Estimating a Preference-Based Single Index from the Overactive Bladder Questionnaire
ABSTRACT
Objectives: The aim of the study is to estimate a preference-based single index for calculating quality-adjusted life years for patients with overactive bladder (OAB), based on a survey of the UK general population using the 5-dimensional health classification system OAB-5D, derived from the validated Overactive Bladder Questionnaire (OAB-q).
Methods: An interview valuation survey of members of public in the South Yorkshire, UK was undertaken using the time-trade-off method. Each respondent was randomly allocated to one of 14 blocks, and valued seven states each plus the “pits” state, so that a total of 99 states were valued. A number of multivariate regression models were estimated for predicting a total of 3125 health state values defined by the classification. Models were compared and selected using a set of criteria, including overall diagnosis by adjusted R-squared, the sign and significance of individual parameter estimates, the relative size of coefficients within a given dimension and predictive ability.
Results: The mean model was recommended for use in economic evaluation.
Conclusion: This will permit the cost-effectiveness of new interventions to be assessed in patients with OAB using existing and future OAB-q data sets.
Introduction
The use of economic evaluation to aid health decision-making, especially health priority setting and health resources allocation, has become widespread. Government agencies have been established to examine the cost-effectiveness of health-care interventions, such as the National Institute for Health and Clinical Excellence (NICE) [1] in England and Wales, the Scottish Medicines Consortium [2] for Scotland and similar agencies in Australia (Pharmaceutical Benefits Advisory Committee) [3] and Canada (Canadian Coordinating Office for Health Technology Assessment) [4]. Cost-effectiveness analyzes increasingly use quality-adjusted life years (QALYs) as the measure of effectiveness to allow comparisons across different health-care interventions for different medical conditions. The QALY is able to achieve this by capturing the impact of interventions on the length of life (in the form of “years”) and/or the quality of life (in the form of “health state values”) into a single summary measure, based on people's preference.
A common way to obtain health state values is to use one of the “off-the-shelf” generic preference-based instruments, such as the EQ-5D [5], HUI3 [6], and SF-6D [7]. These instruments typically consist of two parts: A general health descriptive system and a ready-to-use scoring algorithm based on public preference. These measures produce health state values that lie on a scale, where zero denotes equivalence to being dead and one denotes equivalence to being in full health. The overall gain from treatment in terms of QALYs is calculated as a function of these health state values and their associated durations. Owing to their ease to use and generic nature, which permits comparisons across different health-care interventions and/or different medical conditions, generic preference-based instruments have been widely used. Nevertheless, for some medical conditions, the generic dimensions may be considered to be irrelevant or insensitive in terms of capturing small but important clinical changes [8].
Overactive bladder (OAB), a syndrome characterized by urinary urgency with or without urge incontinence, usually with increased urinary frequency and nocturnal urine [9], is one such medical condition. Given the symptoms are related to urinary urgency, frequency, urgency incontinence, and sleep, generic utility measures such as EQ-5D or SF-6D may not adequately capture the impact on patient's health-related quality of life or be sensitive to change. There is some empiric evidence for this concern where for example, the SF-36 was found to be unresponsive to improvements after pharmacologic treatment for OAB [10].
The purpose of this study was to develop an OAB-specific preference-based measure with a population value set. To do this, the methods which have been successfully used to derive the generic preference-based index SF-6D from the SF-36 [7] were employed using an OAB-specific symptom bother and health-related quality of life instrument, the Overactive Bladder Questionnaire (OAB-q) [11]. Basically, this methodology consists of three stages with the first stage being to select a subset of OAB-q items to serve as the dimensions of a trimmed down descriptive system, and then to decrease the number of severity levels within each dimension. The first stage produces a health state classification system, that is not only amenable to being valued using a preference elicitation technique, but also suffers minimum information loss relative to the original instrument. The second stage is to undertake a survey which asks a representative sample of the general population to value a selection of states defined by the reduced classification system. The third stage is modeling the sample health state values and using these models to predict health state values for all possible states defined by the new classification system. The resulting scoring algorithm or population value set will provide a preference-based single index measure for different OAB-q states and hence for any OAB-q data sets, contributing toward the calculation of QALYs. The first stage is briefly summarized in the next section, and reported elsewhere [12] in more detail, produces a health state classification system OAB-5D. This article will provide the details of the second and the third stage of this three-stage process.
Reducing the OAB-q to Derive a Health State Classification
The OAB-q is a 33-item OAB-specific questionnaire that consists of an eight-item Symptom Bother scale and a 25-item Health-Related Quality of Life scale that has four subscales: coping, concern, sleep, and social interaction [11]. Responses are based on a six-point Likert scale. Among continent and incontinent OAB patients, the OAB-q has demonstrated good internal consistency, reliability, test–retest reliability, concurrent validity, discriminate validity, and responsiveness to treatment-related change [11,13,14]. The OAB-q has been widely used in patients with OAB, and for instance a PUBMED search using OAB-q as the keyword results in 219 articles for years between 1956 and 2007. Nevertheless, the OAB-q cannot be used in cost per QALY analyzes because the scores are not preference-based.
