The Recovering Quality of Life – Utility Index (ReQoL-UI): the Hong Kong valuation study

Abstract

Objective

This study aims to establish a Hong Kong (HK)-specific value set for the ReQoL-UI, an instrument designed to evaluate health-related quality of life with a focus on mental health, using the discrete choice experiment-time trade-off approach to facilitate cost-utility analyses of mental health interventions.

Method

A web-based survey was conducted from April to August 2024, recruiting 1,008 HK residents aged 20–69 years. Quota sampling was used to ensure age and sex representativeness based on 2021 HK census data. The ReQoL-UI’s seven dimensions with a duration item were evaluated across 180 choice sets divided into 15 blocks, with participants completing 12 tasks each. Utility weights were estimated using a conditional logit and mixed logit models, anchored to a 1 (full health) to 0 (dead) scale, and refined for consistency and statistical fit. Preference heterogeneity was tested using a latent class model.

Results

The HK value set yielded a utility score of -0.870 for the worst health state (i.e., all dimensions at level 5), with physical health dimension contributing the largest decrement (-0.675), followed by dimensions well-being (-0.234) and activity (-0.218). The coefficient for the second level of the “activity” and “well-being” dimensions showed no statistically different from the reference level. For the dimensions of “choice, control, and autonomy” and “hope,” the coefficients for the second and third levels were identical. Sensitivity and preference heterogeneity analyses confirmed the overall stability of the value set.

Conclusion

This study presents the first Asian value set for the ReQoL-UI, enabling quality-adjusted life years calculations to support cost-utility analyses of mental health programs and interventions in HK.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12955-025-02464-x.

Introduction

Healthcare providers increasingly assess patient health through patient-reported outcome measures (PROMs), where patients report their own health-related quality of life (HRQoL) [1]. These PROMs help compare the benefits of different treatments in economic evaluations by measuring their impact via utilities as the indicators of HRQoL [2]. The utilities are then converted into quality-adjusted life years (QALYs), a metric that combines both quality and length of life. The quality-of-life component of QALYs typically comes from either generic or condition-specific preference-based measures (PBMs) [3]. However, currently, EQ-5D, the most common generic preference-based measure, focuses primarily on physical health and includes just one mental health dimension. Research indicates that the EQ-5D may not effectively capture changes in mental health status [4, 5]. The question therefore remains of whether an alternative PBM that mainly emphasizes mental health while still addressing physical health would better facilitate cost-effectiveness analyses in mental health-related interventions and policies. Such a measure might offer psychometric benefits, with greater sensitivity to mental health changes over time and treatment variations.

The Recovering Quality of Life (ReQoL) instrument was developed for adults experiencing a broad range of mental health difficulties from common mental health problems to more severe psychotic ones [6]. Currently, the ReQoL instrument has two versions, the 20-item full and 10-item short version. The ReQoL has been translated into Chinese and showed sound psychometric properties in the Hong Kong (HK) population [7]. To support the economic evaluation of mental health-related interventions and strategies, the Recovering Quality of Life-Utility Index (ReQoL-UI), as a new PBM that aims to capture changes in mental HRQoL, was developed [8]. The ReQoL-UI descriptive system consists of six mental health dimensions activity; belonging; choice, control and autonomy; hope; self-perception; and well-being) and one physical health dimension.

Currently, preference weights for ReQoL-UI health states are not available in HK. Developing these weights for the HK population is crucial to ensure that local cultural, social, and health-related values are accurately reflected in health assessments. Moreover, with rising mental health issues in HK [9], there is a need for tools that can accurately measure the impact of interventions. Local preferences will help healthcare providers and policymakers better understand treatment values from the perspective of those directly affected, guiding decisions to direct resources toward interventions that offer the greatest benefit to mental health and overall well-being. Furthermore, creating HK-specific preference weights for the ReQoL-UI will contribute to global research on mental health quality of life measures. Comparing these weights with those from other regions where ReQoL value sets have been established, such as the United Kingdom and Australia, will provide insights into cultural differences in health perceptions and priorities, similar to cross-country comparisons of value sets for other PBMs, like the EQ-5D. Thus, this study aims to develop a scoring system that generates utility values for all health states as defined by the ReQoL-UI classification system.

Methods

Data and participants

Data used in this study were obtained from a web-based survey conducted in HK between April and August 2024. A survey company, Kantar, which had rich experience in conducting discrete choice experiment (DCE) surveys, was invited to carry out the data collection via its online panel. Interested and eligible members of the online panel received an invitation to the survey on their agency portal. The inclusion criteria were as follows: (1) aged between 20 and 69 years; (2) able to understand Chinese; and (3) HK permanent residents; and (4) able to provide informed consent. Quota sampling by gender and age was used to achieve representativeness of the HK general population for these variables, consistent with methods in similar preference-based measure valuation studies [8]. Participants were divided into 10 age groups from 20 to 69 years (e.g., 20–24 [4.4%], 25–29 [6.2%], and 65–69 [6.6%]) according to the proportions reported by the 2021 HK census data [10]. For each age group, participants were recruited stratified by sex based on the sex ratio of that age group as presented in the 2021 HK census data. Recruitment did not stop until quota samples were met. For our formal survey, a minimum sample size of 1,000 was determined (Appendix, Table 1). The Institutional Review Board of Hong Kong Polytechnic University approved the research protocol (Ref: HSEARS20221024005). All the participants provided written informed consent.

