Development and Validation of a Preference-Based Glaucoma Utility Instrument Using Discrete Choice Experiment

Key Points

Question  Can the new Glaucoma Utility Instrument (Glau-U) estimate utilities across the spectrum of glaucoma and glaucoma-related vision impairment severity?

Findings  In this cross-sectional study of 304 adults with mild, moderate, or advanced glaucoma in Singapore, Glau-U utilities decreased as glaucoma severity increased, demonstrating reductions of up to 41% in quality-adjusted life-years compared with the best health state of preperimetric glaucoma.

Meaning  These findings suggest that Glau-U is a valid glaucoma-specific utility instrument that can estimate utility weights that are associated with glaucoma and related vision impairment and can be used in cost-effectiveness analyses of interventions for glaucoma and vision loss.

Importance  A glaucoma-specific instrument for estimating utilities across the spectrum of glaucoma severity is currently lacking, hindering the assessment of the cost-effectiveness of glaucoma treatments.

Objective  To develop and validate the preference-based Glaucoma Utility Instrument (Glau-U) and to ascertain the association between Glau-U utilities and severity of glaucoma and vision impairment.

Design, Setting, and Participants  This cross-sectional study was conducted in 2 stages at the Singapore National Eye Centre glaucoma clinics. Stage 1 focused on the identification and pretesting of the Glau-U attributes and was carried out between June 2009 and May 2016. Stage 2 involved the development and administration of the discrete choice experiment (DCE) survey and tasks and was conducted between May 7, 2018, and December 11, 2019. Stage 2 participants were English- or Mandarin-speaking Singaporean citizens or permanent residents of Chinese, Malay, or Indian ethnicity who were 40 years or older and had a clinical diagnosis of glaucoma in at least 1 eye.

Exposures  Glau-U comprised 6 quality-of-life attributes: activities of daily living, lighting and glare, movement, eye discomfort, other effects of glaucoma, and social and emotional effects. The descriptions or response options for these attributes were no difficulty or never, some difficulty or sometimes, or severe difficulty or often.

Main Outcomes and Measures  Utility weights for Glau-U were developed using a DCE questionnaire, which was interviewer administered to participants. Mixed logit regression determined utility weights for each health state. Glau-U utility weights across better- or worse-eye glaucoma and vision impairment severity were calculated using 1-way analysis of variance. Correlations between Glau-U utilities and better- or worse-eye visual fields and EuroQol 5-Dimension utilities were ascertained to assess convergent and divergent validity.

Results  Of the 304 participants (mean [SD] age, 68.3 [8.7] years; 182 men [59.9%]), 281 (92.4%) had no vision impairment in the better eye, 13 (4.3%) had mild impairment, and 10 (3.3%) had moderate to severe vision impairment. Mean (SD) Glau-U utilities decreased as better-eye glaucoma severity increased (none: 0.73 [0.21]; mild: 0.66 [0.21]; moderate: 0.66 [0.20]; severe: 0.60 [0.28]; and advanced or end-stage: 0.22 [0.38]; P < .001), representing reductions of 20.7% to 76.1% in quality-adjusted life-years compared with a health state that included preperimetric glaucoma. Mean (SD) Glau-U utilities also decreased as better-eye vision impairment worsened from 0.67 (0.23) for none to 0.58 (0.32) for mild to 0.46 (0.29) for moderate to severe vision impairment. Glau-U utilities demonstrated moderate correlations with better-eye (r = 0.34; P < .001) and worse-eye (r = 0.33; P < .001) mean deviation scores and low correlations with EuroQol 5-Dimension utilities (r = 0.22; P < .001), supporting convergent and divergent validity.

Conclusions and Relevance  Use of Glau-U revealed large decrements in utility that were associated with late-stage glaucoma, suggesting that this new instrument may be useful for cost-effectiveness analyses of interventions and informing resource allocation policies for glaucoma and vision loss.

