Attribute Selection for a Discrete Choice Experiment Incorporating a Best-Worst Scaling Survey

Liu

JMIR Public Health Surveill

et al. 2022

Background China and the United States play critical leading roles in the global effort to contain the COVID-19 virus. Therefore, their population’s preferences for initial diagnosis were compared to provide policy and clinical insights. Objective We aim to quantify and compare the public’s preferences for medical management of fever and the attributes of initial diagnosis in the case of presenting symptoms during the COVID-19 pandemic in China and the United States. Methods We conducted a cross-sectional study from January to March 2021 in China and the United States using an online discrete choice experiment (DCE) questionnaire distributed through Amazon Mechanical Turk (MTurk; in the United States) and recruited volunteers (in China). Propensity score matching (PSM) was used to match the 2 groups of respondents from China and the United States to minimize confounding effects. In addition, the respondents’ preferences for different diagnosis options were evaluated using a mixed logit model (MXL) and latent class models (LCMs). Moreover, demographic data were collected and compared using the chi-square test, Fisher test, and Mann-Whitney U test. Results A total of 9112 respondents (5411, 59.4%, from China and 3701, 40.6%, from the United States) who completed our survey were included in our analysis. After PSM, 1240 (22.9%) respondents from China and 1240 (33.5%) from the United States were matched for sex, age, educational level, occupation, and annual salary levels. The segmented sizes of 3 classes of respondents from China were 870 (70.2%), 270 (21.8%), and 100 (8.0%), respectively. Meanwhile, the US respondents’ segmented sizes were 269 (21.7%), 139 (11.2%), and 832 (67.1%), respectively. Respondents from China attached the greatest importance to the type of medical institution (weighted importance=40.0%), while those from the United States valued the waiting time (weighted importance=31.5%) the most. Respondents from China preferred the emergency department (coefficient=0.973, reference level: online consultation) and fever clinic (a special clinic for the treatment of fever patients for the prevention and control of acute infectious diseases in China; coefficient=0.974, reference level: online consultation), while those from the United States preferred private clinics (general practices; coefficient=0.543, reference level: online consultation). Additionally, shorter waiting times, COVID-19 nucleic acid testing arrangements, higher reimbursement rates, and lower costs were always preferred. Conclusions Improvements in the availability of COVID-19 testing and medical professional skills and increased designated health care facilities may help boost potential health care seeking during COVID-19 and prevent unrecognized community spreading of SARS-CoV-2 in China and the United States. Moreover, to better prevent future waves of pandemics, identify undiagnosed patients, and encourage those undiagnosed to seek health care services to curb the pandemic, the hierarchical diagnosis and treatment system needs improvement in China, and the United States should focus on reducing diagnosis costs and raising the reimbursement rate of medical insurance.

Section: Methodsmentioning

confidence: 99%

Preferences for Attributes of Initial COVID-19 Diagnosis in the United States and China During the Pandemic: Discrete Choice Experiment With Propensity Score Matching

Liu

JMIR Public Health Surveill

et al. 2022

“…The questionnaire can also have simple rank ordering questions applied to focus groups (Morgan et al, 2000). Webb et al (2021) used Best-Worst scaling as a strategy to reduce the number of attributes. Three stages of reduction are proposed in table 2.…”

Section: Attribute Reductionmentioning

confidence: 99%

“…The latter normally elicit fewer attributes. Notwithstanding this, in Health Economics the reduction of attributes is also necessary, with inadequate information (Webb et al, 2021). There are important exceptions to this in health economics, for example, Helter andBoehler 2016 andObadha et al, 2019.…”

Section: Introductionmentioning

confidence: 99%

“…These contributions in healthcare can also be useful for consumer research studies as well, and for that reason, worthy of discussion in this paper. The author deemed convenient to find a method to elicit attributes in the context of restaurants and later found that this method could be shared to enlighten consumer researchers and address Webb et al's (2021) about the lack of information on this topic.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Determining Attributes for Discrete Choice Experiments in Consumer Research

