Interaction Effects in Health State Valuation Studies: An Optimal Scaling Approach.

OBJECTIVES: This study aimed to introduce a parsimonious modeling approach that enables the estimation of interaction effects in health state valuation studies. METHODS: Instead of supplementing a main-effects model with interactions between each and every level, a more parsimonious optimal scaling approach is proposed. This approach is based on the mapping of health state levels onto domain-specific continuous scales. The attractiveness of health states is then determined by the importance-weighted optimal scales (ie, main effects) and the interactions between these domain-specific scales (ie, interaction effects). The number of interaction terms only depends on the number of health domains. Therefore, interactions between dimensions can be included with only a few additional parameters. The proposed models with and without interactions are fitted on 3 valuation data sets from 2 different countries, that is, a Dutch latent-scale discrete choice experiment (DCE) data set with 3699 respondents, an Australian time trade-off data set with 400 respondents, and a Dutch DCE with duration data set with 788 respondents. RESULTS: Important interactions between health domains were found in all 3 applications. The results confirm that the accumulation of health problems within health states has a decreasing marginal effect on health state values. A similar effect is obtained when so-called N3 or N5 terms are included in the model specification, but the inclusion of 2-way interactions provides superior model fits. CONCLUSIONS: The proposed interaction model is parsimonious, produces estimates that are straightforward to interpret, and accommodates the estimation of interaction effects in health state valuation studies with realistic sample size requirements. Not accounting for interactions is shown to result in biased value sets, particularly in stand-alone DCE with duration studies.

Two for the price of one: If moving beyond traditional single-best discrete choice experiments, should we use best-worst, best-best or ranking for preference elicitation?

This study undertook a head-to-head comparison of best-worst, best-best and ranking discrete choice experiments (DCEs) to help decide which method to use if moving beyond traditional single-best DCEs. Respondents were randomized to one of three preference elicitation methods. Rank-ordered (exploded) mixed logit models and respondent-reported data were used to compare methods and first and second choices. First choices differed from second choices and preferences differed between elicitation methods, even beyond scale and scale dynamics. First choices of best-worst had good choice consistency, scale dynamics and statistical efficiency, but this method's second choices performed worst. Ranking performed best on respondent-reported difficulty and preference; best-best's second choices on statistical efficiency. All three preference elicitation methods improve efficiency of data collection relative to using first choices only. However, differences in preferences between first and second choices challenge moving beyond single-best DCE. If nevertheless doing so, best-best and ranking are preferred over best-worst DCE.

Mimicking Real-Life Decision Making in Health: Allowing Respondents Time to Think in a Discrete Choice Experiment.

OBJECTIVE: To empirically test the impact of allowing respondents time to think (TTT) about their choice options on the outcomes of a discrete choice experiments (DCE). METHODS: In total, 613 participants of the Swedish CArdioPulmonary bioImage Study (SCAPIS) completed a DCE questionnaire that measured their preferences for receiving secondary findings of a genetic test. A Bayesian D-efficient design with 60 choice tasks divided over 4 questionnaires was used. Each choice task contained 2 scenarios with 4 attributes: type of disease, disease penetrance probability, preventive opportunities, and effectiveness of prevention. Respondents were randomly allocated to the TTT or no TTT (NTTT) sample. Latent class models (LCMs) were estimated to determine attribute-level values and their relative importance. In addition, choice certainty, attribute-level interpretation, choice consistency, and potential uptake rates were compared between samples. RESULTS: In the TTT sample, 92% of the respondents (245 of 267) indicated they used the TTT period to (1) read the information they received (72%) and (2) discuss with their family (24%). In both samples, respondents were very certain about their choices. A 3-class LCM was fitted for both samples. Preference reversals were found for 3 of the 4 attributes in one class in the NTTT sample (34% class-membership probability). Relative importance scores of the attributes differed between the 2 samples, and significant scale effects indicating higher choice consistency in TTT sample were found. CONCLUSIONS: Offering respondents TTT influences decision making and preferences. Developers of future DCEs regarding complex health-related decisions are advised to consider this approach to enhance the validity of the elicited preferences.

Are Healthcare Choices Predictable? The Impact of Discrete Choice Experiment Designs and Models.

