Predicting Patient-Level 3-Level Version of EQ-5D Index Scores From a Large International Database Using Machine Learning and Regression Methods

ObjectivesThis study aimed to evaluate the performance of machine learning and regression methods in the prediction of 3-level version of EQ-5D (EQ-5D-3L) index scores from a large diverse data set.MethodsA total of 30 studies from 3 countries were combined. Predictions were performed via eXtreme Gradient Boosting classification (XGBC), eXtreme Gradient Boosting regression (XGBR) and ordinary least squares (OLS) regression using 10-fold cross-validation and 80%/20% partition for training and testing. We evaluated 6 prediction scenarios using 3 samples (general population, patients, total) and 2 predictor sets: demographic and disease-related variables with/without patient-reported outcomes. Model performance was evaluated by mean absolute error and percent of predictions within clinically irrelevant error range and within correct health severity group (EQ-5D-3L index <0.45, 0.45-0.926, >0.926).ResultsThe data set involved 26 318 individuals (clinical settings n = 6214, general population n = 20 104) and 26 predictor variables plus diagnoses. Using all predictors and the total sample, mean absolute error values were 0.153, 0.126, and 0.131, percent of predictions within clinically irrelevant error range were 47.6%, 39.5%, and 37.4%, and within the correct health severity group were 56.3%, 64.9%, and 63.3% by XGBC, XGBR, and OLS, respectively. The performance of models depended on the applied evaluation criteria, the target population, the included predictors, and the EQ-5D-3L index score range.ConclusionsRegression models (XGBR and OLS) outperformed XGBC, yet prediction errors were outside the clinically irrelevant error range for most respondents. Our results highlight the importance of systematic patient-reported outcome (EQ-5D) data collection. Dialogs between artificial intelligence and outcomes research experts are encouraged to enhance the value of accumulating data in health systems.     

A Comparison of a Preliminary Version of the EQ-HWB Short and the 5-Level Version EQ-5D

ObjectivesThe EQ Health and Wellbeing Short (EQ-HWB-S) is a new broad generic measure of health and wellbeing for use in economic evaluations of interventions across healthcare, social care, and public health. This measure conceptually overlaps with the 5-level version EQ-5D (EQ-5D-5L), while expanding on the coverage of health and social care related dimensions. This study aims to examine the extent to which the EQ-HWB-S and EQ-5D-5L overlap and are different.MethodsA sample of US-based respondents (n = 903; n = 400 cancer survivors and n = 503 general population) completed a survey administered via an online panel. The survey included the EQ-HWB item pool (62 items, including 11 items used in this analysis), EQ-5D-5L, and questions about sociodemographic and health characteristics. The analysis included (Spearman’s) correlations, the comparison of patterns of response (distributions and ceiling effects), and the ability to discriminate between known groups.ResultsModerate to strong associations were found between conceptually overlapping dimensions of the EQ-5D-5L and the EQ-HWB-S (r_s > 0.5, P < .001). Among respondents reporting full health on the EQ-5D-5L (n = 161, 18.23%), the EQ-HWB-S identified ceiling effects, particularly with the item “feeling exhausted.” Most EQ-5D-5L and EQ-HWB-S items demonstrated discriminative ability among those with and without physical and mental conditions, yielding medium (> 0.5) to large effect sizes (> 0.8). Nevertheless, only EQ-HWB-S items distinguished between caregivers and noncaregivers and those with low and high caregiver burden, albeit with small effect sizes (0.2-0.5).ConclusionsResults indicate a convergence between the measures, especially between overlapping dimensions, lending support to the validity of the EQ-HWB-S. The EQ-HWB-S performed similarly or better than the EQ-5D-5L among patient groups and is better able to differentiate among caregivers and respondents closer to full health.     

Cost-Utility Analysis Using EQ-5D-5L Data: Does How the Utilities Are Derived Matter?

Objectives

To explore how the use of EQ-5D-5L value set and crosswalk from EQ-5D-5L to EQ-5D-3L (and use of 3L value set) would affect cost-effectiveness analysis results for England and six other countries (Canada, the Netherlands, China, Japan, South Korea, and Singapore).