Design of valuation study

DCE was used to develop an algorithm to estimate the preference weights of the HK ReQoL-UI. To model the latent values from the DCE directly anchored onto the 0 (dead) to 1 (full health) scale required for QALY calculations, an additional duration attribution was included in the DCE survey. This technique, called the DCE-time trade-off (DCE-TTO), has been successfully used in several web-based valuation studies [3, 4]. The seven dimensions of ReQoL-UI combined with duration (1, 5, 10, and 15 years) would result in 312,500 possible ReQoL health profiles (5⁷ × 4). The severity levels for each dimension were treated as categorical, while the duration dimension was treated as a continuous variable. The duration levels were determined based on the discussion with local psychologists and psychiatrists, and were selected to balance plausible life expectancies with values spread enough to ensure respondents traded-off between health and life expectancy.

Choice sets were selected using JMP software. First, a number of 200 potential profiles and choice sets were randomly generated. Then, a subset with 3 overlapping items from the classification system was selected, meaning 5 items were different in both profiles in each choice set, to make the choice task easier. These potential choice sets were then optimized and selected using a design that maximizes D-efficiency and accounts for the model specification required to generate preference weights [11, 12]. In total, 180 choice sets (15 blocks) were determined. Each respondent was randomly allocated 12 (one block) of the 180 choice-sets in the survey. The design was tested by examining correlations between attributes within pairs, frequency of each attribute level, and regression results from models estimated on random responses. A sample DCE question is presented in Appendix.

The DCE survey

A pilot with a sample of 100 participants selected from the same online panel of the survey company was conducted to assess the survey formatting and wording. At the start of formal survey, participants were presented with the study information sheet and invited to complete informed consent. After that, participants were asked to provide information about their sociodemographic and health characteristics, including whether they had received mental health services in the last one year or have been formally diagnosed with mental health problems in the last 12 months. Then, participants completed a series of questions, including ReQoL-10, EQ-5D-5 L, and Kessler-10. After that, participants were asked to complete a practice DCE question featuring a dominant profile, with open-ended interpretation provided to explain their response if an incorrect selection was made. After that, participants completed one block of 12 DCE tasks randomly selected from 15 blocks. Finally, participants completed questions about their understanding and what they thought of the survey.

Data quality control

To ensure respondent engagement and understanding, we worked with survey company to incorporate several methods to check the data quality in the questionnaire. First, we used the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) to prevent bot submissions, ensuring that only genuine human respondents participated. We also conducted a time analysis to exclude implausibly rapid responses (≤ 6 or ≥ 20 min), which indicated a lack of thoroughness. Additionally, we limited submissions to one per IP address within a set timeframe to prevent duplicate entries. To further enhance data integrity, we used Qualtrics bot-detection feature to identify and filter out parallel response patterns, such as consistent and repetitive responses, which might indicate inattentive or automated participation.

Data analysis

The DCE data were analyzed using well-established methods widely used in previous studies [13–16]. The analysis used the conditional logit model (CLM) and mixed logit model (MXL) featuring a model specification that incorporated dummy variables. These variables represented each severity level of every item of the ReQoL-UI, interacting with duration (life years), where level 0 (i.e., no problems) served as the reference baseline, alongside an additional duration term.

To derive utility weights on the standardized full health-to-dead scale (ranging from 1 to 0), the marginal rate of substitution was used. Based on the selected model, attribute-level interactions were subsequently anchored to a utility scale of 0 to 1 by dividing the estimated attribute-level interaction coefficients (β) by the coefficient estimated for duration (α). The resulting coefficient ratios (β/α) represent the relative decrement in value compared to the optimal ReQoL state across all dimensions, which served as the reference levels. Consequently, negative coefficients were anticipated, exhibiting a logical progression of decreasing magnitude. From these calculations, ReQoL health state utilities were constructed by adding 1 to the corresponding (negative) utility weights, thereby generating a comprehensive set of values that reflect perceived health states.

Following the methodology of the Australian ReQoL-UI valuation study [17], we developed several models. Model 1, an unadjusted CLM, allowed for illogical orderings of attribute levels, where greater health severity could counter-intuitively increase utility rather than decrease it, defying expectations. To address these inconsistencies, we merged disordered levels with adjacent ones, creating a coherent and consistent Model 2. For Model 3, we introduced a dummy variable (N5), assigned a value of 1 if any attribute reached its worst level and 0 otherwise. Model 4 incorporated three-factor interaction terms, combining duration with two-factor interactions between attributes at level 5 (e.g., AC5 × HO5). Model 5 was an unadjusted model estimated using MXL, similar to Model (1) Model 6 merged disordered levels with adjacent ones in Model 5, similar to Model (2) Model performance was evaluated based on statistical significance, logical consistency of coefficients, and goodness-of-fit metrics, including the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The optimal model was selected for producing more statistically significant coefficients with expected signs (i.e., fewer illogical coefficients) and lower AIC or BIC values.

