD	Not reported	Not reported	Pre- and post-intervention study design	EQ-5D-5L (Vietnam version)	Questionnaire	Not reported
Nolan et al. 2016 [45]	UK	Stable COPD outpatients	70.4	59.7	Cross-sectional cohort study	EQ-5D-5L	Questionnaire	46.1
Paly et al. 2019 [46]	Global	Participants in TORCH study	64.4 (8.8)	79	Economic cost and utility comparison	EQ-5D	–	Not reported
Richeldi et al. 2024 [47]	Italy	Stable COPD	71.2 (9)	68.2	Prospective observational study	EQ-5D-5L	Questionnaire
Rivera et al. 2016 [48]	Spain	Stable COPD from the outpatient clinic	66.9 (9.7)	93	Multicentre, observational and cross-sectional study	EQ-5D	Questionnaire	44.4
Roncero et al. 2016 [49]	Spain	Stable COPD patients in respiratory medicine and primary care centres	67.7	81.6	Post hoc analysis of a cross-sectional, observational study	EQ-5D	Questionnaire	55.2
Sohanpal et al. 2024 [50]	UK	Stable COPD	68 (10.2)	53.7	RCT	EQ-5D-5L	Interview	Not reported
Sundh et al. 2015 [51]	Sweden	Secondary care respiratory units	70.5	0	Cross-sectional study	Swedish version EQ-5D	Questionnaire	34.5
Talboom-Kamp et al. 2017 [52]	Netherlands	Stable COPD in primary care	66.6	52.1	Implementation study	EQ-5D	Automatically extracted from log files	Not reported
Thuppal. et al. 2019 [53]	USA	Patients undergoing surgery for severe stable COPD	66	57	Utility elicitation study comparing SF-6D and EQ-5D-3L	EQ-5D-3L	Questionnaire	26.7
Verberkt et al. 2021 [54]	Netherlands	Stable COPD patients experiencing moderate to very severe impairment due to breathlessness	65.6	55	Cost-effectiveness study alongside RCT	EQ-5D-5L	Questionnaire	34.5
Wacker et al. 2016 (55)	Germany	Stable COPD	65.1 (8)	60.9	Cross-sectional study	EQ-5D-3L	Questionnaire	52.5
Wacker et al. 2016b [56]	Germany	Stable COPD	65.1 (8.4)	61.1	Cross-sectional study	EQ-5D-3L	Questionnaire	Not reported
Wu et al. 2015 [57]	China	Stable COPD patients from community health centres or hospitals	70.4 (11.73)	72.9	Cross-sectional study	EQ-5D	Questionnaire	54.5
Xiao et al. 2020 [58]	China	Mild stable COPD in urban communities	61.5 (7.7)	48.7	Cross-sectional study	EQ-5D	Questionnaire	Not reported
Zanforlini et al. 2021 [59]	Italy	Stable COPD	72.6	77.6	RCT	EQ-5D	Questionnaire	Not reported

RCT: random controlled trial; K-LCADL: Korean London Chest Activity of Daily Living Scale.

Table 2 details the estimated mean utility value per study data point. Most data points were collected in Asia (n = 54) and the most common method of elicitation was EQ-5D-3L (n = 79). There was an almost equal spread across reporting of comorbidities (52 reported, whereas 51 did not). For data points associated with the disease stage, the largest category was mild, and the smallest category was very severe.