Estimating a Preference-Based Index for Mental Health From the Recovering Quality of Life Measure: Valuation of Recovering Quality of Life Utility Index

BackgroundThere are increasing concerns about the appropriateness of generic preference-based measures to capture health benefits in the area of mental health.ObjectivesThe aim of this study is to estimate preference weights for a new measure, Recovering Quality of Life (ReQoL-10), to better capture the benefits of mental healthcare.MethodsPsychometric analyses of a larger sample of mental health service users (n = 4266) using confirmatory factor analyses and item response theory were used to derive a health state classification system and inform the selection of health states for utility assessment. A valuation survey with members of the UK public representative in terms of age, sex, and region was conducted using face-to-face interviewer administered time-trade-off with props. A series of regression models were fitted to the data and the best performing model selected for the scoring algorithm.ResultsThe ReQoL-Utility Index (UI) classification system comprises 6 mental health items and 1 physical health item. Sixty-four health states were valued by 305 participants. The preferred model was a random effects model, with significant and consistent coefficients and best model fit. Estimated utilities modeled for all health states ranged from ?0.195 (state worse than dead) to 1 (best possible state).ConclusionsThe development of the ReQoL-UI is based on a novel application of item response theory methods for generating the classification system and selecting health states for valuation. Conventional time-trade-off was used to elicit utility values that are modeled to enable the generation of QALYs for use in cost-utility analysis of mental health interventions.     

A Comparison of Generic and Condition-Specific Preference-Based Measures Using Data From Nivolumab Trials: EQ-5D-3L,Mapping to the EQ-5D-5L,and European Organisation for Research and Treatment of Cancer Quality of Life Utility Measure-Core 10 Dimensions

ObjectivesThere is growing interest in condition-specific preference measures, including the European Organisation for Research and Treatment of Cancer Quality of Life Utility Measure-Core 10 Dimensions (QLU-C10D). This research assessed the implications of using utility indices on the basis of the EQ-5D-3L, a mapping of EQ-5D-3L to the EQ-5D-5L, and the QLU-C10D, and compared their psychometric properties.MethodsData were taken from 8 phase 3 randomized controlled trials of nivolumab with or without ipilimumab for the treatment of solid tumors. Utilities for progression-related states were calculated using the UK and English value sets and incremental quality-adjusted life-years (QALYs) derived from established UK cost-effectiveness models. The psychometric properties of the utility indices were assessed using pooled trial data.ResultsCompared with the EQ-5D-3L index, the mapped EQ-5D-5L index yielded an average of 6% more and the QLU-C10D index an average of 2% fewer incremental QALYs for nivolumab versus comparators. All indices could differentiate between groups defined by performance status, cancer stage, or self-reported health status at baseline and detect meaningful changes in performance status, tumor response, health status, and quality of life over approximately 12 weeks of treatment.ConclusionsThe lower QALY yield of the QLU-C10D was balanced by evidence of greater validity and responsiveness. Benefits gained from using the QLU-C10D may be apparent when treatments affect targeted symptoms and functional aspects, including sleep, bowel function, appetite, nausea, and fatigue. The observed differences in QALYs may not be sufficiently large to affect health technology assessment decisions.     

Estimating Health-State Utility Values in Kidney Transplant Recipients and Waiting-List Patients Using the EQ-5D-5L

Objectives

To report health-state utility values measured using the five-level EuroQol five-dimensional questionnaire (EQ-5D-5L) in a large sample of patients with end-stage renal disease and to explore how these values vary in relation to patient characteristics and treatment factors.

Instrument-Defined Estimates of the Minimally Important Difference for EQ-5D-5L Index Scores

Background

The five-level EuroQol five-dimensional questionnaire (EQ-5D-5L) is a preference-based measure of health-related quality of life (HRQOL), which yields an index score anchored at 0 (dead) and 1 (full health). We lack evidence on estimates for the minimally important difference (MID) of the EQ-5D-5L that will help in interpreting differences or changes in HRQOL measured by this scale score.