Robustness of the results was assessed by comparing models estimated using subsamples of the participants, following the method introduced by a previous study [18]. Three types of subsamples were generated: first, excluding participants who reported being diagnosed with mental health problems; second, removing those who may not have understood or engaged (assessed separately using self-reported understanding and difficulty); and last, removing those who completed tasks in less than the 5th and more than 95th percentile of the duration distribution. Preference heterogeneity was evaluated by estimating models that included interaction effects for sex, age, educational level, employment status, self-reported mental health problems, and chronic conditions. The sign and significance of model coefficients were examined and compared across models. These analyses help understand the data but are not intended for inclusion in the value set.

A latent class model (LCM) was also estimated to account for unobserved preference heterogeneity across respondents by identifying distinct classes with different preference patterns. This model assumes that the population consists of a finite number of classes, each with homogeneous preferences, and respondents are assigned to these classes based on their response patterns. The anchored coefficients for best fit model and latent class models were compared for the fully consistent models.

Results

Participants’ demographics

Table 1 presents the sociodemographic characteristics and health status of the participants. The distribution of sex, age, and marital status did not significantly differ from that of the general population. However, compared to the general population in HK, our sample showed significant differences in educational attainment and family income. Participants were more highly educated and had relatively higher incomes. Additionally, 24.3% and 16.2% of respondents reported diagnosed with chronic disease and mental health problems, respectively.

Table 1.

Demographic characteristics of respondents (N = 1008)

Characteristics	N (%)	HK general Population (%)	Difference	P-value of χ2
Age groups (years)				1.000
20–24	62 (6.2)	6.1	0.1
25–29	85 (8.4)	8.5	-0.1
30–34	98 (9.7)	9.8	-0.1
35–39	110 (10.9)	10.9	0
40–44	110 (10.9)	10.9	0
45–49	110 (10.9)	10.8	0.1
50–54	108 (10.7)	10.7	0
55–59	118 (11.7)	11.7	0
60–64	115 (11.4)	11.4	0
65–69	92 (9.1)	9.2	0
Gender				1.000
Male	450 (44.6)	45.6	-1
Female	558 (55.4)	54.4	1
Marital status				0.220
Married	599 (59.4)	58.6	0.8
Never married	349 (34.6)	28.6	6.0
Divorced/Widowed	60 (6.0)	12.7	-6.8
Education level				< 0.001
Primary and below	12 (1.2)	18.4	-17.2
Secondary	402 (39.9)	47	-7.1
Post-secondary	594 (58.9)	34.6	24.3
Family income (1 HKD = 0.13 USD)				< 0.001
< 10,000	46 (4.6)	19.5	-14.9
10,001–20,000	79 (7.8)	18.3	-10.5
20,001–30,000	122 (12.1)	14.8	-2.7
30,001–40,000	147 (14.6)	11.5	3.1
40,001–60,000	290 (28.8)	14.9	13.9
60,001–80,000	172 (17.1)	7.9	9.2
80,001–100,000	84 (8.3)	4.3	4.0
≥ 100,000	68 (6.8)	8.7	-1.9
Employment
Full-time	750 (74.4)
Part-time	94 (9.3)
Unemployed	164 (16.3)
Diagnosed with chronic disease	245 (24.3)
Diagnosed with mental health problem	163 (16.2)
EQ-5D-5 L (No problem)
Mobility	881 (87.4)
Self-care	969 (96.1)
Usual activities	895 (88.8)
Pain/discomfort	433 (43.0)
Anxiety/depression	513 (50.9)
EQ VAS, mean (SD)	76.5 (16.8)
K-10, mean (SD)	19.6 (7.5)

Note: χ^{2 =} Chi-squared test

Feedback on the survey

A total of 2,709 participants clicked the link and joined the survey. Of these, 225 partially completed it, 1,061 completed outside the designated time frame, 315 exceeded the quota, and 100 were identified as bots by Qualtrics platform or failed to correctly answer the trap question. Finally, a sample of 1,008 participants was included in the data analysis. The median completion time was 13 minutes. For the DCE survey, the median duration to complete the 15 DCE questions was 6.3 minutes (range: 2.5–19.9 minutes), with no respondents failing the dominant choice task. A post-experiment survey evaluating participants’ perceptions of the DCE choice tasks (Appendix) indicated minimal comprehension difficulties (5.86%). Additionally, 24.5% of participants rated the choice complexity as ‘Difficult,’ while 1.88% found it ‘Very difficult.’

Estimation of utility score for ReQoL-UI

The estimated coefficients for the DCE-TTO data models are presented in Table 2. In Model 1 (unadjusted conditional logistic model), 22 out of the 29 coefficients were statistically significant. The coefficient for duration was positive, while the remaining 26 coefficients, corresponding to ReQoL-UI dimension levels, were negative. Three coefficients displayed logical inconsistencies. After merging disordered levels, Model 2 achieved logical consistency across all attributes, with coefficients decreasing monotonically as symptom severity increased relative to the reference level (level 1). Model 3 retained coefficient consistency but revealed non-significant interaction terms (e.g., level 5 of symptom severity had no statistically significant effect on Model 2 outcomes). In Model 4, the inclusion of three-way interaction terms altered the significance signs (positive or negative) or disrupted monotonicity for certain attributes, suggesting instability in higher-order interactions. The MXL models outperformed the CML models based on AIC and BIC criteria. In Model 5, 27 of the 29 coefficients were significant, but two were not negative as expected, and three exhibited logical inconsistencies. Model 6 addressed these issues by merging the disordered levels from Model 5, achieving the anticipated logical consistency. The Standard deviation of the coefficients for MXL models are presented in Appendix (Table A2).

