The Garbage Class Mixed Logit Model: Accounting for Low-Quality Response Patterns in Discrete Choice Experiments

A Direct Comparison Between Discrete Choice With Duration and Composite Time Trade-Off Methods: Do They Produce Similar Results?

OBJECTIVES

Discrete choice experiments including a duration attribute (DCEd) represent a promising candidate method for valuing health-related quality-of-life instruments. However, it has not been established that DCEd can produce similar results as composite time trade-off (cTTO) or EuroQol Valuation Technology (EQ-VT) valuations of the EQ-5D-5L instrument. This study provides a direct comparison between cTTO and EQ-VT, and DCEd valuation methods.

METHODS

An EQ-VT study was conducted in Trinidad and Tobago to value the EQ-5D-5L. 1079 respondents each completed 10 cTTO tasks and 12 discrete choice experiments tasks without a duration attribute. A separate sample of 970 respondents each completed 18 split-triplet DCEd tasks. Several regression models were applied to the EQ-VT data, and the DCEd data were analyzed using mixed logit models with an exponential discount rate. The estimated values were compared using scatterplots and Bland-Altman plots.

RESULTS

The ordering of dimensions was identical in level 5 for cTTO/EQ-VT and DCEd models, with pain/discomfort being the most important dimension and usual activities being least important. cTTO/EQ-VT models produced a value for state 55555 ranging between -0.52 and -0.69, whereas this was -0.543 for the nonlinear mixed logit model for the DCEd data. Scatterplots and Bland-Altman plots suggested excellent agreement between cTTO/EQ-VT and DCEd-based estimates.

CONCLUSIONS

CTTO/EQ-VT and DCEd valuations produce similar results when correcting DCEd for nonlinear time preferences. The ordering of importance of the dimensions and scale are identical, suggesting that the 2 methods measure the same construct and produce similar results.

Navigating Public Policy Responses to a Pandemic: The Balancing Act Between Physical Health, Mental Health, and Household Income

OBJECTIVE

During COVID-19, governments imposed restrictions that reduced pandemic-related health risks but likely increased personal and societal mental health risk, partly through reductions in household income. This study aimed to quantify the public's willingness to accept trade-offs between pandemic health risks, household income reduction, and increased risk of mental illness that may result from future pandemic-related policies.

METHODS

A total of 547 adults from an online panel participated in a discrete choice experiment where they were asked to choose between hypothetical future pandemic scenarios. Each scenario was characterized by personal and societal risks of dying from the pandemic, experiencing long-term complications, developing anxiety/depression, and reductions in household income. A latent class regression was used to estimate trade-offs.

RESULTS

Respondents state a willingness to make trade-offs across these attributes if the benefits are large enough. They are willing to accept 0.8% lower household income (0.7-1.0), 2.7% higher personal risk of anxiety/depression (1.8-3.6), or 3.2% higher societal rate of anxiety/depression (1.7-4.7) in exchange for 300 fewer deaths from the pandemic.

CONCLUSION

Results reveal that individuals are willing to accept lower household income and higher rates of mental illness, both personal and societal, if the physical health benefits are large enough. Respondents placed greater emphasis on maintaining personal, as opposed to societal, mental health risk and were most interested in preventing pandemic-related deaths. Governments should consider less restrictive policies when pandemics have high morbidity but low mortality to avoid the prospect of improving physical health while simultaneously reducing net social welfare.

What Can Discrete-Choice Experiments Tell Us about Patient Preferences? An Introduction to Quantitative Analysis of Choice Data

This paper provides an introduction to statistical analysis of choice data using example data from a simple discrete-choice experiment (DCE). It describes the layout of the analysis dataset, types of variables contained in the dataset, and how to identify response patterns in the data indicating data quality. Model-specification options include linear models with continuous attribute levels and non-linear continuous and categorical attribute levels. Advantages and disadvantages of conditional logit, mixed logit, and latent-class analysis are discussed and illustrated using the example DCE data. Readers are provided with links to various software programs for analyzing choice data. References are provided on topics for which there currently is limited consensus and on more advanced techniques to guide readers interested in exploring choice-modeling challenges in greater depth. Supplementary materials include the simulated example data used to illustrate modeling approaches, together with R and Matlab code to reproduce the estimates shown.

Scale and rate heterogeneity in the EQ-5D-5L valuation

OBJECTIVES

To estimate values on a quality-adjusted life year (QALY) scale using individual preference evidence, choice analyses typically include ancillary parameters, such as scale factors and discount rates. These parameters potentially differ among respondents. In this study, we investigated how allowing heterogeneity in scale and rate affects the estimation of EQ-5D-5L values.