BACKGROUND: Lack of evidence about the external validity of discrete choice experiments (DCEs) is one of the barriers that inhibit greater use of DCEs in healthcare decision making. OBJECTIVES: To determine whether the number of alternatives in a DCE choice task should reflect the actual decision context, and how complex the choice model needs to be to be able to predict real-world healthcare choices. METHODS: Six DCEs were used, which varied in (1) medical condition (involving choices for influenza vaccination or colorectal cancer screening) and (2) the number of alternatives per choice task. For each medical condition, 1200 respondents were randomized to one of the DCE formats. The data were analyzed in a systematic way using random-utility-maximization choice processes. RESULTS: Irrespective of the number of alternatives per choice task, the choice for influenza vaccination and colorectal cancer screening was correctly predicted by DCE at an aggregate level, if scale and preference heterogeneity were taken into account. At an individual level, 3 alternatives per choice task and the use of a heteroskedastic error component model plus observed preference heterogeneity seemed to be most promising (correctly predicting >93% of choices). CONCLUSIONS: Our study shows that DCEs are able to predict choices-mimicking real-world decisions-if at least scale and preference heterogeneity are taken into account. Patient characteristics (eg, numeracy, decision-making style, and general attitude for and experience with the health intervention) seem to play a crucial role. Further research is needed to determine whether this result remains in other contexts.

Attribute level overlap (and color coding) can reduce task complexity, improve choice consistency, and decrease the dropout rate in discrete choice experiments.

A randomized controlled discrete choice experiment (DCE) with 3,320 participating respondents was used to investigate the individual and combined impact of level overlap and color coding on task complexity, choice consistency, survey satisfaction scores, and dropout rates. The systematic differences between the study arms allowed for a direct comparison of dropout rates and cognitive debriefing scores and accommodated the quantitative comparison of respondents' choice consistency using a heteroskedastic mixed logit model. Our results indicate that the introduction of level overlap made it significantly easier for respondents to identify the differences and choose between the choice options. As a stand-alone design strategy, attribute level overlap reduced the dropout rate by 30%, increased the level of choice consistency by 30%, and avoided learning effects in the initial choice tasks of the DCE. The combination of level overlap and color coding was even more effective: It reduced the dropout rate by 40% to 50% and increased the level of choice consistency by more than 60%. Hence, we can recommend attribute level overlap, with color coding to amplify its impact, as a standard design strategy in DCEs.

Effect of Level Overlap and Color Coding on Attribute Non-Attendance in Discrete Choice Experiments.

OBJECTIVE: The aim of this study was to test the hypothesis that level overlap and color coding can mitigate or even preclude the occurrence of attribute nonattendance in discrete choice experiments. METHODS: A randomized controlled experiment with five experimental study arms was designed to investigate the independent and combined impact of level overlap and color coding on respondents' attribute nonattendance. The systematic differences between the study arms allowed for a direct comparison of observed dropout rates and estimates of the average number of attributes attended to by respondents, which were obtained by using augmented mixed logit models that explicitly incorporated attribute non-attendance. RESULTS: In the base-case study arm without level overlap or color coding, the observed dropout rate was 14%, and respondents attended, on average, only two out of five attributes. The independent introduction of both level overlap and color coding reduced the dropout rate to 10% and increased attribute attendance to three attributes. The combination of level overlap and color coding, however, was most effective: it reduced the dropout rate to 8% and improved attribute attendance to four out of five attributes. The latter essentially removes the need to explicitly accommodate for attribute non-attendance when analyzing the choice data. CONCLUSIONS: On the basis of the presented results, the use of level overlap and color coding are recommendable strategies to reduce the dropout rate and improve attribute attendance in discrete choice experiments.

Advocating a Paradigm Shift in Health-State Valuations: The Estimation of Time-Preference Corrected QALY Tariffs.

BACKGROUND: Despite evidence of nonproportional trade-offs in time trade-off exercises and the explicit incorporation of exponential discounting in health technology assessment calculations, quality-adjusted life-year (QALY) tariffs are currently still established under the assumption of linear time preferences. OBJECTIVES: The aim of this study was to introduce a general method of accommodating for nonlinear time preferences in discrete choice experiment (DCE) duration studies and to evaluate its impact on estimated QALY tariffs. METHODS: A parsimonious utility function is proposed that accommodates any discounting function and preserves linear time preferences as a special case. Based on an efficient DCE design and 1775 respondents from a nationally representative scientific household panel, preferences and QALY tariffs for the Dutch SF-6D were estimated while accommodating for nonlinear time preferences via exponential and hyperbolic discounting functions. RESULTS: When the discount rate was estimated directly, we found strong evidence of nonlinear time preferences (with an exponential and hyperbolic discount rate of 5.7% and 16.5%, respectively). When the discount rate was estimated as a function of health state severity, we found that years lived in better health states are discounted minus years lived in impaired health states. Finally, the best statistical fit was obtained when using a hyperbolic discount function, which resulted in smaller QALY decrements and fewer health states classified as worse than immediate death. CONCLUSIONS: Our results highlight the relevance and even necessity of a paradigm shift in health valuation studies in favor of time-preference corrected QALY tariffs, with potentially important implications for health technology assessment calculations and regulatory decisions.

