This observational study was conducted using the Diagnosis Procedure Combination database, a nationwide Japanese administrative inpatient database. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

The database contains discharge summaries and administrative claims from more than 1500 acute-care voluntarily participating hospitals and data from approximately 50% of all acute hospitals and 90% of all tertiary emergency hospitals in Japan [19]. It includes the following patient-level data for all hospitalizations: demographic characteristics; primary diagnoses, comorbidities, and complications recorded with the International Classification of Diseases, 10th Revision (ICD-10) codes; daily procedures; daily drug administrations; daily blood product administrations; and admission and discharge status. A previous study that validated this database showed high specificity and moderate sensitivity for the recorded diagnoses and high specificity and sensitivity for the recorded procedures, although the trauma-specific diagnostic codes were not examined [20].

We identified patients who were hospitalized for trauma (ICD-10 codes: S00–T14 for the primary diagnosis) on an emergency basis by ambulance or walk-in between January 1, 2011, and December 31, 2020. We enrolled patients who received massive transfusion, defined as administration of at least 20 units of RBC in Japan (equivalent to 10 units of RBC in the USA or UK) within the first 2 days of admission. One unit of packed RBCs is equal to approximately 140 mL in Japan, 250–350 mL in the USA, and 280 mL in the UK.

Data on the following characteristics were collected from the database: calendar year of admission, i.e., 2011 to 2020; hospital characteristics (tertiary emergency or teaching hospital); age, sex, and body mass index at admission; Japan Coma Scale at admission [21], Charlson comorbidity index score [22]; ambulance use, regions to which injury was sustained, ICD-10-based injury severity score [23]; and treatments administered within the first 2 days of admission. The ICD-10 codes for the injured regions are listed in Additional file 1: Table S1. The severity of trauma was assessed using a validated ICD-10-based injury severity score [23]. Unavailable values for the body mass index at admission were treated as a missing category.

The study outcomes were in-hospital mortality and incidence of adverse events. Adverse events were defined as a composite of cardiac failure, respiratory failure, hepatic failure, renal failure, sepsis, thrombosis, transfusion transmitted viral infections, allergic/anaphylactic reactions, hemolytic transfusion reaction, and volume overload (besides the above) based on the definition of transfusion-related adverse events in previous studies [24, 25]. The ICD-10 codes used to identify adverse events are shown in Additional file 1: Table S2. Data on death in the emergency room, death within 24 h of admission, duration of hospitalization, and hospitalization costs were also collected.

The trends in the incidence and practice patterns of massive transfusion for patients with trauma were described by calendar year at admission from 2011 to 2020, and analyzed using the Cochran–Armitage trend test for binary variables and Jonckheere–Terpstra trend test for continuous variables [26]. The incidence of massive transfusion was calculated using the “number of hospitalizations for trauma that received at least 20 units of RBC within the first 2 days of admission” as the numerator and “the number of hospitalizations for (i) all trauma; (ii) trauma in a tertiary emergency hospital; (iii) trauma requiring admission to the intensive care unit or high-dependency care unit; and (iv) trauma requiring at least one unit of RBCs” as the denominator. The trends in massive transfusion-related procedures within the first 2 days of admission and consumption rate of blood products during hospitalization for patients who received massive transfusion from among the entire trauma population were also examined in a similar manner.

Restricted cubic spline analyses were performed to assess the non-linear association between the outcomes and transfusion ratios (FFP-to-RBC ratio and platelet-to-RBC ratio) [27]. Five transfusion ratio points (0.50, 0.75, 1.00, 1.25, and 1.50) were denoted as the knots. We fitted generalized estimating equations to the restricted cubic spline analyses with individual hospitals as the cluster and calculated the adjusted odds ratios and their 95% confidence intervals for each transfusion ratio relative to the reference point of 1.00. In a different analysis, transfusion ratios were categorized into four groups: 0.75 or less; 0.75 to 1.00; 1.00 to 1.25; over 1.25, and generalized estimating equations with individual hospitals as the cluster were created to assess the association between the four transfusion ratio categories and the outcomes, using 0.75 to 1.00 as the reference category. All adjusted analyses included the calendar year at admission, hospital characteristics, age, sex, body mass index at admission, Japan Coma Scale at admission, Charlson comorbidity index, ambulance use, injured regions, and ICD-10-based injury severity score as covariates.

Sensitivity analyses were performed by excluding patients who died in the emergency room in order to reduce survivor bias, because patients requiring massive transfusion often die during the early hours of admission before receiving substantial quantities of FFP or platelets [15, 28, 29]. Furthermore, post hoc sensitivity analyses were performed (i) by altering the definition of massive transfusion to patients who received at least 20 units of RBC on the day of admission; (ii) by altering the definition of massive transfusion to patients who received at least 60 total units of RBC, FFP, and platelets within the first 2 days of admission; (iii) by restricting the sample to patients admitted to tertiary emergency centers; and (iv) by restricting the sample to patients who were admitted to hospitals that had continuously provided data to the database from 2011 to 2020.

All analyses were performed using Stata/SE 17.0 software (StataCorp, College Station, TX, USA). Continuous variables were presented as means and standard deviations or medians and interquartile ranges as appropriate, and categorical variables were presented as numbers and percentages. All reported P-values were two-sided, and P-values < 0.05 were considered statistically significant.