The number of possible health states that the OAB-q can distinguish from each other in theory is 633, which is too large for a valuation survey in practice. Thus, it is necessary to derive a new classification system from the OAB-q which is amenable for valuation mainly using Rasch analysis [15]. Rasch analysis uses a logit model to convert ordinal data into a continuous latent scale. Rasch models were applied to OAB-q data collected from patients with the aim of selecting a number of items for a reduced health state classification. Five items were chosen to construct a new health state classification system (OAB-5D, see Table 1) and they are: Urge to urinate, urine loss, sleep impact, coping strategy, and concern with OAB, where each dimension has five levels of severity with level 1 denoting no problem and level 5 indicating an extreme problem. A total of 3125 (= 55) health states can be defined by the OAB-5D classification system, and all OAB-q data which contain these five items can be mapped to a specific OAB-5D health state (for further details of this article, see the study by Young et al.) [12].
| Urge |
| 1. Not at all bothered by an uncomfortable urge to urinate |
| 2. Bothered by an uncomfortable urge to urinate a little bit or somewhat |
| 3. Bothered by an uncomfortable urge to urinate quite a bit |
| 4. Bothered by an uncomfortable urge to urinate a great deal |
| 5. Bothered by an uncomfortable urge to urinate a very great deal |
| Urineloss |
| 1. Not at all bothered by urineloss associated with a strong desire to urinate |
| 2. Bothered by urineloss associated with a strong desire to urinate a little bit or somewhat |
| 3. Bothered by urineloss associated with a strong desire to urinate quite a bit |
| 4. Bothered by urineloss associated with a strong desire to urinate a great deal |
| 5. Bothered by urineloss associated with a strong desire to urinate a very great deal |
| Sleep |
| 1. Bladder symptoms interfered with your ability to get a good night's rest none of the time |
| 2. Bladder symptoms interfered with your ability to get a good night's rest a little of the time |
| 3. Bladder symptoms interfered with your ability to get a good night's rest some of the time |
| 4. Bladder symptoms interfered with your ability to get a good night's rest a good bit or most of the time |
| 5. Bladder symptoms interfered with your ability to get a good night's rest all of the time |
| Coping |
| 1. Bladder symptoms caused you to plan “escape routes” to restrooms in public places none of the time |
| 2. Bladder symptoms caused you to plan “escape routes” to restrooms in public places a little of the time |
| 3. Bladder symptoms caused you to plan “escape routes” to restrooms in public places some of the time |
| 4. Bladder symptoms caused you to plan “escape routes” to restrooms in public places a good bit or most of the time |
| 5. Bladder symptoms caused you to plan “escape routes” to restrooms in public places all of the time |
| Concern |
| 1. Bladder symptoms caused you embarrassment none of the time |
| 2. Bladder symptoms caused you embarrassment a little of the time |
| 3. Bladder symptoms caused you embarrassment some of the time |
| 4. Bladder symptoms caused you embarrassment a good bit or most of the time |
| 5. Bladder symptoms caused you embarrassment all of the time |
- OAB, overactive bladder.
Methods
Valuation Survey
To elicit preference values for a sample of health states defined by the OAB-5D, a valuation survey was undertaken. The basic design of the survey was that selected OAB-5D health states were valued using the time-trade-off (TTO) technique by a sample of the general public randomly selected from the electoral register in South Yorkshire, UK to be representative in terms of age and sex. General public preferences were sought, because health policy organizations such as NICE recommend using general public values.
Given the logistical difficulties of valuing all 3125 health states, a sample of states were selected for direct valuation, using a balanced design method. In terms of the sample size of health states needed for this kind of work, researchers have not reached an agreement. In this study, 98 health states were selected on the basis of the authors' previous experience with SF-6D and resource constraints. Two criteria were used in the selection process to select those 98 states:
- 1
Any dimension level has an equal chance of being combined with all levels of the other dimensions;
- 2
The resulting health states cover a wide range of the scale with a mix of good, mild, moderate, and severely ill health states.
Furthermore, because the modeling process required a larger number of states to be directly valued than a single respondent can value in one interview, respondents were divided into 14 subgroups, where each subgroup valued a different subset of seven health states. To allocate the 98 states to these 14 respondents subgroups, each state was given a severity score, calculated as the simple sum of the levels across the OAB-5D dimensions, grouped into seven severity groups (each consisting of 14 states of similar severity), and then randomly allocated to one of the 14 respondent subgroups. In addition, the worst possible state was allocated to all 14 respondent subgroups. This way, each respondent valued a set of states balanced in terms of severity, and each state is valued the same number of times (except for the worst possible state valued by all respondents).