Are Health State Valuations from the General Public Biased? A Test of Health State Reference Dependency Using Self-assessed Health and an Efficient Discrete Choice Experiment.

Health state valuations of patients and non-patients are not the same, whereas health state values obtained from general population samples are a weighted average of both. The latter constitutes an often-overlooked source of bias. This study investigates the resulting bias and tests for the impact of reference dependency on health state valuations using an efficient discrete choice experiment administered to a Dutch nationally representative sample of 788 respondents. A Bayesian discrete choice experiment design consisting of eight sets of 24 (matched pairwise) choice tasks was developed, with each set providing full identification of the included parameters. Mixed logit models were used to estimate health state preferences with respondents' own health included as an additional predictor. Our results indicate that respondents with impaired health worse than or equal to the health state levels under evaluation have approximately 30% smaller health state decrements. This confirms that reference dependency can be observed in general population samples and affirms the relevance of prospect theory in health state valuations. At the same time, the limited number of respondents with severe health impairments does not appear to bias social tariffs as obtained from general population samples. Copyright © 2016 John Wiley & Sons, Ltd.

Estimating the Impact of Health-related Behaviors on Geographic Variation in Cardiovascular Mortality: A New Approach Based on the Synthesis of Ecological and Individual-level Data.

BACKGROUND: Incidence of and mortality from cardiovascular disease (CVD) exhibit a strong geographical pattern, with inhabitants of more affluent neighborhoods showing a substantially lower risk of CVD mortality than inhabitants of deprived neighborhoods. Thus far, there is insufficient evidence as to what extent these differences can be attributed to differences in health-related behaviors. METHODS: Using a Hierarchical Related Regression approach, we combined individual and aggregate (ecological) data to investigate the extent to which small-area variation in CVD mortality in Dutch neighborhoods can be explained by several behavioral risk factors (i.e., smoking, drinking, overweight, and physical inactivity). The proposed approach combines the benefits of both an ecological analysis (in terms of data availability and statistical power) and an individual-level analysis (in terms of identification of the parameters and interpretation of the results). RESULTS: After correcting for differences in age and sex, accounting for differences in the behavioral risk factors reduces income-related inequalities in CVD mortality by approximately 30%. CONCLUSIONS: Direct targeting of the excess prevalence of unhealthy behaviors in deprived neighborhoods is identified as a relevant strategy to reduce inequalities in CVD mortality. Our results also show that the proposed Hierarchical Related Regression approach provides a powerful method for the investigation of small-area variation in health outcomes.

Should I stay or should I go home? A latent class analysis of a discrete choice experiment on hospital-at-home.

OBJECTIVES: This study aimed 1) to quantify the strength of patient preferences for different aspects of early assisted discharge in The Netherlands for patients who were admitted with a chronic obstructive pulmonary disease exacerbation and 2) to illustrate the benefits of latent class modeling of discrete choice data. This technique is rarely used in health economics. METHODS: Respondents made multiple choices between hospital treatment as usual (7 days) and two combinations of hospital admission (3 days) followed by treatment at home. The latter was described by a set of attributes. Hospital treatment was constant across choice sets. Respondents were patients with chronic obstructive pulmonary disease in a randomized controlled trial investigating the cost-effectiveness of early assisted discharge and their informal caregivers. The data were analyzed using mixed logit, generalized multinomial logit, and latent-class conditional logit regression. These methods allow for heterogeneous preferences across groups, but in different ways. RESULTS: Twenty-five percent of the respondents opted for hospital treatment regardless of the description of the early assisted discharge program, and 46% never opted for the hospital. The best model contained four latent classes of respondents, defined by different preferences for the hospital and caregiver burden. Preferences for other attributes were constant across classes. Attributes with the strongest effect on choices were the burden on informal caregivers and co-payments. Except for the number of visits, all attributes had a significant effect on choices in the expected direction. CONCLUSIONS: Considerable segments of respondents had fixed preferences for either treatment option. Applying latent class analysis was essential in quantifying preferences for attributes of early assisted discharge.

