Explanation of the emergency medical system in South Korea
The promulgation of the Emergency Medical Service Act in 1994 became the starting point for the establishment of the modern emergency medical system in South Korea; the current system was established in the 2000s [23]. This is more than 30 years behind the United States of America (USA) and Western European countries, where separate organizations for the management of emergency patients have operated with national support since the 1960s. The system was created by rating three levels of EMIs according to the level of available resources or specialized care. These EMIs included the Regional Emergency Medical Center (REMC), Local Emergency Medical Center (LEMC), and the Local Emergency Medical Institution (LEMI). The REMC and LEMI represent the highest and lowest levels of EMIs, respectively. The LEMCs are also subdivided into tertiary hospitals with ≥ 500 beds and general hospitals with 300–499 as well as < 300 beds. In 2017, there were 36 REMCs, 118 LEMCs, and 261 LEMIs, and the funding received by EMIs amounted to 250 million United States dollars (USD) [18].
Master plan for the establishment of a national trauma system in South Korea
The South Korean government and the medical society created a master plan for NTS establishment in 2012 (Figure 1a) [24]. The law on the establishment and operation of RTCs was enacted in the Emergency Medical Service Act and candidate institutions for RTCs, among REMCs and LEMC with ≥ 500 beds that met the criteria, were publicly recruited and selected after evaluation. Essentially, each institution would be equipped with facilities and equipment dedicated to the management of trauma patients, including ≥ 2 resuscitation rooms, ≥ 1 operating room always available for emergency surgery, an intensive care unit (ICU) with ≥ 20 beds, and a general ward with ≥ 40 beds. The South Korean government provided 67 million USD per institution for the construction these facilities and equipment and also supported labor cost for 25 dedicated trauma doctors per institution since designation. By 2017, 16 RTCs were designated and 9 RTCs were officially opened (Figure 1b) [18, 24].
Database used for the study
We used data from the National Emergency Department Information System (NEDIS), which collected healthcare-related information registered by EMIs nationwide in real time based on the Emergency Medical Service Act. In 2017, 413 of 416 EMIs nationwide transmitted data, and the total number of transmissions was approximately 9,100,000 (See Additional file 1). The Korean Trauma Data Bank (KTDB) was established in 2013 to prepare the base for the trauma system by collecting information transmitted from RTCs. The KTDB has injury severity scores registered by full-time trauma coordinators. In 2017, 14 institutions registered information for 2,000–3,000 trauma admissions per RTC.
National survey of the preventable trauma death rate by multi-panel review
We performed a national survey on PTDR every 2 years beginning from 2015. The investigation was conducted in the following order: design/extraction of the sample population, data collection, panel review, reliability test for the review process, and result analysis. The criteria and review process for the decision on preventability were mainly based on the WHO guidelines for trauma quality improvement programs [4].
The target population for sampling was selected from trauma deaths with at least one diagnostic codes of S, T based on the Korean Standard Classification of Diseases (the Korean version of the International Classification of Disease) of 2015 and 2017 according to the NEDIS (Figure 2a). After obtaining approval (E-1811-005-982) from the Institutional Review Board of Seoul National University Hospital, data were collected. All medical records, including imaging studies, were obtained from the EMIs with cooperation from the central and local governments, in accordance with the Medical Service Act. Pre-hospital information including the data of hospitals that transferred patients were also included.
We used a structured review form, including audit filters, for the review of medical records (see Additional file 2). The review form was based on the data sheet for use in preventable death panel reviews with embedded audit filters as in the WHO guidelines for the Trauma Quality Improvement Program (TQIP). Designated assistants (pre-reviewers), comprising trauma coordinators (overall, 5 in 2015 and 12 in 2017) working in RTCs nationwide, investigated and recorded the general characteristics of patients, injury-related information, as well as transport and treatment-related information, using the review form before the preventable death panel review. The case review panels mainly comprised trauma specialists working at RTCs. A total of 10 teams in 2015 and five teams in 2017 were formed, and each team consisted of two general surgeons, one thoracic surgeon, one neurosurgeon, and one emergency physician. Moreover, a committee that comprised five trauma specialties, was responsible for developing guidelines for the entire review process and for training reviewers. When preventability was not decided by the multi-panel review, the committee reviewed and confirmed the final decisions. Three teams were selected to evaluate the reliability of the panel review. They repeated reviews for 5% of the overall cases that had already been reviewed by other teams (see Additional file 3).
National evaluation of the performance and outcomes of the new trauma system
An analysis was conducted to compare the performances and outcomes of trauma care between two survey periods. International Classification of Disease Injury Severity Score (ICISS), extended ICISS, and Trauma and Injury Severity Score (TRISS) models were created for severity adjustments. We calculated the probability of survival (Ps) from these models for the outcome analysis (Figure 2b). The extended ICISS indicated an ICISS model adjusted for age and the Revised Trauma Score (RTS). To calculate the RTS, the initial physiologic parameters on ED admission were used. The LEMIs were excluded from the extended ICISS model because physiologic parameters in LEMIs were not registered to NEDIS. Since the injury severity scores required for the TRISS model are only registered in the KTDB, it was possible to create TRISS only with data collected from RTCs.
Statistical analysis
To calculate the nation’s representative PTDR and improve the efficiency of the panel review, we selected the target of the trauma death review through stratified two-stage cluster sampling. The stratification was designed as a double layer (see Additional file 4); the primary stratified variables were region and EMI level, and the secondary stratified variables were place (timing) of death and patient age. Our initial targeted sample sizes were 1,000 in 2015 and 1,300 in 2017; however, considering the cases to be excluded from the panel review, the survey sample sizes were determined to be 1,131 and 1,862, respectively (see Additional file 5). The sample size was targeted such that stable estimation would be possible to meet a limit of error of approximately ±4.5%p (2015) and ±3.8%p (2017) at 95% confidence levels for population ratio estimation.
To estimate the population PTDR, the sample weights of each hospital level and death were calculated according to the sample design method and applied to analyze the sample-designed survey data. For continuous data, normality testing was performed using the Kolmogorov-Smirnov test. Categorical variables were compared using the Chi-Square and Fisher’s exact tests accordingly. Comparison between groups was performed using the t-test and analysis of variance. The paired t-test was used to compare differences in 2015 and 2017. To examine the association between the likelihood of preventable death and its relating factors, we constructed multivariate logistic regression models. The goodness of fit for the models was confirmed through the Hosmer-Lemeshow test. The agreement between panel teams was evaluated using Cohen’s Kappa index. All statistical analyses were performed using SPSS (Version 18.0; IBM Corp., Armonk, NY), and two-sided p values < 0.05 were considered to indicate statistical significance.