Table 2. Mean utility values per study data point.^a

Author, year	Number of participants	Country	Disease stage	Reported comorbidity	QoL measurement instrument	Utility value, mean (SD)
Ahn et al. 2020 [17]	261	Korea	Overall	No	EQ-5D	0.86 (0.11)
Alcazar-Navarrete et al. 2020 [18]	50	Spain	Overall	No	EQ-5D-5L	0.82 (0.1)
	43	Spain	Overall	No	EQ-5D-5L	0.74 (0.23)
Altawalbeh et al. 2022 [19]	127	Jordan	Overall	Yes	EQ-5D-5L	0.69 (0.33)
	76	Jordan	Overall	No	EQ-5D-5L	0.66 (0.39)
	25	Jordan	Overall	Yes	EQ-5D-5L	0.67 (0.34)
	177	Jordan	Overall	No	EQ-5D-5L	0.68 (0.36)
	192	Jordan	Overall	No	EQ-5D-5L	0.68 (0.36)
	11	Jordan	Overall	Yes	EQ-5D-5L	0.69 (0.23)
Bae et al. 2020 [20]	298	Korea	Overall	Yes	EQ-5D-3L	0.83 (0.15)
	32	Korea	Mild	Yes	EQ-5D-3L	0.91 (0.11)
	156	Korea	Moderate	Yes	EQ-5D-3L	0.86 (0.12)
	90	Korea	Severe	Yes	EQ-5D-3L	0.79 (0.16)
	20	Korea	Very severe	Yes	EQ-5D-3L	0.66 (0.17)
Baltasar-Fernandez et al. 2023 [21]	13	Spain	Overall	No	EQ-5D-5L	0.9 (0.12)
Boland et al. 2016 [22]	171	Netherlands	Overall	No	EQ-5D-3L	0.74 (0.26)
	294	Netherlands	Overall	No	EQ-5D-3L	0.74 (0.27)
Boven et al. 2016 [23]	88	Netherlands	Overall	Yes	EQ-5D-3L	0.74 (0.25)
Burns et al. 2016 [24]	75	UK	Overall	No	EQ-5D	0.7 (0.2)
Chen et al. 2016 [25]	142	Taiwan	Overall	Yes	EQ-5D	0.84 (0.21)
	40	Taiwan	Mild	Yes	EQ-5D	0.88 (0.2)
	53	Taiwan	Moderate	Yes	EQ-5D	0.89 (0.16)
	30	Taiwan	Severe	Yes	EQ-5D	0.79 (0.2)
	19	Taiwan	Very severe	Yes	EQ-5D	0.65 (0.32)
Choi et al. 2018 [26]	94	Korea	Overall	No	EQ-5D (Korean version)	0.58 (0.16)
Ding et al. 2017 [27]	2419	France, Germany, Italy, Spain, UK, China, and USA	Overall	No	EQ-5D-3L	0.7 (0.3)
Fishwick et al. 2014 [28]	148	England	Overall	No	EQ-5D	0.52 (0.32)
Garcia-Gordillo et al. 2017 [29]	1130	Spain	Overall	No	EQ-5D-5L	0.74 (0.31)
Hansen et al. 2021 [30]	50	Denmark	Overall	No	EQ-5D-3L	0.66 (0.17)
Hoogendoorn et al. 2021 [31]	1008	UK	Overall	Yes	EQ-5D-5L	0.7 (0.29)
Hoyle et al. 2016 [32]	1138	UK	Not clear	No	EQ-5D-3L	0.91 (0.13)
	2093	UK	Not clear	No	EQ-5D-3L	0.8 (0.17)
	1080	UK	Not clear	No	EQ-5D-3L	0.67 (0.2)
	135	UK	Not clear	No	EQ-5D-3L	0.43 (0.31)
Husain et al. 2021 [33]	293	Pakistan	Overall	Yes	EQ-5D-3L	0.61 (0.22)
Jiménez-Reguera et al. 2020 [34]	36	Spain	Overall	No	EQ-5D-3L	0.5 (0.2)
Kokturk et al. 2018 [35]	405	Turkey, Saudi Arabia	Overall	Yes	EQ-5D-3L	0.59 (0.33)
	206	Saudi Arabia	Overall	Yes	EQ-5D-3L	0.52 (0.37)
	199	Turkey	Overall	Yes	EQ-5D-3L	0.66 (0.27)
	190	Turkey, Saudi Arabia	Overall	No	EQ-5D-3L	0.54 (0.35)
	196	Turkey, Saudi Arabia	Overall	No	EQ-5D-3L	0.64 (0.3)
Koo et al. 2017 [36]	1227	Korea	Overall	Yes	EQ-5D	0.92 (0.14)
Kwon and Kim 2016 [37]	2734	Korea	Overall	No	EQ-5D	0.92 (0)
	1129	Korea	Mild	No	EQ-5D	0.92 (0.01)
	1434	Korea	Moderate	No	EQ-5D	0.92 (0)
	156	Korea	Severe	No	EQ-5D	0.87 (0.02)
	16	Korea	Very Severe	No	EQ-5D	0.78 (0.06)
	2734	Korea	Overall	Yes	EQ-5D	0.94 (0)
Kwon 2018 [38]	424	Korea	Mild	No	EQ-5D-3L	0.93 (0.1)
