Health state utilities associated with weight loss: preferences of people with type 2 diabetes and obesity in Japan

Abstract

Aims

Health state utilities associated with weight change are needed for cost-utility analyses (CUAs) examining the value of treatments for type 2 diabetes and obesity. Previous studies have estimated the utility benefits associated with various amounts of weight reduction in the US and Europe, but preferences for weight change in Asian cultures may differ from these published values. The purpose of this study was to estimate utilities associated with reductions in body weight based on preferences of individuals with type 2 diabetes and obesity in Japan.

Methods

Health state vignettes represented type 2 diabetes with respondents’ own current weight and weight reductions of 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, and 20%. Utilities were elicited in time trade-off interviews with a sample of respondents in Japan with type 2 diabetes and body mass index (BMI) ≥25 kg/m² (the cutoff for obesity in Japan).

Results

Analyses were conducted with data from 138 respondents (84.8% male; mean age = 58.0 years; mean BMI = 29.4 kg/m²) from all eight regions of Japan. Utility gains gradually increased with rising percentage of weight reductions ranging from 2.5% to 15%. Weight reductions of 2.5% to 15% resulted in utility increases of 0.013 to 0.048. The health state representing a 20% weight reduction yielded a wide range of preferences (mean utility increase of 0.044). Equations are recommended for estimating utility change based on any percentage of weight reduction (up to 20%) in Japanese people with type 2 diabetes and obesity.

Limitations

This study was conducted in a sample with limited representation of patients with BMI >35 kg/m² (n = 13) and relatively few women (n = 21).

Conclusion

Results may be used to provide inputs for CUAs examining the value of treatments that are associated with weight loss in patients with type 2 diabetes and obesity in Japan.

Keywords:

• Health state utilities
• time trade-off
• type 2 diabetes
• weight
• BMI
• cost-utility analysis

JEL Classification Codes:

• I12
• I1
• I
• I19

Introduction

The association between type 2 diabetes and obesity has been established by a substantial body of research^1–3, and therefore, weight management is considered to be a key element of type 2 diabetes management⁴. Consequently, there has been growing interest in medications for diabetes that also promote weight reduction^5,6. Several pharmaceutical treatments that are currently available or in development for type 2 diabetes have been shown to be associated with substantial amounts of weight reduction, often with decreases exceeding 10% of a patient’s body weight^7–10. Cost-utility analyses (CUAs) examining the value of these treatments are needed to inform healthcare resource allocation decisions¹¹. To quantify benefits of weight reductions and calculate quality-adjusted life years (QALYs) associated with these medications, CUAs will require health state utilities representing the strength of preference for weight changes observed in clinical trials.

The link between body weight and health state utility in type 2 diabetes has been demonstrated in a range of previous studies suggesting that reductions in weight tend to be associated with increases in utilities. These studies have primarily focused on utility changes associated with weight loss up to approximately 7% of one’s body weight^12–17. However, economic modeling of newer treatments for type 2 diabetes would require utilities representing larger amounts of weight loss. Therefore, a study was conducted in the United Kingdom (UK) to estimate utility gains associated with weight decreases of up to 20% of one’s body weight¹⁸. Because this study was conducted with a sample consisting only of people with type 2 diabetes and obesity in the UK, generalizability to other countries and regions is unknown.

Newer treatments for type 2 diabetes associated with significant weight loss have been examined for use in Japan^19–22, and CUAs will be needed to examine their potential value in this country. Utilities from the published UK study¹⁸ may not apply in Japan and other Asian countries for several reasons. First, utilities and health preferences often vary by culture and geographic region, and Japan has been highlighted as a country where utilities may differ from those in European countries^23,24. Second, there is some evidence that perceptions of weight and body image may differ in Asian countries, and Japan in particular, compared with European countries²⁵. Third, the body mass index (BMI) threshold for obesity varies across countries. The UK study included patients with BMI ≥30 kg/m^218, which is the commonly accepted threshold in European countries. In contrast, the commonly accepted threshold for obesity in Japan is a BMI ≥25 kg/m^225–30. Therefore, there may be important differences between the populations of people with obesity in the UK and Japan.

The purpose of this study was to estimate change in utility associated with reduction in body weight based on preferences of people with type 2 diabetes and obesity in Japan. This time trade-off (TTO) utility elicitation was conducted with vignette-based methods similar to those used in the UK study¹⁸. This is the first utility elicitation study focused on preferences for weight change among people with type 2 diabetes and obesity in Japan, and the study was designed to yield utilities that may be useful in CUAs of treatments associated with a wide range of reductions in weight.

Methods

Study design

The study design was based on a previous vignette-based utility valuation study conducted in the UK¹⁸. Like the UK study, health state vignettes were developed to represent type 2 diabetes with respondents’ own current weight and weight reductions ranging from 2.5% to 20%, and the vignettes were valued in TTO interviews. There were three key differences between the original UK study and the study replication in Japan. First, due to the COVID-19 pandemic, all interviews were conducted by videoconference rather than in person. Second, whereas the study in the UK required that all participants had a BMI of at least 30 kg/m² to meet criteria for obesity, the current study had a BMI cutoff of 25 kg/m², which is the commonly accepted threshold for obesity in Japan^25,26,28,29. Third, whereas participants in the UK study valued six health states (current weight and reductions of 2.5%, 5%, 10%, 15%, and 20%), the Japanese study included two additional health states (reductions of 7.5% and 12.5%) to increase sensitivity to preferences for lower levels of weight loss, resulting in a total of eight health states.

