Optimal and near-optimal pairs for the estimation of effects in 2-level choice experiments

Discrete choice experiments: An overview on constructing D-optimal and near-optimal choice sets

Discrete choice experiments (DCEs) are frequently used to estimate and forecast the behavior of an individual's choice. DCEs are based on stated preference; therefore, underlying experimental designs are required for this type of study. According to psychologists, DCE designs consist of a small number of choice sets with a limited size in the number of alternatives within a choice set to increase the response efficiency in the questionnaire. Even though algorithmic constructions (known as efficient designs) become quite common for practitioners, optimal designs (sometimes so-called orthogonal designs) continue to be used in choice experiment studies, particularly in the case that prior information about the extent of the population preference is not available. Various approaches have been developed to construct DCE designs with fewer choice sets. However, the question in many practitioners' minds is which techniques perform better (i.e. given small designs with high efficiency) in a given circumstance. In this paper and to address these concerns, we conducted an overview of the constructions of discrete choice experiments in the literature for models with only main effects. The various ways of constructing optimal and near-optimal designs were compared in terms of their ability to minimize the number of choice sets in the survey. Our findings shed light on the optimal sample sizes needed for efficient experimentation which then can help the researchers to design more effective experiments in this area.

Preferences for interventions designed to increase cervical screening uptake in non-attending young women: How findings from a discrete choice experiment compare with observed behaviours in a trial

BACKGROUND

Young women's attendance at cervical screening in the UK is continuing to fall, and the incidence of invasive cervical cancer is rising.

OBJECTIVES

We assessed the preferences of non-attending young women for alternative ways of delivering cervical screening.

DESIGN

Postal discrete choice experiment (DCE) conducted during the STRATEGIC study of interventions for increasing cervical screening uptake. Attributes included action required to arrange a test, location of the test, availability of a nurse navigator and cost to the National Health Service.

SETTING AND PARTICIPANTS

Non-attending young women in two UK regions.

MAIN OUTCOME MEASURES

Responses were analysed using a mixed multinomial logit model. A predictive analysis identified the most preferable strategy compared to current screening. Preferences from the DCE were compared with observed behaviours during the STRATEGIC trial.

RESULTS

The DCE response rate was 5.5% (222/4000), and 94% of respondents agreed screening is important. Preference heterogeneity existed around attributes with strong evidence for test location. Relative to current screening, unsolicited self-sampling kits for home use appeared most preferable. The STRATEGIC trial showed this same intervention to be most effective although many women who received it and were screened, attended for conventional cytology instead.

CONCLUSIONS

The DCE and trial identified the unsolicited self-sampling kit as the most preferred/effective intervention. The DCE suggested that the decision of some women receiving the kit in the trial to attend for conventional cytology may be due to anxieties around home testing coupled with a knowledge that ignoring the kit could potentially have life-changing consequences.

Using a stated preference discrete choice experiment to assess societal value from the perspective of patients with rare diseases in Italy

Background

Decision makers have huge problems when attempting to attribute social value to the improvements achieved by new drugs, especially when considering the use of orphan drugs for rare diseases. We present the results of a pilot study aimed to investigate patient preferences regarding public funding for drugs used to treat rare diseases.

Methods

An online questionnaire was used as a discrete choice experiment (DCE) survey to explore the preferences of patients with cystic fibrosis and haemophilia in Italy. The questionnaire focused on relevant issues that were defined in a review of the literature. A conditional logistic model showed preferences for specific attributes.

Results

A total of 54 questionnaires (20% response rate) were completed. The issues that received the greatest attention were improvement in health, treatment cost and value for money. However, disease severity and the availability of other treatments were important social values that could not be ignored.

Conclusions

The findings presented here provide evidence as to what patients with cystic fibrosis or haemophilia think are the most important considerations on which to base decisions in health technology scenarios, and regarding the priorities for funding.

Comparison of Equity Preferences for Life Expectancy Gains: A Discrete Choice Experiment Among the Japanese and Korean General Public

BACKGROUND

Setting priorities for limited public resources has become a topic of heated discussion the world over. Assigning different weights for the health gains of different population groups allows for equity considerations in cost-effectiveness analysis. However, only a few empirical works have elicited the preferences of the general public.

OBJECTIVE

To compare the equity preferemce assigned by Japanese and Koreans.

METHODS

We conducted a Web-based survey in March 2013, including a discrete choice experiment, to elicit the equity preferences of the general public for the life expectancy gains of different population groups. We selected attributes and designed the experiment following Norman et al.'s study (Norman R, Hall J, Street D, Viney R. Efficiency and equity: a stated preference approach. Health Econ 2013;22:568-81). Accordingly, we analyzed preference for sex, smoking status, lifestyle, caring status, income, and age.

