The RESTRIC trial was a cluster-randomized, crossover, non-inferiority multicenter trial of patients with severe trauma and was registered with the UMIN Clinical Trials Registry (UMIN000034405) on October 8, 2018. The protocol (V.1.3) was initially approved on October 11, 2018. The detailed trial protocol was published in July 2020 [5]. The original protocol is in Japanese but was translated into English (Additional file 1). This pragmatic trial aimed to reproduce real-world management of how a transfusion strategy is applied upon patient arrival at the ED based as far as possible on the physician’s judgment. In this trial, we applied a cluster-randomized design to enable the study intervention initiation upon patient arrival at the ED; furthermore, the crossover design was implemented to reduce the confounding effects of different hospitals. The study protocol and statistical analysis plan were previously published [5]. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study design was approved by the Ethics Committee of each participating institution (Additional file 2) and that of the Japanese Association for the Surgery of Trauma.

Japanese tertiary emergency medical centers participated in the RESTRIC trial. The participating institutions were randomized to two schedules (restrictive or liberal RBC transfusion strategy [target hemoglobin levels: 7–9 or 10–12 g/dL, respectively]) with a 1:1 ratio based on a pre-created random assignment table. After randomization, the centers applied the first transfusion strategy for 1 year (first study period). After a 1-month washout period following the first study period, the second transfusion strategy was applied for another year (second study period) (Fig. 1).

Study design. A total of 22 emergency medical centers participated in the RESTRIC trial. The participating institutions were randomized to implement either the restrictive or liberal RBC transfusion strategy at a ratio of 1:1. Hb hemoglobin, RBC red blood cell, RESTRIC Restrictive Transfusion Strategy for Critically Injured Patients

The allocated transfusion strategy was posted at each center to provide opt-out opportunities to patients and their next-of-kin. The allocated transfusion strategy was applied to all patients during the initial trauma resuscitation phase and upon arrival at the ED. Written informed consent was obtained as soon as possible from the patients or their next-of-kin, after which the patients were enrolled in the trial; thereafter, the transfusion strategy was applied until a predefined initiation and follow-up period. The applied transfusion strategy was selected at the physician’s discretion for patients who declined enrolment in the trial.

The need for RBC transfusion is not always apparent at the time of arrival at the ED. Therefore, we included patients with trauma, aged ≥ 20 years, who had one of the following complications or conditions, based on the physician’s judgment: severe bleeding that can result in circulatory shock; suspected severe bleeding after arrival at the ED; and the potential for severe bleeding postoperatively during the acute phase of trauma. Furthermore, we excluded patients based on the following criteria: cardiac arrest before or upon arrival at the ED; transfer from another hospital; the physician’s decision to withdraw active treatment at initial assessment; complications of severe burns (≥ 15% body surface area); pregnancy; chronic anemia, as determined by the attending physician based on medical history (hemoglobin level: ≤ 7 g/dL); and objection to blood transfusion.

RBC transfusion is often initiated in patients with severe trauma with active bleeding before confirming a decrease in hemoglobin levels. Therefore, each RBC transfusion strategy was defined based on the target hemoglobin level rather than the threshold hemoglobin level. The attending physician determined the RBC transfusion initiation timing in patients with active bleeding based on hemoglobin levels and the presence of hemodynamic instability. Notably, the restrictive RBC transfusion strategy was not permissive hypotension and hypovolemia resuscitation strategy. When the hemoglobin level was sufficiently high as the target of each strategy, a crystalloid and/or colloid were administered for resuscitation. One RBC transfusion strategy was applied until 7 days after hospital admission, discharge from the ICU, decision to withdraw active treatment, or death. Patients were followed up for 28 days. Investigators contacted patients (or their representatives, as appropriate) discharged from the hospital prior to 28 days after arrival at the ED by telephone to collect information regarding patient status.

A safety monitoring board, comprising two independent experts not involved in the trial, was responsible for monitoring trial safety. Significant adverse events were immediately recorded in the medical record and electronic data capture system, the same system that recorded the assessment data. The treating physician reported significant adverse events to the site investigator, who reported them to the chief investigator of each site and the principal investigator. The principal investigator consulted the safety monitoring board. The board reviewed and examined the report and sent written recommendations in response to the principal investigator.