Glaucoma has a substantial role in quality of life (QOL), being independently associated with reduced vision-specific functioning,¹ mobility,² and emotional well-being.^3,4 Management of glaucoma relies on topical medication, laser procedure (eg, laser trabeculoplasty), and surgery (eg, trabeculectomy or minimally invasive glaucoma surgery) along with novel treatments, such as a sustained-release drug delivery system (eg, Bimatoprost implant).⁵ The direct and indirect costs of glaucoma and its treatments are substantial,⁶ but the cost-effectiveness and quality-adjusted life-year (QALY) gains associated with existing and emerging treatments are unclear because of the lack of a valid glaucoma-specific utility instrument.

Several utility instruments, which are used to calculate QALYs and contribute to cost-utility analyses, are available for measuring the burden of glaucoma and its associated treatments. The EuroQol 5-Dimension (EQ-5D),⁷ for example, is a widely used generic, health-related utility system; however, it lacks sensitivity to quantify the burden of eye conditions, particularly glaucoma.^8,9 Vision-specific instruments (eg, Visual Function Questionnaire Utility Index) are also available, but they lack precision in glaucoma populations.¹⁰ Health utility weights can also be estimated by time trade-off and standard gamble techniques; however, these methods are complex and may not be suitable for older adults or those with limited education.¹¹

Ordinal approaches, such as the discrete choice experiment (DCE), have shown promise for health state valuation because they offer a flexible method for estimating which attributes are important in decision-making (eMethods in the Supplement).¹² The Glaucoma Utility Index (GUI) was developed by Burr and colleagues¹³ using a DCE questionnaire; however, because the GUI does not capture the mental health impact of glaucoma and its utility values are anchored in best to worst glaucoma health states rather than the generic QALY scale of perfect health to death, the use of the GUI in public resource allocation, which typically requires utilities that are based on the generic QALY scale, incites debate.⁶ Recently, Kennedy et al¹⁴ and Muratov et al¹⁵ developed and validated the Health Utility for Glaucoma-5 descriptive system in a Canadian population; however, utilities have not yet been developed. Therefore, a glaucoma-specific utility measure is urgently needed.

This cross-sectional study aimed to develop and validate the preference-based Glaucoma Utility Instrument (Glau-U) and to ascertain the association between Glau-U utilities and the severity of glaucoma and vision impairment. Specifically, we identified glaucoma-specific QOL attributes and levels using a qualitative method and estimated Glau-U utility weights using DCE questionnaires in people with glaucoma.

We conducted this cross-sectional study in 2 stages. Both stages received ethical approval from the SingHealth Centralised Institutional Review Board and adhered to the tenets of the Declaration of Helsinki.¹⁶ All participants provided written informed consent.

Identification of the attributes for the Glau-U descriptive system was informed by the GUI,¹³ focus groups with 6 glaucoma ophthalmologists from the Singapore National Eye Centre, and semistructured interviews with 39 older adults (mean [SD] age, 67.5 [9.9] years) with mild, moderate, or severe glaucoma in Singapore.¹⁷ After the QOL attributes were established, a DCE survey was developed and pretested by 70 older adults with varying severity of glaucoma. Stage 1 work was conducted between June 2009 and May 2016. Supporting quotes for each attribute are shown in eTable 1 in the Supplement, and more details of the pretesting phase are provided in the eMethods in the Supplement.

For stage 2, English- or Mandarin-speaking Singaporean citizens or permanent residents of Chinese, Malay, or Indian ethnicity who were 40 years or older and had a clinical diagnosis of primary open-angle glaucoma, normal-tension glaucoma, or primary angle-closure glaucoma in at least 1 eye were recruited from Singapore National Eye Centre glaucoma clinics between May 7, 2018, and December 11, 2019. Those with symptoms suggestive of glaucoma, those with secondary glaucoma or other late-stage ocular retinal comorbidities affecting visual functioning, and/or those with a history of intraocular surgery for other retinal conditions were excluded. People with cognitive,¹⁸ hearing, or physical impairments that prevented study participation were also excluded.

Included individuals received $30 Singapore dollars as compensation for participating in the study. After providing written informed consent, participants underwent a standardized protocol at the Singapore Eye Research Institute research clinic that was interviewer-administered in English or Mandarin by a bilingual clinical research coordinator. The protocol included the DCE tasks, a sociodemographic and medical history questionnaire, the 5-level EQ-5D,⁷ visual acuity testing to ascertain the severity of vision impairment, and visual field testing to ascertain the severity of glaucoma (eMethods and eTable 2 in the Supplement).