Mean utility change was calculated for each of the seven levels of reduction in body weight (i.e. 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, and 20%) represented in the health states. In addition, regression analyses were performed to model the relationship between weight reduction and utility change, and two equations were derived to estimate utility for any percentage of weight change between 2.5% and 20%.

All participants provided informed consent prior to participation. Study procedures and materials were approved by an ethics committee in Japan (Research Institute of Healthcare Data Science [RIHDS] RI20210278).

Health state vignettes

The health states valued in this Japanese study were based on those used in the prior study in the UK¹⁸. Each health state included two sections with headings for “Type 2 Diabetes” and “Weight.” The description of type 2 diabetes was a translated version of language used in the UK study’s health states¹⁸, which were adapted from health states used in previously published utility elicitation studies^{13,14,31–35}. The same description of type 2 diabetes appeared at the beginning of every health state with the following statements in four bullet points: 1) “You have had type 2 diabetes for several years; 2) You receive treatment that may include medication for your diabetes. Your blood sugar levels are usually in control, but sometimes your blood sugar is too high or too low; 3) If your blood sugar level is too low, you may experience dizziness/light-headedness, sweating, or shaking; 4) If your blood sugar level is too high, you may experience tiredness, blurred vision, thirst, or frequent urination.”

This description of type 2 diabetes was followed by a description of body weight, ranging from the participant’s self-reported current weight to a reduction of 20% of their current weight. In addition to the five weight reduction health states valued in the UK study¹⁸, two additional weight reduction health states (7.5% and 12.5%) were added in this study for greater sensitivity to utility change. The “Weight” section of the current weight health state included one bullet with the statement “You weigh [participant’s current weight].” The weight reduction health states included one bullet with the statement “You weigh ____ (which is ____ less than you weigh now).” In the first blank space of the statement, the interviewer inserted the participant’s current body weight reduced by 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, or 20% depending on the health state. In the second blank space, the interviewer inserted the amount of weight decrease from the participant’s current weight. For example, the 10% weight reduction health state for a participant with a current weight of 80 kg would read: “You weigh 72 kg (which is 8 kg less than you weigh now).” All weights were listed in kilograms, which is the most commonly understood unit of weight measurement in Japan, and all text was presented in Japanese.

Translation of study materials

Translation of study materials from English to Japanese was conducted in several steps. Study materials other than the health states (e.g. interview guide, consent form, demographic form) were first translated from English to Japanese by a native Japanese speaker. These translated materials were then proofread and edited by a second native Japanese speaker who was not involved in the initial translation. The materials were then reviewed by the initial translator, the second translator, and Japanese-speaking members of the study team to ensure that the final materials were clear and comprehensible to Japanese-speaking study participants.

The translation of the health states followed the same steps, but with an additional step of back-translation performed by a native English speaker fluent in Japanese. The back-translation was then reviewed by a native Japanese speaker not involved in the forward translation.

Participants

Participants were recruited throughout Japan using multiple methods, including a market research consumer panel, referrals from diabetes patient associations, physician referrals, and social media (e.g. Facebook support groups for people with diabetes). Potential participants were screened by telephone to ensure they met eligibility criteria for the study. To be eligible for inclusion, participants were required to be between the ages of 18 and 75 years old (inclusive), a current resident of Japan, and fluent in Japanese. Participants were also required to have a BMI ≥25 kg/m² (based on self-reported current height and weight) and a diagnosis of type 2 diabetes by a medical doctor. Participants currently receiving medication for type 2 diabetes were required to provide proof of their medication treatment such as the medication packaging, a prescription note, or a letter from their doctor. Participants not receiving medication treatment for their type 2 diabetes were asked to describe their diabetes symptoms and the diagnosis process in a way that clearly demonstrated personal experience with the disease. Because interviews were conducted virtually, all participants were required to have access to a computer, tablet, or smartphone with videoconference capabilities.

This replication of a UK study was conducted in Japan with a sample size target of 130 participants. The results of most vignette-based utility studies are descriptive (i.e. the mean utility values). Unlike a clinical trial, there is not usually a primary statistical analysis of a between-group difference that can be used to inform a power analysis to determine a sample size target. Therefore, determination of the sample size target was based on review of recently published vignette-based TTO studies. In these previous studies, sample sizes varied widely, but most had between 100 and 200 participants. Because this study was a replication, a sample size in the lower part of this range was considered to be adequate for the study purpose.

Utility interview procedures and scoring

Due to the status of the COVID-19 pandemic in Japan at the time this study was conducted (November 2022), it was decided not to conduct interviews in person. Instead, interviews were conducted one-on-one via videoconference using Zoom. Study materials such as the health states, TTO choices, and questionnaires were shown to participants using screen sharing.

