Long-term projections of clinical and economic outcomes were made by using a patient-level cost-utility model developed in Microsoft Excel, based on risk equations and risk factor progression estimations from the United Kingdom Prospective Diabetes Study Outcomes Model 2 (UKPDS OM2) [22, 23]. The model is based on an integrated system of parametric equations that predict annual probabilities of disease-related complications and uses Monte Carlo methods to predict the occurrence of events.

Complication incidence (congestive heart failure, ischemic heart disease, myocardial infarction, stroke, blindness, ulcer, amputation, and renal failure) was driven by multivariate semi-parametric proportional hazards survival models capturing time-invariant factors, time-variant clinical risk factors and time-variant comorbidities based on the UKPDS OM2 complication risk equations. Diabetes-related mortality was driven by two logistic regression models to capture mortality in first year of event (i.e., the year in which simulated patients experienced their first diabetes-related complication) and subsequent years of events (i.e., the year in which simulated patients experienced a further diabetes-related complication) according to UKPDS OM2 risk equations [22]. For years in which no complications occurred, standard mortality rates derived from Japanese life tables, indexed by age and sex, were applied [24].

The base case analysis was run over a 50-year time horizon to capture the full duration of patient lifetimes, with model outputs including survival curves, cumulative incidence of complications, life expectancy, quality-adjusted life expectancy, disaggregated and total costs as well as incremental cost-effectiveness ratios (ICERs). ICERs were compared against the commonly used willingness-to-pay threshold of JPY 5,000,000 to assess whether treatment with tirzepatide versus dulaglutide represented good value for money [25, 26].

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

The simulated population for the base case analysis was based on cohort characteristics from the SURPASS J-mono randomized controlled trial, a head-to-head trial of Japanese patients comparing tirzepatide 5, 10, and 15 mg with dulaglutide 0.75 mg (Table 1) [18]. Treatment effects associated with tirzepatide 5 mg and dulaglutide 0.75 mg therapy on physiological parameters were taken from the SURPASS J-mono trial at 52 weeks, evaluated by efficacy estimand for the intention-to-treat population (Table 2) [18].

Long-term changes in glycemic control were based on UKPDS OM2 HbA1c progression equation, with HbA1c levels increasing over time in all treatment arms, and the difference between arms gradually diminishing [23]. Treatment switching occurred at a HbA1c threshold of > 8.0%, a commonly used threshold in Japan, at which point patients discontinued their treatment and initiated basal insulin therapy for the rest of the simulation. This simplifying assumption was applied as there is no evidence on the impact of the addition of treatments to tirzepatide, and incorporating the continuation of initial therapies alongside newly added treatments would have required significant assumptions. Upon switching treatment to basal insulin, HbA1c was assumed to decrease by a mean of 1.22% based on a published formula to determine changes in glycemic control when insulin-naïve patients initiate insulin therapy [27]. After this decrease, HbA1c continued to progress in line with the UKPDS OM2 parameter progression equation [23]. Changes in other risk factors associated with treatment (BMI, systolic blood pressure, LDL cholesterol, HDL cholesterol) were applied in the first year of analysis according to changes reported in SURPASS J-mono, and remained constant until switching treatment, at which point they returned to baseline (Table 2) [18]. The exception was estimated glomerular filtration rate (eGFR), which did not return to baseline after switching treatment, as an assumption of improving renal function upon initiation of insulin therapy seemed implausible. Therefore, no change was applied to eGFR on treatment switching. No further treatment-switching steps were modeled in the base case analysis.

Rates of severe and non-severe hypoglycemia on tirzepatide and dulaglutide were taken from the SURPASS J-mono trial [18]. Severe hypoglycemia was defined as an episode with severe cognitive impairment requiring the assistance of another person to actively administer carbohydrate, glucagon, or other resuscitative actions, with a blood glucose threshold of < 54 mg/dl used to define non-severe (clinically significant) hypoglycemia. For basal insulin therapy, non-severe hypoglycemia event rates were assumed to be 1400 per 100 patient-years in all groups, based on a real-world analysis of hypoglycemic events in patients receiving insulin [28].

