Overview of study design

Because generic preference-based measures such as the EQ-5D do not specifically assess the impact of treatment process or treatment convenience, generic instruments are unlikely to be sensitive to differences related to treatment process attributes. Therefore, utilities were estimated using vignette-based methods, which are well-suited for isolating the impact of specific treatment-related attributes on utility. These methods were similar to those of two previously published studies that estimated the utility impact of treatment-related attributes in the context of type 2 diabetes [22, 35]. The utilities derived in these previous studies have been used in a range of published CUAs of treatments for type 2 diabetes [36–39]. The health states in the current study used the same basic description of type 2 diabetes that had been used in these two previous utility studies, with additional content describing weekly injectable treatment and variations in the injection devices.

Like the two previous studies, the current study estimated utilities for treatment-related attributes based on a valuation study in a sample of patients with type 2 diabetes. First, health state vignettes were developed and refined based on clinician interviews, published literature, injection device instructions for use, and a pilot study. Then, the health states were valued by patients in a time trade-off (TTO) task with a 20-year time horizon. By comparing utility values across the health states, it was possible to estimate the disutility associated with various combinations of treatment process attributes.

Health state development

Health state descriptions (often called vignettes or scenarios) were drafted based on health states administered in two previous studies [22, 35], published literature [40–42], interviews with four clinicians who had research and clinical experience with type 2 diabetes (all with MD degrees), and instructions for use of the three devices associated with weekly GLP-1 receptor agonists [28–30]. First, health states administered in the two previous studies were adapted for the description of type 2 diabetes in all seven health states. Then, a literature search and clinician interviews were conducted to identify injection device attributes that were likely to be important to patients and differentiate among devices used to inject weekly GLP-1 receptor agonists. In addition, descriptions provided by clinicians were used to inform the development of health state content. Finally, instructions for use of the three available GLP-1 receptor agonist devices were used to finalize details of health state content, including the images and descriptions presented on the device display page (described below).

All health states were presented on individual cards, each with a series of bullet point descriptions. Health state text is presented in the paragraphs below. To ensure that respondents had a clear understanding of the health states, a device display page was given to each participant. This page summarized and illustrated characteristics of an injection pen and the three injection process attributes. The descriptions were derived from the instructions for use of each injection pen (Appendix A), and the illustrations were taken directly from these instructions [28–30]. Prior to presenting health states, the interviewer reviewed this device display page with each participant to ensure that he/she/they understood the relevant concepts. Respondents also referred to the device display page as needed throughout the interview.

The first health state (health state A), called the basic health state, described a patient with type 2 diabetes on oral treatment: “You have had type 2 diabetes for several years. You are at your current weight. You take oral medication (pills or tablets). Your blood sugar levels are usually in control, but sometimes your blood sugar is too high or too low. If your blood sugar level is too low, you may experience dizziness/light-headedness, sweating, or shaking. If your blood sugar level is too high, you may experience tiredness, blurred vision, thirst, or frequent urination.” This health state, which provided context for the injection-related attributes described in subsequent health states, was based on health states administered in two previous studies [22, 35].

Six additional health states (i.e., health states B to G or “the injectable health states”) started with the same content as health state A, but added a weekly injectable treatment. After the descriptions from health state A, each of these six health states included the same bullet point: “In addition to taking oral medication (pills or tablets), you give yourself an injection ONCE EACH WEEK using a device called an injection pen.” Then, these health states included three statements, each representing an injection process attribute.

The three injection process attributes (reconstitution, waiting, and needle handling) were selected because they were likely to differentiate among the three currently available once-weekly GLP-1 receptor agonists: albiglutide [30], dulaglutide [29], and once-weekly exenatide [28]. The first attribute was the requirement for reconstitution, referred to as “mixing.” Albiglutide and once-weekly exenatide both require reconstitution of the medication prior to the injection. It was hypothesized that some participants would prefer health states without this requirement. The description of reconstitution on the device display page was designed to be general enough to apply to both albiglutide (which requires “rocking” the pen) and once-weekly exenatide (which requires “tapping” the pen). Each of the injection health states included one of the following two descriptions: “You follow a sequence of several steps to mix the medication before injecting” or “When you open the package, the medication is ready to be injected. You do NOT need to mix the medication prior to injection.”

The second attribute was called “waiting.” As part of the reconstitution process, albiglutide requires a waiting period of 15 to 30 minutes, while neither of the other two weekly treatments require waiting. It was hypothesized that some participants would prefer not to have this extra requirement. The waiting requirement was represented by one of the following two statements: “During the steps required for mixing the medication, you have to stop and wait 15 to 30 minutes for the medication to mix” or “During the steps required for mixing the medication, you do NOT have to stop and wait for the medication to mix.”