Table 2.

Estimated coefficients (standard errors) of the fitted models on DCE-TTO data

Variable	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
Duration	0.308***	0.011	0.309***	0.010	0.310***	0.011	0.294***	0.014	0.519***	0.023	0.531***	0.023
Activity (I enjoyed what I did)
AC2	0.017**	0.006	0.000	NA	0.000	NA	0.000	NA	0.023**	0.009	0.000	NA
AC3	-0.005	0.006	-0.015**	0.005	-0.016**	0.005	-0.017**	0.005	-0.017*	0.008	-0.029***	0.007
AC4	-0.010	0.005	-0.017***	0.005	-0.018***	0.005	-0.014**	0.005	-0.022**	0.008	-0.035***	0.007
AC5	-0.053***	0.007	-0.063***	0.007	-0.063***	0.007	-0.013	0.015	-0.095***	0.011	-0.116***	0.010
Belonging and relationship (I felt lonely)
BE2	-0.017**	0.006	-0.009	0.005	-0.009	0.006	0.001	0.006	-0.025**	0.009	-0.021*	0.009
BE3	-0.022***	0.006	-0.014*	0.006	-0.014*	0.006	0.000	0.007	-0.033***	0.009	-0.030**	0.009
BE4	-0.051***	0.006	-0.041***	0.006	-0.041***	0.005	-0.035***	0.006	-0.077***	0.009	-0.074***	0.009
BE5	-0.062***	0.006	-0.057***	0.005	-0.056***	0.006	-0.067***	0.015	-0.104***	0.009	-0.104***	0.010
Choice, control, and autonomy (I felt unable to cope)
CH2	-0.030***	0.005	-0.025***	0.005	-0.025***	0.005	-0.034***	0.005	-0.048***	0.008	-0.045***	0.007
CH3	-0.010	0.006	-0.025***	0.005	-0.025***	0.005	-0.034***	0.005	-0.034***	0.008	-0.045***	0.007
CH4	-0.049***	0.005	-0.050***	0.005	-0.050***	0.005	-0.052***	0.005	-0.090***	0.008	-0.086***	0.008
CH5	-0.055***	0.007	-0.060***	0.007	-0.059***	0.007	-0.089***	0.015	-0.091***	0.011	-0.099***	0.011
Hope (I thought my life was not worth living)
HO2	-0.035***	0.005	-0.033***	0.005	-0.033***	0.005	-0.026***	0.006	-0.046***	0.008	-0.038***	0.008
HO3	-0.029***	0.007	-0.047***	0.005	-0.047***	0.005	-0.061***	0.006	-0.042***	0.010	-0.038***	0.008
HO4	-0.060***	0.006	-0.047***	0.005	-0.047***	0.005	-0.061***	0.006	-0.094***	0.009	-0.091***	0.009
HO5	-0.054***	0.006	-0.055***	0.006	-0.055***	0.006	-0.115***	0.014	-0.100***	0.010	-0.103***	0.010
Self-perception (I felt confident in myself)
SE2	-0.009	0.006	-0.001	0.005	-0.001	0.005	0.009	0.005	-0.012	0.009	-0.004	0.009
SE3	-0.010	0.006	-0.001	0.005	-0.001	0.005	0.009	0.005	-0.025**	0.009	-0.019**	0.007
SE4	-0.003	0.006	-0.001	0.005	-0.001	0.005	0.009	0.005	-0.016*	0.008	-0.019**	0.007
SE5	-0.054***	0.005	-0.048***	0.005	-0.048***	0.005	0.001	0.013	-0.086***	0.009	-0.088***	0.009
Well-being (I felt happy)
WE2	0.000	0.006	0.000	NA	0.000	NA	0.000	NA	0.007	0.010	0.000	NA
WE3	-0.024***	0.006	-0.023***	0.005	-0.023***	0.005	-0.015**	0.005	-0.048***	0.009	-0.051***	0.008
WE4	-0.031***	0.005	-0.037***	0.004	-0.038***	0.004	-0.039***	0.005	-0.056***	0.008	-0.060***	0.007
WE5	-0.074***	0.006	-0.074***	0.005	-0.074***	0.005	-0.081***	0.014	-0.120***	0.010	-0.124***	0.009
Physical health item
PH2	-0.038***	0.006	-0.041***	0.006	-0.040***	0.006	-0.029***	0.007	-0.056***	0.009	-0.058***	0.009
PH3	-0.085***	0.006	-0.086***	0.006	-0.086***	0.006	-0.077***	0.007	-0.136***	0.010	-0.132***	0.010
PH4	-0.176***	0.006	-0.177***	0.007	-0.177***	0.007	-0.177***	0.007	-0.289***	0.013	-0.295***	0.013
PH5	-0.201***	0.007	-0.205***	0.007	-0.204***	0.008	-0.203***	0.016	-0.349***	0.015	-0.358***	0.015
Interaction terms
N5					-0.001	0.006
Duration: AC5: BE5							-0.003	0.003
Duration: AC5: CH5							0.008	0.004
Duration: AC5: HO5							0.001	0.005
Duration: AC5: SE5							0.003	0.003
Duration: AC5: WE5							-0.001	0.004
Duration: AC5: PH5							0.019***	0.005
Duration: BE5: CH5							0.001	0.004
Duration: BE5: HO5							0.004	0.003
Duration: BE5: SE5							-0.006*	0.003
Duration: BE5: WE5							-0.014***	0.004
Duration: BE5: PH5							0.009*	0.004
Duration: CH5: HO5							-0.018***	0.003
Duration: CH5: SE5							0.018***	0.004
Duration: CH5: WE5							0.009*	0.004
Duration: CH5: PH5							-0.021***	0.006
Duration: HO5: SE5							0.016***	0.003
Duration: HO5: WE5							-0.012***	0.003
Duration: HO5: PH5							-0.014**	0.004
Duration: SE5: WE5							-0.004	0.003
Duration: SE5: PH5							0.008*	0.003
Duration: WE5: PH5							0.007	0.004
AIC	17576.46		17620.82		17622.76		17526.35		16673.84		16661.61
BIC	17817.65		17812.11		17822.37		17892.30		17116.02		17027.55
N	30,240		30,240		30,240		30,240		30,240		30,240
n	1008		1008		1008		1008		1008		1008