	354	Korea	Moderate	No	EQ-5D-3L	0.93 (0.11)
	31	Korea	Severe	No	EQ-5D-3L	0.88 (0.2)
	475	Korea	Overall	Yes	EQ-5D-3L	0.91 (0.13)
	340	Korea	Overall	No	EQ-5D-3L	0.96 (0.87)
	815	Korea	Overall	Yes	EQ-5D	0.93 (0.11)
Lee and Kwon 2017 [39]	191	Korea	Overall	Yes	EQ-5D (Korean version) specific	0.84 (0.02)
	890	Korea	Overall	No	EQ-5D (Korean version)	0.92 (0.01)
Lee et al. 2022 [40]	93	Korea	Overall	Yes	EQ-5D (Korean version)	0.89 (0.17)
	58	Korea	Overall	Yes	EQ-5D (Korean version)	0.84 (0.21)
Lim et al. 2019 [41]	299	Korea	Overall	No	EQ-5D-3L	0.83 (0.15)
Merino et al. 2019 [42]	386	Spain	Overall	Yes	EQ-5D-5L	0.72 (0.31)
	41	Spain	Mild	No	EQ-5D-5L	0.74 (0.35)
	79	Spain	Moderate	No	EQ-5D-5L	0.77 (0.26)
	35	Spain	Severe	No	EQ-5D-5L	0.58 (0.31)
	5	Spain	Very Severe	No	EQ-5D-5L	0.48 (0.45)
Miravitlles et al. 2015 [43]	1864	Spain	Overall	Yes	EQ-5D	0.69 (0.29)
	346	Spain	Overall	Yes	EQ-5D-3L with weighted Spanish societal preferences	0.73 (0.29)
Nguyen et al. 2019 [44]	211	Vietnam	Overall	No	EQ-5D-5L Vietnam version	0.47 (0.2)
Nolan et al. 2016 [45]	616	UK	Overall	No	EQ-5D-5L	0.68 (0.24)
	324	UK	Overall	No	EQ-5D-5L	0.697 (0.22)
Paly et al. 2019 [46]	1832	Global	Overall	No	EQ-5D	0.76 (0.2)
Richeldi et al. 2024 [47]	655	Italy	Overall	Yes	EQ-5D-5L	0.7 (0.2)
Rivera et al. 2016 [48]	115	Spain	Overall	Yes	EQ-5D	0.72 (0.31)
Roncero et al. 2016 [49]	940	Spain	Overall	Yes	EQ-5D	0.6 (0.2)
Sohanpal et al. 2024 [50]	242	UK	Overall	Yes	EQ-5D-5L	0.552 (0.23)
	181	UK	Overall	Yes	EQ-5D-5L	0.55 (0.27)
Sundh et al. 2015 [51]	165	Sweden	Overall	Yes	EQ-5D (Swedish version)	0.71 (0.23)
	208	Sweden	Overall	Yes	EQ-5D (Swedish version)	0.6 (0.3)
Talboom-Kamp et al. 2017 [52]	215	Netherlands	Mild	No	EQ-5D	0.86 (0.82)
Thuppal et al. 2019 [53]	94	USA	Severe	Yes	EQ-5D-3L	0.66 (0.2)
Verberkt et al. 2021 [54]	50	Netherlands	Overall	No	EQ-5D-5L	0.6 (0.19)
	56	Netherlands	Overall	No	EQ-5D-5L	0.58 (0.22)
Wacker et al. 2016b [55]	2275	Germany	Overall	Yes	EQ-5D-3L	0.82 (0.21)
	204	Germany	Mild	Yes	EQ-5D-3L	0.85 (0.18)
	954	Germany	Moderate	Yes	EQ-5D-3L	0.84 (0.19)
	869	Germany	Severe	Yes	EQ-5D-3L	0.81 (0.21)
	248	Germany	Very Severe	Yes	EQ-5D-3L	0.74 (0.24)
Wacker et al. 2016 [56]	2291	Germany	Overall	Yes	EQ-5D-3L	0.82 (0.2)
	206	Germany	Mild	Yes	EQ-5D-3L	0.85 (0.18)
	962	Germany	Moderate	Yes	EQ-5D-3L	0.84 (0.19)
	874	Germany	Severe	Yes	EQ-5D-3L	0.81 (0.21)
	249	Germany	Very Severe	Yes	EQ-5D-3L	0.74 (0.24)
Wu et al. 2015 [57]	678	China	Overall	Yes	EQ-5D	0.73 (0.15)
	140	China	Mild	No	EQ-5D	0.79 (0.09)
	296	China	Moderate	No	EQ-5D	0.73 (0.16)
	181	China	Severe	No	EQ-5D	0.69 (0.16)
	61	China	Very Severe	No	EQ-5D	0.66 (0.15)
Xiao et al. 2020 [58]	275	China	Mild	Yes	EQ-5D	0.89 (0.08)
	134	China	Mild	No	EQ-5D	0.9 (0.07)
	141	China	Mild	No	EQ-5D	0.88 (0.1)
	275	China	Mild	No	EQ-5D	0.89 (0.08)
	275	China	Overall	Yes	EQ-5D	0.89 (0.09)
	275	China	Overall	Yes	EQ-5D	0.88 (0.09)
Zanforlini et al. 2021 [59]	49	Italy	Overall	Yes	EQ-5D	0.9 (0.54)