Specially trained and experienced interviewers conducted interviews in the respondents' own home during October 2005. The interview began with the respondents being asked to describe their own current state of health, which included completing the EQ-5D questionnaire and the OAB-5D. This procedure collected information about respondents' general health and OAB-related health which could be used in the modeling to estimate the impact of respondents' own health on their valuation of other health states, and also familiarized the respondents with the health state description of the OAB-5D.
The TTO technique was used for the valuation task. This asks respondents to trade off between length of life and quality of life. The TTO-prop method developed and used in the UK valuation of EQ-5D by the Measurement and Valuation of Health (MVH) project [16] was used in this survey. This has been shown to be more reliable than a nonprops version [17].
Modeling Health State Values
The overall aim of modeling is to predict preference values for all 3125 health states defined by the OAB-5D, based on the values obtained on the selected 98 states and the “pits” state over the valuation survey. The data from this type of valuation survey are normally skewed and clustered by respondent. A number of alternative models (taken from the study by Brazier et al.) [7] were explored for predicting the TTO scores generated in the valuation survey. The TTO scores were placed on a scale where 0 denotes dead and 1 denotes full health. Respondents can value a health state as being worse than dead, with a negative value as low as −1.
The general model is:
where: yij = TTO score for health state i valued by respondent j;
-
i = Individual health state values (1, 2, . . . , 99);
-
j = Individual respondents (1, 2, . . . , m);
-
g = Function specifying the appropriate functional form;
-
x = A vector of binary dummy variables (xδλ) for each level λ of dimension δ of the classification(level λ = 1 acts as the baseline for each dimension);
-
r = A vector of terms to account for interactions;
-
z = A vector of personal characteristics of respondents (i.e., sex, age, etc.);
-
εij = Error term. Its autocorrelation structure and distributional properties depend on the assumptions underlying the particular model used.
Given the main purpose of this article is to provide a general population value set for use in economic evaluation, the z term has not been explored in this article.
Time-trade-off values can be negative (although bounded to −1) if respondents value the health states under study as worse than death. When estimating models, the transformed variable of dis_TTO (1 − TTO) was used as dependent variable simply to ensure the coefficients are expected to have a more intuitive sign (namely positive and larger with increasing severity). Given 1 denotes full health, the dependent variable dis_TTO indicates the extent to which a given health state moves away from full health. Thus, the more severe the ill health state, the greater the coefficient should be, and the expected signs of the dummy coefficients are positive.
The data were modeled at both an individual and an aggregate level. The individual level model was a one-way error components random effects model that takes account of variation both within and between respondents, where the dependant variable is the disvalue (dis_TTO) for health state i valued by respondent j, with a vector of dummy explanatory variables defined as previously and an error term which is subdivided into a respondent-specific variation and a random error term. Ordinary least squares was used to estimate the aggregate model, where mean health state values were the dependent variable and the independent variables were a series of dummy explanatory variables representing each level of the five dimensions of the OAB-5D. There are 20 of these terms in total with level 1 acting as a baseline for each dimension.
OABq-5D full health 11111 was used as the upper anchor for TTO method. Therefore, each model was estimated without a constant term as dis_TTO was the dependent variable (this is equivalent to forcing the constant equal 1). There is evidence that preferences for different dimensions of health may not be additive [5,7]. Therefore, it is important to try to estimate interactions. We did not consider analysis of first-order interactions because of the very large number of such coefficients, and the collinearity problem that may occur [7]. Adapting the approach used in other studies (see, for example, the study by Brazier et al.) [7], an interaction variable “N2_severe” was created as a dummy variable which takes a value of 1 if two or more dimensions in the health state are at level 4 or 5, and 0 if otherwise. The creation of N2_severe was based on the assumption that possible interactions would happen between severe levels of different dimensions rather than between mild levels.
In addition, a series of empiric criteria were used to identify the best model: Models were compared in terms of their adjusted R-squared, the sign and significance of individual parameter estimates, the relative size of coefficients within a given dimension (in other words, logical consistency), as well as their predictive ability measured by mean absolute error (MAE) and the numbers of absolute errors greater than 0.05 and 0.10.
All analysis and modeling was carried out using SPSS 12.0 for Windows (SPSS, Chicago, IL) and STATA 9.0 (StataCorp, College Station, TX).