Comparing Outcomes of a Discrete Choice Experiment and Case 2 Best-Worst Scaling: An Application to Neuromuscular Disease Treatment.

BACKGROUND AND OBJECTIVES: Case 2 best-worst scaling (BWS-2) is an increasingly popular method to elicit patient preferences. Because BWS-2 potentially has a lower cognitive burden compared with discrete choice experiments, the aim of this study was to compare treatment preference weights and relative importance scores. METHODS: Patients with neuromuscular diseases completed an online survey at two different moments in time, completing one method per occasion. Patients were randomly assigned to either first a discrete choice experiment or BWS-2. Attributes included: muscle strength, energy endurance, balance, cognition, chance of blurry vision, and chance of liver damage. Multinomial logit was used to calculate overall relative importance scores and latent class logit was used to estimate heterogeneous preference weights and to calculate the relative importance scores of the attributes for each latent class. RESULTS: A total of 140 patients were included for analyses. Overall relative importance scores showed differences in attribute importance rankings between a discrete choice experiment and BWS-2. Latent class analyses indicated three latent classes for both methods, with a specific class in both the discrete choice experiment and BWS-2 in which (avoiding) liver damage was the most important attribute. Ex-post analyses showed that classes differed in sex, age, level of education, and disease status. The discrete choice experiment was easier to understand compared with BWS-2. CONCLUSIONS: This study showed that using a discrete choice experiment and BWS-2 leads to different outcomes, both in preference weights as well as in relative importance scores, which might have been caused by the different framing of risks in BWS-2. However, a latent class analysis revealed similar latent classes between methods. Careful consideration about method selection is required, while keeping the specific decision context in mind and pilot testing the methods.

Comparison of Bayesian random-effects and traditional life expectancy estimations in small-area applications.

There are several measures that summarize the mortality experience of a population. Of these measures, life expectancies are generally preferred based on their simpler interpretation and direct age standardization, which makes them directly comparable between different populations. However, traditional life expectancy estimations are highly inaccurate for smaller populations and consequently are seldom used in small-area applications. In this paper, the authors compare the relative performance of traditional life expectancy estimation with a Bayesian random-effects approach that uses correlations (i.e., borrows strength) between different age groups, geographic areas, and sexes to improve the small-area life expectancy estimations. In the presented Monte Carlo simulations, the Bayesian random-effects approach outperforms the traditional approach in terms of bias, root mean square error, and coverage of the 95% confidence intervals. Moreover, the Bayesian random-effects approach is found to be usable for populations as small as 2,000 person-years at risk, which is considerably smaller than the minimum of 5,000 person-years at risk recommended for the traditional approach. As such, the proposed Bayesian random-effects approach is well-suited for estimation of life expectancies in small areas.

What Factors Influence Non-Participation Most in Colorectal Cancer Screening? A Discrete Choice Experiment.

BACKGROUND AND OBJECTIVE: Non-participation in colorectal cancer (CRC) screening needs to be decreased to achieve its full potential as a public health strategy. To facilitate successful implementation of CRC screening towards unscreened individuals, this study aimed to quantify the impact of screening and individual characteristics on non-participation in CRC screening. METHODS: An online discrete choice experiment partly based on qualitative research was used among 406 representatives of the Dutch general population aged 55-75 years. In the discrete choice experiment, respondents were offered a series of choices between CRC screening scenarios that differed on five characteristics: effectiveness of the faecal immunochemical screening test, risk of a false-negative outcome, test frequency, waiting time for faecal immunochemical screening test results and waiting time for a colonoscopy follow-up test. The discrete choice experiment data were analysed in a systematic manner using random-utility-maximisation choice processes with scale and/or preference heterogeneity (based on 15 individual characteristics) and/or random intercepts. RESULTS: Screening characteristics proved to influence non-participation in CRC screening (21.7-28.0% non-participation rate), but an individual's characteristics had an even higher impact on CRC screening non-participation (8.4-75.5% non-participation rate); particularly the individual's attitude towards CRC screening followed by whether the individual had participated in a cancer screening programme before, the decision style of the individual and the educational level of the individual. Our findings provided a high degree of confidence in the internal-external validity. CONCLUSIONS: This study showed that although screening characteristics proved to influence non-participation in CRC screening, a respondent's characteristics had a much higher impact on CRC screening non-participation. Policy makers and physicians can use our study insights to improve and tailor their communication plans regarding (CRC) screening for unscreened individuals.

