Sage Journals: Discover world-class research

Abstract

Introduction

The Dental Caries Utility Index (DCUI) is a new oral health–specific health state classification system for adolescents, consisting of 5 domains: pain/discomfort, difficulty eating food/drinking, worried, ability to participate in activities, and appearance. Each domain has 4 response levels. This study aims to generate an Australian-specific utility algorithm for the DCUI.

Methods

An online survey was conducted using a representative sample of the adult Australian general population. The discrete choice experiment (DCE) was used to elicit the preferences on 5 domains. Then, the latent utilities were anchored onto the full health-dead scale using the visual analogue scale (VAS). DCE data were modeled using conditional logit, and 2 anchoring procedures were considered: anchor based on the worst health state and a mapping approach. The optimal anchoring procedure was selected based on the model parsimony and the mean absolute error (MAE).

Results

A total of 995 adults from the Australian general population completed the survey. The conditional logit estimates on 5 dimensions and levels were monotonic and statistically significant, except for the second level of the “worried” and “appearance” domains. The mapping approach was selected based on a smaller MAE between the 2 anchoring procedures. The Australian-specific tariff of DCUI ranges from 0.1681 to 1.

Conclusion

This study developed a utility algorithm for the DCUI. This value set will facilitate utility value calculations from the participants’ responses for DCUI in economic evaluations of dental caries interventions targeted for adolescents.

Highlights

Preference-based quality-of-life measures (PBMs), which consist of a health state classification system and a set of utility values (a scoring algorithm), are used to generate utility weights for economic evaluations.

This study is the first to develop an Australian utility value set for the Dental Caries Utility Index (DCUI), a new oral health–specific classification system for adolescents.

The availability of a utility value set will enable using DCUI in economic evaluations of oral health interventions targeted for adolescents and may ultimately lead to more effective and efficient planning of oral health care services.

Keywords

dental caries adolescents valuation discrete choice experiments preference based visual analogue scale

Introduction

Cost-utility analysis (CUA) is the preferred type of full economic evaluation by most health technology assessment authorities to assess health care interventions across different disease areas.^1,2 Quality-adjusted life-years (QALYs) are the most common outcome measure in CUA³ and allow for a comparison of health interventions in terms of QALY gain or loss for an incremental cost. The QALY is a summary outcome measure that combines utility values assigned to the health states and the duration of time spent in each health state in the condition of interest.⁴ These health state utility values represent individuals’ preferences for different health states and are presented on a full health-dead (1–0) scale.⁵

Preference-based quality-of-life measures (PBMs) can be used to derive utility values for QALYs in economic evaluations.^5,6 PBMs consist of a health state classification system and a set of utility values (a scoring algorithm). The classification system includes health states described in terms of dimensions with response levels relevant to general health or specific disease conditions. The scoring algorithm enables the generation of utility values for these health states.⁵ The algorithms are generated using the values for selected health states in the classification system, estimated from a sample of patients or general population preferences using a preference-elicitation method. Once these values are obtained, statistical modeling techniques are used to generate the scoring algorithm and estimate the utility values for all possible health states derived from the classification system.

Most of the existing pediatric oral health–related quality-of-life (OHRQoL) instruments are non–preference based (non-PBM). They cannot be used to calculate utility values in health economic evaluations.⁷ The available oral health–related pediatric PBMs, the Early Childhood Oral Health Impact Scale–4D (ECOHIS-4D), and the Caries Impacts and Experiences Questionnaire for Children classification system (CARIES-QC-U) are targeted at children of very young age and a wider age group of 5 to 16 years, respectively.^8,9 Therefore, in this study’s first phase, a classification system for dental caries, the Dental Caries Utility Index (DCUI), was developed targeting adolescents older than 12 y.¹⁰ The aim of the present study is to value health states defined by the DCUI classification system. A detailed protocol for this study has been published elsewhere.¹¹

Methods

DCUI Classification System

The DCUI classification system was developed based on a literature review, qualitative interviews, and expert opinion, the accepted methodologies to develop a classification system for a PBM. A systematic literature review of pediatrics OHRQoL instruments⁷ and a quick literature search on studies eliciting utility values for oral health outcomes were performed to identify possible domains and items to develop a draft DCUI classification system. The draft classification system was then refined based on semi-structured interviews with a convenience sample of fifteen 12- to 17-year-old adolescents with dental caries experience and a group of dental experts using the modified Delphi technique. A detailed description of DCUI development is reported elsewhere.¹⁰ The finalized DCUI classification system consists of 5 items: pain/discomfort, difficulty eating food/drinking, worried, ability to participate in activities, and appearance, and each item comprises 4 levels (Table 1).

Table 1

Dental Caries Utility Index (DCUI) Classification System^a

Dimension	Description
Pain/discomfort	1. I have no pain or discomfort 2. I have a little pain or discomfort 3. I have quite a lot of pain or discomfort 4. I have lots of pain or discomfort
Difficulty eating food/drinking	1. I have no difficulty in eating food/drinking 2. I have a little difficulty in eating food/drinking 3. I have quite a lot of difficulty in eating food/drinking 4. I have lots of difficulty in eating food/drinking
Worried (e.g., about losing a tooth, etc.)	1. I am not worried 2. I am a little bit worried 3. I am quite worried 4. I am very worried
Ability to participate in activities (e.g., playing with your friends, sports, school work, etc.)	1. I have no difficulty participating in activities 2. I have a little difficulty participating in activities 3. I have quite a lot of difficulty participating in activities 4. I have lots of difficulty participating in activities
Appearance	1. I am not concerned about my appearance 2. I am a little concerned about my appearance 3. I am quite concerned about my appearance 4. I am very concerned about my appearance