Results
Respondents
A random sample of the general population from South Yorkshire in the UK was sent letters of invitation to take part in the survey. The relevant numbers are reported in Table 2. The rate of successful interviews to potential interview respondents is 27%, although the rate of successful interviews to eligible respondents is 52%. A total of 312 members of the public were successfully interviewed, and their data are included in the subsequent analysis. The description of the sample is shown in Table 3 in terms of age, sex, education, general health, and position on the OAB-5D dimension scale. This sample was shown to be representative of the South Yorkshire general population in terms of age and sex.
| Letters of invitation sent: 1440 |
| Letters returned with vacant properties: 24 |
| Refusals at this stage: 20 |
| Properties called where resident were not at home: 415 |
| Refusals when interviewer calls: 268 |
| Respondents unsuitable for interview: 172 |
| Other reasons for interview not going ahead: 179 |
| Completed interviews: 312 |
| Properties not visited: 61 |
| Rate 1 = successful interviews/basic sample |
| = 312/1440 − (addressed vacant + not visited + unsuitable respondents) |
| = 312/1440 − (24 + 61 + 172) |
| = 312/1183 = 27% |
| Rate 2 = successful interviews/total number of eligible respondents |
| = 312/(successful interviews + total number of respondents refused to participate) |
| = 312/(312 + 20 + 268) = 52% |
| Survey sample | South Yorkshire* | ||||
|---|---|---|---|---|---|
| Count | Percentage | Percentage | |||
| Age 18–25 | 37 | 11.9 | N/A | ||
| 26–35 | 57 | 18.3 | 18.5 | ||
| 36–45 | 61 | 19.6 | 19.7 | ||
| 46–55 | 51 | 16.4 | 17.3 | ||
| 56–65 | 45 | 14.5 | 14.5 | ||
| >66 | 60 | 19.3 | 21.6 | ||
| Female | 160 | 51.4 | 51.2 | ||
| Married or living with partner | 217 | 69.8 | |||
| Experienced serious illness | |||||
| In family | 176 | 56.6 | |||
| Themselves | 94 | 30.2 | |||
| Degree or equivalent | 85 | 27.3 | |||
| Education after 17 | 182 | 58.5 | |||
| Renting property | 63 | 20.2 | |||
| Found valuation task difficult | |||||
| Very difficult | 13 | 4.2 | |||
| Quite difficult | 80 | 25.9 | |||
| Neither difficult nor easy | 70 | 22.7 | |||
| Self-reported OAB-5D health state | Level 1 (%) | Level 2 (%) | Level 3 (%) | Level 4 (%) | Level 5 (%) |
| Urge | 235 (75.8) | 48 (15.5) | 21 (6.8) | 4 (1.3) | 2 (0.6) |
| Urine loss | 258 (83.2) | 34 (11.0) | 10 (3.2) | 4 (1.3) | 4 (1.3) |
| Sleep | 206 (66.5) | 62 (20.0) | 25 (8.1) | 14 (4.5) | 3 (1.0) |
| Coping | 242 (78.1) | 40 (12.9) | 17 (5.5) | 8 (2.6) | 3 (1.0) |
| Concern | 264 (85.2) | 33 (10.6) | 9 (2.9) | 2 (0.6) | 2 (0.6) |
| Self-reported EQ-5D scores† | Male | Female | |||
| Respondent sample | 0.88 | 0.88 | |||
| UK population | 0.86 | 0.85 | |||
- * UK census 2001.
- † Self-reported EQ-5D scores are based in the EQ-5D tariff of the York value and measure of health project.
- N/A, not available; OAB, overactive bladder.
Health State Values
Table 4 presents health state values for the 98 health states valued in the survey, which formed the basis for modeling. In total, there were 2487 health state values generated by the 311 respondents. Each intermediate health state was valued on average 22 times (range from 17 to 29) and the worst possible state (OAB-5D 55,555) was valued 310 times (one missing value). Among the 2487 valuations, the proportion of health states been given a value of 1 was 32% and the proportion of health states been valued less than 0 was around 2%. Among the 310 valuations for the pits state, 13 valuations were valued as less than 0 and 65 were given a value of 1. The mean health state values ranged from 0.55 for state 32,435, to 0.91 for state 13,321, with an average standard deviation of 0.28. The standard deviations of states with lower TTO values tend to be larger than those states with higher values. Figure 1 shows the distribution of 2487 health state values to be negatively skewed, which is the same as was found in other valuation studies (e.g., SF-6D and EQ-5D).