All interviews were conducted in Japanese by a team of six Japanese interviewers. The interviewers were trained and supervised by a team of three researchers who were a part of the team that conducted the UK study that served as the model for this Japanese study¹⁸. Two translators participated in all aspects of data collection to facilitate training and supervision. Both translators had lived in both Japan and England and were fluent in both Japanese and English.

To ensure effective interviewer training, the two translators were trained in the study methods prior to the larger training session with the interviewers. Then, the interviewers were trained in a three-hour session (directed by the lead author and translated by one of the two translators), followed by several hours of practice with the TTO interview procedures prior to conducting interviews with study participants. To ensure interview quality and consistency with the TTO methods used in the prior study in the UK¹⁸, the three researchers from the original study observed interviews (with the help of both translators), provided guidance as necessary, and responded to queries from the interviewers.

The first 15 interviews were considered “pilot” interviews, which were used to assess whether the health states and methods were clear to participants. Data collection was temporarily paused after these interviews were completed so that the study team (i.e. the three researchers from the UK study, the two translators, and the six interviewers) could discuss whether any clarifications or edits to study materials and methods might be needed. All interviewers agreed that the study participants were consistently understanding the health states and TTO procedures. In addition, utility results appeared logical, and no participants or interviewers suggested any changes to the health states or procedures. Therefore, the team decided to continue with the remainder of the interviews. Because no changes were made to the methods or health states following the pilot interviews, data from these 15 participants were included in the final data set for analysis.

In each interview, the interviewer began by introducing the eight health states and asking the participant to rank the health states in order of their preference (i.e. from most preferable to least preferable). Health states were presented to participants in random order and displayed on the screen. Using the “slide sorter” display in Microsoft PowerPoint, the interviewer reordered the health states on the screen according to the participant’s order of preference. During the ranking task, the participants were asked to explain the reasons for the order of preference among the health states.

Utilities for each health state were elicited using the TTO method. For each health state, participants were offered a choice between spending a 20-year period in the health state versus spending varying amounts of time in full health. Choices were presented in one-year increments, alternating between longer periods of time and shorter periods of time (i.e. 20 years, 0 years [dead], 19 years, 1 year, etc.). Utility scores (u) were assigned based on the point of indifference between the years in the health state being valued (y) and years in full health (x), where u = x/y. Procedures for valuing health states with negative utility scores were not necessary since none of the participants perceived any of the health states to be worse than dead. During the interviews, interviewers transcribed some key quotes from participants when they explained their preferences among the health states. These statements were translated from Japanese into English, and some of the relevant statements are reported below in the results section.

EQ-5D-5L

The EQ-5D-5L was administered to characterize the overall health status of the sample. The EQ-5D-5L is a generic, self-administered, preference-based measure assessing five dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Respondents select one of five response options for each dimension, and scores on the five dimensions were converted into an EQ-5D index score using published scoring tariffs derived for use in Japan³⁶.

Statistical analysis procedures

Statistical analyses were completed using SAS (version 9.4). Continuous variables (including utilities and utility difference scores) were summarized as means and standard deviations, and categorical variables were presented as frequencies and percentages. Paired t-tests were conducted to test whether there were statistically significant pairwise differences between health state utilities (e.g. differences between utilities associated with current weight and utilities associated with various levels of reduced weight). Independent t-tests were conducted to test for subgroup differences in utility scores with participants categorized based on age, gender, employment status, and BMI. Subgroup differences based on current medications were examined with analysis of variance.

The mean utility increases for each weight reduction health state may be used in cost-utility models to represent utility change associated with the specific percentage of reductions in body weight that appear in the health states (i.e. 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, and 20% decreases in body weight). However, in some cases, it may be necessary to estimate utility change associated with percentage of weight reductions between those represented in the health states (e.g. 9%). In these instances, utility associated with weight reductions can be estimated by modeling the relationship between weight change and utility change. To derive equations that could be used for this approach, a series of regression models were run to determine the optimal model for predicting utility change. In every regression model, the dependent variable was change in utility.

The relationship between percentage of reduction in body weight and utility was analyzed in three ways including linear, logarithmic, and quadratic models. A repeated-measures linear regression model was run (i.e. a repeated-measures linear regression model with compound symmetry correlation structure for within-individual observations), first with percentage of weight reduction as the independent variable with values of weight reduction corresponding to the health states (i.e. 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, 20%). To allow for a possible non-linear relationship between percentage of weight reduction and utility change, the models were also run with logarithmic and quadratic terms. In the logarithmic models, the natural logarithm (log) of percentage of weight reduction was included as an independent variable. In the quadratic models, a quadratic term for percentage of weight decrease was added to the linear model (i.e. percentage of weight decrease and percentage of weight reduction squared were included in the same model). Models were run with and without covariates. Covariates hypothesized to modify the relationship between weight change and utility change (i.e. gender, age [continuous], and BMI [continuous]) were added to all models as main effects as well as in interactions with percentage of weight reduction. The optimal modeling approach was selected based on model results while considering the expected shape of the relationship between percentage of weight reduction and utility change.