To evaluate the non-inferiority of the restrictive RBC transfusion strategy to the liberal RBC transfusion strategy, 28-day survival after arrival at the ED was used as the primary outcome. Patients with incomplete information regarding survival or death 28 days after arrival at the ED were considered dropouts and excluded from the analysis. Secondary outcomes included the following: time to death during the first 28 days; cumulative RBC concentrate, FFP, and platelet concentrate volumes transfused on Days 1, 7, and 28; ventilator-, catecholamine-, and ICU-free days during the first 28 days; organ (renal, hepatic, and respiratory) failure during the first 7 days; complication (deep venous thrombosis, pulmonary embolism, cerebral infarction, myocardial infarction, bowel ischemia, transfusion-associated lung injury [6], and sepsis [7]) rates during the first 28 days; and Glasgow Outcome Scale scores at hospital discharge. The number of event-free days for patients who died during the first 28 days after arrival at the ED was zero. Renal failure was defined as Stage III, according to the Kidney Disease: Improving Global Outcomes guidelines [8]. Hepatic failure was defined as a total bilirubin level ≥ 6 mg/dL, based on the Sequential Organ Failure Assessment score [9]. Respiratory failure was defined as moderate acute respiratory distress syndrome, according to the Berlin definition [10].

Based on our previous retrospective multicenter observational study [11,12,13,14,15,16,17,18], we assumed a 25% mortality rate at 28 days after arrival at the ED in patients exposed to the liberal RBC transfusion strategy. The inter-class and inter-period correlation coefficients were set at 0.05, and the non-inferiority margin was set at 3%. The non-inferiority margin was defined based on statistically and clinically acceptable tolerance margins, as referenced in previous large-scale clinical trials [4, 19,20,21]. In the RESTRIC trial protocol, we calculated a sample size of 170 patients for each of the transfusion (restrictive and liberal RBC) strategy groups to reach a power of 80% and a one-sided alpha of 2.5%, assuming that 17 centers participated as a cluster [5, 22]. Therefore, we set the target sample size at 400 patients, considering possible cluster size variation, including non-eligible patients and dropouts during follow-up [5]. However, the actual sample size required (6214 patients) was much larger than the target sample size (400 patients) because of an error in the sample size calculation [22]. Because the error was not discovered until the end of the study, the study could not be terminated during its course.

Continuous variables are expressed as medians (interquartile ranges) and were compared using the Wilcoxon rank sum test. Categorical variables are expressed as numbers and percentages and were compared using the χ2 test or Fisher’s exact test if the expected count was < 5. The primary outcome analysis was adjusted for clustering within sites. The analysis used a mixed model, with intervention (restrictive or liberal RBC transfusion strategy) and period (order of transfusion strategy allocation) as fixed effect factors. Site and interaction between site (participating institution) and period (order of transfusion strategy allocation) were incorporated as random effect factors [23]. Furthermore, the non-inferiority margin was set at 3%. The null hypothesis was P1–P0 ≤ –0.03 (P0, 28-day survival rate [liberal RBC transfusion strategy]; P1, 28-day survival rate [restrictive RBC transfusion strategy]). Therefore, we evaluated whether the lower limit of the P1–P0 95% confidence interval (CI) exceeded –0.03. However, a logistic regression model was used for the primary analysis. Thus, we converted the non-inferiority margin into a certain value in terms of the odds ratio, which was determined based on the actual value of P0. We evaluated whether the lower limit of the 95% CI of the odds ratio exceeded this value (Results section). After excluding cases with missing primary outcome data, we used the full analysis set for the primary outcome analysis. In particular, we followed the intention-to-treat and per-protocol analysis principles for the primary and sensitivity analyses, respectively. The per-protocol analysis excluded cases in which transfusions intentionally deviated from the target hemoglobin level. Subgroup analysis was performed to investigate the effect of the intervention in terms of sex, age (< 60 vs. ≥ 60 years), Injury Severity Score (< 16 vs. ≥ 16 points), head trauma, and the performance of definitive surgery within 6 h of ED arrival.

Secondary outcomes were evaluated as follows. Time to death during the first 28 days was estimated using the Kaplan–Meier method and compared using the log-rank test. Hazard ratios were calculated using a Cox regression model. In addition, for changes in hemoglobin levels, the P-value at each timepoint was calculated using a mixed model adjusted for the initial hemoglobin level, intervention, period as a fixed effect, and site and interaction of the site with the period as a random effect. A P-value < 0.05 was considered statistically significant. All statistical analyses were conducted using R statistical software (version 3.6.3; R Foundation for Statistical Computing).