Grading of glaucoma severity was performed by 2 of us (glaucoma clinicians M.B. and M.E.N.) using both the Glaucoma Staging System and the Advanced Glaucoma Intervention Study software.^19,20 The better or worse eye was rated with the following severity levels: none, mild, moderate, severe, and advanced or end-stage. Given the low number of end-stage glaucoma cases, we collapsed the advanced and end-stage categories.

A description of the DCE questionnaire is provided in the eMethods in the Supplement. The final version of the Glau-U descriptive system comprised 6 QOL attributes (eTable 3 in the Supplement): activities of daily living (performing day-to-day tasks, such as reading and housework), lighting and glare (challenging lighting conditions, such as dim or bright light), movement (getting out and about, such as crossing a road), eye discomfort (ocular comfort symptoms, such as dry or watery eyes), other effects of glaucoma and its treatment (nonocular adverse effects, such as tiredness and headaches), and social and emotional effects of glaucoma (eg, feelings of frustration or social isolation). Each attribute was described according to these 3 levels of difficulty: no difficulty, some difficulty, or severe difficulty for the attributes of activities of daily living, lighting and glare, and movement. Never, sometimes, or often was used to describe the attributes of eye discomfort, other effects of glaucoma and its treatment, and social and emotional effects of glaucoma. In addition, we included a survival attribute (duration in this health state) with 4 possible levels (5, 10, 15, and 20 years) to account for the quantity of life lived in a certain health state.¹² The 4 levels were based on the mean age (67.5 years) of the glaucoma population in stage 1 and the maximum projected life expectancy (mean [range], 84 [81-86] years) of adults in Singapore.²¹ This survival attribute allowed people to realistically imagine themselves living in each particular health state for the stated duration. These 7 attributes were used to construct the DCE tasks.

The DCE questionnaire was interviewer administered by trained clinical research coordinators who used a standardized script that guided participants through a sample DCE task. When the clinical research coordinator was confident that the participant understood the task sufficiently, the coordinator commenced the study protocol in which the participant was asked to choose a preferred health state between 2 health state options presented in a DCE task (eTable 3 and the eMethods in the Supplement provide more details on DCE design and sample size). All study materials were professionally translated and then back-translated into Mandarin.

The data were analyzed using a mixed logit regression model that provides population estimates that allow for random variables (eg, people taking into account different factors or the same factors but in different ways²²). The survival attribute was estimated as a linear variable after testing for nonlinearity. All other attributes were included as categorical variables, and the no difficulty or never response option was used as the base comparator. All QOL attribute levels were interacted with the survival attribute because health is usually defined as a product of both quality and quality of life. This interaction produced 13 explanatory variables (2 for each of the 6 attributes interacted with the survival attribute vs no difficulty or never, and 1 for the linear survival attribute). Theoretical validity was supported if the coefficients moved in the expected direction. A priori, we expected coefficients to decrease with increased levels of difficulty.

We calculated QALY or Glau-U utility weights for each health state using the method developed by Bansback and colleagues.¹² This method assigned a utility weight of 1 to the best possible glaucoma health state (defined as preperimetric glaucoma) and a utility weight of 0 to death. To convert the Glau-U descriptive system to the QALY scale (a utility weight of 1 for perfect health and 0 for death), we assumed that the weight of the best possible glaucoma health state (no difficulty or never in each attribute [111111; ie, a score of 1 of a possible score of 3]) was equivalent to 0.92 based on the mean EQ-5D QALY score for preperimetric glaucoma in a population-based cohort of the 3 major ethnic groups (Chinese, Malay, and Indian) in Singapore²³ and on the utility study by Lee and colleagues.²⁴ The QALY weights for each health state were calculated by subtracting the loss of utility associated with moving from the best possible glaucoma health state (0.92) to the relevant health state.