Results

Sample characteristics

A total of 604 patients with type 2 diabetes and obesity (BMI ≥25 kg/m²) were screened to assess whether they met the eligibility criteria, and 382 of those did not meet criteria. Of the 222 who were eligible, 140 were scheduled, and all 140 attended their interviews. Two of the 140 participants had difficulty understanding the utility interview procedures and were therefore unable to provide valid data. Therefore, the analysis includes data from 138 interviews.

The sample was 84.8% male (N = 117), with a mean age of 58.0 years (Table 1). All participants were Japanese, and the sample included participants from all eight regions of Japan. Most participants reported being employed (44.9% full-time, 10.9% part-time, and 14.5% self-employed), and the majority (62.3%) had a university degree.

Table 1. Summary of participant characteristics.

Characteristics	N = 138
Age (years), mean (SD) [Range]	58.0 (9.5) [28–75]
Sex, n (%)
Male	117 (84.8%)
Female	21 (15.2%)
Ethnicity, n (%)
Japanese	138 (100.0%)
Marital status, n (%)
Single	36 (26.1%)
Married/Cohabitating/Living with partner	84 (60.9%)
Other^a	18 (13.0%)
Employment status, n (%)
Employed^b	97 (70.3%)
Other^c	41 (29.7%)
Education level, n (%)
University degree	86 (62.3%)
No university degree	52 (37.7%)
Region of Japan^d, n (%)
Hokkaido	6 (4.3%)
Tohoku	1 (0.7%)
Kanto	68 (49.3%)
Chubu	17 (12.3%)
Kansai	29 (21.0%)
Chugoku	3 (2.2%)
Shikoku	2 (1.4%)
Kyushu and Okinawa	12 (8.7%)
Height, weight, and BMI
Height (centimeters), Mean (SD) [Range]	168.1 (6.5) [153–182]
Weight (kg), Mean (SD) [Range]	83.2 (11.9) [61–128]
BMI (kg/m^2), Mean (SD) [Range]	29.4 (3.9) [25.0–45.2]
Duration of diabetes diagnosis (years)
Mean (SD) [Range]	10.8 (7.5) [0.6–34.6]
Current diabetes medication, n (%)
No medication treatment	8 (5.8%)
Oral medication only	103 (74.6%)
Injectable medication only	6 (4.3%)
Oral and injectable medication	21 (15.2%)

Abbreviations. BMI, body mass index; SD, standard deviation.

^aOther marital status reported by participants includes divorced (n = 13), separated (n = 2), and widowed (n = 3).

^bEmployed includes full-time, part-time, and self-employed.

^cOther employment status reported by participants includes retired (n = 26), unemployed (n = 13), voluntary work (n = 1), and not able to work due to disability (n = 1).

^dEight Regions of Japan: 1. Hokkaido (Prefectures: Hokkaido), 2. Tohoku (Prefectures: Miyagi), 3. Kanto (Prefectures: Ibaraki, Tochigi, Gunma, Saitama, Chiba, Tokyo, and Kanagawa), 4. Chubu (Prefectures: Niigata, Toyama, Fukul, Nagano, Gifu, Shizuoka, and Aichi), 5. Kansai (Prefectures: Mie, Shiga, Kyoto, Osaka, Hyogo, and Nara), 6. Chugoku (Prefectures: Okayama and Hiroshima), 7. Shikoku (Tokushima, and Kagawa), 8. Kyushu and Okinawa (Fukuoka, Saga, Nagasaki, Oita, Miyazaki, and Okinawa).

Participants reported being diagnosed with type 2 diabetes an average of 10.8 years prior to their interview. Most participants were currently receiving medication to treat their diabetes (94.2%), with the majority of the sample taking oral medication (89.9%), either alone (74.6%) or in combination with injectable medication (15.2%). All participants provided at least one form of proof of a type 2 diabetes diagnosis, including proof of prescription medication (95.7%) or a detailed description of diabetes symptoms and/or diagnosis indicating that their report of a type 2 diabetes diagnosis was genuine (4.3%). Two of the participants who provided documentation of a prescription for oral medication reported that they were not taking the prescribed medication at the time of their interview.

The mean BMI was 29.4 kg/m² (range: 25.0 to 45.2), with a mean height of 1.68 meters and a mean weight of 83.2 kg. Nearly all participants (99.3%) reported wanting to lose weight, with an average desired weight loss of 9.9 kg. The most commonly reported comorbidities included hypertension (50%), depression (11.6%), arthritis (9.4%), and asthma (9.4%). The mean EQ-5D-5L index score was 0.90 (SD = 0.11), which is similar to scores for patients with type 2 diabetes in Japan in a previous study³⁷. To examine overall health status among subgroups varying in BMI, participants were categorized into three BMI subgroups (BMI <27.5 kg/m² n = 48; 27.5 – <30 kg/m² n = 48; ≥30 kg/m² n = 42). Mean EQ-5D-5L index score was highest for the lowest BMI group (0.913) and lowest for the highest BMI group (0.882), but between-group differences in EQ-5D-5L index scores were not statistically significant.