The third attribute was the requirement to handle the needle. Whereas albiglutide and once-weekly exenatide require the patient to attach a needle to the injection pen for each injection, the dulaglutide pen includes a pre-attached needle that automatically retracts into the device after use. It was hypothesized that some respondents may have a preference for an injection device that does not require needle handling. Each of the injection health states included one of the following two descriptions: “You have to handle the needle and attach it to the pen” or “When you open the injection pen package, the needle is already included as a part of the injection pen. You do NOT need to handle the needle or attach it to the pen.”

Participants

Participants were required to meet these inclusion criteria: (1) diagnosed with type 2 diabetes by a recognized medical professional; (2) be between the ages of 30 and 75 years old; (3) able to identify the age at which they were first diagnosed with diabetes; (4) able to read and understand English; (5) willing and able to give informed consent. Participants were not eligible if they had a cognitive impairment, hearing difficulty, or severe pathology that could interfere with their ability to complete the interview. To verify diagnosis of type 2 diabetes, participants receiving pharmaceutical treatment were required to bring proof of medication to the interviews (e.g., medication packaging or a letter from a doctor). Participants who were not taking medication were required to describe their symptom history, diagnosis process, and disease management strategies at a level of detail suggesting that they were honestly reporting their diagnoses.

Participants were recruited via newspaper advertisements, online advertisements, fliers posted near interview sites, and advertisements in newsletters distributed by patient advocacy groups. A total of 301 individuals were screened to assess whether they met inclusion criteria. Of the 301 screened participants, 44 were ineligible because they did not meet inclusion criteria. Of the 257 who were eligible, 235 were scheduled, and 214 attended their interviews. Three of the 214 participants were unable to sufficiently understand the utility interview procedures in order to provide valid data, and two participants were determined to be ineligible upon rescreening prior to consent (one was unable to complete protocol requirements due to visual impairment, and the other was unable to provide proof of treatment or diagnosis). Thus, a total of 209 (150 Scotland, 59 England) valid interviews were completed.

Pilot study

To assess the comprehensibility of the draft health states and determine whether the TTO methods were feasible for valuing these particular health states, a pilot study was conducted with 26 patients with type 2 diabetes in London, United Kingdom (UK) (84.6% male; mean age = 57.7 years; age range = 35 to 73). Participants valued the health states using TTO methods with 10-year and 20-year time horizons. The TTO method with both time horizons was easy for participants to understand and complete. The 20-year time horizon was selected over 10 years because it better represents the chronic nature of type 2 diabetes.

During the pilot study, the health states were revised based on patient feedback in order to maximize clarity and comprehension. To shorten health states, some of the details about injection pens and the treatment process attributes were deleted from the health states and presented instead on the device display page. This change appeared to make the task easier for participants. After finalizing the health states and methods, participants consistently reported that the health states, device display page, and TTO procedures were clear and easy to understand.

Utility interview procedures and scoring

After the pilot study, health state utilities were elicited in a TTO valuation study in four UK locations: London, England, and three locations in Scotland (Inverness, Portree, and Edinburgh). All participants provided written informed consent. Procedures and materials (for both the pilot study and the full utility valuation study) were approved by an independent Institutional Review Board (Ethical & Independent Review Services; Study Number 15022). In each geographic location, interviews were conducted at a facility with multiple private offices so that every one-on-one interview could be conducted in a quiet private room. There were five interviewers (including three of the study authors and two additional interviewers mentioned in the acknowledgments section). The utility assessment was conducted by following a semi-structured interview script in order to standardize the utility assessment procedures. The interview team was trained by the principal investigator, who also observed each interviewer multiple times to ensure that procedures were followed consistently.

In each interview, the basic health state (health state A) was presented first. Then, the device display page was presented, followed by the six health states describing treatment with injectable medication (B to G) in random order. Every participant valued all seven health states. During this process, interviewers reviewed the treatment process attributes using the standardized language on the device display page. As an introductory task, participants were asked to rank the health states in order of preference. Then, participants valued the health states in a TTO task with a 20-year time horizon and 1-year (i.e., 5%) trading increments. Following commonly used procedures for each health state [4], participants were offered a choice between spending 20 years in the health state versus spending varying amounts of time in full health. Choices alternated between longer and shorter time periods (i.e., 20 years, 0 years [i.e., dead], 19 years, 1 year, 18 years, 2 years, 17 years, 3 years…). For each health state, the utility value was calculated based on the point of indifference between y years in the health state being evaluated (i.e., 20 years) and x years in full health (followed by dead). The resulting utility estimate (u) was calculated as u = x/y. For example, if 20 years in a health state was perceived to be equally preferable to 16 years in full health, the utility of that health state would be 0.80 (i.e., 16/20).