Retrieved from Hettiarachchi et al.¹¹

Preference Elicitation Technique

Discrete choice experiments (DCEs) have recently emerged as a preferred method for health state valuations.¹² The methodological aspects of DCE valuations have been tested widely and used in health state valuation studies. DCE surveys are typically compatible with online surveys; thus, they are less time- and resource-consuming than traditional valuation methods, such as the standard gamble (SG) and the time tradeoff (TTO) methods.¹³ Furthermore, compatibility with the online mode allows large representative samples to be collected within a shorter duration, and DCE methods have moved forward methodologically for the health state valuation over the recent past. However, DCEs produce utility values on a latent scale. These values should be anchored onto the full health-dead scale to calculate the QALYs.¹⁴ Several methods, such as DCE with duration, use of external data from a concurrent TTO study, anchoring with visual analogue scale (VAS), and use of immediate death as a health state, have been applied to anchor latent DCE values onto the full health-dead scale.¹² VAS health state valuation tasks are comparatively easy to understand and have less burden on participants than the SG and TTO methods do.¹⁵ The VAS approach has been used to generate utility values in several health state valuation studies^16,17 and oral health studies.¹⁸ It has also been used solely as an anchoring approach to rescale DCE utility values onto a full heath–dead scale.¹⁹ However, the lack of a tradeoff between life-years and risk, which does not require preferences of life-years over health, is a concern for VAS. Therefore, 2 online pilot surveys were conducted prior to the main survey to evaluate the feasibility and suitability of DCE anchoring with a VAS approach (DCE_VAS) and DCE with duration approach (DCE_TTO). A sample of the adult Australian general population completed the 2 online surveys for DCE_TTO (n = 200) and DCE_VAS (n = 191). Pilot data analyses indicated more ordered and significant coefficients from the DCE_VAS approach compared with the DCE_TTO models. The utility decrements obtained from the DCE_TTO approach were more dispersed. Moreover, the DCE_TTO approach valued health states lower than the DCE_VAS approach; in particular, severe health states were valued worse than death. DCE anchoring with a VAS approach is preferable over DCE_TTO for a dental caries health state valuation and was consequently chosen for the main survey. The detailed results of this comparison study will be published elsewhere.²⁰

DCE Experimental Design and Choice Tasks

The DCUI classification system has 1,024 (4⁵) possible health states (Table 1). As the full factorial design is not feasible, a widely used D-efficient design was adopted with zero priors to select a manageable 200 pairwise choice tasks of these health states, which were further blocked into 25 versions using Ngene software.²¹ Therefore, each participant completed 8-DCE choice tasks from the design. In addition, the survey included a practice choice task and a task with a dominant choice at the beginning to familiarize the participants with DCE tasks. Although there is no consensus in the literature about the number of choice sets per respondent, most DCE studies have used 8 choice tasks per respondent.²² Also, the DCE design in the present study is in line with a recent structured review, which identified that most health state valuations studies using DCE included more than 151 choice tasks,¹² the number of choice tasks per respondent ranged from 2 to 108,¹² and also nonzero prior values are not essential for optimal study design.²³ An example of a DCE choice task is presented in Figure 1.

Figure 1

An example choice set from the discrete choice experiment valuation task.

VAS Task

The VAS task included 6 health states in a single VAS scale: best health state (best level, level 1 of each attribute denoted by 11111), worst health state (worst level, level 4 of each attribute denoted by 44444), mild health state (22211), moderate health state (33341) severe health state (44431) defined by the DCUI classification system, and death (Figure 2). The mild, moderate, and severe health states were selected to represent combinations of attribute levels with differing severity. The 2 extreme endpoints of the VAS scale were calibrated as “best imaginable oral health” (score 100) and “worst imaginable oral health” (score 0).

Figure 2

The visual analogue scale (VAS) valuation task.

Study Sample

Several options exist to generate utility algorithms for pediatric PBMs, including eliciting preferences from children and adolescents, or from the general adult population, or based on proxy responses from parents/guardians.⁶ Eliciting preferences from a pediatric sample may be a better option to understand their preference given they tend to have different preferences for health states than adults do.^24,25 However, it is methodologically challenging, as the DCE elicitation tasks require higher cognitive ability compared with the other methods, such as best-worst scaling.²⁶ Another major concern is the ethical issues associated with presenting tasks with the concept of “dead.” Consequently, it is more common to elicit preferences from adult general population samples to generate utility algorithms for pediatric PBMs.^27,28 This approach is often justifiable, as taxpayers (adult general population members) should have greater influence in deciding which health interventions are funded through public health systems. Furthermore, the anchoring task based on the VAS also required participants to value “death” in relation to the 5 other health states defined by the DCUI classification system. One of the main concerns over general population health state valuation is that they do not have any experience with the disease condition.⁶ However, dental caries is one of the most prevalent conditions among children and adults. According to the National Adult Oral Health Study 2017–18 data, 32% of Australian adults older than 15 y had untreated dental caries, and 77% of those 15 y and older had 1 or more filled teeth.²⁹ Hence, adults are more likely to experience dental caries and would provide more reliable values for the dental caries health states defined by the DCUI, compared with a disease condition for which they potentially have no experience. Furthermore, choice tasks were designed from the participants’ own health perspective to allow adults to call on their own experiences, as imagining a young person experiencing the state is challenging.³⁰ The DCUI is intended for adolescents older than 12 y, and most children have all of their permanent teeth by the age of 12 y. Therefore, the health states defined by the DCUI are relevant for the permanent dentition status of both adolescents and adults. Moreover, the dental caries symptoms do not differ across age groups. Considering these facts, the present study elicited preferences for health states from an adult general population sample and their “own health” perspective.