Health state rankings

As expected, lower body weight was generally preferred over higher body weight (Table 2). All participants preferred at least one of the weight reduction health states over their current weight, and the current weight health state was ranked as the least preferred health state by most participants (81.9%). Participants reported a range of reasons for preferring health states with reduced weight. Some noted the impact on their health (“Healthier body.” “It would reduce the risk of developing other illness”), and others believed that they may live longer with reduced weight (“I want to lose weight…I might be able to live longer if I am lighter”). Participants also reported that some activities are easier at a lower weight (“The lighter I am, the better my life quality, whether I’m walking around outside or other things.” “If I lose weight I would be able to cut my nails.” “If I lost 10 kg I would be able to move around more easily.” “When I was a student and turned 20, I weighed 66–67 kg. It was easy to move around and I felt at my best.” “I have been told by my doctor to lose weight. Better mobility. Sleep apnea, airways are pressured with fat. I would like to resolve this”).

Table 2. Health state rankings^a.

	Health State Rankings (n, %)
Health States	1 = Most preferred	2	3	4	5	6	7	8 = Least preferred
A. Current weight	–	–	1 (0.7%)	1 (0.7%)	3 (2.2%)	3 (2.2%)	17 (12.3%)	113 (81.9%)
B. 2.5% weight reduction	2 (1.4%)	1 (0.7%)	3 (2.2%)	5 (3.6%)	7 (5.1%)	21 (15.2%)	99 (71.7%)	–
C. 5% weight reduction	3 (2.2%)	9 (6.5%)	6 (4.3%)	8 (5.8%)	20 (14.5%)	92 (66.7%)	–	–
D. 7.5% weight reduction	11 (8.0%)	12 (8.7%)	7 (5.1%)	24 (17.4%)	84 (60.9%)	–	–	–
E. 10% weight reduction	17 (12.3%)	9 (6.5%)	24 (17.4%)	86 (62.3%)	2 (1.4%)	–	–	–
F. 12.5% weight reduction	9 (6.5%)	25 (18.1%)	85 (61.6%)	8 (5.8%)	6 (4.3%)	5 (3.6%)	–	–
G. 15% weight reduction	21 (15.2%)	78 (56.5%)	9 (6.5%)	4 (2.9%)	8 (5.8%)	10 (7.2%)	8 (5.8%)	–
H. 20% weight reduction	75 (54.3%)	4 (2.9%)	3 (2.2%)	2 (1.4%)	8 (5.8%)	7 (5.1%)	14 (10.1%)	25 (18.1%)

^aRankings had a possible range from 1 to 8, with lower numbers indicating more preferred health states. The most frequent ranking position for each health state is bolded.

The health state representing the greatest percentage of weight reduction (i.e. 20%) yielded a wide range of preferences. This health state was most preferred by 75 (54.3%) participants, but least preferred by 25 (18.1%) participants. The 75 participants who ranked this health state as most preferred had a significantly higher BMI than the other 63 participants in the sample (31.2 kg/m² vs. 27.3 kg/m²; p < 0.001).

Participants who ranked the 20% reduction as most preferred reported wanting to be “as light as possible” or had an ideal weight near or lower than the 20% weight reduction (e.g. “I want to be as light as possible. I was about 60 kg…and my doctor told me that is about my ideal weight.” “Going too low is worrying, but within this range, I want to be as low as possible.” “I felt my best at this weight. It is also the standard weight for my height”).

Participants who did not rank the 20% reduction as most preferred typically perceived this amount of weight reduction to be excessive. These participants tended to believe that their health, strength, or energy could be negatively affected if they were to lose this much weight (e.g. “I would lose strength and energy,” “[20% weight decrease] feels too low and may affect my life negatively,” “I feel like my health would be negatively affected if the weight drops below [12.5% weight decrease]”).

Health state utilities

Respondents’ preference for reduction in body weight was reflected in the mean health state utilities. All health states representing weight reduction (i.e. health states B to H representing weight reductions of 2.5% to 15% of current body weight) had greater mean utilities than the health state representing the respondent’s own current weight (i.e. health state A). For weight reductions up to 15%, utility was inversely related to weight, with a gradual increase in utility associated with lower body weight (Table 3). The mean utility for every weight reduction health state was significantly greater than the mean utility of the current weight health state (p < 0.001 for all comparisons).

Table 3. Health state utilities.