Acquisition costs of the comparator (dulaglutide) and concomitant medications (metformin and basal insulin) were taken from the database of pharmaceutical costs in Japan, based on the update released from the MHLW in April 2023 [29]. In addition to the study medication, it was assumed that patients took 1000 mg metformin per day, as it is a representative maintenance dose in Japan [30]. Following treatment switching, basal insulin use was 40 IU per day in line with the World Health Organization defined daily dose for insulin glargine [31]. The analysis captured physician fees associated with tirzepatide and dulaglutide treatment. Visits were assumed to occur bimonthly in all treatment arms according to MHLW recommendations, and separate costs were applied in the year of treatment initiation and subsequent years [32]. Instructional training for the administration of insulin injections for patients switching treatment occurred in hospitals, and costs, therefore, were estimated based on costs from multiple Japanese institutions.

Costs of treating diabetes-related complications were estimated from panel data regression models using data from Japan Medical Data Center (JMDC) claims database, constructed by the JMDC Inc. (Tokyo, Japan) [33, 34]. The JMDC database contains de-identified inpatient, outpatient, dispensing and medical examination records pooled from different health insurance associations. The database captured information from 17 million Japanese people as of January 2024 [34]. For this analysis, data from the past 3 years of available data were extracted for costs associated with treatment of diabetes-related complications and are presented in the supplementary material (Table S1) [33, 35,36,37,38]. Non-severe hypoglycemia was assumed to be associated with no costs, while costs of severe events were taken from a previous cost-effectiveness analysis of orally administered GLP-1 receptor agonists in Japan [39].

Utility scores associated with diabetes and related complications were combined using an additive approach, with utility scores based on published data in Japanese populations [39,40,41]. Baseline quality of life for individuals with diabetes in Japan was based on the previous cost-effectiveness analysis of an orally administered GLP-1 receptor agonist in Japan and was assumed to be 0.901, with utility decrements applied for diabetes-related complications [39, 40]. Quality of life utilities and disutilities associated with diabetes-related complications and adverse events used in the modeling analysis are summarized in the Supplementary Material (Table S2).

Different utilities were applied to estimate the quality of life impact of BMI over the simulated time horizon. In the first year of the analysis, utility scores based on change in BMI were applied, while in subsequent years a BMI health state approach was used, as the quality of life impact of BMI changes is greater than the corresponding BMI state utilities. Therefore, in the first year of treatment, treatment-dependent increases in quality of life in relation to changes in BMI were calculated based on a recent publication in Japan (tirzepatide: + 0.034; dulaglutide: + 0.004) while for each subsequent year of the analysis a utility decrement per BMI unit above 25 kg/m2 of  0.0061 was applied in all treatment arms [42,43,44]. For each patient experiencing nausea, a disutility of 0.043 was applied in the first year of treatment, based on data from Matza et al. in 2007 [45]. The percentages of patients with nausea in each treatment arm were taken from the SURPASS J-mono trial (tirzepatide 5 mg: 11.9%; 10 mg: 19.6%; 15 mg: 20.0%; dulaglutide 0.75 mg: 7.5%) and the consequent utility impacts were  − 0.005 for tirzepatide and  − 0.003 for dulaglutide [18].

Sensitivity analyses were performed around the base case analysis to investigate the impact of alternative model inputs and assumptions on projected cost-effectiveness outcomes. To examine the effect of discounting on cost-effectiveness outcomes, simulations were performed with (symmetric) discount rates of 0% and 4%. The influence of time horizon on the outcomes projected by the model was investigated by running an analysis over 15 years, rather than 50 years in the base case analysis.

A series of analyses were performed to assess the impact of treatment-switching assumptions on the results. In the first, it was assumed that no treatment switching occurred. Analyses were then conducted with alternative HbA1c thresholds used to trigger the change to basal insulin, with thresholds of 7.5% and 8.5% used, compared with 8.0% in the base case analysis. A final scenario assumed that patients remained on tirzepatide or dulaglutide for 3 years, with the HbA1c difference maintained for that period and abolished on treatment switching.

Two scenarios were prepared to assess the impact of variation in the costs of treating diabetes-related complications. In the first scenario, the costs of complications were increased by 10%, while in the second scenario the costs were decreased by 10%.

As supporting scenarios, long-term projections of clinical and economic outcomes were estimated for tirzepatide 10 and 15 mg compared with dulaglutide 0.75 mg. The modeling approach followed the same one described in the base case analysis, using equivalent inputs and assumptions. Treatment effects for the two dosages are shown in the Supplementary Material (Table S3).