Sample Recruitment and Survey Administration

According to Lancsar and Louviere,³¹ 20 respondents per choice set and a sample size of 1,000 to 2,000 is sufficient to estimate a reliable model. Although the use of a larger sample was planned during the study protocol, due to time and financial limitations, the health state valuation study aimed to recruit a sample size of 1,000 members from the Australian general population. This sample size was compatible with the sample size requirement and also with other Australian online DCE surveys.³² A representative sample of the adult Australian general population in relation to age and gender was recruited from an Australian online panel via a research company SurveyEngine (http://www.surveyengine.com). Australian general population age and sex distribution were obtained from Australian Bureau of Statistics data.³³

Quotas were set for age and sex based on these data during recruitment to derive an age- and sex-representative sample of the adult Australian general population. Invitations were sent to the potential participants registered with the survey company, and interested participants accessed the survey using the provided link. The online survey started with the introduction page, where participants were given all the necessary details about the research project and then requested to provide consent for data collection and use. Those who consented were guided through the online survey by screen prompts. The next section of the survey included screening questions (age and sex) to ensure the inclusion of a representative sample of the Australian general population. The participants were then asked to complete the DCUI classification system concerning their oral health as a warm-up task to familiarize them with the words used in the choice tasks. The next section included a set of DCE choice tasks, a practice choice task and a task with a dominant choice, to familiarize the participants with DCE tasks. These were followed by the 8-DCE tasks and the VAS anchoring task. The participants were then asked to complete a generic preference-based quality-of-life (QoL) instrument (EQ-5D-5L),³⁴ an oral health–specific QoL instrument (Oral Health Impact Profile-14; OHIP-14),³⁵ and questions regarding their sociodemographic characteristics, oral health status, and frequency of dental visits. The OHIP-14 and EQ-5D-5L are the most frequently used instruments in oral health research to evaluate oral health and general health–related QoL in Australian adults.^36,37

Data Analysis

Sample characteristics

The sociodemographic characteristics of the sample (e.g., age, sex, education level, and income) were compared with those of the Australian general population. Chi-square tests were used to assess the representativeness of the study sample against the Australian general population, which were drawn from Australian Bureau of Statistics data^33,38 as well as the Household, Income and Labour Dynamics in Australia (HILDA) Survey wave 16.³⁹

Utility estimation

The data were analyzed using the STATA version 15.1⁴⁰ statistical software package. Data from DCE choice tasks were modeled using a conditional logit model under a random utility framework, as specified in equation 1. The random utility framework assumes respondents choose the option that maximizes their utility.⁴¹ The utility function consists of a vector of observable attributes and a random error term.⁴¹

μ_{ijk} = β'_{ijk} x_{ijk} + ε_{ijk}

(1)

where μ_ijk is the the utility individual i would get from choosing option k in the choice set j; $β_{i j k}$ is an estimate of the utility associated with the level of each dimension in x_ijk; $x_{ijk}$ is a vector of binary dummy variables $(x_{ijk}^{12}, x_{ijk}^{13}, \dots, x_{ijk}^{54})$ , representing each level of 5 DCUI health attributes (with “no problems” serving as the reference level within each attribute); and $ε_{ijk}$ is the unobservable random error term. Furthermore, a model was also estimated using mixed logit (equation 2) to evaluate the preference heterogeneity.

μ_{ijk} = η'_{ijk} x_{ijk} + β'_{ijk} x_{ijk} + ε_{ijk}

(2)

where μ_ijk is the the utility individual i would get from choosing option k in the choice set j; η_i is the variability term; $β_{i j k}$ is an estimate of the utility associated with the level of each dimension in x_ijk; $x_{ijk}$ is a vector of binary dummy variables $(x_{ijk}^{12}, x_{ijk}^{13}, \dots, x_{ijk}^{54})$ , representing each level of 5 DCUI health attributes (with “no problems” serving as the reference level within each attribute); and $ε_{ijk}$ is the unobservable random error term.

The best-fit model was selected based on the statistical significance, logical order of the coefficients, and Akaike’s information criterion and Bayesian information criterion. The base level of the effect-coded model was recovered using the equation 3 prior to anchoring.⁴²

[L_{1} = - 1 * (L_{2} + L_{3} + L_{4}); L = level]

(3)

Anchoring

The VAS data first underwent a quality check to ensure meaningful results were attained; that is, participants needed to give logical answers regarding the death and best or worst health states. Strict exclusion criteria were applied to remove all relevant logical inconsistencies to achieve meaningful results in the anchoring procedure. The participants were excluded if they met any of the following criteria:

Participants valued best ≤ death or worst or mild or moderate or severe.

Participants valued death ≥ mild or moderate.

Participants valued worst ≥ mild or moderate, or severe.

Participants were missing either death or best or worst VAS data (due to a technical issue in the VAS slider task, some missing data exist for participants who initially completed the survey).

The raw VAS scores of the worst, mild, moderate, and severe health states were converted to the full health–dead scale based on equation 4 at the participant level proposed by Brazier et al.,¹⁵ as used in previous studies.¹⁶

V_{h} = \frac{S_{h} - S_{dead}}{S_{11111} - S_{dead}}

(4)

where V_h is the adjusted VAS rating for health state h, S_h is the respondent’s unadjusted VAS score for state h, S_dead is the respondent’s assigned VAS score for the health state “death,” and S₁₁₁₁₁ is the respondent’s assigned VAS score for the health state 11111 (best state).