Model 1, conditional logit model; Model 2, adjusted conditional logit model; Model 3, conditional logit model with N5; Model 4, conditional logit model with three-factors interactions; Model 5, mixed logit model; Model 6, adjusted mixed logit model

Significance level: p < 0.05 *, p < 0.01 **, p < 0.001 ***

Model 6 was preferred for generating the value set due to its superior performance in terms of monotonicity, statistical significance of coefficients, and model fit (AIC/BIC). The value set, derived from Model 6, is anchored at 1 for full health and 0 for a state equivalent to death (Table 3). The ReQoL-UI value sets for Hong Kong are presented in Table 4; Fig. 1. The relevant importance of each dimension is presented in Appendix. An individual’s health state utility score is calculated by subtracting the sum of the corresponding value set decrements from 1. The worst state (i.e., 5555555) has a value of -0.870 (calculated as 1–0.218–0.197 − 0.187–0.194 − 0.165–0.234–0.675), indicating a condition worse than death.

Table 3.

The Hong Kong ReQoL-UI value set

Variable	Value set decrement
	Coef.	SE
Activity (I enjoyed what I did)
Most of the time	0.000	NA
Often	0.000	NA
Sometimes	-0.055***	0.007
Only occasionally	-0.066***	0.007
Never	-0.218***	0.010
Belonging and relationship (I felt lonely)
Never	0.000	NA
Only occasionally	-0.040*	0.009
Sometimes	-0.056**	0.009
Often	-0.139***	0.009
Most of the time	-0.197***	0.010
Choice, control, and autonomy (I felt unable to cope)
Never	0.000	NA
Only occasionally	-0.084***	0.007
Sometimes	-0.084***	0.007
Often	-0.162***	0.008
Most of the time	-0.187***	0.011
Hope (I thought my life was not worth living)
Never	0.000	NA
Only occasionally	-0.071***	0.008
Sometimes	-0.071***	0.008
Often	-0.172***	0.009
Most of the time	-0.194***	0.010
Self-perception (I felt confident in myself)
Most of the time	0.000	NA
Often	-0.007	0.009
Sometimes	-0.036**	0.007
Only occasionally	-0.036**	0.007
Never	-0.165***	0.009
Well-being (I felt happy)
Most of the time	0.000	NA
Often	0.000	NA
Sometimes	-0.097***	0.008
Only occasionally	-0.114***	0.007
Never	-0.234***	0.009
Physical health item
Never	0.000	NA
Only occasionally	-0.109***	0.009
Sometimes	-0.249***	0.010
Often	-0.556***	0.013
Most of the time	-0.675***	0.015

Significance level: p < 0.05 *, p < 0.01 **, p < 0.001 ***

Table 4.

