We conducted a model-based cost-utility analysis from the perspective of the third-party healthcare payer in Japan. The intervention of interest was the novel trauma workflow using a hybrid ER and the comparator was the standard trauma workflow based on the ATLS guidelines without angiography-CT equipment. Outcomes analyzed were quality-adjusted life years (QALYs), costs, and the incremental cost-effectiveness ratio (ICER), which is the incremental cost associated with a new therapy needed to generate one additional QALY, since this methodology for economic evaluation is commonly used in health system payers and health technology assessment organizations.[9,10] As the costs were recorded in Japanese yen (JPY), we converted them into US dollars (105 JPY = $1 USD). The willingness-to-pay for one additional QALY gained was set to $47,619, which is equivalent to 5 million JPY, the current threshold willingness-to-pay for a QALY in Japan. The annual discount rate was set at 2% in both costs and utilities based on Japanese guidelines for economic evaluations of healthcare.
The study was approved by the Institutional Review Board of the Osaka General Medical Center (S201916010). The board waived the need for informed consent, as this was a modeling study based on retrospective data collection only.
The modeled population was severe blunt trauma (injury severity score ≥ 16) patients without severe traumatic brain injury (TBI) (Glasgow Coma Scale ≤8 with intracranial hemorrhage demonstrated by CT). According to our previous study, cases of traumatic cardiopulmonary arrest on arrival, pediatric patients younger than 15 years of age, patients who were transferred to other hospitals within 24 hours after admission, penetrating trauma patients, and pregnant women were excluded.
Figure 2 shows model structure. We constructed a short-term decision tree and a long-term Markov model to determine the QALYs, life years (LYs), and costs associated with the conventional ER system and the hybrid ER system. All patients started with the “severe trauma” state and then transitioned to the “survived” state or “dead” state at 28 days after injury. Patients in the “survived” state either stayed in the “survived” state or moved to the “dead” state. The “dead” state was defined as the absorbing state. We set the initial age as 50 years according to the mean age of the previous study population. The length of time for the first decision tree was defined as 28 days and the cycle length for the Markov model was set as 1 year. The model was run until either death or fiftieth year, assuming that no patients survived after the age of 100 years. The model was developed and analyzed using TreeAge Pro 2019 (TreeAge Software, Williamstown, MA, USA).
The initial transition probabilities from the “severe trauma” state to the “dead” state, which were the 28-day mortalities in the two groups, were derived from our previous study cohort. In the conventional group, 28-day mortality was directly calculated from the observed data. We conducted a multivariable logistic regression to estimate the odds ratio (OR) and its 95% confidence interval (CI) adjusting for clinically plausible or known confounders: heart rate, body temperature, hemoglobin, lactate, prothrombin time-international normalized ratio, and probability of survival using Trauma and Injury Severity Score. The 28-day mortality in the hybrid ER group (P1) was obtained from the 28-day mortality in the conventional group (P0) and OR as follows.
For the transition probabilities from the “survived” state to the “dead” state, the same probabilities were used in both groups. Mortalities at first, second, and third year were extrapolated from an observational study that reported the 1-year and 3-year mortalities of major adult blunt trauma survivors. A transition probability (P) of death occurring over a time interval (t) with hazard rate (r) was calculated according to the following formula.
The transition probabilities in the fourth year and later were based on Japanese life tables and calculated with a weighted average of males and females in the general population using the proportion of male and female patients in the cohort.
Healthcare related costs for the initial admission in each group were obtained from the claims data in our hospital. We categorized them into surgical costs, transfusion costs, and hospitalization costs including pharmaceutical and procedural costs. These costs were analyzed in the short-term decision tree. We also investigated annual follow-up costs of the patients that survived. First to fifth year follow-up costs were directly obtained from the claims data. For the sixth year and later, we used the same costs as for the fifth year.
Costs for installation of the hybrid ER were provided by a manufacturer (Canon Medical Systems Corp., Tochigi, Japan). The costs consisted of two different parts: the price of the equipment and initial reconstruction costs of the ER. As the angiography-CT machine is a long-lasting resource, a depreciation period of 6 years was chosen as the life of the investment. The amortized yearly expenditure for the angiography-CT machine (M) was calculated from the initial price of the machine (Pr) of $1,714,000, the annual interest rate of 1% (i), and the depreciation period of 6 years (N) using the following formula:
In addition to the amortized cost, we included annual maintenance costs. The residual value of the angiography-CT machine was subtracted from the expenditure during the year following the depreciation period. We assumed that these costs were used to treat severe trauma patients who were transferred to the hybrid ER. As 270 patients were treated during a 4-year study period in the hybrid ER group, the discounted capital investment costs were divided by 405 patients (the estimated number of patients in the 6-year depreciation period) and added to the admission costs of patients in the hybrid ER group.
We did not include time costs, productivity costs, and other non-healthcare costs in the analysis.
To calculate QALYs, we extrapolated utility values from the literature.[17,18] The utility in the first 28-day hospitalization period was derived from a study that assessed quality of life of critical care patients using the six-dimensional short-form health state questionnaire (SF-6D). Another study that used the EuroQol – Five Dimensions questionnaire (EQ-5D) for the health state assessment was selected to determine the utility in the follow-up period.
We performed deterministic sensitivity analyses to assess the impact of various key parameters. The ranges for each parameter were determined by 95% CIs derived from the cohort data or publications if available. Otherwise, the plausible range was decided based on expert opinion. Moreover, we conducted a probabilistic sensitivity analysis using second order Monte Carlo simulations to explore uncertainty in the input parameters. Values of parameters were randomly selected from the distribution of the input parameters and the model was run, which was repeated for 1000 simulations. We plotted the results on the cost-effectiveness plane and described the cost-effectiveness acceptability curve to estimate the proportion of simulations that the novel trauma workflow using hybrid ER would be preferred in terms of cost-effectiveness as function of the willingness-to-pay threshold.