Background and Objectives:
Understanding patient and public values and preferences is essential to healthcare and
policy decision making. Discrete choice experiments (DCEs) are a common tool used to capture and quantify these preferences. Recent technological advances allow for a variety of approaches to create and analyze DCEs. However, there is no optimal DCE design, nor analysis method.
Our objectives were to (i) survey DCE simulation studies to determine what design features
affect statistical efficiency, and assess their reporting, (ii) further investigate these findings with a de novo simulation study, and (iii) explore the sensitivity of individuals’ preference of attributes to several methods of analysis.
Methods:
We conducted a systematic survey of simulation studies within the health literature, created
a DCE simulation study of 3204 designs, and performed two empirical comparison studies. In one empirical comparison study, we determined addiction agency employees’ preferences on
knowledge translation attributes using four models, and in the second, we determined elementary school children’s choice of bullying prevention programs using nine models.
Results and Conclusions:
In our evaluation of DCE designs, we identified six design features that impact the
statistical efficiency of a DCE, several of which were further investigated in our simulation study. The reporting quality of these studies requires improvement to ensure that appropriate inferences can be made, and that they are reproducible. In our empirical comparison of statistical models to explore the sensitivity of individuals preferences of attributes, we found similar rankings in the relative importance measures of attributes’ mean part-worth utility estimates, which differed when using latent class models.
Understanding the impact of design features on statistical efficiency are useful for
designing optimal DCEs. Incorporating heterogeneity in the analysis of DCEs may be important to make appropriate inferences about individuals’ preferences of attributes within a population. / Thesis / Doctor of Philosophy (PhD) / This thesis focuses on the design and analysis of preference surveys, which are referred to
as discrete choice experiments. These surveys are used to capture and quantify individuals’
preferences on various characteristics describing a product or service. They are applied in various health settings to better understand a population. For example, clinicians may want to further understand a patient population’s preferences in regards to multiple treatment alternatives. Currently, there is no optimal approach for designing or analyzing preference surveys. We investigated what factors help improve the design of a preference survey by exploring the literature and conducting our own simulation study. We also investigated how sensitive the results of a preference survey were based on the statistical model used. Overall, we found that (i) increasing the amount of information presented and reducing the number of variables to explore will maximize the statistical optimality of the survey; and (ii) analyzing the data with different statistical models will yield similar results in the ranking of individuals’ preferences of the variables explored.
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/23642 |
| Date | 22 November 2018 |
| Creators | Vanniyasingam, Thuvaraha |
| Contributors | Thabane, Lehana, Clinical Epidemiology/Clinical Epidemiology & Biostatistics |
| Source Sets | McMaster University |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.002 seconds