	Health State Utility^a		Utility Increases^b Associated with Weight Reduction
	Mean (SD)	95% CI	Mean Utility Increase (SD)	95% CI
A. Current weight	0.783 (0.209)	0.747 − 0.818	–	–
B. 2.5% weight reduction	0.796 (0.197)	0.763 − 0.829	0.013 (0.042)	0.006 − 0.021
C. 5% weight reduction	0.804 (0.191)	0.772 − 0.836	0.022 (0.053)	0.013 − 0.030
D. 7.5% weight reduction	0.816 (0.181)	0.786 − 0.846	0.033 (0.079)	0.020 − 0.047
E. 10% weight reduction	0.820 (0.180)	0.790 − 0.850	0.037 (0.084)	0.023 − 0.051
F. 12.5% weight reduction	0.825 (0.178)	0.795 − 0.855	0.043 (0.098)	0.026 − 0.059
G. 15% weight reduction	0.830 (0.174)	0.801 − 0.860	0.048 (0.107)	0.030 − 0.066
H. 20% weight reduction	0.827 (0.173)	0.798 − 0.856	0.044 (0.111)	0.025 − 0.063

Abbreviations. CI, confidence interval; SD, standard deviation; TTO, time trade-off.

^a TTO scores are on a scale anchored with 0 representing dead and 1 representing full health.

^bCalculated by subtracting the utility of health state A from the utility of each other health state.

Mean utilities ranged from 0.783 for the health state with the respondents’ current weight to 0.830 for the 15% weight reduction. Utility increases associated with weight reductions ranged from 0.013 for a 2.5% weight reduction to 0.048 for a 15% weight reduction. The mean utility of the 20% weight reduction health state (0.827) was slightly lower than that of the 15% weight reduction (0.830), but there was no significant difference between utilities of health states with 15% and 20% weight reduction.

There were no significant differences in utility by age, gender, employment status, BMI, or current medication type. Although there were no statistically significant differences between groups, participants with BMI <30 kg/m² had numerically higher mean utilities than participants with BMI ≥30 kg/m² for three of the eight health states (current weight, 2.5% weight loss, and 5% weight loss health states). This trend reversed after the first three health states, and participants with BMI ≥30 kg/m² had higher mean utility for all health states representing weight loss of ≥7.5%. The difference in utility between the current weight health state and the 20% weight decrease health state was larger among participants with BMI ≥30 kg/m² (0.071 vs. 0.032) than among participants with lower BMI, but this difference was not statistically significant (p = 0.055).

The majority of participants (>62%) were willing to trade time to avoid living in all of the health states. No participants perceived any of the health states to be worse than dead, and therefore, there were no negative utility scores.

Estimating utility change for weight loss percentages not represented in the health states

A series of regression models were run to derive an equation for estimating the utility change associated with any percentage of weight change between 2.5% and 20%. In every regression model, the dependent variable was change in utility, and the key independent variable was percentage of weight reduction as represented in the health states. All models found a highly significant relationship between percentage of weight reduction and utility change, including linear, logarithmic, and quadratic models, with and without covariates. Therefore, it seems reasonable to predict utility change based on percentage of weight reduction.

Initial models were run without covariates, and the ordinary least squares regression model was run first. Although this linear model found a statistically significant relationship between weight change and utility change, the relationship between these two variables did not appear to be linear when data were examined graphically. Therefore, the model was re-run using the natural logarithm of percentage weight change to allow for a non-linear relationship between of weight change and utility change. The log of percentage of weight change was a significant predictor of utility change, and natural logarithm of percentage of weight reduction appears to be a useful way to represent this independent variable in these models. The logarithmic approach allows for a steeper curve across the health states representing smaller percentages of weight decrease, gradually flattening as the amount of weight reduction increases.

Three covariates hypothesized to modify the relationship between weight reduction and utility change were then added to models: gender, age (continuous), and BMI (continuous). The main effect of BMI was significant, but the other variables did not have a significant effect. It appeared, based on examination of the data, that the impact of weight reduction on utility may depend on the BMI of the respondent. In general, respondents with higher BMI tended to have stronger preference for the health states representing the greatest levels of weight reduction. Therefore, a percentage of weight reduction by BMI interaction term was added to the models. In addition, to simplify the model, the nonsignificant age and gender terms were dropped. Results of the two regression models predicting utility change based on the natural log of percentage of weight reduction, with and without BMI as a covariate, are presented in Table 4.

Table 4. Two repeated-measures regression models predicting utility change based on percentage of weight reduction.

Effect	Estimate	Standard Error	Degrees of Freedom	t Value	p Value
Model with BMI
Intercept	0.062	0.059	136	1.06	0.29
Natural log of weight reduction	−0.053	0.014	826	−3.67	<0.001
BMI	−0.0022	0.0020	136	−1.12	0.27
Interaction term: Natural log of weight reduction * BMI	0.0024	0.00049	826	4.91	<0.001
Model without BMI
Intercept	−0.0028	0.0078	137	−0.36	0.72
Natural log of weight reduction	0.017	0.0019	827	8.94	<0.001

Abbreviations: BMI, body mass index; log, logarithm.

Based on regression models run for the current study, two equations are proposed for estimating utility change associated with of weight reduction (Table 5). In the primary equation, utility change is estimated based on the natural log of percentage of weight reduction and BMI (Figure 1). The second equation is an alternate option to estimate utility change based on percentage weight reduction in situations when BMI is not available in the data set. This alternate equation should be used with caution because it does not account for the potentially important interaction between BMI and preference for weight reduction. Table 6 presents hypothetical examples of how these equations can be applied to estimate utility change associated with the percentage of weight reduction in samples with varying baseline BMI.