Latent class model analysis

Class share	Class 1			Class 2			Class 3			Class 4
	0.220			0.423			0.258			0.098
	Coef.	95%CI		Coef.	95%CI		Coef.	95%CI		Coef.	95%CI
Duration	1.363	-1.143	3.868	0.287	-2.218	2.793	0.298	-2.207	2.804	4.589***	2.084	7.095
Activity (I enjoyed what I did)
AC2	0.000	NA	0.000	NA	0.000	NA	0.000	NA	0.000	NA	0.000	NA
AC3	0.179	-0.014	0.372	-0.060	-0.253	0.133	0.011	-0.182	0.203	-0.107	-0.300	0.086
AC4	-0.087*	-0.169	-0.005	-0.032	-0.113	0.050	-0.004	-0.086	0.078	0.063	-0.019	0.144
AC5	-0.482**	-0.819	-0.146	-0.122	-0.459	0.214	-0.015	-0.352	0.321	-0.284	-0.621	0.052
Belonging and relationship (I felt lonely)
BE2	0.193	-0.050	0.435	-0.041	-0.284	0.202	-0.021	-0.264	0.222	-0.255*	-0.498	-0.012
BE3	0.225*	0.007	0.443	-0.040	-0.258	0.177	-0.035	-0.253	0.183	-0.071	-0.288	0.147
BE4	0.115	-0.100	0.330	-0.078	-0.293	0.137	-0.057	-0.272	0.158	-0.291**	-0.506	-0.076
BE5	0.102	-0.282	0.486	-0.078	-0.462	0.306	-0.073	-0.458	0.311	-0.644**	-1.028	-0.260
Choice, control, and autonomy (I felt unable to cope)
CH2	-0.108	-0.282	0.065	-0.019	-0.192	0.154	-0.032	-0.205	0.142	0.229**	0.055	0.402
CH3	-0.108	-0.282	0.065	-0.019	-0.192	0.154	-0.032	-0.205	0.142	0.229**	0.055	0.402
CH4	-0.176*	-0.315	-0.037	-0.051	-0.190	0.088	-0.062	-0.201	0.077	0.085	-0.054	0.224
CH5	-0.236	-0.517	0.044	-0.069	-0.350	0.211	-0.061	-0.342	0.219	0.313*	0.032	0.593
Hope (I thought my life was not worth living)
HO2	-0.009	-0.110	0.092	-0.042	-0.143	0.059	-0.012	-0.113	0.089	-0.191***	-0.292	-0.090
HO3	-0.009	-0.110	0.092	-0.042	-0.143	0.059	-0.012	-0.113	0.089	-0.191***	-0.292	-0.090
HO4	-0.224	-0.556	0.108	-0.067	-0.399	0.266	-0.043	-0.376	0.289	-0.620***	-0.952	-0.287
HO5	-0.297**	-0.509	-0.086	-0.091	-0.303	0.120	-0.022	-0.234	0.190	0.025	-0.187	0.236
Self-perception (I felt confident in myself)
SE2	0.059	-0.129	0.247	0.009	-0.180	0.197	-0.023	-0.212	0.165	-0.296**	-0.485	-0.108
SE3	0.067	-0.146	0.280	-0.011	-0.224	0.202	-0.009	-0.222	0.204	-0.340**	-0.554	-0.127
SE4	0.067	-0.146	0.280	-0.011	-0.224	0.202	-0.009	-0.222	0.204	-0.340**	-0.554	-0.127
SE5	-0.173**	-0.286	-0.059	-0.064	-0.177	0.049	-0.050	-0.163	0.064	-0.200***	-0.313	-0.087
Well-being (I felt happy)
WE2	0.000	NA	0.000	NA	0.000	NA	0.000	NA	0.000	NA	0.000	NA
WE3	-0.072	-0.308	0.164	-0.022	-0.258	0.215	-0.014	-0.250	0.222	-0.430***	-0.667	-0.194
WE4	-0.054**	-0.093	-0.016	-0.027	-0.066	0.011	-0.043*	-0.081	-0.005	0.016	-0.022	0.055
WE5	-0.575*	-1.025	-0.124	-0.071	-0.521	0.379	-0.054	-0.504	0.397	-0.515*	-0.965	-0.065
Physical health item
PH2	-0.132	-1.258	0.994	-0.052	-1.177	1.074	-0.015	-1.141	1.111	-1.986***	-3.112	-0.860
PH3	-0.607	-1.938	0.724	-0.114	-1.445	1.217	-0.027	-1.358	1.304	-2.371***	-3.702	-1.040
PH4	-1.268	-2.869	0.333	-0.215	-1.816	1.386	-0.030	-1.631	1.572	-2.677**	-4.278	-1.076
PH5	-1.516	-3.368	0.335	-0.237	-2.088	1.615	-0.053	-1.905	1.798	-3.064**	-4.916	-1.213
AIC	16219.00
BIC	16705.66
N	30,240
n	1008

Significance level: p < 0.05 *, p < 0.01 **, p < 0.001 ***

Fig. 1.

Utility decrements for the HK version of the ReQoL-UI

Preference heterogeneity and sensitivity analysis

In the LCA model, the three-class solution (class share = 22.0%, 42.3%, 25.8%, and 9.9% for Class 1, 2, 3, and 4) demonstrated the lowest BIC and was thus selected for heterogeneity testing (Table 4). Overall, the results indicated limited preference heterogeneity across classes, with physical health consistently emerging as the most prioritized attribute, consistent with the main model. The characteristics for each class are present in the appendix (Table A3). The sensitivity analysis confirmed the robustness of the coefficients, further validating the stability of the model’s parameter estimates (Table 5).

Table 5.

Sensitivity analysis

Duration	Full sample		Sub-sample
	Coef.	SE	Coef.	SE
	0.531***	0.023	0.620***	0.032
Activity (I enjoyed what I did)
AC2	0.000	NA	0.000	NA
AC3	-0.029***	0.007	-0.029**	0.010
AC4	-0.035***	0.007	-0.036***	0.010
AC5	-0.116***	0.010	-0.130***	0.014
Belonging and relationship (I felt lonely)
BE2	-0.021*	0.009	-0.027*	0.013
BE3	-0.030**	0.009	-0.037**	0.013
BE4	-0.074***	0.009	-0.072***	0.013
BE5	-0.104***	0.010	-0.117***	0.014
Choice, control, and autonomy (I felt unable to cope)
CH2	-0.045***	0.007	-0.050***	0.010
CH3	-0.045***	0.007	-0.050***	0.010
CH4	-0.086***	0.008	-0.101***	0.011
CH5	-0.099***	0.011	-0.099***	0.015
Hope (I thought my life was not worth living)
HO2	-0.038***	0.008	-0.044***	0.011
HO3	-0.038***	0.008	-0.044***	0.011
HO4	-0.091***	0.009	-0.104***	0.012
HO5	-0.103***	0.010	-0.118***	0.014
Self-perception (I felt confident in myself)
SE2	-0.004	0.009	0.008	0.013
SE3	-0.019**	0.007	-0.024*	0.011
SE4	-0.019**	0.007	-0.024***	0.011
SE5	-0.088***	0.009	-0.096***	0.012
Well-being (I felt happy)
WE2	0.000	NA	0.000	NA
WE3	-0.051***	0.008	-0.061***	0.011
WE4	-0.060***	0.007	-0.069***	0.011
WE5	-0.124***	0.009	-0.142***	0.012
Physical health item (Please describe your physical health: problems with pain, mobility, difficulties caring for yourself, or feeling physically unwell)
PH2	-0.058***	0.009	-0.066***	0.012
PH3	-0.132***	0.010	-0.156***	0.014
PH4	-0.295***	0.013	-0.344***	0.019
PH5	-0.358***	0.015	-0.423***	0.023
AIC	16661.61		9582.43
BIC	17027.55		9923.39
N	30,240		30,240
n	1008		1008

