Study design, setting, and ethical concerns
This population-based retrospective cohort study was conducted in accordance with the guidelines of Strengthening the Reporting of Observational Studies in Epidemiology [11]. The Institutional Review Board (IRB) of Seoul National University Bundang Hospital exempted deliberation of the study protocol, because we used public data open to all researchers (IRB number: X-2102–666-904). The IRB also waived the requirement of informed consent, because data analysis was performed retrospectively in an anonymized form.
Database
The national health insurance service (NHIS) database of South Korea was used as the data source. As the NHIS is the sole public health insurance system in South Korea, it contains all data on disease diagnoses according to the International Diseases and Related Health Issues 10th edition (ICD-10) codes and prescription information on all drugs and/or procedures. The NHIS permitted data sharing after approving the study protocol (NHIS-2021–1-620).
Study population
We included all adult patients (≥ 20 years) who were admitted to the ICU from 2016 to 2019 in South Korea. The prescription code of ICU admission during hospitalization was used for data extraction; no patient was admitted to the ICU in South Korea without registering the prescription code. If a patient was admitted to the ICU twice or more during the study period, only the last ICU admission on the latest date was included in this study, because mortality after ICU admission was one of the endpoints of our study. Patients with missing data on important demographic variables (age and sex) were excluded from the analysis.
Intensivist coverage ICU in South Korea
The South Korean government established a special payment system by law only for hospitals that employed trained intensivists for ICU staffing. This payment system is implemented for hospitals to employ an intensivist for ICU staffing under the condition that the intensivist should work only in the ICU for ≥ 8 h/day and ≥ 5 days/week. Moreover, the law requires that there should be at least one trained intensivist per ICU. The intensivist has to be a certified specialist physician from the Korean Society of Critical Care Medicine after specific fellowship training in critical care medicine. Doctors specializing in internal medicine, anesthesiology and pain medicine, pediatrics, neurology, neurosurgery, emergency medicine, general surgery, and thoracic surgery can apply for the fellowship training course for trained intensivists, which lasts for 1 year. In 2022, there were 1774 trained intensivists in South Korea, comprising 316 (18%) anesthesiologists, 546 (30%) doctors of internal medicine, 201 (12%) neurosurgeons, 196 (11%) doctors of emergency medicine, 196 (11%) thoracic surgeons, 129 (7%) general surgeons, 118 (7%) doctors of neurology, and 72 (4%) pediatricians. As this special payment system for intensivist coverage in South Korea was initiated in August 2015, our study commenced on January 1, 2016. Patients who were admitted to the ICU in a hospital that employed trained intensivists were designated as the intensivist group, whereas the other patients were denoted the non-intensivist group.
Study endpoints
The primary endpoint was in-hospital mortality after ICU admission. The secondary endpoint was the 1-year all-cause mortality after ICU admission, defined as any death within 1 year after the date of ICU admission.
Study parameters
Age and sex were collected as demographic variables. Socioeconomic status-related information, such as the employment status, national household income level, and residence at ICU admission, were collected. Residence was classified into two groups: urban (Seoul and other metropolitan cities) and rural (all other areas). The NHIS contains data on patients’ household income level to determine insurance premiums for the year, and approximately 67% of medical expenses are subsidized by the government [12]. However, the Medical Aid Program includes individuals who cannot afford insurance premiums or have difficulty supporting themselves financially. In this program, the government covers nearly all medical expenses to minimize the financial burden. All patients were divided into five groups using the quartile ratio, in addition to the Medical Aid Program group. The length of hospital stay (LOS) (days) and ICU stay were collected. The admitting departments were classified into two groups [internal department (IM) or non-IM]. We determined whether or not patients were admitted to the hospital through the emergency room (ER). If patients underwent surgery during hospitalization, it was considered a surgery-associated hospital admission. Data regarding whether the patient was admitted to the isolated ICU were collected. The Charlson comorbidity index (CCI) was calculated using ICD-10 codes according to previous study, to reflect the patient’s comorbid disease status [13]. Patients were classified into three groups according to the level of the hospital to whose ICU they were admitted. The results of hospitalization were classified into four groups as follows: 1) same hospital follow-up, 2) transfer to a long-term care center facility, 3) death during hospitalization, and 4) discharge and other outpatient clinic follow-ups. The date of death during hospitalization or hospital discharge was also collected. The total cost for hospitalization was recorded in United States Dollars (USD).
Statistical analysis
The clinicopathological characteristics of patients were presented as the mean with standard deviation (SD) for continuous variables and numbers with percentages for categorical variables. The clinicopathological characteristics of the intensivist and non-intensivist groups were compared using the t test and chi-squared test for continuous and categorical variables, respectively. A hierarchical approach was used to account for clustering of the covariates at the level of the hospital, where patients were admitted to the ICU. For hierarchical cluster analysis, agglomerative clustering was performed using hospital-related variables, including hospital location, total number of doctors, type of hospital (general tertiary hospital, general hospital, and other hospitals), specialist doctors, nurses, and pharmacists, total number of hospital beds, and total number of operating rooms. Three groups were created based on the results of hierarchical clustering analysis; the characteristics of the three hospital groups are presented in Additional file 1: Table S1.
Thereafter, as the hospital level variable could be interdependent with patient-related variables as clusters, we constructed a mixed effect logistic regression model for in-hospital mortality in patients admitted to the ICU. All covariates were included in the mixed effect logistic regression model for adjustment, and the results were presented as odds ratios (ORs) with 95% confidence intervals (CIs), considering the random effect of the hospital level.
We also constructed a mixed effects Cox regression model for 1-year mortality in patients admitted to the ICU considering the random effect of the hospital level. The results were presented as hazard ratios (HRs) with 95% CIs. Moreover, we performed subgroup analyses according to age (> 65 or ≤ 65 years), CCI (> 3 or ≤ 3), surgery or non-surgery associated admission, and invasive treatment during ICU stay, such as mechanical ventilation, continuous renal replacement therapy (CRRT), and extracorporeal membrane oxygenation (ECMO) support. There was no issue regarding multi-collinearity between variables in all models with the criterion of variance inflation factors < 2.0. All statistical analyses were performed using R software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria), and statistical significance was set at P < 0.05.