To assess the convergent and divergent validity of Glau-U, we hypothesized moderate correlation (0.3 > r < 0.69) with better-eye visual field (continuous mean deviation scores) and low correlation (r <0.3) with EQ-5D utilities given the known insensitivity of EQ-5D utilities in detecting the burden associated with glaucoma.^8,9 For criterion validity, we calculated Glau-U utility values across mild, moderate, severe, and advanced or end-stage glaucoma as well as across 3 levels of better-eye vision impairment using 1-way analysis of variance with Tukey post hoc tests. Percentage decreases in the QALY weights for each level of glaucoma severity were calculated with reference to the QALY weight of 0.92 for preperimetric glaucoma, which was considered as a control group.

All statistical tests performed were 2-sided, and a P < .05 was considered statistically significant. All analyses were conducted with Stata, version 14 (StataCorp LLC).

Of the 304 participants (mean [SD] age, 68.3 [8.7] years; 182 men [59.9%], 122 women [40.1%]; and 283 with Chinese ethnicity [93.1%]), 288 (94.7%) were using topical medication, 67 (22.0%) had laser treatment, and 92 (30.3%) had surgery for glaucoma (Table 1). Ninety-five participants (31.2%) had no glaucoma, 111 (36.5%) had a mild case, 60 (19.7%) had moderate severity, 26 (8.6%) had severe glaucoma, and 12 (3.9%) had advanced or end-stage disease in the better eye. The mean (SD) presenting visual acuity in the better eye was 0.14 (0.14) logMAR or 20/25 Snellen, with 281 participants (92.4%) having no vision impairment in the better eye (visual acuity in the better eye: ≤0.3 logMAR or ≤20/40 Snellen), 13 (4.3%) having mild vision impairment (visual acuity: >0.3 to ≤0.48 logMAR or >20/40 to ≤20/60 Snellen), and 10 (3.3%) having moderate to severe vision impairment (visual acuity: >0.48 logMAR or >20/60 Snellen).

The analysis included responses to 1406 DCE choice tasks from the 304 participants in the study. Twenty-one participants (6.9%) did not pass the attention-test task, suggesting that they may not have understood the DCE tasks. We excluded these data in a sensitivity analysis; however, because utility weights were unaffected, these data were included in the final calculations.

The coefficients from the mixed logit model were in the expected direction. As hypothesized, the absolute value of the coefficient for the most severe level of an attribute was larger than the value for the middle level (ie, utility decreased as the severity of a health state increased) (Table 2), confirming the theoretical validity of the model. However, the coefficients for eye discomfort were 0.00 (95% CI, – 0.03 to 0.02; P = .77) for the severe level and 0.00 (95% CI, – 0.03 to 0.02; P = .90) for the middle level, and the coefficients for other glaucoma effects were –0.02 (95% CI, – 0.05 to 0.01; P = .14) for the severe level and 0.00 (95% CI, – 0.03 to 0.02; P = .96) for the middle level (Table 2).

The largest QALY differences between the best and worst levels for each attribute (while holding other attributes at their best level) was observed for movement (0.64), activities of daily living (0.54), and social and emotional effects (0.41), suggesting that these 3 attributes were the most important to patients with glaucoma (Figure). In contrast, the smallest changes were observed for eye discomfort (0.02) and other effects of glaucoma and its treatment (0.08), suggesting that participants valued these 2 attributes the least.

The QALY weights for each health state can be calculated against the weight assigned to the best possible glaucoma health state (0.92). For example, a health state with some difficulty with activities of daily living, lighting and glare, and movement as well as rating of sometimes experiencing eye discomfort, other effects of glaucoma and its treatment, and social and emotional effects would result in a total loss of utility of 0.57 (calculated by summing the individual utility loss associated with each attribute state). The final QALY weight for this particular health state was, therefore, 0.35 (0.92 minus 0.57) (Table 3). A QALY weight for all 729 possible profile combinations can be calculated in this way: subtract the known utility loss associated with each level of each attribute. Of the 729 profile combinations, 299 (41.0%) had negative QALY weights (which meant worse than death). The QALY weights for the Glau-U were normally distributed (eFigure 1 in the Supplement) and ranged from –0.98 to 0.92.