Figure 1. Change in utility as a function of percentage of reduction in weight^a,b.

^aThis is a plot of the regression model from Table 5 for the log-linear regression of change in utility as a function of the natural log of the weight reduction represented in the health states. Model: utility = 0.062 - 0.053(lnWR) - 0.0022(BMI) + 0.0024(lnWR*BMI).

^bHealth states valued in this study included current weight and weight reductions of 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, and 20%. Weight reduction of 17.5% was not valued as a health state in this study. Abbreviations: lnWR, natural log of weight reduction; BMI, body mass index

Table 5. Equations proposed for estimating utility change based on percentage of weight reduction.

Intended use	Independent variables, covariates, and interaction terms	Equation for predicting utility change based on percentage of weight reduction
Primary equation to be used when BMI is known	Natural logarithm of percentage of weight reduction BMI Interaction: natural logarithm of percentage of weight reduction * BMI	Utility change = 0.062 − 0.053 (lnWR) − 0.0022 (BMI) + 0.0024 (lnWR*BMI)
Alternate equation to be used when BMI is not known^a	Natural logarithm of percentage of weight reduction	Utility change = −0.0028 + 0.017 (lnWR)

^aThis equation should be used with caution because it does not account for the interaction between BMI and preference for weight reduction.

Abbreviations: BMI, body mass index; lnWR, natural log of percentage of weight reduction.

Table 6. Examples of how the equations can be used to estimate utility increase associated with various percentages of weight reduction.

Model	Weight Reduction	BMI (kg/m^2)	Equation	Utility Increase
Primary Equation: Natural Log of weight reduction with terms representing interactions between natural log of weight reduction and BMI	4%	25.0	Utility change = 0.062 − 0.053 (ln(4)) − 0.0022 (25.0) + 0.0024 (ln(4)*25.0)	0.017
	4%	29.4^a	Utility change = 0.062 − 0.053 (ln(4)) − 0.0022 (29.4) + 0.0024 (ln(4)* 29.4)	0.022
	4%	36.0	Utility change = 0.062 − 0.053 (ln(4)) − 0.0022 (36.0) + 0.0024 (ln(4)*36.0)	0.029
	17%	25.0	Utility change = 0.062 − 0.053 (ln(17)) − 0.0022 (25.0) + 0.0024 (ln(17)*25.0)	0.027
	17%	29.4^a	Utility change = 0.062 − 0.053 (ln(17)) − 0.0022 (29.4) + 0.0024 (ln(17)* 29.4)	0.047
	17%	36.0	Utility change = 0.062 − 0.053 (ln(17)) − 0.0022 (36.0) + 0.0024 (ln(17)*36.0)	0.077
Alternate equation: Natural log of weight reduction	4%	n/a	Utility change = −0.0028 + 0.017 * (ln(4))	0.021
	17%	n/a	Utility change = −0.0028 + 0.017 * (ln(17))	0.045

^a 29.4 kg/m² is the mean BMI of the study sample. At this BMI, the predicted utility change will align with the regression curve graphed in Figure 1 for the logarithmic model with terms representing interactions between the natural log of weight reduction and BMI.

Abbreviations: ln, natural log; BMI, body mass index.

Discussion

Health states representing reduction in body weight were associated with utility increases based on preferences of people with type 2 diabetes and obesity in Japan. Utility gains gradually increased with rising percentage of weight reduction from 2.5% to 15%. While all participants preferred health states with lower weight over the health state with their current weight, the health state representing 20% weight reduction yielded a wide range of preferences. This was the most preferred health state for 75 (54.3%) participants, but the least preferred for 25 (18.1%) participants, resulting in a mean utility increase of 0.044.

Perceptions of the 20% weight reduction health state appear to be related to the BMI of the respondent. Participants who ranked the 20% weight reduction health state as the most preferred had a significantly higher mean BMI than participants who preferred lower percentages of weight reduction. Essentially, participants with lower BMI were less likely to prefer the greatest amounts of weight loss. It should be noted that there were no statistically significant differences in utilities by BMI subgroups (i.e. BMI above and below 30 kg/m²).

Six of the eight health states valued in this Japanese study were also valued by a sample with type 2 diabetes and obesity in the UK¹⁸. The inverse relationship between weight and utility was similar in these two samples for weight reduction up to 15%. However, the two samples differed in perceptions of the health state representing a 20% weight decrease. In the current Japanese sample, 54.3% of participants ranked this as the most preferred health state, compared with 83.7% of the sample in the UK. As a result, the mean utility increase associated with a 20% weight reduction was somewhat larger in the UK than in Japan (0.060 vs. 0.044, respectively). The difference is likely due to the lower average BMI in the Japanese sample (mean BMI = 29.4 kg/m² in Japan vs. 36.1 kg/m² in the UK), resulting from differences in the BMI threshold for obesity in the two countries (≥25 kg/m² in Japan and ≥30 kg/m² in the UK). The subgroup of the Japanese sample who would meet the UK’s obesity criteria (i.e. BMI ≥30 kg/m²) tended to have a strong preference for the 20% weight reduction health state (e.g. 85.7% of this subgroup ranked this health state as most preferred, and this subgroup had a mean utility improvement of 0.071).