If a participant indicated that a health state was worse than dead, the task and scoring procedures were altered as described in previous literature [43, 44]. Participants were offered a choice between dead (choice 1) and 20 years (choice 2) beginning with varying amounts of time in the health state being rated, followed by full health. Then, utility scores were calculated with a bounded scoring approach (u = −x / t, where x is the time in full health, and t is the total life span of choice 2), which is commonly used to avoid highly skewed distributions for negative utility scores.

EQ-5D-3L and EQ-5D-5L

The EQ-5D-3L and EQ-5D-5L were administered to characterize the sample in terms of quality of life and overall health status. Both versions were administered to all participants because there is interest in comparing results of the newer EQ-5D-5L with results from the more established EQ-5D-3L [45].

The EQ-5D is a self-administered, generic, preference weighted measure designed to assess health status [46–48]. The original version with three levels per domain (EQ-5D-3L) and the newer version with five levels per domain (EQ-5D-5L) were both administered. In each version, the first section consists of five dimensions to assess HRQL (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression). These five dimensions are scored based on preference weightings to obtain an “index score” that is often used as a utility in economic modeling. The second section consists of a visual analogue scale on which respondents rate their current health, with anchors of 0 (“worst imaginable health state”) and 100 (“best imaginable health state”). Half the participants were randomized to complete the EQ-5D-3L first, while the other half completed the EQ-5D-5L first. The EQ-5D-3L index score was computed using the UK “MVH” tariffs [49], and the EQ-5D-5L index score was computed with the recently published value set derived from a general population sample in England [50].

Statistical analysis procedures

Statistical analyses were completed using SAS version 9.4 (SAS Institute, Cary, NC). Continuous variables were summarized in terms of means and standard deviations. Categorical demographic variables were summarized as frequencies and percentages. Demographic subgroups were compared with chi-square analyses (for categorical variables) and t-tests (for continuous variables). Pairwise comparisons between health state utilities were performed using t-tests. Disutilities (i.e., decreases or differences in utility) of injection process attributes were calculated by subtracting the mean utility of one health state from another.

Sample description

The EQ-5D-3L and EQ-5D-5L mean index were scores similar to those published previously for patients with type 2 diabetes without complications [51]. The score ranges and standard deviations suggest that the sample was diverse in terms of health status (Table 2).

Health state utilities

In the introductory task, respondents generally ranked health states in a logical order. Health states with more treatment administration steps were typically ranked as less preferable, with rankings ranging from 1 (most preferable health state) to 7 (least preferable health state). The oral-only health state (A) had a mean ranking of 1.08, followed by G (2.01), E (3.54), F (3.59), D (5.24), C (5.57), and B (6.98).

The mean utility scores followed the same order as the health state rankings, with greater numbers of treatment administration steps associated with lower utility (Table 3). The mean utility for health state A describing diabetes with oral treatment was 0.888, while adding the weekly injection in health state G reduced utility to 0.878. Each additional inconvenience associated with the weekly injection decreased utility further, and the lowest utility value was for health state B, which described reconstitution, waiting, and needle handling (0.858). T-tests found no statistically significant differences in utility between men and women; between older and younger respondents (categorized by median split); or between respondents from England and Scotland.

All but one participant rated every health state as better than dead (i.e., utility score > 0). One participant rated every health state as worse than dead. When asked to explain these TTO choices, the participant explained that she had negative experiences with oral medications in the past, and therefore she strongly preferred not to live in a health state requiring oral medication.

Disutilities and comparisons among health states

Disutility scores (i.e., decreases in utility) were calculated relative to both oral-only treatment as described in health state A and oral plus weekly injection treatment as described in health state G (Table 3). Disutilities for the three injection process attributes (i.e., reconstitution, waiting, needle handling) can be obtained based on the difference between health state G and health states B to F. For example, health states G and F are identical except for the needle handling requirement in F. Therefore, the difference between G and F (-0.004) represents the disutility of the needle handling requirement in the context of a weekly injection for treatment of type 2 diabetes. Disutilities of individual injection attributes were of a relatively small magnitude, such as the disutility of the requirement for reconstitution (-0.004). However, health states representing combinations of multiple injection attributes had larger disutilities, including a disutility of -0.020 for health state B representing the combination of reconstitution, waiting, and needle handling requirements.

T-tests were conducted to explore whether differences among health state utilities were statistically significant. The utility of the oral-only health state (health state A) was found to be significantly different from all other utilities (p < 0.0001 for health state A vs B, C, D; p = 0.0007 vs. E and F; p = 0.0098 vs. G). The injection health state without any of the three inconveniences (health state G) was compared to each of the other injection health states (B to F). The utility of G was found to be significantly different from utilities of all health states that included reconstitution, waiting, and/or mixing (p < 0.0001 vs. B, C, D; p = 0.0003 vs. F; p = 0.0007 vs E).