DCE data were anchored to the full health–dead scale using adjusted VAS scores via 2 methods: 1) anchoring with the worst (PITS) health state and 2) anchoring by mapping DCE onto VAS. DCE data were anchored to the worst-state value obtained in VAS using the equation 5 adopted from Rowen et al.⁴³ (in which the TTO instead of VAS was used):

β_{r λ \partial} = β_{λ \partial} * \frac{wVAS}{wDCE}

(5)

where β_rλ_ə is the rescaled coefficient for level λ of dimension ə, β_λə is the coefficient for level λ of dimension ə, wVAS is the estimated VAS value for the worst state generated using equation 4, and wDCE is the DCE value for the worst state estimated using the DCE model.

For the anchoring with mapping DCE onto VAS using linear regression, the mean VAS values obtained for the worst health state (44444), a mild health state, a moderate health state, and a severe health state of the DCUI were used as shown in equation 6:

\begin{matrix} dVA S_{j} = 1 - VA S_{j} \\ dDC E_{j} = 1 - DC E_{j} \\ dVA S_{j} = f (dDCEj) + ε_{j} \end{matrix}

(6)

where VAS_j is the mean VAS value of health state j, DCE_j is the modeled latent utility value of health state j, dVAS_j is the mean VAS disutility value of health state j, dDCE_j is the modeled latent disutility value of health state j, and ε_j is the error term.

The optimal anchoring procedure among anchoring with the worst (PITS) health state and anchoring by mapping DCE onto VAS was selected based on the model parsimony (e.g., logical order of the coefficients) and the goodness-of-fit mean absolute error (MAE). The MAE provides a measure of the fit of predicted values relative to observed or criterion values, with a small MAE being preferred.⁴⁴ Since there are no directly elicited health state utilities for the vast majority of DCUI health states, VAS health state utilities for the small number of health states included in the VAS task were used as the criterion to evaluate the MAE for the 2 anchoring approaches. The utility algorithm for the DCUI health states was developed based on the approach with a smaller MAE.⁴⁵

Results

A total of 995 participants completed the survey. Table 2 shows the sociodemographic characteristics, health characteristics, and self-reported oral health data of the study participants. The sample was more educated and had a higher weekly household income than the Australian general population.

Table 2

Characteristics of the Study Sample (n = 995)

Characteristic	Sample Number	%	Population Value^a	χ² Statistic	P Value^b
Mean age, y (SE)	46.87 (0.583)
Age, y
18–29	210	21.11	21.76	0.83	0.98
30–39	192	19.30	18.55
40–49	164	16.48	16.62
50–59	161	16.18	15.60
60–69	129	12.96	13.22
70 or older	139	13.97	14.25
Sex
Male	479	48.14	49.2	0.45	0.50
Female	516	51.86	50.8
Highest level of education
Year 11 or below	115	11.56	17.45	600.56	<0.001
Completed high school	155	15.58	12.95
Trade certificate	136	13.67	25.45
Diploma/advanced diploma	160	16.08	10.8
Bachelor’s degree and above	429	43.12	33.4
Gross weekly household income, AUD
Less than $1,000	329	33.07	29.23	128.22	<0.001
$1,000–2,999	360	36.18	47.46
$3,000–4,499	139	13.97	14.69
$4,500–5,999	75	7.54	4.91
$6,000 or more	92	9.25	3.71
Marital status,^bn (%)
Never married	269	27.04	35.03	57.91	<0.001
Married/de facto	590	59.30	48.05
Divorced/separated	96	9.65	11.74
Widowed	29	2.91	5.18
Prefer not to say	11	1.11	—
DCUI score, $\bar{x}$ (s)	8.01 (3.12)		n/a
EQ5D utility score, $\bar{x}$ (s)^c	0.74 (0.30)		n/a
EQ-5D VAS score, $\bar{x}$ (s)	73.84 (19.03)		n/a
OHIP14 score, $\bar{x}$ (s)	26.21 (13.45)		n/a
Oral health status
Excellent	145	14.57
Very good	277	27.84
Good	315	31.66
Fair	173	17.39
Poor	85	8.54
Degree of tooth or mouth bother in everyday life
Not at all	381	38.29
A little	363	36.48
Some	131	13.17
A lot	58	5.83
Very much	62	6.23
Dental visits			n/a
Once every 6 mo	442	44.42
When there is an oral health problem	459	46.13
Never	94	9.45
Private insurance for dental treatments			n/a
Yes	484	48.64
No	511	51.36
Experienced toothache in the past 6 mo			n/a
Yes	326	32.76
No	669	67.24

Australian age and sex distribution, gross weekly household income, and registered marital status were derived from Australian Bureau of Statistics data.^33,38 Education levels were derived from the Household, Income and Labour Dynamics in Australia Survey wave 16.³⁹

Chi-square goodness-of-fit test was used to compare observed values with general population values with continuity correction for categorical data.

EQ-5D-5L utility weights were derived from Australian utility weights for EQ-5D-5L.⁵¹

DCE Data Estimates

DCE choice data were modeled using conditional logit. Supplementary Table 1 reports the parameter estimates from the unadjusted dummy coded, conditional logit model, including all participants (n = 995). All coefficients were in the expected direction (all coefficient moves from level 1 of each dimension were valued negatively) and in the expected order for all 5 dimensions. All coefficients were significant except for level 2 of the dimensions “worried” and “appearance.”

Analysis of VAS Data

In total, 521 participants were included in the VAS data analysis. The participants included in the VAS data analysis were significantly different from those who met the VAS exclusion criteria concerning age, education level, and oral health status (Supplementary Table 2). Raw VAS scores for the worst, mild, moderate, and severe health states were adjusted to the full health–dead scale in VAS at the individual level using equation 4, and the mean scores were taken as the adjusted VAS scores for each health state. Supplementary Table 3 shows the unadjusted and adjusted mean VAS values for each health state.