Japanese Population Norms of EQ-5D-5L and Health Utilities Index Mark 3: Disutility Catalog by Disease and Symptom in Community Settings

ObjectivesThis study aimed to establish the Japanese population norms of the EQ-5D-5L and Health Utilities Index Mark 3 (HUI3) and estimate the disutility associated with diseases and symptoms.MethodsWe performed a door-to-door survey of the general population by random sampling. The planned sample size was 10 000 residents (age ≥16 years) of 334 districts in Japan. In addition to the EQ-5D-5L and HUI3 questionnaires, questions regarding demographic factors and self-reported main diseases and symptoms were asked. The EQ-5D-5L and HUI3 responses were converted to index values on the basis of Japanese value sets. Summary values by age and sex were calculated to obtain Japanese normative values. A multiple linear model was used to examine relationships between these values and diseases and symptoms.ResultsWe collected 10 183 responses from 334 districts. The mean EQ-5D-5L index values were 0.821 (male) and 0.774 (female) in the age group of 80 to 89 years, which were lower compared with 0.978 (male) and 0.967 (female) in the age group of 16 to 19 years. Similar trends were observed for the HUI3 values. Age, sex, household income, and education level had a significant influence on the values of both instruments. When measured with the EQ-5D-5L, Parkinson disease, dementia, and stroke were associated with the largest disutility (>0.2), and the disutility for depression was approximately 0.18. In contrast, the HUI3 disutility values for Parkinson disease and dementia were approximately 0.4.ConclusionsThis study established the Japanese population norms of the EQ-5D-5L and HUI3, which can be used in healthcare decision making and contribute to a more reliable analysis of economic evaluations.     

Quebec Health-Related Quality-of-Life Population Norms Using the EQ-5D-5L: Decomposition by Sociodemographic Data and Health Problems

ObjectivesPopulation norms for the EQ-5D-5L were published in Canada but only for Alberta province. The purpose of this study was to derive Quebec population norms from the EQ-5D-5L.MethodsThe data came from a larger study conducted between September 2016 and March 2018 using elicitation techniques for a quality-adjusted life-year project. The online survey was distributed randomly in the province of Quebec. To best describe the entire population, data were stratified by various sociodemographic characteristics such as age, gender, urban and rural populations, whether disadvantaged or not, immigrant or nonimmigrant, and health problems.ResultsA total of 2704 (53.8%) respondents completed the EQ-5D-5L. Mean (95% confidence interval) and median (interquartile range) utility scores were 0.824 (0.818-0.829) and 0.867 (0.802-0.911), respectively. The EQ-VAS scores were estimated at 75.9 (75.2-76.6) and 80 (69-90). Subjects with lower scores were those who had a low or high body mass index; were smokers; were single, divorced, or widowed; had no children; were unemployed or sick; had lower education or lower annual income; and had a family or personal history of serious illness. Immigrants had higher scores. There was no difference in gender and urban or rural population. The score logically decreased with worsening health status, from a mean score of 0.896 (0.884-0.908) to 0.443 (0.384-0.501; P < .0001. Similar results were observed for subjects’ satisfaction with their health or life. Subjects with lower scores were less willing to take risks. Subjects who declared they were affected by health problems presented significant lower utility scores, ranging from 0.554 (nervous problem) to 0.750 (cancer), compared with those without health problems (0.871; confidence interval: 0.867-0.876).ConclusionThis is the first study to present utility score norms for EQ-5D-5L for the Quebec population. These results will be useful for comparison with quality-adjusted life-year studies to better interpret their results. Moreover, utility norms were provided for 21 health problems, which was rarely done.     

