This study was exempt from requiring ethical approval from the Institutional Review Board of the University of Tohoku (reference no. 2022-1-444). The requirement for informed patient consent was also waived because of the anonymized nature of the data. This study was reported in accordance with the Consolidate Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) checklists [10].

This retrospective observational study analyzed inpatient data from the DPC database in Japan. The DPC database contains the clinical and medical expenditure information of over seven million patients admitted annually to the hospital, collected from nearly 1100 healthcare facilities. The database includes the following data for all inpatients: age; sex; diagnostic record with the International Classification of Diagnosis, 10th Revision (ICD-10) code, admission type (emergency or elective), daily procedures recorded using Japanese medical procedure codes, SOFA score during ICU admission, and discharge status.

ICU patients enrolled in the DPC database between April 2020 and March 2021 were included in this study. Patients with ICU management fees 1, 2, 3, or 4 were eligible. The exclusion criteria were as follows: (1) requirement of multiple intensive care treatments under ICU1, 2, 3, and/or 4 in the same admission period; (2) readmission to the ICU after having previously been discharged; (3) age < 15 years; (4) ICU admission > 14 days; (5) missing or unclear SOFA score data; and (6) the surgery date did not correspond with the recorded anesthesia date. The selected patients were categorized into two groups: patients who received ICU management fees 1 or 2 (ICU1/2 group) and those who received ICU management fees 3 or 4 (ICU3/4 group). The following baseline patient information was collected: age at admission, sex, information on surgery, admission type, academic hospital or non-academic hospital, and usage of blood transfusion therapy.

The following clinical outcomes were compared between the ICU1/2 and 3/4 groups: ICU all-cause mortality, in-hospital all-cause mortality, length of ICU stay, and length of hospital stay. Patients who were discharged alive and those who died in the ICU or hospital were included in the evaluation of the length of hospital stay.

The medical costs for each patient from the day of ICU admission to the day of ICU discharge were obtained from the DPC data. The medical costs covered procedures, surgeries, anesthesia, blood transfusion, drugs, and hospital fees but excluded service fees of meals, transportation, and family care. The costs (which does not include the cost of surgery and anesthesia) and the total costs (which include the cost of surgery and anesthesia) were determined. The difference in the total costs between the ICU1/2 and ICU3/4 groups was used to analyze cost-effectiveness. Cost-effectiveness was determined based on the total costs, because the intention was to be close to real-world settings. All costs were obtained in Japanese Yen (JPY).

For patients who were alive at hospital discharge, the expected life expectancy after hospital discharge was calculated using the life expectancy table obtained from the Japanese Ministry of Health, Labour and Welfare [11]. Patients who died during their hospital stay were recorded as having zero life expectancy. The life expectancy of the ICU-discharged population was estimated to be shorter than that of the general population of the same age [12,13,14]. Based on previous reports, the reduction rates according to the age group (≤ 51, 52–63, 64–74, ≥ 75 years) for patients who were discharged from the ICU were set to 0.66, 0.67, 0.56, and 0.71, respectively [15, 16]. The value of 0.65 was used as the overall reduction rate for patients who were discharged from the ICU. This value was calculated using a weighted average of the respective reduction rates in the percentage of the population by age in the present study. These reduction rates were used to calculate the life-year gained (LYG) by multiplying the life expectancy by reduction rates, as shown in the following formula [17]:

Although a decline in the quality of life (QOL) of the patients after discharge from the ICU was expected, QOL cannot be assessed directly with the DPC data [18,19,20,21,22,23,24,25]. Therefore, LYG was multiplied with utility values (derived from the EuroQol 5-dimensions [EQ-5D]) to estimate quality-adjusted life-year (QALY), as shown in the following formula [26]:

A systematic review and previous literature reported that the utility values of patients admitted to the ICU ranged from 0.63 to 0.81 (Additional file 2) [18,19,20,21,22,23,24,25]. In this study, the utility values for ICU stay, emergency surgery, elective surgery, and non-surgical settings were 0.63, 0.70, 0.71, and 0.61, respectively [25]. Although several studies reported utility values classified by ICU stay, emergency surgery, elective surgery, and non-surgery, the lowest utility values reported were used in this study [19, 25].

The incremental cost-effectiveness ratio (ICER) was calculated as the difference in the total costs of ICU stay divided by QALY, as shown in the following formula:

$$}\left( } - }} \right)\left( } - }} \right).$$

The cost-effectiveness cutoff was specified as an ICER value of < 5 million JPY/QALY, according to previous studies conducted in Japan [27, 28]. The cost-effectiveness of ICU1/2 compared with that of ICU3/4 was analyzed from a health policy perspective.

Costs were not discounted as only the intensive care expenses were considered. Similarly, the time point was not considered, as only intensive care can be objective.

Subgroup analyses of clinical outcomes and costs were conducted. Patients were stratified into subgroups by (1) age (≦ 51, 52–63, 64–74, and ≥ 75 years); (2) type of admission (elective admission, emergency admission); (3) type of surgery (emergency surgery, elective surgery, and non-surgery); and (4) SOFA scores (0–2, 3–5, 6–8, 9–11, 12–14, 15–24). The ICU mortality rates, in-hospital mortality rates, and cost-effectiveness for ICU1/2 and ICU3/4 were then compared according to these subgroups. The SOFA scores were calculated to determine the impact of the severity of organ damage [9, 29]. A surgery was selectively defined as a surgical operation performed on the same day with anesthesia (including general, intravenous, epidural, and spinal anesthesia). Emergency surgery was defined as emergency admission and surgery. Elective surgery was defined as elective admission and surgery. The category of non-surgical operation was used when a patient was not registered for surgery. The SOFA scores on the first day of ICU admission were used.

Several sensitivity analyses were conducted to investigate the validity of the main analysis. The adjustments mentioned above that were used for the calculation of LYG and QALY (0.65 and 0.63) were obtained from the published literature; however, they were based on old studies; hence, the values were likely to be inaccurate. The sensitivity analysis was performed to investigate this uncertainty for the calculated factors of LYG and QALY. In this scenario, the variation of ICERs was examined when the adjustment factors were changed. In the worst case, a sensitivity analysis was performed when reduction rates were 0.5 for the LYG calculation and 0.5 for the QALY calculation. In the best case, a sensitivity analysis was also performed when reduction rates were 0.8 for the LYG calculation and 0.9 for the QALY calculation. The sensitivity analysis set values to a range covering previously reported utility values for ICU stay of 0.63 to 0.81 (Additional file 2) [18,19,20,21,22,23,24,25]. The results were compared with the values obtained under the standard condition (basic case).

All analyses were performed using Python (version 3.7.13) software. Continuous variables are expressed as means and standard deviations, while categorical variables are expressed as numbers and percentages. The Chi-squared test and Wilcoxon rank-sum test were used to compare the two groups. For all analyses, statistical significance was defined as p < 0.05.