This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.
Model Structure and Analytical Methods
Using TreeAge Pro Version 2019 R2.1 (Boston, MA, USA), a microsimulation model was developed and a base case analysis was performed to assess the cost-effectiveness of linagliptin plus SoC compared with SoC, as per the two patient groups included in the CARMELINA trial (ClinicalTrials.gov NCT01897532) [7, 8].
The six-step simulation process using the microsimulation model is shown in Fig. 1. The microsimulation model was used to simulate the process of repeated experiences of multiple events. Microsimulation tracks patients at an individual level and therefore can take into account the impact of each individual medical history and time spent; additionally, using endpoint data from an outcomes trial, the microsimulation model allows estimation of cost-effectiveness beyond the trial follow-up period to a lifetime horizon. According to the International Society for Pharmacoeconomics and Outcomes Research–Society for Medical Decision Making modeling task force, a microsimulation model enables the estimation of the long-term impact of intervention for people with T2D . Furthermore, the United Kingdom Prospective Diabetes Study (UKPDS) regression model, which is widely used for cost-effectiveness analysis for antidiabetic drugs, was also constructed on the basis of a microsimulation model . An analysis cycle of 1 month was chosen to minimise the influence of competing risks occurring when multiple events were simultaneously considered and quality-adjusted life years (QALYs) were used as an effectiveness measure. The time horizon was defined as a lifetime. Only direct medical costs were included and the discount rate for costs and effectiveness was 2% per year. The analyses were conducted according to current Japanese guidelines for cost-effectiveness evaluations .
The present analysis used data from the Asian subpopulation (n = 555) included in the CARMELINA trial [7, 8]; patients were from Japan, China, South Korea, Taiwan, and Malaysia, and self-identified as being of Asian race . During a 2.2-year median follow-up period, linagliptin treatment had a neutral cardiorenal profile and was not found to be associated with 3P-MACE in Asian patients, with a similar event occurrence rate in linagliptin plus SoC- and SoC-treated patients (10.7% and 11.7%, respectively; hazard ratio [HR] 0.90; 95% confidence intervals [CI] 0.55–1.48). Linagliptin was associated with a nominal decrease in the risk of hospitalisation for heart failure (HR 0.47; 95% CI 0.24–0.95) .
The mean age of the Asian subpopulation in the CARMELINA study used for the present analysis was 65 years. The incidence of events included in this analysis (nonfatal myocardial infarction, nonfatal stroke, hospitalisation due to unstable angina, hospitalisation due to heart failure, albuminuria progression, renal failure, and CV death) was reported as event incidence per month. This was calculated from the events observed in the analysis of the Asian subgroup of the CARMELINA trial  using the person-year (PY) method.
To estimate the risk for each event, the baseline event risk was estimated from the data of the SoC group using data from the CARMELINA Asian subgroup analysis (Table 1). The event risk for linagliptin was estimated by multiplying the event risk calculated for the SoC group by the HR observed for the linagliptin/SoC group in the CARMELINA Asian subgroup analysis.
Additionally, this analysis attempted to evaluate how the risk of subsequent CV or kidney events would be increased in the lifetime simulation. To estimate the effect of medical history on the risk of a subsequent CV or kidney event, the regression coefficient, which was estimated from a previously published cost-effectiveness analysis in Japan using data from another Asian subgroup analysis of a CV outcome trial , was converted to an HR for use in this analysis. Supplementary Table S1 presents the assessment of the increased risk of an event due to a medical history of events included in this analysis.
The present analysis was performed from the perspective of the public healthcare payer, using only direct medical costs, including patient co-payments (reported in yen). The drug cost of linagliptin was set according to the National Health Insurance Drug Price List at the time of analysis, and the treatment costs for each event were obtained from a previous study .
The treatment costs for each event were estimated on the basis of the results calculated from data in the EBM Provider®—a Japanese Claims database provided by Medical Data Vision Co., Ltd.—in a previously published cost-effectiveness analysis in Japan  (Supplementary Table S2). For the costs associated with renal failure, only dialysis was considered. This analysis did not take into account the impact of the medical payment system revision rate that was implemented every other year in Japan because the difference in the medical payment rate between that used in this analysis  and the 2018 implementation was only + 0.55%, and the impact on the results was negligible. The cost of SoC was not considered because SoC treatment was received by both groups in the CARMELINA trial.
The QALYs of linagliptin added to SoC or SoC alone were estimated by subtracting the disutility of the occurrence of each event during the time horizon from the QALYs calculated by accumulating baseline utility weights of patients included in the analysis. The baseline utility weight and the disutility at the occurrence of each event were obtained from a previously published analysis using the EuroQol 5-dimension 3-level (EQ-5D-3L) measure of health value in an American civilian population . This study did not include the decrease in utility with hospitalisation for unstable angina and therefore this was assumed to be the same as that for nonfatal myocardial infarction (Supplementary Table S2).
The impacts of uncertainty and variability around the model inputs were tested using a one-way sensitivity analysis series and a probabilistic sensitivity analysis. To evaluate the magnitude of the effect of each parameter on the analysis results, a one-way sensitivity analysis was performed. To display the results of the one-way sensitivity analysis pictorially, a tornado diagram was constructed using each parameter. The range of change in each parameter was 0–4% for the discount rate and ± 20% of default values for other parameters with the associated 95% CI, when available.
In addition, a probabilistic sensitivity analysis was performed with 10,000 Monte Carlo simulations in 10,000 patients to evaluate the uncertainty of the results. The stochastic parameters and utility parameters were assumed to have a beta distribution, the ratio data (HR) were assumed to be logarithmically normally distributed, and the cost parameters were assumed to have a gamma distribution .
A sensitivity analysis was conducted on the base case analysis to consider the effects of increased event risk due to medical history, as these factors will result in a conservative estimate for linagliptin.
To evaluate the impact of using the point estimates of HR data, which were not tested for statistical significance, a scenario analysis was performed in cases where the HR of linagliptin on each event was set to 1.0 and where the costs for renal failure were assumed to be 0 yen.
A scenario analysis was conducted on the base case analysis considering medical history for the same reasons as the sensitivity analysis.