Labeled versus unlabeled discrete choice experiments in health economics: an application to colorectal cancer screening.

OBJECTIVES: Discrete choice experiments (DCEs) in health economics commonly present choice sets in an unlabeled form. Labeled choice sets are less abstract and may increase the validity of the results. We empirically compared the feasibility, respondents' trading behavior, and convergent validity between a labeled and an unlabeled DCE for colorectal cancer (CRC) screening programs in The Netherlands. METHODS: A labeled DCE version presented CRC screening test alternatives as "fecal occult blood test,""sigmoidoscopy," and "colonoscopy," whereas the unlabeled DCE version presented them as "screening test A" and "screening test B." Questionnaires were sent to participants and nonparticipants in CRC screening. RESULTS: Total response rate was 276 (39%) out of 712 and 1033 (46%) out of 2267 for unlabeled and labeled DCEs, respectively (P<0.001). The labels played a significant role in individual choices; approximately 22% of subjects had dominant preferences for screening test labels. The convergent validity was modest to low (participants in CRC screening: r=0.54; P=0.01; nonparticipants: r=0.17; P=0.45) largely because of different preferences for screening frequency. CONCLUSION: This study provides important insights in the feasibility and difference in results from labeled and unlabeled DCEs. The inclusion of labels appeared to play a significant role in individual choices but reduced the attention respondents give to the attributes. As a result, unlabeled DCEs may be more suitable to investigate trade-offs between attributes and for respondents who do not have familiarity with the alternative labels, whereas labeled DCEs may be more suitable to explain real-life choices such as uptake of cancer screening.

Summarizing Patient Preferences for the Competitive Landscape of Multiple Sclerosis Treatment Options.

Objective. Quantitatively summarize patient preferences for European licensed relapsing-remitting multiple sclerosis (RRMS) disease-modifying treatment (DMT) options. Methods. To identify and summarize the most important RRMS DMT characteristics, a literature review, exploratory physician interviews, patient focus groups, and confirmatory physician interviews were conducted in Germany, the United Kingdom, and the Netherlands. A discrete choice experiment (DCE) was developed and executed to measure patient preferences for the most important DMT characteristics. The resulting DCE data (n=799 and n=363 respondents in the United Kingdom and Germany, respectively) were analyzed using Bayesian mixed logit models. The estimated individual-level patient preferences were subsequently summarized using 3 additional analyses: the quality of the choice data was assessed using individual-level R2 estimates, individual-level preferences for the available DMTs were aggregated into DMT-specific preference shares, and a principal component analysis was performed to explain the patients' choice process. Results. DMT usage differed between RRMS patients in Germany and the United Kingdom but aggregate patient preferences were similar. Across countries, 42% of all patients preferred oral medications, 38% infusions, 16% injections, and 4% no DMT. The most often preferred DMT was natalizumab (26%) and oral DMT cladribine tablets (22%). The least often preferred were mitoxantrone and the beta-interferon injections (1%-3%). Patient preferences were strongly correlated with patients' MS disease duration and DMT experience, and differences in patient preferences could be summarized using 8 principle components that together explain 99% of the variation in patients' DMT preferences. Conclusion. This study summarizes patient preferences for the included DMTs, facilitates shared decision making along the dimensions that are relevant to RRMS patients, and introduces methods in the medical DCE literature that are ideally suited to summarize the impact of DMT introductions in preexisting treatment landscapes.

A labelled discrete choice experiment adds realism to the choices presented: preferences for surveillance tests for Barrett esophagus.

BACKGROUND: Discrete choice experiments (DCEs) allow systematic assessment of preferences by asking respondents to choose between scenarios. We conducted a labelled discrete choice experiment with realistic choices to investigate patients' trade-offs between the expected health gains and the burden of testing in surveillance of Barrett esophagus (BE). METHODS: Fifteen choice scenarios were selected based on 2 attributes: 1) type of test (endoscopy and two less burdensome fictitious tests), 2) frequency of surveillance. Each test-frequency combination was associated with its own realistic decrease in risk of dying from esophageal adenocarcinoma. A conditional logit model was fitted. RESULTS: Of 297 eligible patients (155 BE and 142 with non-specific upper GI symptoms), 247 completed the questionnaire (84%). Patients preferred surveillance to no surveillance. Current surveillance schemes of once every 1-2 years were amongst the most preferred alternatives. Higher health gains were preferred over those with lower health gains, except when test frequencies exceeded once a year. For similar health gains, patients preferred video-capsule over saliva swab and least preferred endoscopy. CONCLUSION: This first example of a labelled DCE using realistic scenarios in a healthcare context shows that such experiments are feasible. A comparison of labelled and unlabelled designs taking into account setting and research question is recommended.