Glau-U demonstrated good criterion validity, with the mean (SD) QALY weights decreasing as the level of better-eye glaucoma severity increased (none: 0.73 [0.21]; mild: 0.66 [0.21]; moderate: 0.66 [0.20]; severe: 0.60 [0.28]; and advanced or end-stage: 0.22 [0.38]; P < .001) (Table 4). Based on the reference QALY weight (0.92), the corresponding mean values decreased by 20.7% for none, by 28.3% for mild, by 28.3% for moderate, by 34.8% for severe, and by 76.1% for advanced or end-stage glaucoma. The mean (SD) Glau-U QALY weights also decreased from 0.67 (0.23) for none to 0.58 (0.32) for mild to 0.46 (0.29) for moderate to severe vision impairment in the better eye. Similar associations between Glau-U QALY weights and worse-eye glaucoma and vision impairment were also observed. In contrast, the EQ-5D utilities did not significantly differ across better- or worse-eye glaucoma severity or vision impairment. Convergent validity of the Glau-U was supported by moderate correlations with both better-eye (r = 0.34; P < .001) and worse-eye (r = 0.33; P < .001) mean deviation scores. Divergent validity was supported by low correlations with the EQ-5D (better eye: r = 0.22; P < .001).

In this cross-sectional study of a clinical, multiethnic Singaporean population with glaucoma, we developed and validated the preference-based Glau-U, which comprised 6 glaucoma-specific QOL attributes. Glau-U was found to be capable of estimating disutility associated with mild, moderate, and advanced stages of glaucoma and glaucoma-related vision impairment (criterion validity) and revealed substantial decrements in QALYs across the spectrum of glaucoma severity, particularly the late stages of disease (76.1%). These results suggest that Glau-U may be used to generate health utilities associated with glaucoma and to analyze the cost-effectiveness of novel glaucoma treatments.

We observed substantial decreases in the mean Glau-U QALY weights as glaucoma severity worsened, particularly for those with late-stage disease in which the weights were 41% lower (0.22 vs 0.73) compared with patients with preperimetric glaucoma (ie, the best health state). Similar to findings from other studies,^1,3 the results of the present study emphasize the importance of early detection and intervention to prevent or slow glaucoma progression and avert substantial QOL loss.

As expected,^8,9 the EQ-5D did not significantly discriminate between either glaucoma or vision impairment severity in this study, highlighting the limitations of using generic utilities in eye-related studies and supporting the need for a valid glaucoma-specific utility instrument. The debate over generic vs condition-specific utility instruments is ongoing, and Health Technology Assessment guidelines recommend that health economists use both types of instruments to provide well-rounded information to decision makers, especially if generic utility instruments are not sufficiently sensitive.^25,26 We believe that the present study addressed a key limitation of the GUI by including a survival attribute and rescaling the best glaucoma health state to perfect health. Doing so enabled us to match the death and perfect health anchors of the QALY scale, ensuring that the QALYs from Glau-U were comparable to those of generic instruments and, in turn, were appropriate for guiding allocation decisions.

Unlike the GUI¹³ but similar to the Health Utility for Glaucoma-5,¹⁵ Glau-U contains a psychosocial attribute for assessing the socioemotional issues highlighted by the patients with glaucoma in the stage 1 semistructured interviews. We observed a large decrease in QALYs (0.41) between best and worst levels in the social and emotional effects attribute, highlighting its importance to people with glaucoma and supporting previous work.^3,27 Unlike the Health Utility for Glaucoma-5, Glau-U has a separate attribute for lighting and glare given that previous studies in developing QOL item banks in diabetic retinopathy²⁸ and glaucoma²⁹ suggested that lighting was a separate and unidimensional construct. The difference between best and worst levels in lighting and glare was 0.21, suggesting that it is an important attribute to capture.

We found that changes in eye discomfort and other effects of glaucoma and its treatment compared with other Glau-U attributes were not valued by participants. This finding is similar to that in the GUI study, in which these 2 attributes displayed smaller utility differences between levels than other attributes.¹³ However, results of the present study were somewhat unexpected given the known burden of topical medications on ocular comfort, such as stinging, and systemic adverse effects.³⁰ Further research is needed to better understand patient preferences for these attributes.