Model Selection

For sensitivity analysis, DCE data were modeled by excluding the data of the participants who met VAS exclusion criteria, and the coefficients were estimated using a conditional logit model (Supplementary Table 1). As can be seen, except for the coefficient for level 2 of “appearance,” all coefficients estimated in model excluding participants who met VAS exclusion criteria were in the expected direction. Coefficients for each dimension were in the expected order. Of the 15 coefficients estimated from this model, 2 coefficients (level 2 of the dimensions “worried” and “appearance”) were nonsignificant. The exclusion of the DCE data of the participants who met VAS exclusion criteria did not improve the model estimates substantially, indicating that it was not necessarily correct to assume that those who failed the VAS task could not complete the DCE tasks. Furthermore, exploring other model types with total sample data (mixed logit; Supplementary Table 1) also did not differ substantially from conditional logit model and evaluating preference heterogeneity is a secondary concern of this study.

Therefore, anchoring of the DCE coefficients onto a full health–dead scale was performed based on conditional logit model.

Anchoring Modeled DCE Coefficients to the Full Health–Dead Scale Using VAS Data

The steps of rescaling based on the worst health state and mapping were derived from the studies to develop a new scoring algorithm for CHU9D.^45,46 The base level was recovered in the effect coded, conditional logit model (Supplementary Table 4). To rescale it based on the PITS health state, effect-coded coefficients in the DCE latent scale were rescaled based on the adjusted mean VAS score for the worst health state, 0.1264 (Supplementary Table 3). The calculated utility values for 4 health states (22211, 33341, 44431, and 44444) in the DCE latent scale (Supplementary Table 4) were 0.8263, 0.2610, 0.0946, and 0.0000, respectively. To map VAS onto DCE, these calculated utility values were regressed onto the adjusted VAS scores shown in Supplementary Table 3, as outlined in equation 6. The estimated coefficient was 0.8319. Table 3 shows the coefficient estimates from effect-coded model and rescaled coefficients based on the worst health state of VAS data and mapping VAS onto DCE.

Table 3

Estimated and Rescaled Coefficients

	Level	Estimated Coefficient^a	Rescaled Coefficient Based on Worst State	Rescaled Coefficient Based on Mapping Approach
Pain/discomfort	1	0.7409	0.3399	0.2827
	2	0.4682	0.2615	0.2081
	3	–0.5473	−0.0301	−0.0695
	4	–0.6617	−0.0629	−0.1008
Difficulty eating/drinking	1	0.3468	0.2103	0.1750
	2	0.1737	0.1606	0.1276
	3	–0.2376	0.0425	0.0152
	4	–0.2828	0.0295	0.0028
Worried	1	0.1696	0.1521	0.1265
	2	0.0861	0.1281	0.1037
	3	–0.0485	0.0894	0.0669
	4	–0.2071	0.0439	0.0235
Ability to participate activities	1	0.2188	0.1682	0.1400
	2	0.0539	0.1209	0.0949
	3	–0.1278	0.0687	0.0452
	4	–0.1449	0.0638	0.0405
Appearance	1	0.1020	0.1299	0.1080
	2	0.1010	0.1296	0.1078
	3	–0.0344	0.0907	0.0707
	4	–0.1686	0.0521	0.0340

Conditional logit model, effect coded (Supplementary Table 2).

The utility decrements from Table 3 provided the weights for calculating utility values for the health states defined by the DCUI. The goodness-of-fit MAE value indicated that the DCE estimates based on the mapping approach had the lowest MAE and thus performed better than the estimates based on the PITS approach (Table 4). Therefore, the utility algorithm was developed based on the rescaled coefficients from the mapping approach.

Table 4

Comparison of 2 Rescaling Approaches

Health State	DCUI Classification	VAS Scores	Latent DCE Estimates	Rescaled Estimates Based on PITS Approach	Rescaled Estimates Based on Mapping Approach
1	22211	0.7253	0.8263	0.8483	0.8555
2	33411	0.4333	0.2610	0.3544	0.3852
3	44431	0.2786	0.0946	0.2091	0.2468
4	44444	0.1263	0.0000	0.1264	0.1681
MAE	—	—	—	0.0679	0.0629

DCE, discrete choice experiment; DCUI, Dental Caries Utility Index; MAE, mean absolute error; PITS, worst health state; VAS, visual analogue scale.

Figure 3 shows the rescaled coefficients (attribute weights) of each dimension based on the mapping approach. The “pain/discomfort” domain was associated with the highest attribute weight decrements for each level, followed by “difficulty eating food/drinking.” Smaller decrements were observed in the “appearance” domain (mainly between levels 1 and 2). The utility value for a health state can be calculated as the sum of the constant and the relevant coefficient for each dimension. For example, the utility value for the worst health state based on the mapping approach was 0.1681, and it was calculated using a regression coefficient of 0.8319 and the constant 0.1681 (1-0.8319) as

\begin{matrix} HS (44444) = 0.1681 + (- 0.1008) + 0.0028 \\ + 0.0235 + 0.0405 + 0.0340 = 0.1681 \end{matrix}

Figure 3

Rescaled coefficients for the Dental Caries Utility Index.

Discussion

We estimated a utility algorithm for the newly developed PBM for dental caries (i.e., DCUI) using an online survey with DCE and VAS approaches and an age- and sex-representative sample of the adult general population in Australia. The value set estimated from this utility algorithm is the first value set for this newly developed PBM for dental caries. We used conditional logit, and the resulting model was logically consistent for all dimensions in which utility decreases as severity increases. A logically consistent model is required to apply the utility algorithm in policy decisions.²⁷ Therefore, this value set enables the use of DCUI in the economic evaluation of dental caries interventions among Australian adolescents.