| Health state | N | N* (%) | Minimum | Maximum | Mean | SD | Median |
|---|---|---|---|---|---|---|---|
| 32435 | 24 | 0.13 | −0.35 | 1 | 0.55 | 0.35 | 0.5 |
| 55555 | 310 | 0.21 | −0.88 | 1 | 0.56 | 0.39 | 0.63 |
| 34554 | 24 | 0.13 | −0.25 | 1 | 0.59 | 0.32 | 0.57 |
| 31155 | 24 | 0.17 | 0 | 1 | 0.61 | 0.35 | 0.64 |
| 51454 | 24 | 0.21 | −0.68 | 1 | 0.61 | 0.49 | 0.75 |
| 42554 | 21 | 0.14 | 0 | 1 | 0.62 | 0.32 | 0.68 |
| 54245 | 24 | 0.25 | −0.88 | 1 | 0.63 | 0.43 | 0.74 |
| 42245 | 24 | 0.29 | −0.88 | 1 | 0.66 | 0.44 | 0.82 |
| 23235 | 24 | 0.29 | −0.73 | 1 | 0.66 | 0.51 | 0.85 |
| 53525 | 25 | 0.24 | 0.17 | 1 | 0.67 | 0.27 | 0.63 |
| 53325 | 25 | 0.16 | 0.17 | 1 | 0.67 | 0.24 | 0.7 |
| 34225 | 29 | 0.28 | −0.57 | 1 | 0.67 | 0.39 | 0.82 |
| 43234 | 24 | 0.25 | −0.43 | 1 | 0.67 | 0.46 | 0.88 |
| 23154 | 21 | 0.19 | 0.13 | 1 | 0.68 | 0.29 | 0.73 |
| 42214 | 24 | 0.25 | −0.47 | 1 | 0.68 | 0.47 | 0.86 |
| 33245 | 28 | 0.25 | −0.47 | 1 | 0.69 | 0.38 | 0.77 |
| 11445 | 28 | 0.29 | −0.63 | 1 | 0.69 | 0.39 | 0.79 |
| 14444 | 21 | 0.14 | 0.32 | 1 | 0.7 | 0.24 | 0.7 |
| 42542 | 24 | 0.21 | 0.03 | 1 | 0.7 | 0.28 | 0.74 |
| 24422 | 24 | 0.25 | 0.03 | 1 | 0.7 | 0.3 | 0.77 |
| 54152 | 21 | 0.19 | 0.03 | 1 | 0.7 | 0.28 | 0.77 |
| 15553 | 28 | 0.29 | −0.68 | 1 | 0.7 | 0.37 | 0.82 |
| 54333 | 24 | 0.29 | −0.63 | 1 | 0.7 | 0.44 | 0.89 |
| 23534 | 25 | 0.16 | 0.08 | 1 | 0.71 | 0.26 | 0.8 |
| 45532 | 20 | 0.35 | 0.03 | 1 | 0.71 | 0.32 | 0.83 |
| 32235 | 24 | 0.38 | −0.88 | 1 | 0.71 | 0.44 | 0.88 |
| 13434 | 23 | 0.17 | 0.28 | 1 | 0.72 | 0.24 | 0.77 |
| 35453 | 23 | 0.22 | 0.22 | 1 | 0.72 | 0.27 | 0.77 |
| 51214 | 20 | 0.30 | 0.03 | 1 | 0.72 | 0.36 | 0.93 |
| 24524 | 23 | 0.26 | 0.28 | 1 | 0.73 | 0.24 | 0.77 |
| 24335 | 20 | 0.30 | 0 | 1 | 0.73 | 0.29 | 0.81 |
| 31531 | 24 | 0.17 | 0.03 | 1 | 0.73 | 0.3 | 0.86 |
| 42325 | 24 | 0.38 | −0.88 | 1 | 0.73 | 0.47 | 0.93 |
| 41125 | 25 | 0.20 | 0.08 | 1 | 0.74 | 0.24 | 0.8 |
| 44114 | 21 | 0.29 | 0 | 1 | 0.74 | 0.3 | 0.9 |
| 33323 | 21 | 0.14 | 0.22 | 1 | 0.75 | 0.24 | 0.88 |
| 25425 | 22 | 0.36 | 0.08 | 1 | 0.75 | 0.29 | 0.93 |
| 31215 | 27 | 0.30 | −0.68 | 1 | 0.75 | 0.37 | 0.93 |
| 34254 | 20 | 0.35 | 0 | 1 | 0.75 | 0.32 | 0.93 |
| 54123 | 20 | 0.30 | 0.03 | 1 | 0.76 | 0.27 | 0.83 |
| 45253 | 21 | 0.29 | 0.03 | 1 | 0.76 | 0.28 | 0.93 |
| 55521 | 23 | 0.22 | 0.3 | 1 | 0.77 | 0.2 | 0.77 |
| 25313 | 24 | 0.38 | −0.88 | 1 | 0.77 | 0.39 | 0.86 |
| 15131 | 23 | 0.30 | 0.03 | 1 | 0.77 | 0.3 | 0.93 |
| 44135 | 20 | 0.50 | 0.17 | 1 | 0.77 | 0.3 | 0.96 |
| 33132 | 24 | 0.50 | −0.4 | 1 | 0.77 | 0.45 | 1 |
| 52444 | 20 | 0.35 | 0.25 | 1 | 0.78 | 0.24 | 0.85 |
| 41153 | 25 | 0.28 | 0.08 | 1 | 0.78 | 0.25 | 0.88 |
| 55424 | 17 | 0.29 | 0 | 1 | 0.78 | 0.27 | 0.88 |