The Impact of Subclinical Hypothyroidism on the Quality of Life During Pregnancy: Mapping 5-Level Version of EQ-5D and ThyPRO-39

ObjectivesTo describe how subclinical hypothyroidism (SubHypo) influences the quality of life (QoL) during pregnancy.MethodsIn primary data collection (NCT04167423), thyroid stimulating hormone (TSH), free thyroxine (FT4), thyroid peroxidase antibodies, generic quality of life (QoL; 5-level version of EQ-5D [EQ-5D-5L]), and disease-specific QoL (ThyPRO-39) were measured among pregnant women. SubHypo during each trimester was defined according to the 2014 European Thyroid Association guidelines (TSH > 2.5, 3.0, and 3.5 IU/L, respectively; with normal FT4). Path analysis described relationships and tested mediation. Linear ordinary least squares, beta, tobit, and two-part regressions were used to map ThyPRO-39 and EQ-5D-5L. Alternative SubHypo definition was tested in sensitivity analysis.ResultsA total of 253 women at 14 sites (31 ± 5 years old, 15 ± 6 weeks pregnant) completed the questionnaires. Sixty-one (26%) had SubHypo and differed from 174 (74%) euthyroid women in smoking history (61% vs 41%), primiparity (62% vs 43%) and TSH level (4.1 ± 1.4 vs 1.5 ± 0.7 mIU/L, P < .001). EQ-5D-5L utility in SubHypo (0.89 ± 0.12) was lower than that in euthyroid (0.92 ± 0.11; P = .028) even after adjustment (difference −0.04, P = .033), whereas ocular (P = .001, ThyPRO-39), cognitive symptoms (P = .043), anxiety (P < .0001), and the composite score were higher. The impact of SubHypo on utility was mediated by anxiety. Results were confirmed by sensitivity analysis. Final mapping equation (ordinary least squares) includes goiter symptoms, anxiety, upset stomach, composite score (ThyPRO-39), FT4 levels, and week of pregnancy (determination coefficient 0.36).ConclusionThis is the first QoL mapping of SubHypo during pregnancy and the first evidence that SubHypo is associated with a negative impact on QoL. The effect is mediated by anxiety. EQ-5D-5L utilities can be generated based on ThyPRO-39 scores collected in pregnant euthyroid and patients with SubHypo.     

基于EQ-5D-3L和ICECAP-A量表的我国普通人群生命质量研究

目的:分析EQ-5D-3L和ICECAP-A量表评价我国普通人群生命质量的差异以及对干预方案价值评价的影响,为研究者选择合适的生命质量测量工具提供参考。方法:采用配额抽样选取802名受访者进行生命质量评价,并分别采用因子分析、多分格相关性和Bland-Altman plot一致性分析等方法探讨两量表测量结果的差异。结果:Wilcoxon秩和检验表明EQ-5D-3L量表的健康效用均值高于ICECAP-A量表的测量结果;ICECAP-A量表五个维度均主要加载于反映社会心理健康的因子1,而EQ-5D-3L量表的大部分维度主要加载于反映生理健康的因子2。部分维度之间也存在显著的相关性,但均较弱。两量表效用值的ICC为0.32,一致性分析显示5.74%的受访者超出了95%的一致性界限。结论:前者的测量内涵是健康相关生命质量,而后者则反映的是幸福感、可行能力等更广义的生命质量,其对于旨在提升公众广义幸福感和社会福祉的干预措施效果评估方面具有较好的适用性。研究者可根据测量目的及量表属性选择合适的量表,鉴于两个量表在测量内涵中的互补性,也可以在研究中同时采用两种量表以便更全面地反映干预措施的效果或受访者的生命质量。     

Generating EQ-5D-3L Utility Scores from the Dermatology Life Quality Index: A Mapping Study in Patients with Psoriasis

Objectives

To develop an algorithm to predict the three-level EuroQol five-dimensional questionnaire (EQ-5D-3L) utility scores from the Dermatology Life Quality Index (DLQI) in psoriasis.