^a Studies that report multiple utility values have been reported here as multiple data points.

Meta-analyses

Meta-analyses were conducted for groups of observations that were agreed by the authors on a priori grounds. Specifically, based on existing literature [13], studies were grouped according to region, stage of COPD, elicitation method (EQ-5D-5L vs. EQ-5D-3L), and the presence of comorbidities. These groupings are referred to as moderators below. Table 3 shows the results of random-effects models for each subgroup, with k representing the number of data points.

Table 3. Estimated mean utility values by region, disease severity, method of elicitation, and reported comorbidities.

Subgroup	Moderator	Utility value (95% CI)	Heterogeneity chi-squared/Cochrane’s Q-test			Tau-squared	I^2 heterogeneity statistics (%)
			X^2	df	p-Value
Region	Asia (k = 54)	0.793 (0.758–0.828)	67,048.07	53	<0.001	0.0161	99.9
	Europe (k = 35)	0.739 (0.706–0.773)	1905.54	34	<0.001	0.0083	98.2
	UK (k = 11)	0.656 (0.565–0.747)	2496.82	10	<0.001	0.0181	99.6
Disease stage	Overall (k = 62)	0.730 (0.697–0.762)	74,769.63	61	<0.001	0.0157	99.9
	Mild (k = 13)	0.875 (0.846–0.903)	480.01	12	<0.001	0.0017	97.5
	Moderate (k = 8)	0.849 (0.792–0.907)	821.45	7	<0.001	0.0672	99.1
	Severe (k = 9)	0.769 (0.695–0.844)	529.83	8	<0.001	0.0082	98.5
	Very severe (k = 7)	0.712 (0.658–0.7663)	32.24	6	<0.001	0.0023	81.4
Method of elicitation	3-L (k = 79)	0.780 (0.753–0.808)	79,008.18	78	<0.001	0.0148	99.9
	5-L (k = 24)	0.676 (0.636–0.717)	590.19	23	<0.001	0.0084	96.9
Comorbidities	Comorbidities reported (k = 52)	0.768 (0.737–0.800)	19,070.97	51	<0.001	0.0123	99.7
	No comorbidities reported (k = 51)	0.746 (0.707–0.785)	14,188.99	50	<0.001	0.0182	99.9

Figure 2 below shows the mean utility value for each subgroup and mean utility value overall for each model: common (fixed) effect and random-effects. Both mean values are reported here, but the main analysis focuses on the random-effects model. This decision was made as a result of substantial heterogeneity (I² = 89%) between data points.