| 34351 | 20 | 0.40 | 0.03 | 1 | 0.78 | 0.3 | 0.93 |
| 35422 | 21 | 0.33 | 0 | 1 | 0.79 | 0.3 | 0.93 |
| 45143 | 21 | 0.38 | 0.03 | 1 | 0.79 | 0.28 | 0.93 |
| 53242 | 20 | 0.50 | 0.17 | 1 | 0.79 | 0.29 | 0.96 |
| 13514 | 20 | 0.30 | 0.32 | 1 | 0.8 | 0.21 | 0.81 |
| 14341 | 21 | 0.29 | 0.15 | 1 | 0.8 | 0.23 | 0.88 |
| 13431 | 25 | 0.28 | 0.08 | 1 | 0.8 | 0.25 | 0.9 |
| 32414 | 29 | 0.34 | 0 | 1 | 0.8 | 0.27 | 0.93 |
| 45553 | 21 | 0.33 | 0.13 | 1 | 0.8 | 0.27 | 0.93 |
| 53411 | 20 | 0.45 | 0.22 | 1 | 0.81 | 0.24 | 0.89 |
| 15355 | 22 | 0.41 | 0.08 | 1 | 0.81 | 0.28 | 0.93 |
| 52141 | 23 | 0.35 | 0.3 | 1 | 0.81 | 0.22 | 0.93 |
| 41322 | 24 | 0.50 | −0.77 | 1 | 0.81 | 0.38 | 0.99 |
| 41123 | 23 | 0.39 | 0.3 | 1 | 0.82 | 0.22 | 0.93 |
| 53532 | 21 | 0.38 | 0.03 | 1 | 0.82 | 0.29 | 0.95 |
| 12511 | 28 | 0.36 | −0.68 | 1 | 0.82 | 0.35 | 0.96 |
| 32412 | 20 | 0.50 | 0.13 | 1 | 0.82 | 0.28 | 0.96 |
| 12314 | 19 | 0.47 | 0.03 | 1 | 0.82 | 0.29 | 0.99 |
| 25543 | 20 | 0.50 | 0.08 | 1 | 0.82 | 0.28 | 1 |
| 25421 | 20 | 0.45 | 0.13 | 1 | 0.83 | 0.24 | 0.91 |
| 13251 | 22 | 0.41 | 0.08 | 1 | 0.83 | 0.25 | 0.93 |
| 15251 | 22 | 0.41 | 0.08 | 1 | 0.83 | 0.25 | 0.93 |
| 32441 | 21 | 0.29 | 0.43 | 1 | 0.83 | 0.2 | 0.93 |
| 52314 | 18 | 0.44 | 0.43 | 1 | 0.83 | 0.23 | 0.93 |
| 14225 | 21 | 0.43 | 0.03 | 1 | 0.83 | 0.25 | 0.95 |
| 12543 | 20 | 0.50 | 0.13 | 1 | 0.83 | 0.26 | 0.96 |
| 21113 | 25 | 0.40 | 0.08 | 1 | 0.84 | 0.23 | 0.93 |
| 24133 | 20 | 0.40 | 0.03 | 1 | 0.84 | 0.24 | 0.93 |
| 41442 | 22 | 0.36 | 0.35 | 1 | 0.84 | 0.21 | 0.93 |
| 51451 | 22 | 0.45 | 0.13 | 1 | 0.85 | 0.26 | 0.99 |
| 25112 | 25 | 0.36 | −0.4 | 1 | 0.85 | 0.3 | 1 |
| 41112 | 20 | 0.55 | 0.13 | 1 | 0.85 | 0.25 | 1 |
| 11233 | 21 | 0.33 | 0.38 | 1 | 0.86 | 0.18 | 0.93 |
| 21332 | 20 | 0.25 | 0.47 | 1 | 0.86 | 0.18 | 0.93 |
| 24433 | 21 | 0.33 | 0.32 | 1 | 0.86 | 0.21 | 0.93 |
| 31143 | 20 | 0.50 | 0.22 | 1 | 0.86 | 0.21 | 0.97 |
| 45341 | 20 | 0.45 | 0.32 | 1 | 0.86 | 0.21 | 0.97 |
| 15331 | 23 | 0.35 | 0.3 | 1 | 0.87 | 0.17 | 0.93 |
| 22242 | 21 | 0.38 | 0.47 | 1 | 0.87 | 0.17 | 0.95 |
| 12144 | 21 | 0.43 | 0.17 | 1 | 0.87 | 0.24 | 1 |
| 24352 | 18 | 0.50 | 0.22 | 1 | 0.87 | 0.24 | 1 |
| 41211 | 22 | 0.45 | 0.08 | 1 | 0.88 | 0.23 | 0.95 |
| 21223 | 21 | 0.52 | 0.17 | 1 | 0.88 | 0.22 | 1 |
| 33552 | 18 | 0.50 | 0.32 | 1 | 0.88 | 0.2 | 1 |
| 51522 | 21 | 0.38 | 0.47 | 1 | 0.88 | 0.16 | 1 |
| 15311 | 25 | 0.40 | 0.38 | 1 | 0.89 | 0.16 | 0.95 |
| 52112 | 18 | 0.56 | 0.2 | 1 | 0.89 | 0.22 | 1 |
| 23312 | 19 | 0.53 | 0.47 | 1 | 0.9 | 0.16 | 1 |
| 13321 | 20 | 0.55 | 0.17 | 1 | 0.91 | 0.2 | 1 |
| 33511 | 18 | 0.50 | 0.5 | 1 | 0.91 | 0.14 | 1 |
- The table is ranked by means of health state values.