The “pain/discomfort” domain had larger attribute weight decrements than the others did, followed by “difficulty eating food/drinking.” This indicates that the average respondent opposed having more severe pain or discomfort. The domains “worried” and “appearance” showed the smallest decrement from level 1 to 2, which indicated that moving between these levels does not significantly affect the utility value. Similar observations were reported in the health state valuation of other oral health–specific PBMs. For example, the ECOHIS-4D health state valuation also reported the highest utility decrement for the “pain” attribute among the 4 attributes of ECOHIS 4D, indicating “pain” as the most important factor for trading off attributes between health states.⁸ Moreover, previous health state valuation studies conducted using TTO and SG methods also showed that pain was an important factor influencing dental health state utility values.^18,47

The DCUI worst health state (lots of pain or discomfort, lots of difficulties eating food/drinking, very worried, lots of difficulties participating in activities, and very concerned about appearance) has a utility weight of 0.1681. Compared with the utility value of the other oral health–specific PBMs, the DCUI worst health state was valued higher than the worst health state of the ECOHIS-4D (0.0376).⁸ Nevertheless, the utility value for the DCUI worst state was lower than the worst health state of the CARIES-QC-U (CARIES-QC 33333 valued at −0.326 in the adolescent value set and −0.402 in the adult value set).⁹ The DCUI worst health state utility weight was also lower than the worst health state values in other CSPBMs developed for nonfatal disease conditions. For example, a general population sample valued the worst health state of the PBM developed for atopic dermatitis in children as 0.36.⁴⁸ The PITS state of the Asthma Quality of Life-5D (AQL-5D) was 0.39.⁴⁹ However, it should be noted that the valuation techniques used in these CSPBMs are quite different from those used in DCUI health state valuation.

The suitability of using VAS in health state valuation and cost-utility analysis is often a debatable topic. VAS is comparatively simple, easy to use, and reliable.¹⁵ However, VAS is often criticized because it does not involve any tradeoff of life-years or risk as in TTO or SG methods.¹⁵ Nevertheless, Parkin and Devlin⁵⁰ suggested that VAS has many favorable features as a health state valuation technique when compared with the other methods based on theoretical and empirical evidence related to VAS. The VAS approach has been used in previous oral health studies as a direct health state valuation method to value dental health states’ utilities.¹⁸ However, the health states valued in these oral health studies were not based on a classification system. Although the VAS approach has been applied in previous studies to derive preference weights for health states defined by classification systems,¹⁶ it has not been used to anchor the utility values obtained from DCE studies, except for in a recent valuation study for EQ-5D health states.¹⁹ In this study, the authors identified the VAS approach as a feasible method to anchor DCE results to a full health–dead scale using the worst health state and noted that it is more appropriate for a disease condition in which participants may be reluctant to trade off life-years to achieve a better health state.¹⁹ Dental caries is not a life-threatening condition under normal circumstances. Thus, the present study supported VAS as a feasible and reliable approach to anchor the utility values obtained from DCE in a latent scale onto a full health–dead scale. Furthermore, the present study adds that in addition to the rescaling with the VAS worst health state, mapping VAS onto DCE is also another possible method to rescale DCE coefficients onto a full health–dead scale.

This study has some limitations. The valuation sample was large, and an age- and sex-representative sample was taken from the adult Australian general population. However, the sample’s other sociodemographic characteristics significantly differed from the Australian population average. This is commonly reported in other online studies as well.^8,9 The CARIES-QC-U PBM reported both adult and adolescent value sets. The adolescent value set was derived by mapping adolescent best-worst scaling data onto the QALY scale using adult DCE_TTO values.⁹ Although the DCUI measure is intended for adolescents, health state valuations were conducted among an adult sample due to the methodological constraints described in the “Methods” section. The perspective used here is adults on their own health. Therefore, future research is needed to assess whether these health state utility values significantly differ from the preferences elicited from an adolescent sample.

Furthermore, the study was conducted as an online survey. One of the limitations of online surveys is that it is difficult to control the environment and participant engagement; hence, we could not establish real reasons for implausible answers.¹² Participants can complete the surveys in any environment at their convenient time. The implausible answers could be due to the complex nature of the question or simply the lack of participant engagement for a particular question at the time of completion. Some participants in our study also provided logically inconsistent answers for the VAS task, and we were unable to establish reasons for these anomalous responses. However, we performed sensitivity analysis by removing the participants who met VAS exclusion criteria, and modeling suggests data quality was not an issue for the DCE tasks.

Conclusion

This study provides the first utility algorithm for the DCUI PBM for dental caries. This value set will facilitate future utility value calculations from participants’ responses for DCUI in economic evaluations of dental caries interventions.

Supplemental Material

sj-docx-1-mdm-10.1177_0272989X231197149 – Supplemental material for Valuing the Dental Caries Utility Index in Australia

Supplemental material, sj-docx-1-mdm-10.1177_0272989X231197149 for Valuing the Dental Caries Utility Index in Australia by Ruvini M. Hettiarachchi, Sanjeewa Kularatna, Joshua Byrnes, Brendan Mulhern, Gang Chen and Paul A. Scuffham in Medical Decision Making

Footnotes

Acknowledgements

The authors would like to acknowledge the participants of the online survey.

Author’s Note

This research work is a part of Ruvini Hettiarachchi’s doctoral research project submitted to Griffith University, Australia.

Ethics Approval

Ethical approval for this study was obtained from the Human Research Ethics Committee, Griffith University (HREC/2019/550).

Patient Consent

Not applicable as no identifiable data is included in the manuscript.

