Study design and setting
We conducted a nationwide registry-based retrospective cohort study, wherein we analyzed data from the Japan Trauma Data Bank (JTDB) between April 2009 and March 2019. The details of all trauma patients who suffered a severe injury at any region of the body, with an abbreviated injury scale (AIS) score of ≥3, were registered in the JTDB. During the study period, the JTDB received records from 280 secondary or tertiary emergency hospitals in the country. The database includes information on injury mechanisms, prehospital times (including the times of paramedic dispatch, physician contact, and hospital arrival), patient baseline characteristics (including vital signs at the scene of injury and upon arriving at an emergency department [ED]), procedures performed, and survival status at hospital discharge.
In Japan, the operation of prehospital physician teams, such as dispatch criteria and operating time), varies according to the medical control area. The coverage area also varies largely depending on whether it is an urban or rural area. The physicians are delivered in a car or a helicopter according to the system of the medical control area. They are not always trauma surgeons but are those usually working at an ED and trained to provide basic prehospital trauma management such as assessment with sonography, tracheal intubation, chest drainage, intraosseous infusion, and temporal hemostatic maneuver using a tourniquet. Regarding fluid resuscitation, prehospital blood transfusion is not common in Japan, and only the administration of the crystalloid solution is provided in many cases. In contrast, the medical interventions allowed to Japanese paramedics responding to trauma patients without cardiac arrest are limited to performing spinal motion restriction, external fixation of bone fractures, oxygen administration using a mask, and administration of Ringer’s solution (only to patients with shock).
This study complied with the principles of the 1964 Helsinki Declaration and its later amendments. The Ethics Committee of Tokyo Medical and Dental University approved this study (#2192). The requirement for informed consent from each patient was waived because of the study’s retrospective design and the use of anonymized patient data.
Study population
Patients who met all of the following criteria were included: (1) patients who aged more than 15 years and suffered blunt injuries of Injury Severity Score (ISS) ≥16, (2) patients who were transferred directly from the scene of injury, and (3) patients whose specific information regarding times of injury, physician contact, and hospital arrival were available. We excluded patients from the analysis if they met at least one of the following criteria: (1) cardiac arrest at the scene of injury, (2) unsalvageable injury defined as AIS = 6, (3) missing data required for analyses (i.e., complete case analysis), and (4) unrealistic or outlier values on prehospital time course, such as time from injury to hospital arrival and time from injury to physician contact, in which outlier values were detected statistically using a single-sample robust linear regression analysis with M estimator[9] and then removed.
Variables
We collected information on the following items from the JTDB: age, sex, mechanism of injury, year of injury, season of injury, time of injury, time of physician contact, time of hospital arrival, vital signs at the scene of injury (systolic blood pressure, heart rate, and respiratory rate), consciousness level at the scene of injury recorded using the Japan Coma Scale[10] (Supplementary table 1), vital signs upon hospital arrival (systolic blood pressure, heart rate, and respiratory rate), consciousness level upon hospital arrival recorded using the Glasgow coma scale (GCS), the highest score of AIS values for each region of the body, ISS, and patient survival status at hospital discharge.
Eligible patients were divided into the two groups: patients who received physician-led prehospital management (physician-led group) and the patients who received paramedic-led prehospital management (paramedic-led group). Patients who received physician-led prehospital management were identified by comparing time of physician contact (i.e., the time that the physician started the assessment of the patients) and time of hospital arrival. Season of injury was divided into four categories by quarter, beginning in January. Time of injury was divided into four zones every 6 hours, beginning at 0:00. The study outcome was defined by in-hospital mortality.
Statistical analysis
The present study analyzed non-randomized data in which patient characteristics were not equally distributed between the physician-led and the paramedic-led groups. Considering the unbalanced characteristics between the two groups, we used a propensity score matching analysis[11] to compare the outcome. In this analysis, a logistic regression model was applied to estimate the propensity score for each patient, predicting physician-led prehospital management based on age, sex, mechanism of injury, year of injury, systolic blood pressure and respiratory rate at the scene of injury, consciousness level at the scene of injury, and ISS, in addition to prehospital transport time (from injury onset to hospital arrival). Both the time and season categories of injury were also incorporated into the model. Since the availability of emergency physician or trauma surgeon varies depends on working hours, and the prehospital transport time varies according to weather or road conditions depends on season, these variables could affect the patient outcome in severe trauma. These variables were chosen based on the clinical perspective and subject matter knowledge. The accuracy of a logistic regression model predicting in-hospital mortality with these variables was assessed using C-statistics. Propensity score matching extracted 1:4 matched pairs from the physician-led and paramedic-led groups; this ratio was determined based on the feasibility of match balance and maximum use of patient data. Match balance between the groups was assessed by the absolute standardized mean difference (ASMD); values <0.1 were considered acceptable[12]. The caliper width was set as the standard deviation of the logit-transformed propensity score multiplied by 0.1 to achieve well-matched balance between the two groups. The chi-square test was used for intergroup comparison in the propensity score-matched cohort. As a sensitivity analysis, we also evaluated the effectiveness of physician-led prehospital management using a multivariate logistic regression model in an overall study cohort (i.e., not the propensity score-matched cohort). In this model, the aforementioned variables used in the propensity score calculation were used as the covariates. Multicollinearity was assessed by the variance inflation factor, with the tolerance value set at <2.
Subgroup analysis was performed in the propensity score–matched cohort to explore potential patients who were likely to benefit from physician-led prehospital management. We evaluated the p values for the interaction between physician-led prehospital trauma management and the following dichotomized categories for in-hospital mortality: age (<65 vs. ≥65), sex (male vs. female), blood pressure at the scene of injury (<90 mmHg vs. ≥90 mmHg), shock index defined by the heart rate/systolic blood pressure ratio (<1 vs. ≥1), presence or absence of coma (defined by Japan Coma Scale >30 at the scene of injury), ISS (<25 vs. ≥25), the highest AIS scores on the head, chest, abdomen, and pelvis and lower extremities (<3 vs. ≥3), and the time lapse between the time of injury and the time of hospital arrival (<60 min vs. ≥60 min).
Descriptive statistics were reported as counts and percentages for categorical variables and medians and the 25th–75th percentiles for numeric or ordered variables. Predictive statistics were reported as odds ratios (ORs) and 95% confidence intervals (CIs). The level of significance was defined as two-sided p <0.05 for all statistical analyses. All analyses were performed using R 3.5.3 (R Foundation for Statistical Computing, Vienna, Austria) with add-on packages of “Matching[13]” for propensity score matching and “robustbase[14]” for robust regression analysis.