For this study, a cross-sectional, web-based DCE survey was developed and administered to PwT2D. The study design and analyses were developed following good research practices as defined by ISPOR (The Professional Society for Health Economics and Outcomes Research) guidelines and standard practice in preference studies [10,11,12].

The study was reviewed and deemed exempt from full review by the independent RTI International institutional review board. This study was performed in accordance with the Helsinki Declaration of 1964 and its later amendments. All participants gave written informed consent. Respondents who participated in this study were informed that the results may be published and were reassured that, if the results of this study were to be presented at scientific meetings or published in scientific journals, the data would be presented in summary form and no information would be included that could identify them or their answers personally.

To identify relevant attributes and levels for the DCE, we first conducted in-depth qualitative telephone interviews with 21 PwT2D (15 insulin experienced and 6 insulin naive) to better understand their experiences with diabetes and its treatments. Specifically, these interviews focused on participants’ experiences with currently available daily basal insulin treatments and their preferences for a once-weekly, long-acting basal insulin. The qualitative interviews found that many PwT2D who were insulin naive were resistant to initiating insulin, and some felt it would be an “inconvenient” treatment. All participants were generally receptive to the possibility of a once-weekly basal regimen, reporting that fewer injections and greater convenience/flexibility would be less stressful and would relieve worries about taking or remembering insulin administrations.

Using the results of the qualitative interviews, clinical trial data, published data, information about long-acting basal insulins that are currently available or under development, and information from subject matter experts, we identified treatment attributes and levels that were clinically relevant to PwT2D. The six attributes were reduction in hemoglobin A1c (A1c) level after 6 months, time spent in optimal blood sugar range each day, number of serious low blood sugar events, number of nighttime low blood sugar events, change in weight because of insulin over 6 months, and frequency of taking long-acting insulin. The levels for each attribute were selected to span a clinically relevant range of outcomes (Table 1).

A draft version of the DCE survey instrument was evaluated in nine individual pretest interviews conducted by telephone. Pretest interviews used a cognitive debriefing technique called “think aloud,” [13] in which participants were asked to verbalize their thoughts as they completed the draft survey. After completing the survey in this manner, participants were asked a series of debriefing questions to determine whether they understood the definitions and instructions, accepted the hypothetical context of the survey, and successfully completed the choice questions in the survey instrument as instructed. The pretest interviews resulted in minor revisions to the survey to ensure that survey respondents understood the questions.

In the final survey instrument that was administered online, respondents were presented with a series of eight DCE questions that each asked them to choose between two hypothetical treatment profiles. The profiles did not reflect actual treatments but were rather combinations of attribute levels generated from an experimental design and generically labeled as “Insulin A” and “Insulin B” in each choice question. The final attributes and levels included in the DCE survey to inform the hypothetical treatment profiles are shown in Table 1, and an example of a discrete-choice question is shown in Fig. 1. The survey contained an additional fixed direct-elicitation question to determine whether respondents preferred weekly rather than daily insulin, all else being equal in terms of risks (< 1 clinically significant and < 1 nocturnal hypoglycemic event per year) and benefits (up to 0.5 point reduction in A1c and 18 h spent in range) (see Supplementary Figure 1).

Example of a DCE question. A1c hemoglobin A1c, DCE discrete-choice experiment

The survey instrument contained questions on the respondents’ experiences with T2D, descriptions of the attributes, comprehension questions, the eight DCE choice questions, a direct-elicitation question, and demographic questions.

Potential participants were identified by SurveyEngine through online panels and invited to be screened for study eligibility. If eligible, they were provided with a unique link to the survey. PwT2D were recruited to the study if they were aged ≥ 18 years, were able to read and understand English, provided informed consent, resided in the US at the time of the study, had a self-reported physician diagnosis of T2D (received at least 6 months before the survey), and were not currently using an insulin pump but were currently receiving medication for their diabetes. As no once-weekly insulins were approved at the time of data collection, it was anticipated that participants would only have experience with daily insulin injections. Targets were used to ensure that approximately one-third of study participants were currently treated with basal-bolus insulin injections (with or without other diabetes medicines), one-third were treated with basal insulin (with or without other diabetes medicines), and one-third were treated with a diabetes medicine but had no experience with insulin treatment (i.e., were insulin naive).

Descriptive statistics were used to describe the sample of PwT2D. A regression analysis using a random-parameters logit (RPL) model was performed to estimate preference weights from the attributes in the DCE. Preference weights indicate the relative strength of preference for each attribute level included in the survey. More-preferred outcomes have higher preference weights. The statistical significance of differences between adjacent attribute levels was determined using a Wald χ2 test. Preference weights from the RPL model were used to calculate the conditional relative importance of each attribute, which was calculated as the difference between the most- and least-preferred level for each attribute.

Using a series of χ2 tests, we also explored whether there were systematic differences in preferences between subgroups of participants by characteristics of interest. These characteristics included those with experience with different treatment regimens, those with experience with low blood sugar events, and those with different levels of glycemic control. To analyze responses to the direct-elicitation choice question, a logistic regression model with covariates (i.e., indicators of various statuses related to sociodemographics and disease and treatment experience, as well as other variables of interest) was used to determine the association between respondent characteristics and treatment choices. A p value < 0.05 was considered statistically significant. All analyses were performed in STATA 16 (StataCorp; College Station, Texas). Data from all respondents were analyzed together.