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Ruvini M. Hettiarachchi

Brendan Mulhern

Paul A. Scuffham

Data Availability Statement

The data that support the findings of this study might be available on request from the corresponding author, upon approval by the ethics review committee.

References

National Institute for Health and Care Excellence (NICE). Guide to the Methods of Technology Appraisal 2013. London: National Institute for Health and Care Excellence (NICE); 2013.

Medical Services Advisory Committee. Technical Guidelines for Preparing Assessment Reports for the Medical Services Advisory Committee—Medical Service Type: Therapeutic (Version 2.0). Canberra (Australia): Department of Health, Australian Government; 2016.

Greenberg

Pliskin

JS.

Preference-based outcome measures in cost-utility analyses: a 20-year overview. Int J Technol Assess Health Care. 2002;18:461–6.

Torrance

GW.

Measurement of health state utilities for economic appraisal. J Health Econ. 1986;5:1–30.

Chen

Ratcliffe

A review of the development and application of generic multi-attribute utility instruments for paediatric populations. Pharmacoeconomics. 2015;33:1013–28. DOI: 10.1007/s40273-015-0286-7

Rowen

Rivero-Arias

Devlin

Ratcliffe

Review of valuation methods of preference-based measures of health for economic evaluation in child and adolescent populations: where are we now and where are we going?

Pharmacoeconomics. 2020;38:325–40.

Hettiarachchi

Kularatna

Byrnes

Scuffham

PA.

Pediatric quality of life instruments in oral health research: a systematic review. Value Health. 2019;22:129–35.

Hettiarachchi

Arrow

Senanayake

, et al. Developing an Australian utility value set for the Early Childhood Oral Health Impact Scale-4D (ECOHIS-4D) using a discrete choice experiment. Eur J Health Econ. Epub ahead of print November 2022. DOI: 10.1007/s10198-022-01542-x

Rogers

Sagabiel

Marshman

Rodd

Rowen

Adolescent valuation of CARIES-QC-U: a child-centred preference-based measure of dental caries. Health Qual Life Outcomes. 2022;20:18.

10.

Hettiarachchi

Kularatna

Byrnes

Chen

Mulhern

Scuffham

PA.

Development of a classification (descriptive) system for a preference-based quality of life measure for dental caries (dental caries utility index) among adolescents. J Public Health Dent. 2022;82:253–61. DOI: 10.1111/jphd.12490

11.

Hettiarachchi

Kularatna

Byrnes

Mulhern

Chen

Scuffham

PA.

Valuation study for a preference-based quality of life measure for dental caries (Dental Caries Utility Index - DCUI) among Australian adolescents - study protocol. BMJ Open. 2020;10:e038626. DOI: 10.1136/bmjopen-2020-038626

12.

Mulhern

Norman

Street

Viney

One method, many methodological choices: a structured review of discrete-choice experiments for health state valuation. Pharmacoeconomics. 2019;37:29–43.

13.

Ali

Ronaldson

Ordinal preference elicitation methods in health economics and health services research: using discrete choice experiments and ranking methods. Br Med Bull. 2012;103:21–44. DOI: 10.1093/bmb/lds020

14.

Brazier

Rowen

Yang

Tsuchiya

Comparison of health state utility values derived using time trade-off, rank and discrete choice data anchored on the full health-dead scale. Eur J Health Econ. 2012;13:575–87. DOI: 10.1007/s10198-011-0352-9

15.

Brazier

Green

McCabe

Stevens

Use of visual analog scales in economic evaluation. Expert Rev Pharmacoecon Outcomes Res. 2003;3:293–302. DOI: 10.1586/14737167.3.3.293

16.

Goudarzi

Zeraati

Akbari Sari

Rashidian

Mohammad

Population-based preference weights for the EQ-5D health states using the visual analogue scale (VAS) in Iran. Iran Red Crescent Med J. 2016;18:e21584. DOI: 10.5812/ircmj.21584

17.

Michel

Augestad

Barra

Rand

A Norwegian 15D value algorithm: proposing a new procedure to estimate 15D value algorithms. Qual Life Res. 2019;28:1129–43. DOI: 10.1007/s11136-018-2043-9

18.

Fyffe

Deery

Nugent

Nuttall

Pitts

NB.

The reliability of two methods of utility assessment in dentistry. Community Dent Health. 1999;16:72–9.

19.

Webb

EJD

O’Dwyer

Meads

Kind

Wright

. Transforming discrete choice experiment latent scale values for EQ-5D-3L using the visual analogue scale. Eur J Health Econ. 2020;21:787–800. DOI: 10.1007/s10198-020-01173-0

20.

Hettiarachchi

RM.

Application of Cost-Utility Analysis in Oral Health: A Preference-Based Quality of Life Measure for Dental Caries among Adolescents [PhD thesis]. Queensland (Australia): Griffith University; 2021.

21.

ChoiceMetrics. Ngene 1.2 User Manual & Reference Guide. Sydney (Australia): ChoiceMetrics; 2018.

22.

Bech

Kjaer

Lauridsen

Does the number of choice sets matter? Results from a web survey applying a discrete choice experiment. Health Econ. 2011;20:273–86. DOI: 10.1002/hec.1587

23.

Mulhern

Bansback

Hole

Tsuchiya

Using discrete choice experiments with duration to model EQ-5D-5L health state preferences: testing experimental design strategies. Med Decis Making. 2016;37:285–97. DOI: 10.1177/0272989X16670616

24.

Thorrington

Eames

Measuring health utilities in children and adolescents: a systematic review of the literature. PLoS One. 2015;10:e0135672. DOI: 10.1371/journal.pone.0135672

25.

Montgomery

Kusel

The prevalence of child-specific utilities in NICE appraisals for paediatric indications: rise of the economic orphans?