Significance level: p < 0.05 *, p < 0.01 **, p < 0.001 ***

Discussion

In this study, we developed the HK value set for the ReQoL-UI using a DCE-TTO approach. To our knowledge, this is the first value set established in an Asian country and the third globally, following those in the UK and Australia. Derived from the preferences of HK population, this value set al.lows for estimating utility values to compute QALYs, thereby becoming an essential tool for conducting CUAs of mental health interventions in HK. Given the cultural and ethnic similarities between HK and other Chinese-majority regions (e.g., mainland China, Singapore, and Taiwan), and the lack of region-specific value sets for these areas, our HK value set can also facilitate relative economic evaluations in these regions, thereby amplifying its broader impact [19]. HK’s healthcare system, facing rising costs and demand, is well-placed to increasingly adopt CUA techniques to inform funding decisions, following global trends in health economics. The development of the ReQoL-UI HK value set could have a considerable influence on resource allocation processes for mental health, both regionally and globally.

Given that UK value set was developed using a different method, we limit our discussion to the HK and Australian value sets, both of which used the DCE-TTO method. The preference weights for the ReQoL-UI from the Australian and HK studies show both consistencies and inconsistencies across the two populations, reflecting variations in valuation methods, cultural preferences, and health state perceptions. The Australian study yielded a utility score of -0.5851 for the worst health state, while our study estimated an even lower utility of -0.870 for the same state. This difference may suggest that people in HK perceive this state as far worse than those in the Australia. This discrepancy likely arises from both methodological and cultural factors. Methodologically, the DCE-TTO approach may produce stronger (more negative) valuations for severe health states, possibly because it emphasizes preference intensity and avoids the direct life expectancy trade-offs inherent in the TTO method [20, 21]. Beyond methods, population-specific attitudes are significant. HK’s low utility value might reflect stigma surrounding mental health or a lower tolerance for mental disability [22], while the Australia’s less severe valuation could indicate greater resilience or differing health expectations. These variations highlight the need for additional valuation studies, particularly in Asian populations, to explore these variations.

Physical health consistently emerges as a key dimension of utility decrements across all three studies, though its relative importance differs. In Australia, physical health was identified as the most important dimension for the general population, which contributed 33% of the utility score, whereas the 6 mental-related items contributed 67% [17]. Similarly, in HK, physical health emerged as the most influential single factor, explaining 37% of the utility score, followed by well-being (17%) and activity (14%). This consistent emphasis on physical health likely may stem from its tangible descriptors (e.g., pain, mobility), which resonate universally. However, HK respondents assign it greater weight, possibly due to cultural or contextual factors, such as heightened sensitivity to physical limitations in a densely populated, fast-paced urban setting.

Australian and HK studies reveal notable differences in the coefficients and relative importance of mental health-related dimensions, reflecting distinct cultural preferences. In the HK value set, well-being carries the largest mental health decrement (-0.239, 17% relative importance), followed by hope (-0.177, 12%), choice, control, and autonomy (-0.193, 11%), and belonging (-0.183, 10%), while self-perception (-0.155 at its worst, with milder levels at -0.003) and activity (-0.205, 14%) are less pronounced. In contrast, the Australian value set assigns the greatest weight to well-being (-0.3215, 22%), followed by hope (-0.2301, 15%), with activity (-0.1900, 13%), choice, control, and autonomy (-0.1465, 9%), and belonging (-0.1341, 6%) having moderate impacts, and self-perception (-0.0263 across all levels) contributing minimally at 1%. HK respondents place higher decrements on autonomy and belonging compared to Australia, suggesting a stronger cultural emphasis on these dimensions, while Australia prioritizes well-being and hope more heavily. These differences contribute to HK’s more severe worst-state utility (-0.8139) versus Australia’s (-0.5851), potentially reflecting greater sensitivity to mental health impairments in HK, possibly due to cultural stigma or urban stressors. The UK’s aggregated MH approach obscures such granularity, limiting direct comparison but highlighting a methodological divergence that may mask similar preference variations.