Which preferred providers are really preferred? Effectiveness of insurers' channeling incentives on pharmacy choice.

Efficient contracting of health care requires effective consumer channeling. Little is known about the effectiveness of channeling strategies. We study channeling incentives on pharmacy choice using a large scale discrete choice experiment. Financial incentives prove to be effective. Positive financial incentives are less effective than negative financial incentives. Channeling through qualitative incentives also leads to a significant impact on provider choice. While incentives help to channel, a strong status quo bias needs to be overcome before consumers change pharmacies. Focusing on consumers who are forced to choose a new pharmacy seems to be the most effective strategy.

Methods for exploring and eliciting patient preferences in the medical product lifecycle: a literature review.

Preference studies are becoming increasingly important within the medical product decision-making context. Currently, there is limited understanding of the range of methods to gain insights into patient preferences. We developed a compendium and taxonomy of preference exploration (qualitative) and elicitation (quantitative) methods by conducting a systematic literature review to identify these methods. This review was followed by analyzing prior preference method reviews, to cross-validate our results, and consulting intercontinental experts, to confirm our outcomes. This resulted in the identification of 32 unique preference methods. The developed compendium and taxonomy can serve as an important resource for assessing these methods and helping to determine which are most appropriate for different research questions at varying points in the medical product lifecycle.

The Fold-in, Fold-out Design for DCE Choice Tasks: Application to Burden of Disease.

Background In discrete-choice experiments (DCEs), choice alternatives are described by attributes. The importance of each attribute can be quantified by analyzing respondents' choices. Estimates are valid only if alternatives are defined comprehensively, but choice tasks can become too difficult for respondents if too many attributes are included. Several solutions for this dilemma have been proposed, but these have practical or theoretical drawbacks and cannot be applied in all settings. The objective of the current article is to demonstrate an alternative solution, the fold-in, fold-out approach (FiFo). We use a motivating example, the ABC Index for burden of disease in chronic obstructive pulmonary disease (COPD). Methods Under FiFo, all attributes are part of all choice sets, but they are grouped into domains. These are either folded in (all attributes have the same level) or folded out (levels may differ). FiFo was applied to the valuation of the ABC Index, which included 15 attributes. The data were analyzed in Bayesian mixed logit regression, with additional parameters to account for increased complexity in folded-out questionnaires and potential differences in weight due to the folding status of domains. As a comparison, a model without the additional parameters was estimated. Results Folding out domains led to increased choice complexity for respondents. It also gave domains more weight than when it was folded in. The more complex regression model had a better fit to the data than the simpler model. Not accounting for choice complexity in the models resulted in a substantially different ABC Index. Conclusion Using a combination of folded-in and folded-out attributes is a feasible approach for conducting DCEs with many attributes.

Severity-Stratified Discrete Choice Experiment Designs for Health State Evaluations.

BACKGROUND: Discrete choice experiments (DCEs) are increasingly used for health state valuations. However, the values derived from initial DCE studies vary widely. We hypothesize that these findings indicate the presence of unknown sources of bias that must be recognized and minimized. Against this background, we studied whether values derived from a DCE are sensitive to how well the DCE design spans the severity range. METHODS: We constructed an experiment involving three variants of DCE tasks for health state valuation: standard DCE, DCE-death, and DCE-duration. For each type of DCE, an experimental design was generated under two different conditions, enabling a comparison of health state values derived from current best practice Bayesian efficient DCE designs with values derived from 'severity-stratified' designs that control for coverage of the severity range in health state selection. About 3000 respondents participated in the study and were randomly assigned to one of the six study arms. RESULTS: Imposing the severity-stratified restriction had a large effect on health states sampled for the DCE-duration approach. The unstratified efficient design returned a skewed distribution of selected health states, and this introduced bias. The choice probability of bad health states was underestimated, and time trade-offs to avoid bad states were overestimated, resulting in too low values. Imposing the same restriction had limited effect in the DCE-death approach and standard DCE. CONCLUSION: Variation in DCE-derived values can be partially explained by differences in how well selected health states spanned the severity range. Imposing a 'severity stratification' on DCE-duration designs is a validity requirement.