The mean utility increases associated with change in body weight may be used in CUAs to represent utility change associated with a specific percentage of decrease in body weight in Japan (i.e. 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, and 20%). Regression models were run to derive equations for estimating utility change associated with percentages of weight decrease between those represented in the health states. A primary equation has been proposed for estimating utility change based on the natural log of weight reduction and BMI. An alternative equation is provided for use when baseline BMI is unknown. The proposed equations should be used only to estimate utility change associated with weight loss, not weight gain, since no health states representing weight increase were valued. It is not recommended to use the proposed equations to estimate utility change associated with weight reduction greater than 20%.

The current utilities are intended for use in CUAs in Japan. Although the resulting utilities may also be useful in other Asian countries, the generalizability to samples outside of Japan is not known. Current results suggest that BMI may affect utilities. The threshold for obesity in Japan is BMI ≥25 kg/m², but BMI cutoffs vary in other Asian countries^38,39. Country-specific BMI thresholds for obesity should be considered when applying the current utilities in cost-effectiveness modeling to support decision making in countries other than Japan. Despite the variation across Asian countries, the currently reported utilities are likely to be more appropriate and relevant than previously published European-derived utilities for use in cost-effectiveness models in other Asian countries.

This study has limitations associated with the sample characteristics. This study was conducted in a relatively small sample with limited representation of patients with a BMI >35 kg/m² (N = 13). Because all study participants had type 2 diabetes, data do not provide insight into whether utilities would be different if based on perceptions of people in Japan without type 2 diabetes. Future research could examine utilities associated with weight changes based on preferences of people in Japan with obesity, but without type 2 diabetes.

It should also be noted that the sample included a relatively high percentage of male participants (84.8% male). The proportion of male participants in this sample is consistent with statistics reported by Japan’s Ministry of Health, Labour, and Welfare indicating that type 2 diabetes is more common in men than women⁴⁰. The high proportion of male participants is also consistent with samples in relevant Japanese clinical trials^19–22.

There are also limitations associated with the videoconference interview approach. For example, BMI is derived from self-reported height and weight, which could not be confirmed by interviewers at the time of the interview. Although the in-person mode of administration will always be ideal for TTO interviews, TTO tasks conducted by videoconference have been shown to generate good quality data when valuing relatively brief straightforward health states such as health states derived from the EQ-5D-5L⁴¹. The health state vignettes in the current study were similarly brief and easy to compare, and therefore, the videoconference approach seems sufficient for this study.

Conclusions

In summary, this study provides estimates of utility change associated with weight reductions among people with obesity and type 2 diabetes in Japan. Results were generally similar to those from a previous study in the UK, although some differences did emerge for the largest percentage of weight reduction. Results of this study may be used to provide inputs for CUAs examining and comparing the value of treatments that are associated with substantial amounts of weight loss in patients with type 2 diabetes and obesity in Japan.

Transparency

Declaration of financial/other relationships

LSM, KDS, TH, WM, and JI are employed by Evidera, which received funding support from Eli Lilly and Company to conduct this research. AY is an employee of Breeze/Autumn Research, which received funding support from Evidera to conduct this research. JR, RSN, and KSB are employed by and own stock or hold shares in Eli Lilly.

Author contributions

LSM, KDS, TH, WM, JR, RSN, and KSB designed the study; LSM, KDS, TH, WM, and AY conducted the study; LSM, KDS, TH, and JI wrote the manuscript; JR, RSN, WM, AY, and KSB provided edits and comments on the manuscript drafts; and all authors approved the final version for publication.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Previous presentations

A portion of these study results were presented at ISPOR US in Boston, MA (May 7 – 10, 2023). The citation is: Matza LS, Howell TA, Redig J, Stewart KD, Morris W, Sato M, Newson R, Yasui A, Boye KS. Health State Utilities Associated with Weight Loss in People with Type 2 Diabetes in Japan. Presented at ISPOR; May 7 – 10 2023.

Ethical approval

All participants provided informed consent prior to participation. Study procedures and materials were approved by an ethics committee in Japan (Research Institute of Healthcare Data Science [RIHDS] RI20210278).

Acknowledgements

The authors would like to thank Ines Canada and Paula Gonzalez of Global Perspectives and Keiichi Seki, Nobuko Akiyama, Miaki Tajima, Moe Kikuchi, Kazumi Hori, Kyoko Sato, Reina Kougu, and Akane Betsui of Breeze for assistance with participant recruitment and utility interviews in Japan; Toshihiko Aranishi and Manaka Sato of Eli Lilly Japan K.K for protocol review; Robyn Cyr of Evidera for statistical programming; and Amara Tiebout of Evidera for editing and formatting support.

Data availability statement

Data are available upon reasonable request.

Additional information

Funding

This work and the rapid track service fee were funded by Eli Lilly and Company. The authors had independence in decisions related to the study design, study conduct, interpretation of data, and manuscript content.