METHODS

Using the first wave of the 2016 EQ-5D-5L valuation study (N = 1017), we estimated a conditional logit (CL) model and three mixed logit models: random scale, random rate, and bivariate. Prior to the exploratory study, we hypothesized that scale and rate are correlated and that allowing heterogeneity in both parameters decreases the number of insignificant incremental effects. We confirmed the exploratory findings by re-estimating these models using paired comparison responses from a second wave (N = 1229).

RESULTS

Scale and rate exhibited significant heterogeneity and were positively correlated. As hypothesized, allowing this heterogeneity improved the face validity of the EQ-5D-5L value set by reducing the number of insignificant incremental effects (from 6 to 2 p-values > 0.05; out of 20). Nevertheless, the CL and bivariate mixed logit estimates are highly correlated and concordant (Pearson correlation coefficient of 0.897, Spearman correlation coefficient of 0.888, Lin's concordance coefficient of 0.763).

CONCLUSIONS

Allowing this heterogeneity adds three parameters to the estimation (two variances and a correlation) and improves the face validity of the EQ-5D-5L values. This finding may influence experimental design and choice analysis in health valuation more generally.

Exploring the Factors that Drive Clinical Negligence Claims: Stated Preferences of Those Who Have Experienced Unintended Harm

BACKGROUND

Better understanding of the factors that influence patients to make a financial claim for compensation is required to inform policy decisions. This study aimed to assess the relative importance of factors that influence those who have experienced a patient safety incident (PSI) to make a claim for compensation.

METHOD

Participants completed an online discrete choice experiment (DCE) involving 10 single profile tasks where they chose whether or not to file a claim. DCE data were modelled using logistic, mixed logit and latent class regressions; scenario analyses, external validity, and willingness to accept were also conducted.

RESULTS

A total of 1029 participants in the United Kingdom responded to the survey. An appropriate apology and a satisfactory investigation reduced the likelihood of claiming. Respondents were more likely to claim if they could hold those responsible accountable, if the process was simple and straightforward, if the compensation amount was higher, if the likelihood of compensation was high or uncertain, if the time to receive a decision was quicker, and if they used the government compensation scheme. Men are more likely to claim for low impact PSIs.

DISCUSSION AND CONCLUSIONS

The actions taken by the health service after a PSI, and people's perceptions about the probability of success and the size of potential reward, can influence whether a claim is made. Results show the importance of giving an appropriate apology and conducting a satisfactory investigation. This stresses the importance around how patients are treated after a PSI in influencing the clinical negligence claims that are made.

An Overview of Data Collection in Health Preference Research

This paper focuses on survey administration and data collection methods employed for stated-preference studies in health applications. First, it describes different types of survey administration methods, encompassing web-based surveys, face-to-face (in-person) surveys, and mail surveys. Second, the concept of sampling frames is introduced, clarifying distinctions between the target population and survey frame population. The discussion then extends to different types of sampling methods, such as probability and non-probability sampling, along with an evaluation of potential issues associated with different sampling methods within the context of health preference research. Third, the paper provides information about different recruitment methods, including web-surveys, leveraging patient groups, and in-clinic recruitment. Fourth, a crucial aspect addressed is the calculation of response rate, with insights into determining an adequate response rate and strategies to improve response rates in stated-preference surveys. Lastly, the paper concludes by discussing data management plans and suggesting insights for future research in this field. In summary, this paper examines the nuanced aspects of survey administration and data collection methods in stated-preference studies, offering valuable guidance for researchers and practitioners in the health domain.

Prognostic Information on Progression to Dementia: Quantification of the Impact on Quality of Life

Background

The increasing interest in early identification of people at risk of developing dementia, has led to the development of numerous models aimed at estimating the likelihood of progression from mild cognitive impairment (MCI) to dementia. It is important to study both the need for and possible outcomes related with such prediction models, including the impact of risk predictions on perceived quality of life (QoL).

Objective

This study aimed to quantify the impact that receiving a risk prediction on progression from MCI to dementia has on QoL.

Methods

A Discrete Choice Experiment (DCE) and Time Trade Off (TTO) study were performed. Participants completed choice tasks related to dementia prognosis while imagining having MCI. We collected DCE data by an online survey, and TTO data via videoconferencing interviews. DCE data were analyzed using a mixed multinomial logit model and were anchored to a health state utility scale using mean observed TTO valuations.