Overview,Update, and Lessons Learned From the International EQ-5D-5L Valuation Work: Version 2 of the EQ-5D-5L Valuation Protocol

A standardized 5-level EuroQol 5-dimensional questionnaire (EQ-5D-5L) valuation protocol was first used in national studies in the period 2012 to 2013. A set of problems encountered in this initial wave of valuation studies led to the subsequent refinement of the valuation protocol. To clarify lessons learned and how the protocol was updated when moving from version 1.0 to the current version 2.1 and 2.0, this article will (1) present the challenges faced in EQ-5D-5L valuation since 2012 and how these were resolved and (2) describe in depth a set of new challenges that have become central in currently ongoing research on how EQ-5D-5L health states should be valued and modeled.     

Evaluating the Health-Related Quality of Life of the Rare Disease Population in Hong Kong Using EQ-5D 3-Level

ObjectivesThis study aimed to establish a normative profile of health-related quality of life (HRQOL) of the rare disease (RD) population in Hong Kong (HK) and identify potential predictors.MethodsBetween March 2020 and October 2020, patients with RD and caregivers were recruited through Rare Disease Hong Kong, the largest RD patient group alliance in HK. HRQOL was derived using the EQ-5D 3-Level with reference to the established HK value set. Utility scores were stratified according to demographics and disease-related information. Multiple linear regression was performed to explore the associations between patient characteristics and HRQOL.ResultsA total of 286 patients, covering 107 unique RDs, reported a mean utility score of 0.53 (SD 0.36). Thirty patients (10.5%) reported negative utility scores, indicating worse-than-death health states. More problems were recorded in the “usual activities” and “self-care” dimensions. Univariate analyses revealed that neurologic diseases, high out-of-pocket expenditure, home modification, and living in public housing or subdivided flats/units were significantly associated with lower HRQOL. A total of 99 caregivers reported a mean utility score of 0.78 (SD 0.17), which was significantly associated with the utility score of patients they took care of (r = 0.32; P = .001).ConclusionsThe normative profile of the RD population was established, which revealed lower HRQOL in the RD population than other chronic disease groups and general population in HK. Findings were corroborated by evidence from other cohorts using EQ-5D, combined as part of a meta-analysis. Identifying predictors highlight areas that should be prioritized to improve HRQOL of RD population through clinical and psychosocial dimensions.     

Selecting Bolt-On Dimensions for the EQ-5D: Examining Their Contribution to Health-Related Quality of Life

Background

Generic preference-based measures may miss dimensions important for the health-related quality of life (HRQOL) of patients. When this happens, a possible solution is to add bolt-ons. Finch et al. (Finch AP, Brazier JE, Mukuria C, Bjorner JB. An exploratory study on using principal component analysis and confirmatory factor analysis to identify bolt-on dimensions: the EQ-5D case study. Value Health 2017;10:1362–75) have recently shown that bolt-ons can be systematically identified using factor analysis. Nevertheless, because for each bolt-on option a complete re-evaluation may be required, methods to select between them are needed.

Assessing the Health-Related Quality of Life of Australian Adolescents: An Empirical Comparison of the Child Health Utility 9D and EQ-5D-Y Instruments

ObjectivesTo examine the performance of two recently developed preference-based instruments—the Child Health Utility 9D (CHU9D) and the EuroQol five-dimensional questionnaire Youth version (EQ-5D-Y)—in assessing the health-related quality of life (HRQOL) of Australian adolescents.MethodsAn online survey including the CHU9D and the EQ-5D-Y, self-reported health status, and a series of sociodemographic questions was developed for administration to a community-based sample of adolescents (aged 11–17 years). Individual responses to both instruments were translated into utilities using scoring algorithms derived from the Australian adult general population.ResultsA total of 2020 adolescents completed the online survey. The mean ± SD utilities of the CHU9D and the EQ-5D-Y were very similar (0.82 ± 0.13 and 0.83 ± 0.19, respectively), and the intraclass correlation coefficient (0.80) suggested good levels of agreement. Both instruments were able to discriminate according to varying levels of self-reported health status (P < 0.001). Although exhibiting good levels of agreement overall, some wide divergences were apparent at an individual level.ConclusionsThe study results are encouraging and illustrate the potential for both the CHU9D and the EQ-5D-Y to be more widely used for measuring and valuing the HRQOL of adolescent populations in Australia and internationally. Generating adolescent-specific scoring algorithms pertaining to each instrument and an empirical comparison of the resulting utilities is a natural next step. More evidence is required from the application of the CHU9D and the EQ-5D-Y in specific patient groups in adolescent health settings to inform the choice of instrument for measuring and valuing the HRQOL for the economic evaluation of adolescent health care treatments and services.     