RESULTS

The Japanese assigned a higher preference for males (P < 0.001), nonsmokers (P < 0.001), those with lower income (P < 0.001), and carers (P < 0 .001), and they assigned a lower preference for those with a life expectancy of 60 years (P = 0.002) and 75-year-olds (P < 0.001). Koreans have the same patterns of preference for lower income (P < 0.001), caring (P < 0.001), and smoking status (P = 0.026). However, they prefer both sexes (P = 0.331) and different age groups equally. In both countries, respondents tend to prefer groups with characteristics similar to their own.

CONCLUSIONS

People from the two Asian developed countries, with universal health insurance, show different equity preferences. These may reflect the variations in cultural background and coverage of health care services.

A cluster randomised trial of strategies to increase cervical screening uptake at first invitation (STRATEGIC)

BACKGROUND

Falling participation by young women in cervical screening has been observed at a time that has seen an increase in the incidence of cervical cancer in the UK in women aged < 35 years. Various barriers to screening have been documented, including fear, embarrassment and inconvenience.

OBJECTIVES

To measure the feasibility, clinical effectiveness and cost-effectiveness of a range of interventions to increase the uptake of cervical screening among young women.

DESIGN

A cluster randomised trial based on general practices performed in two phases.

SETTING

Primary care in Greater Manchester and the Grampian region in Scotland.

PARTICIPANTS

Phase 1: 20,879 women receiving their first invitation for cervical screening. Phase 2: 10,126 women who had not attended by 6 months.

INTERVENTIONS

Phase 1: pre-invitation leaflet or not, and access to online booking (Manchester only). Phase 2: (1) vaginal self-sampling kits (SSKs) sent unrequested (n = 1141); or (2) offered on request (n = 1290); (3) provided with a timed appointment (n = 1629); (4) offered access to a nurse navigator (NN) (n = 1007); or (5) offered a choice between a NN or a SSK (n = 1277); and 3782 women in control practices.

MAIN OUTCOME MEASURES

Uplift in screening compared with control practices, cost-effectiveness of interventions, and the women's preferences explored in a discrete choice experiment.

RESULTS

The pre-invitation leaflet and offer of online booking were ineffective when compared with control practices at 3 months, 18.8% versus 19.2% [odds ratio (OR) 0.96, 95% confidence interval (CI) 0.88 to 1.06; p = 0.485] and 17.8% versus 17.2% (OR 1.02, 95% CI 0.87 to 1.20; p = 0.802), respectively. The uptake of screening at 3 months was higher among previously human papillomavirus (HPV)-vaccinated women than unvaccinated women, 23.7% versus 11% (OR 2.07, 95% CI 1.69 to 2.53; p < 0.001). Among non-attenders, the SSK sent intervention showed a statistically significant increase in uptake at 12 months post invitation, 21.3% versus 16.2% (OR 1.51, 95% CI 1.20 to 1.91; p = 0.001), as did timed appointments, 19.8% versus 16.2% (OR 1.41, 95% CI 1.14 to 1.74; p = 0.001). The offer of a NN, a SSK on request, and a choice between timed appointments and NN were ineffective. Overall, there was a gradual rather than prompt response, as demonstrated by uptake among control practices. A discrete choice experiment indicated that women invited who had not yet attended valued the attributes inherent in self-sampling. The health economic analysis showed that both timed appointments and unsolicited SSK sent were likely to be cost-effective at a cost per quality-adjusted life-year (QALY) gained of £7593 and £8434, respectively, if extended across the national 25-year-old cohort throughout the duration of screening. The certainty of these being cost-effective at a ceiling ratio of £20,000 per QALY gained was > 90%.

CONCLUSION

Women receiving their initial screening invitation frequently delay taking up the offer and the net impact of interventions was small. Timed appointments and SSKs sent to non-attenders at 6 months are likely to be a cost-effective means of increasing uptake and should be considered further. HPV vaccination in the catch-up programme was associated with an increased uptake of cervical screening. Future work should focus on optimising self-sampling in terms of age range, timing of offer for non-attenders and use of urine testing instead of vaginal samples.

TRIAL REGISTRATION

Current Controlled Trials ISRCTN52303479.

FUNDING

This project was funded by the NIHR Health Technology Assessment programme and will be published in full in Health Technology Assessment; Vol. 20, No. 68. See the NIHR Journals Library website for further project information.