Figure 2. Summary forest plot of subgroups and moderators. I² represents between data point heterogeneity for all 99 data points.

The random-effects model yielded a mean utility value of 0.761 (95% CI: 0.726–0.795) for the current study which is significantly higher than the mean utility value of 0.673 (95% CI: 0.653–0.693) reported by Moayeri et al. [7]. Notably, the confidence intervals do not overlap, indicating a statistically significant difference between the two mean utility estimates.

Test of moderators

For the analysis of subgroup moderators, four meta-regressions with categorical moderators were used to test the significance of the subgroup moderators in each of the four subgroups.

For the method of elicitation (EQ-5D-3L vs. EQ-5D-5L), disease stage, and region subgroups, it was found that the moderators in each subgroup had a statistically significant impact on the effect sizes in the meta-analysis. In other words, whether the method of elicitation was EQ-5D-3L or EQ-5D-5L is associated with the differences in utility values across the studies in that subgroup. Similarly, whether the disease stage was mild, moderate, severe, or very severe, and whether the region was Asia, UK, or Europe, had a significant effect on utility values across the disease stage and region subgroups, respectively. There was not enough evidence to determine whether the reporting of comorbidities was associated with the differences in utility values for this subgroup.

The meta-regression analyses revealed that the method, disease stage, and region covariates significantly explained a portion of the heterogeneity among data points in each subgroup. Specifically, the residual heterogeneity (defined as the variance in mean utility value) decreased by 15.89% when accounting for the moderators in the disease stage model, by 11.23% in the method model, and by 10.97% in the region model, indicating that the included covariates in each model account for a significant portion of the variability in effect sizes for each subgroup.

Egger’s test was used to assess potential publication bias in the meta-analysis. There was low evidence of potential publication bias (p-value = 0.002); however, when between-study heterogeneity (tau-squared) is large, bias detection tests are not likely to be accurate [60]. As such, Egger’s test of study bias was undertaken again for subgroups where between-study heterogeneity (tau-squared) was smaller than that of all studies taken together. Table 4 below indicates that the hypothesis that there is no publication bias for each subgroup’s meta-analysis is unjustified.

Table 4. Egger’s tests for publication bias per sub-group.

Subgroup	Egger’s test p-value
Region	p < 0.001
Disease stage	p < 0.001
Method of elicitation	p < 0.001
Reported comorbidities	p < 0.001

Discussion

This study aimed to examine whether there have been changes in the mean utility values for COPD reported in the systematic review and meta-analysis conducted by Moayeri et al. [7]. Moayeri et al. included 32 papers spanning from the respective dates of inception before July 2015, whereas the current study included 43 papers spanning over almost nine years thereafter. This indicates a considerable increase in research around health-related QoL of patients with COPD. Moreover, this study found a significantly higher mean utility value than the study conducted by Moayeri et al. [7]. Several potential reasons for the difference observed in mean utility values in more recent studies may be worth noting, for instance, increased awareness and better adherence to treatment plans, early detection, advancements in medical treatments and interventions, thus improving the health-related QoL of patients with COPD [61].

In line with the findings from Moayeri et al. [7], this study also found significant heterogeneity in utility values among patients with COPD. To better understand this variability, the analysis included a set of predefined moderators covering the region, stage of COPD disease, method of measuring utility, and whether or not comorbidities are reported. These moderators were based on previous research on factors that might influence the magnitude of utility values [13]. Age was not included as a potential modifier as the studies included in the current analysis are restricted to the adult population.

Whether or not comorbidities were reported did not exhibit a significant impact on the meta-analysed mean utilities despite previous research suggesting that comorbidities affect health-related QoL, particularly in chronic conditions [62]. However, given that some studies did not provide data on comorbidities, it was not possible to determine whether they could be accurately categorised for an analysis based on actual comorbidities. Therefore, the scope of this meta-regression was limited to reported comorbidities rather than actual comorbidities. The finding that reported comorbidities were not a significant modifier of utility values suggests that consistency of comorbidity reporting across studies is important for future research on the impact of comorbidities on QoL.