- Health state: OAB-5D health states been valued using TTO technique.
- N: Times of health state been valued.
- N*: Percentage of times a health state was given a value of 1 which is equivalent to perfect health.
- OAB, overactive bladder; TTO, time-trade-off.
Distribution of health state values using TTO. TTOC are OAB-5D health state values based on TTO technique. OAB, overactive bladder; TTO, time-trade-off.
The Main Models
The main results of modeling are presented in Table 5, with summary statistics for internal sample predictions presented in the lower part of the table. Model 1 was estimated at the individual observation level although model 2 was estimated at the aggregate level using mean health state values.
| Dimension levels | Complete models | Mean model with merged levels | |
|---|---|---|---|
| RE (1) | Mean (2) | Mean (3) | |
| Urge2 | 0.034 | 0.034 | 0.036 |
| Urge3 | 0.031 | 0.068 | 0.070 |
| Urge4 | 0.068 | 0.081 | 0.078 |
| Urge5 | 0.084 | 0.076 | — |
| Urineloss2 | 0.020 | 0.032 | 0.026 |
| Urineloss3 | 0.049 | 0.036 | 0.038 |
| Urineloss4 | 0.031 | 0.047 | 0.045 |
| Urineloss5 | 0.047 | 0.044 | — |
| Sleep2 | 0.028 | 0.028 | 0.023 |
| Sleep3 | 0.019 | 0.020 | — |
| Sleep4 | 0.056 | 0.049 | 0.049 |
| Sleep5 | 0.052 | 0.050 | — |
| Coping2 | 0.008 | 0.032 | 0.030 |
| Coping3 | 0.021 | 0.073 | 0.060 |
| Coping4 | 0.022 | 0.046 | — |
| Coping5 | 0.066 | 0.081 | 0.080 |
| Concern2 | 0.033 | 0.022 | 0.022 |
| Concern3 | 0.045 | 0.047 | 0.048 |
| Concern4 | 0.088 | 0.110 | 0.113 |
| Concern5 | 0.136 | 0.139 | 0.142 |
| N | 2487 | 99 | 99 |
| Adj. R2 | N/A | N/A | |
| Inconsistencies | 4 | 3 | 0 |
| Significant coefficients | 14 | 13 | 10 |
| Mean absolute error | 0.076 | 0.044 | 0.045 |
| No. > |0.05| | 57 | 39 | 40 |
| No. > |0.10| | 27 | 8 | 6 |
- Independent variable: dis_TTO = 1 − TTO.
- Mean absolute error: mean of difference between observed and predicted values.
- No. > |0.05|: number of observations with absolute difference between observed and predicted values greater than 0.05.
- No. > |0.10|: number of observations with absolute difference between observed and predicted values greater than 0.10.
- Estimates showed in bold are significant at the 0.05 level.
- Inconsistencies: count for significant coefficients.
- N/A, not available; TTO, time-trade-off.
As expected, all the coefficients are positive (in other words, they increase with the TTO disvalue) and most were significant. The coefficients are broadly consistent with the ordering of the OAB-5D levels, so that a severer level within a dimension was associated with a larger decrement. For example, the coefficient of level 5 in urge dimension (0.084 as urge5 in Table 5) is greater than those of level 4 (0.068) and level 3 (0.031). There were some inconsistencies between significant coefficients in each model, but in all case except one were 0.01 or less. Examples include urge2 and urge3, urine loss3 and urine loss4, and sleep4 and sleep5 in the random effect model (model 1) as well as urge4 and urge5, urine loss4 and urine loss5, and coping3 and coping4 in the mean model (model 2).
Compared to model 1, the mean model (model 2) had one fewer inconsistency and performed better in terms of prediction (MAE 0.044 vs. 0.075). Given that the main aim of modeling is prediction, model 2 has been chosen as the better of the two. The addition of the interaction terms N2_severe did not improve either model so it was not included in the final specification of either model.
To overcome the inconsistency problem identified with model 2, we have merged those levels where the coefficients are inconsistent with the OAB-5D. The result is a “consistent” model 3 that required the merging of 10 coefficients to create a model with 15 dummy variables (rather than 20 for the other models). Model 3 has achieved consistency with little reduction in the predictive performance (MAE moves from 0.044 to 0.045). This is the recommended model for use in economic evaluation.