The HK value sets for the ReQoL and EQ-5D-5L [23] instruments reveal both similarities and contrasts in how health states are valued, particularly in a cultural context where physical impairments appear to carry substantial weight. Both yield comparable utilities for their worst health states, -0.870 for ReQoL and − 0.865 for EQ-5D-5 L, which indicating a shared societal aversion to severe health deterioration that extends below zero. The ReQoL, designed primarily for mental health recovery, incorporates six psychosocial domains with decrements ranging from − 0.165 to -0.234 at their worst levels, alongside a single physical health item that dominates with a -0.675 decrement, emphasizing how even in a mental health-focused tool, physical limitations incur the heaviest penalty in HK preferences. In contrast, the EQ-5D-5 L’s four dimensions prioritize physical aspects more evenly, with mobility as the most impactful, followed closely by pain/discomfort, and self-care, this may suggest a broader physical emphasis compared to ReQoL’s integration of nuanced mental health facets, potentially reflecting HK’s cultural valuation where physical functionality underpins overall quality of life, even amid mental health assessments. Additional research could clarify whether HK’s extreme valuation is unique or reflects broader Asian attitudes toward mental versus physical health.

It is important to consider several aspects of the methodological choices in this study, particularly in the context of evolving standards in health state valuation. First, although the conditional logit model was initially used to capture average preferences, which is appropriate for deriving population-level utility scores, it may not fully account for preference heterogeneity among respondents, potentially leading to biased parameter estimates when such variability is present [24]. To address this, a mixed logit model, which demonstrated marginally better performance, was used for final utility calculations; however, the conditional logit model remains valuable and highlights the need for more tests in future valuations to better handle individual differences [25]. Second, the DCE with duration format used here assumes a multiplicative utility function, consistent with traditional QALY models, but recent study suggests that some respondents may employ additive decision rules for duration and health dimensions, potentially violating QALY assumptions and contributing to inconsistent coefficients [26]. While evidence supporting additive rules remains limited, and we retained the multiplicative approach due to its theoretical consistency and prevalence in the literature, such as all studies of the QLU-C10D valuation, this could introduce bias if additive preferences are more common than assumed [26]. Third, the assumption of linear time preferences in our DCE tasks may not fully reflect real-world discounting behavior. A number of studies have indicated nonlinear preferences can yield better results to describe DCE duration data across various formats and tend to produce value sets that are more consistent with time trade-off results when properly modeled [27, 28]. Although this quantitative evidence favors nonlinear specifications, linear time preferences remain the most commonly applied assumption in health utility valuation studies, including the majority of published MAU-based instruments [29]. We, therefore, adopted the linear assumption comparability with the existing literature, but we recognized that incorporating nonlinear discounting could improve accuracy in future valuations. These issues indicate the importance of methodological advancements, though our sensitivity analyses support the stability of the derived value set.

Three limitations should be addressed in this study. First, in the HK valuation study, the online sampling method yielded a sample with good sex and age representativeness, closely mirroring the distribution of the general HK population. However, education presents a limitation: too many participants reported higher educational attainment and income, skewing the sample away from the general population’s profile. This overrepresentation may exclude perspectives from less-educated groups, potentially those less familiar with technology or with limited internet access. While sex and age were well-balanced, the reliance on online recruitment could still pose subtle barriers, such as digital literacy challenges, though these did not significantly skew the profile. Given the high prevalence of mental health issues in HK, often linked to urban stressors like isolation, future research should aim to adjust sampling to better reflect educational diversity and ensure broader representation. Second, this study sample includes only 16.2% of participants (163 out of 1,008) diagnosed with mental health problems in the past 12 months, which may differ from the actual prevalence in HK’s general population, as we lack such data. This underrepresentation may lead to inaccurate estimation of preference weights for the ReQoL-UI, particularly for mental health dimensions. Since the ReQoL-UI is designed to assess HRQoL for mental health conditions, a sample with fewer affected individuals might undervalue the severity of severe mental health states, skewing utility estimates away from the preferences of those with lived experience. Finally, our sample excludes individuals over 70 years old due to feasibility constraints, which may restrict the study’s ability to capture the full range of mental health perceptions in HK. This exclusion could bias the utility valuations and limit their relevance to older adults, potentially undermining the ReQoL-UI’s validity for cost-utility analyses of mental health interventions in HK, where representing perspectives across all age groups is essential.

Conclusion

This study provides HK-specific preference weights for the ReQoL-UI using a DCE-TTO method. The generated scoring algorithm and value set can be used to assess and compare the impact of mental health-related programmes and technologies on HRQoL in CUAs, thereby facilitating evidence-based resource allocation decisions.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

RHX: Conceptualization; Methodology; Software; Validation; Formal analysis; Investigation; Resources; Data Curation; Writing - Original Draft; Writing - Review & Editing; Supervision; Project administration; Funding acquisition. CXY: Methodology; Software; Validation; Formal analysis; Visualization; Writing - Review & Editing. TT: Methodology; Writing - Review & Editing. ELYW: Conceptualization; Project administration; Supervision. SN: Conceptualization; Project administration; Supervision. RN: Conceptualization; Methodology; Software; Formal analysis; Writing - Review & Editing; Supervision.

Funding

This project supported by Hong Kong Health and Medical Research Funding (Ref ID: 20211211).

Data availability

The data that support the findings of this study are available on request from the corresponding author.

Declarations

Ethical approval

The Institutional Review Board of Hong Kong Polytechnic University approved the research protocol (Ref: HSEARS20221024005). This study complied with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration.

Informed consent

All the participants provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.