Using blocked fractional factorial designs to construct discrete choice experiments for healthcare studies

Discrete choice experiments (DCEs) are increasingly used for studying and quantifying subjects preferences in a wide variety of healthcare applications. They provide a rich source of data to assess real-life decision-making processes, which involve trade-offs between desirable characteristics pertaining to health and healthcare and identification of key attributes affecting healthcare. The choice of the design for a DCE is critical because it determines which attributes' effects and their interactions are identifiable. We apply blocked fractional factorial designs to construct DCEs and address some identification issues by utilizing the known structure of blocked fractional factorial designs. Our design techniques can be applied to several situations including DCEs where attributes have different number of levels. We demonstrate our design methodology using two healthcare studies to evaluate (i) asthma patients' preferences for symptom-based outcome measures and (ii) patient preference for breast screening services. Copyright © 2016 John Wiley & Sons, Ltd.

An empirical comparison of methods for analyzing correlated data from a discrete choice survey to elicit patient preference for colorectal cancer screening

Background

A discrete choice experiment (DCE) is a preference survey which asks participants to make a choice among product portfolios comparing the key product characteristics by performing several choice tasks. Analyzing DCE data needs to account for within-participant correlation because choices from the same participant are likely to be similar. In this study, we empirically compared some commonly-used statistical methods for analyzing DCE data while accounting for within-participant correlation based on a survey of patient preference for colorectal cancer (CRC) screening tests conducted in Hamilton, Ontario, Canada in 2002.

Methods

A two-stage DCE design was used to investigate the impact of six attributes on participants' preferences for CRC screening test and willingness to undertake the test. We compared six models for clustered binary outcomes (logistic and probit regressions using cluster-robust standard error (SE), random-effects and generalized estimating equation approaches) and three models for clustered nominal outcomes (multinomial logistic and probit regressions with cluster-robust SE and random-effects multinomial logistic model). We also fitted a bivariate probit model with cluster-robust SE treating the choices from two stages as two correlated binary outcomes. The rank of relative importance between attributes and the estimates of β coefficient within attributes were used to assess the model robustness.

Results

In total 468 participants with each completing 10 choices were analyzed. Similar results were reported for the rank of relative importance and β coefficients across models for stage-one data on evaluating participants' preferences for the test. The six attributes ranked from high to low as follows: cost, specificity, process, sensitivity, preparation and pain. However, the results differed across models for stage-two data on evaluating participants' willingness to undertake the tests. Little within-patient correlation (ICC ≈ 0) was found in stage-one data, but substantial within-patient correlation existed (ICC = 0.659) in stage-two data.

Conclusions

When small clustering effect presented in DCE data, results remained robust across statistical models. However, results varied when larger clustering effect presented. Therefore, it is important to assess the robustness of the estimates via sensitivity analysis using different models for analyzing clustered data from DCE studies.

Discrete choice experiments in health economics: a review of the literature

Discrete choice experiments (DCEs) have become a commonly used instrument in health economics. This paper updates a review of published papers between 1990 and 2000 for the years 2001-2008. Based on this previous review, and a number of other key review papers, focus is given to three issues: experimental design; estimation procedures; and validity of responses. Consideration is also given to how DCEs are applied and reported. We identified 114 DCEs, covering a wide range of policy questions. Applications took place in a broader range of health-care systems, and there has been a move to incorporating fewer attributes, more choices and interview-based surveys. There has also been a shift towards statistically more efficient designs and flexible econometric models. The reporting of monetary values continues to be popular, the use of utility scores has not gained popularity, and there has been an increasing use of odds ratios and probabilities. The latter are likely to be useful at the policy level to investigate take-up and acceptability of new interventions. Incorporation of interactions terms in the design and analysis of DCEs, explanations of risk, tests of external validity and incorporation of DCE results into a decision-making framework remain important areas for future research.

Choice experiments in health: the good, the bad, the ugly and toward a brighter future

Compared to many applied areas of economics, health economics has a strong tradition in eliciting and using stated preferences (SP) in policy analysis. Discrete choice experiments (DCEs) are one SP method increasingly used in this area. Literature on DCEs in health and more generally has grown rapidly since the mid-1990s. Applications of DCEs in health have come a long way, but to date few have been ‘best practice’, in part because ‘best practice’ has been somewhat of a moving target. The purpose of this paper is to briefly survey the history of DCEs and the state of current knowledge, identify and discuss knowledge gaps, and suggest potentially fruitful areas for future research to fill such gaps with the aim of moving the application of DCEs in health economics closer to best practice.

