Survey Development
An online survey was developed that included questions on demographics (age, sex, marital status, level of education, employment status), disease experience [time since diagnosis, current and target hemoglobin A1c (HbA1c) levels, and experience with T2DM treatments], and DCE questions about treatment choice.
The DCE questions posed a choice between two hypothetical treatments for T2DM, with each treatment being defined by a set of seven attributes at varying levels: (1) the chance of achieving the respondent’s HbA1c target; (2) reduced risk of serious heart attack or stroke; (3) frequency of hypoglycemic events; (4) risk of gastrointestinal (GI) problems; (5) changes in body weight; (6) mode of administration; (7) frequency of dosing (Table 1). In the final online survey, each of the seven attributes was described separately, and the descriptions were followed by questions about the attribute to ensure respondents’ understanding. The three probabilistic attributes, namely, the chance of achieving an HbA1c target, reduced risk of serious heart attack or stroke, and risk of GI problems, were explained graphically using risk grids in which colored dots identified the number of treated patients for whom the event occurred (Fig. 1). Research suggests that pictograms such as risk grids are an effective method of communicating risk [14]. Respondents were asked to review a sample risk grid, following which they were asked a comprehension question about the risk grid. Respondents who answered incorrectly were presented with the correct answer and an explanation before proceeding with the survey, while respondents who answered correctly were presented with the correct answer to reinforce their understanding. The survey was pretested to evaluate the survey instrument in 14 face-to-face interviews conducted in Berlin, Germany, and 15 face-to-face interviews conducted in Barcelona, Spain.
Table 1 Attribute levels for discrete-choice experiment
An experimental design was used to create the pairs of hypothetical treatments included in the DCE questions, following good practice guidelines [15]. The design was generated using SAS 9.3 analytics software (SAS Institute Inc., Cary, NC). The final design included 48 DCE questions divided into six blocks of eight questions each. Each patient was randomly assigned one block of DCE questions, and each DCE question asked the patient to indicate which treatment they would choose if the two treatments in the question were the only treatments available [14, 15].
Survey Populations
Patients were recruited from local communities for inclusion in survey pretesting and from an online national consumer panel for inclusion in the online survey. Patients were required to be residents of Germany or Spain, be aged ≥ 18 years, be able to read and understand German or Spanish, and provide informed consent. Eligible patients were also required to have a self-reported physician diagnosis of T2DM, have first started medication at least 2 years prior to taking the survey, and be receiving ongoing treatment with at least one T2DM prescription medication at the time of the survey. For the online survey, ≥ 200 respondents of both sexes in each country were sought. The study was approved by an institutional review board of RTI International (Research Triangle Park, NC, USA). All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1964, as revised in 2013. All patients provided informed consent prior to their inclusion in the study.
Statistical Analysis
The primary endpoint was the average preference weight for each attribute level of T2DM medications included in the survey (Table 1). The data from Germany and Spain were analyzed separately. The DCE choice data were estimated using random-parameters logit, a limited dependent variable regression model that avoids potential estimation bias from unobserved preference heterogeneity among respondents [16, 17]. All treatment attribute levels were included in the model as effects-coded categorical variables, except for the chance of reaching target HbA1c level, which was modeled as a linear continuous variable [12]. The coefficients on the independent variables from the random-parameters logit regression can be interpreted as relative preference weights, indicating the relative strength of preference for each attribute level. Larger positive coefficients indicated that respondents preferred that attribute level to levels that had smaller or negative coefficients. A Wald test was used to test for differences between adjacent attribute levels, and the NLOGIT 5 software package (Econometric Software, Inc., Plainview, NY) was used for the multivariate data analysis.
Separate random-parameters logit models were estimated and tested for differences in preferences between the following subgroups separately in each country: male versus female; age < 65 years versus age ≥ 65 years; self-reported T2DM diagnosis of < 7 years prior to survey versus ≥ 7 years prior to survey; use of injectable versus oral therapy at the time of the survey. The results of these exploratory analyses of preference heterogeneity are shown in the Electronic Supplementary Materials (ESM).
The estimated preference weights from the random-parameters logit model were used to calculate the relative preferences for medication profiles defined by the attributes and levels shown in Table 1. These medication profiles were created to approximate T2DM medications that would be the most likely comparators if a new injectable medication entered the market. For Germany, these were: a sulfonylurea (e.g., glipizide), a prandial insulin (e.g., insulin lispro), a once-daily glucagon-like peptide-1 receptor agonist (GLP-1 RA; e.g., liraglutide), and a once-weekly GLP-1 RA (e.g., albiglutide) (Table 2); for Spain, these were: a once-daily GLP-1 RA (e.g., liraglutide), a basal insulin (e.g., insulin glargine), a prandial insulin (e.g., insulin lispro), and a once-weekly GLP-1 RA (e.g., albiglutide) (Table 2). To calculate the proportion of the sample that may choose each of the four medication profiles, we calculated a net clinical benefit score for each medication as the weighted sum of each medication profile’s attribute levels, where the weights represent the relative strength of preference for the corresponding attribute level in the medication profile.
Table 2 Medication profiles that approximate actual treatments
A minimum acceptable benefit (MAB) for changes in attribute levels was also calculated. The MAB is interpreted as the minimum change in efficacy that respondents would require (on average) to accept changes in other attributes. In this case, it is the minimum increase in the probability that the medication will achieve the target HbA1c level required to compensate respondents for change to a less desirable level in another attribute. MAB was calculated as the difference between the preference weights for two levels of an attribute divided by the preference weight for increasing the probability of achieving the target HbA1c level by 1 percentage point.