A Bayesian Approach for Incorporating the EQ-5D Visual Analog Scale When Estimating the Health-Related Quality of Life

ObjectivesThe EuroQoL 3-level version of EQ-5D and 5-level version of EQ-5D questionnaires are often used to quantify health states. They include ordinal responses across 5 health dimensions (EQ-5D index) and an EQ-visual analog scale (EQ-VAS) overall health rating. We investigated the value of incorporating the EQ-VAS to update health utility estimates using a Bayesian framework.MethodsWe created a joint bivariate normal EQ-VAS and EQ-5D index utility model and compared this to a univariate normal EQ-5D index utility model. We tested these models for 1026 Sri Lankan patients with chronic kidney disease and 94 Australian patients with wounds. We validated our approach by simulating EQ-VAS and EQ-5D index responses and applying our Bayesian model and then comparing the modeled estimates to our observed data.ResultsThe combined model showed a reduction in estimate uncertainty for all respondents. Compared with the EQ-5D index-only model, the mean utility for Sri Lankan respondents dropped from 0.556 (0.534-0.579) to 0.540 (0.521-0.559) in men and increased from 0.489 (0.461-0.518) to 0.528 (0.506-0.550) in women, with reduced credible interval width by 13% and 23%, respectively. The mean utility in Australian respondents moved from 0.715 (0.633-0.800) to 0.716 (0.652-0.782) in men, and 0.652 (0.581-0.723) to 0.652 (0.593-0.711) in women, with reduced credible interval width by 23% and 17%, respectively. The credible interval width for simulated data also narrowed, ranging from 8.3 to 8.5%.ConclusionsIncluding the EQ-VAS through Bayesian methods can add value by reducing requisite sample sizes and decision uncertainty using small amounts of additional data that is often collected but rarely used.     

Analyzing the Pain/Discomfort and Anxiety/Depression Composite Domains and the Meaning of Discomfort in the EQ-5D: A Mixed-Methods Study

ObjectivesThe EQ-5D has 2 composite domains: pain/discomfort (PD) and anxiety/depression (AD). This study aims to explore how respondents use the composites to self-report health and what the meaning of discomfort is in the EQ-5D for the general public.MethodsBoth qualitative and quantitative data were collected in an online cross-sectional survey involving a nationally representative general population sample in Hungary (n = 1700). Respondents completed the 5-level version of EQ-5D, followed by the composites split into individual subdomains. Open-ended questions were asked to explore respondents’ interpretations and experiences of discomfort.ResultsSix different response behaviors were identified in the composites: “uniform” (21%-32%), “most severe” (30%-34%), “least severe” (16%-23%), “average” (2%-4%), “synergistic” (4%-5%), and “inconsistent” (13%-15%). Compared with the individual subdomains, many respondents under-reported their problems on both composites (PD 16%-22% and AD 6%-13%, P < .05). In respondents who scored differently in the 2 separate domains, mainly problems with the first subdomain determined responses in the composites (PD 66% and AD 61%). The discomfort subdomain in the EQ-5D captured more than 100 different problems, including pain, nonpain physical discomfort (eg, tiredness, dizziness, and nausea), and psychological discomfort (eg, anxiety, nervousness, and sadness). Women, older adults, and those in worse general health status more often considered discomfort as pain (P < .05).ConclusionsWe found empirical evidence of measurement error in the composite responses on the EQ-5D, including under- and inconsistent reporting, ordering effects, potential differential item functioning, and interdomain dependency. Our findings contribute new knowledge to the development of new and refinement of existing self-reported health status instruments, also beyond the EQ-5D.