Although the high proportion (41.0%) of negative Glau-U QALY weights (which meant worse than death) was similar to percentages in other DCE studies,^12,31 it was higher than those in most studies in Western populations^32-34 that used time trade-off values for the EQ-5D (5%-20% negative values). A recent study that compared time trade-off values of EQ-5D health states for 7 Asian populations found that Singapore had the lowest QALY weights and the highest proportion (approximately 60%) of negative values.³⁵ This finding may reflect Singaporean concern about the unfavorable consequences of poor health, such as financial burden to families.³⁵ As such, investigators outside Singapore may consider administering their own DCE survey using Glau-U to develop country-specific weights. Future research is needed to compare utilities generated from Glau-U across different treatment groups and to assess change in postintervention utility scores.

This study has several strengths. The study had a well-characterized sample of people with glaucoma, including many with severe disease, objective assessment of glaucoma using validated grading systems, and a multiphased development and validation process. Another strength was the recruitment of people with glaucoma to estimate the utilities (rather than a general population sample, which may struggle to imagine issues such as socioemotional well-being), convenience, and treatment experiences or adverse effects associated with inaccurate valuations.

This study also has several limitations. First, the model assumed that people would remain in the same visual health state for 5, 10, 15, or even 20 years, which may lack generalizability to real-world disease progression. Second, we did not evaluate the test-retest reliability or the responsiveness of Glau-U to treatments, nor did we explore the association between Glau-U and other visual parameters, such as contrast sensitivity; future work is needed to explore these factors. Third, we did not administer the choice sets in Malay or Tamil, which may mean that the findings are not generalizable to these ethnic minority groups. Fourth, some of the health states presented may be considered as implausible (eg, severe problems for all attributes except social and emotional effects, which had no problems). However, we kept all 729 profile combinations so as not to lose statistical efficiency.

The new, validated Glau-U can estimate disutility associated with glaucoma and related vision impairment across the spectrum of the disease, demonstrating large decrements in QALYs that are associated with late-stage glaucoma. Glau-U could be useful in estimating the cost-effectiveness of alternative interventions for glaucoma and in informing the resource allocation policies for glaucoma and vision loss.

Accepted for Publication: April 15, 2021.

Published Online: June 24, 2021. doi:10.1001/jamaophthalmol.2021.1874

Corresponding Author: Ecosse L. Lamoureux, PhD, Singapore Eye Research Institute (SERI), The Academia, 20 College Rd, Level 6, Singapore 169856 (ecosse.lamoureux@seri.com.sg).

Author Contributions: Dr Lamoureux had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Fenwick, Ozdemir, Htoon, M. L. Tey, Finkelstein, Wong, Lamoureux.

Acquisition, analysis, or interpretation of data: Fenwick, Ozdemir, Man, Baid, Gan, M. L. Tey, Aw, Baskaran, Nongpiur, C. S. Tey, Soon, Sabanayagam, Sng, Husain, Perera, Lun, Aung.

Drafting of the manuscript: Fenwick, Baid, Htoon, Gan, M. L. Tey, Finkelstein, Soon, Sng, Lamoureux.

Critical revision of the manuscript for important intellectual content: Fenwick, Ozdemir, Man, Baid, Gan, Aw, Baskaran, Nongpiur, C. S. Tey, Sabanayagam, Sng, Wong, Husain, Perera, Lun, Aung.

Statistical analysis: Fenwick, Ozdemir, Man, Baid, Gan, M. L. Tey, Sabanayagam, Sng.

Administrative, technical, or material support: M. L. Tey, Aw, Baskaran, Nongpiur, Finkelstein, C. S. Tey, Soon, Wong.

Supervision: Sabanayagam, Perera, Aung, Lamoureux.

Other - Pilot study: Htoon.

Conflict of Interest Disclosures: Dr Sng reported receiving personal fees from Santen, Allergan, Ivantis, Glaukos, and Alcon outside the submitted work as well as grants from Allergan and Glaukos and holding a patent for Paul Glaucoma Implant with royalties paid. No other disclosures were reported.

Funding/Support: This study was funded by grant R666/16/2009 from the Singapore National Eye Centre Health Research Endowment Fund and by Health Services Research Grant 0070/2016 from the National Medical Research Council (principal investigator: Dr Lamoureux; co-investigators: Drs Fenwick, Man, Baskaran, and Aung).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Information: Glau-U and the utility conversion table can be accessed by contacting Drs Fenwick and Lamoureux.