Exploring the social value of health-care interventions: a stated preference discrete choice experiment

Much of the literature on distributive preferences covers specific considerations in isolation, and recent reviews have suggested that research is required to inform on the relative importance of various key considerations. Responding to this research recommendation, we explore the distributive preferences of the general public using a set of generic social value judgments. We report on a discrete choice experiment (DCE) survey, using face-to-face interviews, in a sample of the general population (n=259). The context for the survey was resource allocation decisions in the UK National Health Service, using the process of health technology appraisal as an example. The attributes used covered health improvement, value for money, severity of health, and availability of other treatments, and it is the first such survey to use cost-effectiveness in scenarios described to the general public. Results support the feasibility and acceptability of the DCE approach for the elicitation of public preferences. Choice data are used to consider the relative importance of changes across attribute levels, and to model utility scores and relative probabilities for the full set of combinations of attributes and levels in the experimental design used (n=64). Results allow the relative social value of health technology scenarios to be explored. Findings add to a sparse literature on 'social' preferences, and show that DCE data can be used to consider the strength of preference over alternative scenarios in a priority-setting context.

Does attribute framing in discrete choice experiments influence willingness to pay? Results from a discrete choice experiment in screening for colorectal cancer

OBJECTIVE

Recent reviews of discrete choice methodology identified methodological issues warranting further exploration, including the issue of "framing." The objective of this study was to conduct a methodological exploration of the effect of attribute framing on marginal rates of substitution (MRS), including willingness to pay (WTP) from a discrete choice experiment (DCE), within the context of colorectal cancer screening preferences.

METHODS

The survey, a fractional factorial design of a two-alternative, unlabeled experiment, was mailed to a sample of 1920 subjects in NSW, Australia. Participants were randomized to one of four alternative "frames" of information. Attributes included: accuracy of the test for finding cancers, accuracy of the test for finding large polyps, how good the test is at saying you don't have cancer, cost, dietary and medication restrictions and sample collection. A mixed logit model was used to estimate preferences; MRS between attributes, including WTP, was calculated.

RESULTS

A total of 1157 surveys from 1920 (60.2%) were returned. Accuracy of the test for finding cancer was most likely to influence choice of test, followed by accuracy of the test for finding large polyps. Under some circumstances, framing of the attributes (e.g., cancers found vs. cancers missed) influenced the relative importance of attributes. Attribute framing significantly influenced estimates of WTP, and benefit: harm trade-offs that were calculated from MRS.

CONCLUSIONS

Attribute framing can influence willingness to pay and benefit: harm trade-offs from DCEs. Appropriate design and analysis methods should be explored to further characterize the influence and extent of framing in discrete choice studies.

Emotions and decision rules in discrete choice experiments for valuing health care programmes for the elderly

The evaluation of health care programmes is commonly approached with stated preference methods such as contingent valuation or discrete choice experiments. These methods provide useful information for policy decisions involving health regulations and infrastructures for health care. However, econometric modelling of these data usually relies on a number of maintained assumptions, such as the use of the compensatory or random utility maximization rule. On the other hand, health policy issues can raise emotional concerns among individuals, which might induce other types of choice behaviour. In this paper we consider potential deviations from the general compensatory rule, and how these deviations might be explained by the emotional state of the subject. We utilized a mixture econometric model which allows for various potential decisions rules within the sample, such as the complete ignorance, conjunctive rule and satisfactory rules. The results show that deviations from the full linear compensatory decision rule are predominant, but they are significantly less observed for those subjects with a medium emotional state about the issue of caring for the health state of the elderly. The implication is that the emotional impact of health policy issues should be taken into account when making assumptions of individual choice behaviour in health valuation methods.

Empirical investigation of experimental design properties of discrete choice experiments in health care

Experimental design is critical to valid inference from the results of discrete choice experiments (DCEs). In health economics, DCEs have placed limited emphasis on experimental design, typically employing relatively small fractional factorial designs, which allow only strictly linear additive utility functions to be estimated. The extensive literature on optimal experimental design outside health economics has proposed potentially desirable design properties, such as orthogonality, utility balance and level balance. However, there are trade-offs between these properties and emphasis on some properties may increase the random variability in responses, potentially biasing parameter estimates.This study investigates empirically the design properties of DCEs, in particular, the optimal method of combining alternatives in the choice set. The study involves a forced choice between two alternatives (treatment and non-treatment for a hypothetical health care condition), each with three, four-level, alternative-specific attributes. Three experimental design approaches are investigated: a standard six-attribute, orthogonal main effects design; a design that combines alternatives to achieve utility balance, ensuring no alternatives are dominated; and a design that combines alternatives randomly. The different experimental designs did not impact on the underlying parameter estimates, but imposing utility balance increases the random variability of responses.