The method of elicitation, region, and disease stage moderators had a significant impact on utility values. In this study, the meta-analysed mean utility value for EQ-5D-5L (0.676) was substantially lower than the mean utility value for EQ-5D-3L (0.780); and the uncertainty around the EQ-5D-5L estimate was smaller than that of EQ-5D-3L. This finding suggests that the increased sensitivity of EQ-5D-5L allows respondents to express health impacts that may not be adequately captured by the EQ-5D-3L questionnaire; and aligns with previous research on the comparative precision of EQ-5D-5L at both the individual and group level [63]. It may be informative to compare studies using EQ-5D-5L with COPD-specific QoL measurement instruments, such as the St. George Respiratory Questionnaire, to gain further insights.

Despite the broad grouping of studies into continental regions, and the heterogeneity therein given cultural and economic disparities, regional differences were still found to have a significant impact on health-related QoL. Studies conducted in the UK report a mean utility value of 0.656 (95% CI: 0.565 to 0.747), compared to studies based in Asia, which report a mean utility value of 0.793 (95% CI: 0.758 to 0.828). The non-overlapping 95% confidence intervals indicate significant differences between the regions. This finding is consistent with prior literature, which reports that even after adjusting for demographic and clinical characteristics, Asian patients with hepatocellular carcinoma had significantly higher QoL scores than those of European patients [64]. Though cultural values have been hypothesised to explain differences in utility values between geographic regions [65], variation remains largely unexplained though generally substantial.

As utility values are intended to measure patient experience of disease, it is not surprising that the disease stage moderator had a significant impact on the results. The meta-analysed mean utility values for disease stages were 0.875, 0.849, 0.769, and 0.712 for mild, moderate, severe, and very severe, respectively, consistent with the recognised pattern of decreasing utility as disease severity increases. Importantly, the confidence intervals for mild and severe stages do not overlap, indicating a significant distinction between these disease stages. This finding underscores the impact of disease progression on utility values, a point also highlighted in a recent review of utility values for metastatic breast cancer [13], which similarly identified disease progression as a key influencing factor.

The search strategy of this study was not restricted by disease stage or study type, allowing the capture of a wide range of studies. However, despite this broad approach, only five out of the 43 included studies (∼12%) were intervention studies, which predominantly focused on patients with moderate to severe disease. This disparity is likely due to the common practice of excluding patients with mild airflow limitation in intervention studies, which differs from the inclusion of these patients in other observational studies. This limitation is a crucial consideration for modellers who wish to utilise these results in economic evaluations of interventions, as it may impact the accuracy and generalisability of their findings.

In comparison with the findings from Moayeri et al. [7], this study has some notable clinical and research implications. The current study reveals that the method of elicitation, disease stage, and region each had a significant effect on utility values. Notably, the EQ-5D-5L value had a lower point estimate and less uncertainty compared with EQ-5D-3L indicating improved sensitivity; more severe disease stages were associated with lower utility values; and Asian-based studies reported higher utility values than those with European and UK participants. Reporting of co-morbidities on the other hand, did not have a significant effect. These findings emphasise the importance of considering the approach used to identify utility values for parameterisation of health economic models. It is crucial for modellers to systematically examine the specific factors that may influence reported utility values. The results from this review can inform future cost-utility analyses evaluating the cost-effectiveness of interventions; however, given the heterogeneity between studies the different subgroup mean utility values, and the nuances resulting from study type differences, a meta-analysed utility value that combines mean utility values across all subgroups identified here should be approached with caution.

Author contributions

LO, SM, and NS developed the study concept and design. NS drafted the study protocol in conjunction with SM and LO. LA designed the search strategies and conducted the electronic searches. NS and YZ independently screened and appraised the relevant studies and extracted the data from the included studies; any disagreements were resolved by SM. LO, NS, and SM planned the analysis. NS conducted all statistical analyses and drafted the manuscript in conjunction with LO and SM. All authors commented on the draft and read and approved the final manuscript. The authors alone are responsible for the content and writing of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

No additional data is available.

Additional information

Funding

No funding was received for conducting this study.