As the regression equation (Equation 1 below) shows, the preference-based, OAB-specific single index (TTO) can be explained by five dimensions in the OAB-5D classification system (urge, urine loss, sleep, coping, and concern) and each dimension enters the equation as a set of dummy variables with health state 11111 acting as default. Therefore, for health state 11111, all independent variables take a value of 0, which results in preference-based index of 1.0. Taking another health state 12345 as example, dummies will enter Equation 1 and result in Equation 2. Using coefficients of model 3, TTO value of health state 12345 will be calculated as 0.749. Note that in model 3, some coefficients have been merged so these merged levels have the same coefficients, such as urge3 and urge5, urine loss4 and urine loss5. An SPSS syntax file is available from the correspondence on request which can be used to calculate OAB-specific utility values on the basis of the OAB-q.
(1)
(2)Discussion and Conclusion
The methodology used in this study was the same as that used to estimate the SF-6D population value set from the generic SF-36. The study reported in this article is one of a number that have shown it is possible to estimate a preference-based measure for a condition-specific measure [18–20]). The models of the health state valuation data were broadly consistent with the descriptive system and achieved levels of predictive performance as good as or better than those for generic health state descriptive systems, such as the EQ-5D and SF-6D [5,7].
The OAB-5D-described health states would probably be regarded as mild compared to the EQ-5D or SF-6D and this is reflected in the ranges of scores, starting from 0.56 for the worst OAB-5D state compared to 0.21 for the SF-6D and negative values for the EQ-5D and HUI3. For sufferers of OAB, it is a bothersome condition that impacts significantly on their health-related quality of life but it is not as debilitating as some other physical and mental disorders. There have been similar findings for other preference-based measures using descriptions for milder conditions, with the Kings Health Questionnaire for urinary incontinence having a lowest value of 0.77 [18] and the International Index of Erectile Dysfunction a lowest value of 0.73 [21].
The original interest in developing a condition-specific preference-based measure rather than using an existing generic measure was that the latter may be insensitive to important differences or indeed may miss important dimensions altogether [22,23]. Nevertheless, there is a concern that condition-specific preference-based measures will not achieve comparability with other preference-based measures and so can only be used to inform resource allocation within a condition [24]. We have argued elsewhere that the problem of comparability can be partly overcome by: 1) using the same valuation protocol; 2) ensuring common upper and lower anchors; and 3) obtaining values from the same population [8].
We have used the same variant of the TTO as the UK EQ-5D valuation study (the MVH project) on a general population, and the upper anchor of state 11,111 defined by the OAB-5D is arguably similar to state 11111 of the EQ-5D. Differences between the scores produced by the OAB-5D and the EQ-5D should be solely due to differences in the descriptive system. An important rider to this claim, however, is that the upper anchors are not strictly the same, and it assumes that the values given to the dimensions of this instrument are independent of dimensions excluded from OAB-5D. Because we did not cover other dimensions of health in the valuation survey, we do not know what a respondent was considering besides urinary issues when presented with OAB-5D health states. Thus, the values given to the OAB-specific dimensions may not be independent of whether or not a person has comorbidities that result in other health problems like pain. The extent of this problem needs to be explored further in the future.
This article reports on a study that enables the derivation of preference-based single indices from the condition-specific quality of life instrument, OAB-q. Researchers with OAB-q patient data can allocate an OAB-5D health state to each OAB-q observation and use the coefficients in the consistent model 3 to predict a preference-based index for use in economic evaluation. Given findings of differences between the EQ-5D scores across countries (see for example the studies by Tsuchiya et al. and Johnson et al.) [25,26], researchers outside the UK need to use the scoring system presented here with caution, as it is based on the UK general public's preference. Another concern is the relevance of the wording of some items across different cultures. For instance, the “coping” dimension refers to “the restroom” (US/North American terminology taken from the original instrument), whereas the valuation is based on a South Yorkshire UK population, not all of whom may relate to this. The OAB-5D classification system was established by taking five items from the original instrument directly without any change, and cultural adaptation was not pursued, to maintain the maximum link between the two instruments. Nevertheless, feedback from the interviewers suggests that this phrase did not cause any difficulties among the respondents.
The results in this article may contribute toward extending the application of cost per QALY analysis in OAB. The three-stage approach explored in this study offers a feasible approach to estimating disease-specific utilities where there is a validated nonpreference-based HRQoL instrument. Nevertheless, there remains further scope for explore, such as to what is the effect of naming the specific condition on health states valuation, and how to deal with comorbidities and side effects which are not incorporated in a condition-specific descriptive system. Further research is needed to address these issues and make general claims on this approach.
Pfizer Pharmaceuticals funded this study. John Brazier is funded by the Medical Research Council Health Service Research Collaboration. We would like to thank Tracey Young and Jennifer Roberts and the three anonymous reviewers for comments on this study. Also, we would like to thank Jason Leman and colleagues in Sheffield Hallam University for undertaking the interviews for us. We are particularly grateful to those respondents who kindly agreed to participate in the survey.
Source of financial support: Pfizer Pharmaceuticals funded this study.