Results

296 people participated in the DCE and 42 in the TTO. Moderate and high predicted dementia risks were associated with decrements in utility (-0.05 and -0.18 respectively), compared to no prognostic information. Low predicted risk was associated with an increase in utility (0.06), as well as the availability of medication or lifestyle interventions (0.05 and 0.13 respectively).

Conclusions

This study shows a significant impact of dementia risk predictions on QoL and highlights the importance of caution when sharing information about expected MCI disease courses.

This Dead or That Dead: Framing Effects in the Evaluation of Health States

OBJECTIVES

Current approaches to health state valuation rely on credible classification of states as either "better than dead" or "worse than dead" (WTD). We investigate how such evaluations of health states are affected by the framing in pairwise comparison tasks.

METHODS

We conducted an online survey with 361 participants to compare the propensity to evaluate a state as WTD under 6 frames: (A) using a regular time trade-off (TTO) choice task, (B) using a lead-time TTO choice task, (C) excluding the immediacy of death, (D) avoiding reference to the process of dying, (E) focusing on longevity, and (F) focusing on improvements in quality of life. Each participant evaluated 9 EQ-5D-5L health states using 3 frames. The frames were compared using several statistical approaches to confirm robustness to indirect comparisons or respondent heterogeneity and inattentiveness.

RESULTS

The odds of a state being evaluated as WTD, compared with frame A, increase 2.7-fold and 1.5-fold in frame B and E, respectively, and decrease >5-fold in frame F. Frames C and D do not differ significantly from frame A.

CONCLUSIONS

Different framings for questions about whether a state is WTD or better than dead, even if theoretically equivalent, yield substantially different results. Notably, whether a state is evaluated as WTD differs greatly between the regular TTO and lead-time TTO choice tasks and when either final outcomes or improvements over time are considered.

Discrete-Choice Experiment to Understand the Preferences of Patients with Hormone-Sensitive Prostate Cancer in the USA, Canada, and the UK

BACKGROUND

Treatment options for patients with metastatic hormone-sensitive prostate cancer (mHSPC) have broadened, and treatment decisions can have a long-lasting impact on patients' quality of life. Data on patient preferences can improve therapeutic decision-making by helping physicians suggest treatments that align with patients' values and needs.

OBJECTIVE

This study aims to quantify patient preferences for attributes of chemohormonal therapies among patients with mHSPC in the USA, Canada, and the UK.

METHODS

A discrete-choice experiment survey instrument was developed and administered to patients with high- and very-high-risk localized prostate cancer and mHSPC. Patients chose between baseline androgen-deprivation therapy (ADT) alone and experimentally designed, hypothetical treatment alternatives representing chemohormonal therapies. Choices were analyzed using logit models to derive the relative importance of attributes for each country and to evaluate differences and similarities among patients across countries.

RESULTS

A total of 550 respondents completed the survey (USA, 200; Canada, 200; UK, 150); the mean age of respondents was 64.3 years. Treatment choices revealed that patients were most concerned with treatment efficacy. However, treatment-related convenience factors, such as route of drug administration and frequency of monitoring visits, were as important as some treatment-related side effects, such as skin rash, nausea, and fatigue. Patient preferences across countries were similar, although patients in Canada appeared to be more affected by concomitant steroid use.

CONCLUSION

Patients with mHSPC believe the use of ADT alone is insufficient when more effective treatments are available. Efficacy is the most significant driver of patient choices. Treatment-related convenience factors can be as important as safety concerns for patients.

Taking the Shortcut: Simplifying Heuristics in Discrete Choice Experiments

Health-related discrete choice experiments (DCEs) information can be used to inform decision-making on the development, authorisation, reimbursement and marketing of drugs and devices as well as treatments in clinical practice. Discrete choice experiment is a stated preference method based on random utility theory (RUT), which imposes strong assumptions on respondent choice behaviour. However, respondents may use choice processes that do not adhere to the normative rationality assumptions implied by RUT, applying simplifying decision rules that are more selective in the amount and type of processed information (i.e., simplifying heuristics). An overview of commonly detected simplifying heuristics in health-related DCEs is lacking, making it unclear how to identify and deal with these heuristics; more specifically, how researchers might alter DCE design and modelling strategies to accommodate for the effects of heuristics. Therefore, the aim of this paper is three-fold: (1) provide an overview of common simplifying heuristics in health-related DCEs, (2) describe how choice task design and context as well as target population selection might impact the use of heuristics, (3) outline DCE design strategies that recognise the use of simplifying heuristics and develop modelling strategies to demonstrate the detection and impact of simplifying heuristics in DCE study outcomes.