Longart

2022

TMJ

Attribute development for Discrete Choice Experiments (DCEs) through qualitative research is a shallowly covered area. Particularly, research is scant in terms of detail about attribute reduction and development. First of all, the paper reviews the literature and justify DCEs as an appropriate method for determining consumer preferences, with particular focus on services. The researcher conducted a DCE on preferences for restaurants for leisure purposes. Whilst finding there was not a clear guidance, the author proposes a clear methodology for guiding DCE researchers regarding the determination of attributes in consumer research, particularly in the service industry. The case study of restaurant attribute development is used. In this context, the first stage consisted in a thorough review of the literature combined with a thematic analysis of six focus group interviews conducted in the South East of the UK. Attribute reduction took place through constant comparison analysis and the method is comprehensively explained in the paper. The study also looked for control variables and found that occasion for eating out is a very salient variable to be tested in the following DCE which consisted in an Adaptive Choice Based Conjoint online survey (Sawtooth Software). It was found the that combining literature review with focus group interviews results in richer attributes because they are based on more complex and nuanced data. Finally, a proposal with nine steps for attribute development is provided.

“…Researchers are often presented with a list of objects (e.g., from qualitative studies, Delphi process, or review of endpoints) and additional quantification would be beneficial. On other occasions, there are too many concepts being reviewed for a preference study or some attributes may be difficult to assign levels (e.g., stigma), and a prioritization study may be preferable or a first step towards a preference study [ 32 ]. Finally, there are many research questions or study purposes that are better suited to a prioritization task, including the selection of endpoints, optimizing communication in healthcare, assessment of the barriers and facilitators to implementation, and many more.…”

Section: Introductionmentioning

confidence: 99%

Best–Worst Scaling and the Prioritization of Objects in Health: A Systematic Review

et al. 2022

Background and ObjectiveBest-worst scaling is a theory-driven method that can be used to prioritize objects in health. We sought to characterize all studies of best-worst scaling to prioritize objects in health, to assess trends of using best-worst scaling in prioritization over time, and to assess the relationship between a legacy measure of quality (PREFS) and a novel assessment of subjective quality and policy relevance. Methods A systematic review identified studies published through to the end of 2021 that applied best-worst scaling to study priorities in health (PROSPERO CRD42020209745), updating a prior review published in 2016. The PubMed, EBSCOhost, Embase, Scopus, APA PsychInfo, Web of Science, and Google Scholar databases were used and were supplemented by a hand search. Data describing the application, development, design, administration/analysis, quality, and policy relevance were summarized and we tested for trends by comparing articles before and after 1 January, 2017. Multivariate statistics were then used to assess the relationships between PREFS, subjective quality, policy relevance, and other possible indicators. Results From a total of 2826 unique papers identified, 165 best-worst scaling studies were included in this review. Applications of best-worst scaling to study priorities in health have continued to grow (p < 0.01) and are now used in all regions of the world, most often to study the priorities of patients/consumers (67%). Several key trends can be observed over time: increased use of pretesting (p < 0.05); increased use of online administration (p < 0.01), and decreased use of paper selfadministered surveys (p = 0.02); increased use of heterogeneity analysis (p = 0.02); an increase in having a clearly stated purpose (p < 0.01); and a decrease in comparing respondents to non-respondents (p = 0.01). The average sample size has more than doubled, from 228 to 472 respondents, but formal sample size justifications remain low (5.3%) and unchanged over time (p = 0.68). While the average PREFS score remained unchanged at 3.1/5, both subjective quality and policy relevance trended up, but changes were not statistically significant (p = 0.06 and p = 0.13). Most of the variation in subjective quality was driven by PREFS (R 2 = 0.42), but it was also positively assosciated with policy relevance, heterogeneity analysis, and using a balanced incomplete block design, and was negatively associated with not using developmental methods and an increasing sample size. Conclusions Using best-worst scaling to prioritize objects is now commonly used around the world to assess the priorities of patients and other stakeholders in health. Best practices are clearly emerging for best-worst scaling. Although legacy measures (PREFS) to measure study quality are reasonable, there may need to be new tools to assess both study quality and policy relevance.