Expert Rev Pharmacoecon Outcomes Res. 2016;16:347–50. DOI: 10.1080/14737167.2016.1179116

26.

Rogers

Marshman

Rodd

Rowen

Discrete choice experiments or best-worst scaling? A qualitative study to determine the suitability of preference elicitation tasks in research with children and young people. J Patient Rep Outcomes. 2021;5:26. DOI: 10.1186/s41687-021-00302-4

27.

Rowen

Mulhern

Stevens

Vermaire

JH.

Estimating a Dutch value set for the pediatric preference-based CHU9D using a discrete choice experiment with duration. Value Health. 2018;21:1234–42. DOI: 10.1016/j.jval.2018.03.016

28.

Kreimeier

Greiner

EQ-5D-Y as a health-related quality of life instrument for children and adolescents: the instrument’s characteristics, development, current use, and challenges of developing its value set. Value Health. 2019;22:31–7. DOI: 10.1016/j.jval.2018.11.001

29.

Luzzi

Oral health status. In: ARCPOH Australia’s Oral Health: National Study of Adult Oral Health 2017–18. Adelaide: The University of Adelaide, South Australia; 2019. p 38–96.

30.

Stevens

Valuation of the child health utility 9D index. Pharmacoeconomics. 2012;30:729–47. DOI: 10.2165/11599120-000000000-00000

31.

Lancsar

Louviere

Conducting discrete choice experiments to inform healthcare decision making a user’s guide. Pharmacoeconomics. 2008;26:661–77.

32.

Wang

Chen

Ratcliffe

Afzali

HHA

Giles

Marshall

Adolescent values for immunisation programs in Australia: a discrete choice experiment. PLoS One. 2017;12:e0181073. DOI: 10.1371/journal.pone.0181073

33.

Australian Bureau of Statistics. National, state and territory population. Canberra: ABS; 2019 December. Available from: https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/dec-2019.

34.

EuroQol Group. EQ-5D-5L About. Available from: https://euroqol.org/eq-5d-instruments/eq-5d-5l-about/ [Accessed November 2021]

35.

Slade

GD.

Derivation and validation of a short-form oral health impact profile. Community Dent Oral Epidemiol. 1997;25:284–90. DOI: 10.1111/j.1600-0528.1997.tb00941.x

36.

Brennan

Spencer

AJ.

Dimensions of oral health related quality of life measured by EQ-5D+ and OHIP-14. Health Qual Life Outcomes. 2004;2:35. DOI: 10.1186/1477-7525-2-35

37.

Brennan

Spencer

AJ.

Comparision of generic an a specific measure of oral health related quality of life. Community Dent Health. 2005;22:11–18.

38.

Australian Bureau of Statistics. Australia 2016 Census All persons QuickStats. Canberra: ABS; 2017. Available from: https://abs.gov.au/census/find-census-data/quickstats/2016/0

39.

Wilkins

Lass

The Household, Income and Labour Dynamics in Australia Survey: Selected Findings from Waves 1 to 16. Victoria (Australia): Melbourne Institute, Applied Economic & Social Research, University of Melbourne; 2018.

40.

StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC. Available from: http://www.stata.com

41.

McFadden

Conditional logit analysis of qualitative choice behaviour. In: Zarembka

, ed. Frontiers of Econometrics. New York (NY): Academic Press; 1973. p 105–42.

42.

Ryan

Gerard

Amaya-Amaya

, eds. Using Discrete Choice Experiments to Value Health and Health Care. Dordrecht (The Netherlands): Springer; 2008.

43.

Rowen

Brazier

Van Hout

A comparison of methods for converting DCE values onto the full health-dead QALY scale. Med Decis Making. 2015;35:328–40. DOI: 10.1177/0272989X14559542

44.

Lamers

McDonnell

Stalmeier

Krabbe

Busschbach

JJ.

The Dutch tariff: results and arguments for an effective design for national EQ-5D valuation studies. Health Econ. 2006;15:1121–32. DOI: 10.1002/hec.1124

45.

Ratcliffe

Chen

Stevens

, et al. Valuing child health utility 9D health states with young adults: insights from a time trade off study. Appl Health Econ Health Policy. 2015;13:485–92. DOI: 10.1007/s40258-015-0184-3

46.

Ratcliffe

Huynh

Chen

, et al. Valuing the Child Health Utility 9D: Using profile case best worst scaling methods to develop a new adolescent specific scoring algorithm. Soc Sci Med 2016;157:48-59. DOI: 10.1016/j.socscimed.2016.03.042.

47.

Acharya

Pentapati

Thomson

WM.

Dental health state utilities among dental patients. J Public Health Dent. 2019;79:147–53. DOI: 10.1111/jphd.12306

48.

Stevens

Brazier

McKenna

Doward

Cork

MJ.

The development of a preference-based measure of health in children with atopic dermatitis. Br J Dermatol. 2005;153:372–77. DOI: 10.1111/j.1365-2133.2005.06736.x

49.

Yang

Brazier

Tsuchiya

Young

TA.

Estimating a preference-based index for a 5-dimensional health state classification for asthma derived from the asthma quality of life questionnaire. Med Decis Making. 2010;31:281–91. DOI: 10.1177/0272989X10379646

50.

Parkin

Devlin

Is there a case for using visual analogue scale valuations in cost-utility analysis?

Health Econ. 2006;15:653–64. DOI: 10.1002/hec.1086

51.

Norman

Cronin

Viney

. A pilot discrete choice experiment to explore preferences for EQ-5D-5L health states. Appl Health Econ Health Policy. 2013 Jun;11(3):287–98. doi: 10.1007/s40258-013-0035-z.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.05 MB