Massive Blood Transfusion for Trauma Score to Predict Massive Blood Transfusion in Trauma

Background Massive blood loss is the most common cause of immediate death in trauma. A massive blood transfusion (MBT) score is a prediction tool to activate blood banks to prepare blood products. The previously published scoring systems were mostly developed from settings that had mature prehospital systems which may lead to a failure to validate in settings with immature prehospital systems. This research aimed to develop a massive blood transfusion for trauma (MBTT) score that is able to predict MBT in settings that have immature prehospital care. Methods This study was a retrospective cohort that collected data from trauma patients who met the trauma team activation criteria. The predicting parameters included in the analysis were retrieved from the history, physical examination, and initial laboratory results. The significant parameters from a multivariable analysis were used to develop a clinical scoring system. The discrimination was evaluated by the area under a receiver operating characteristic (AuROC) curve. The calibration was demonstrated with Hosmer–Lemeshow goodness of fit, and an internal validation was done. Results Among 867 patients, 102 (11.8%) patients received MBT. Four factors were associated with MBT: a score of 3 for age ≥60 years; 2.5 for base excess ≤–10 mEq/L; 2 for lactate >4 mmol/L; and 1 for heart rate ≥105 /min. The AuROC was 0.85 (95% CI: 0.78–0.91). At the cut point of ≥4, the positive likelihood ratio of the score was 6.72 (95% CI: 4.7–9.6, p < 0.001), the sensitivity was 63.6%, and the specificity was 90.5%. Internal validation with bootstrap replications had an AuROC of 0.83 (95% CI: 0.75–0.91). Conclusions The MBTT score has good discrimination to predict MBT with simple and rapidly obtainable parameters.


Introduction
Trauma is one of the leading causes of death globally. In 2013, trauma accounted for 4.8 million deaths worldwide [1]. Hemorrhage was the most common cause of traumatic death in the first 24 hours [2]. It was found that timely resuscitation with blood components with a ratio of packed red blood cells (PRBCs) to fresh frozen plasma (FFP) between 1 : 1 and 2 : 1 decreased the mortality rate from exsanguination [3]. A massive blood transfusion (MBT) protocol is a preset guideline between the clinicians and the blood bank to prepare blood products in a timely manner. erefore, MBT protocols effectively help to reduce unnecessary blood component usage [4].
Usually, the MBT protocol is triggered by MBT scores derived from clinical parameters, such as low blood pressure or tachycardia. Many MBT scores have been published and validated [4]; however, the scores had good prediction only when they were applied in the same settings where the scores were created [5]. For instance, the assessment of blood consumption (ABC) score, which is one of the popular scores, had good prediction performance with an AuROC of 0.842 in the original setting but had an AuROC of 0.587 when validated in ailand, which was probably caused from different mechanisms of injury of patients [6,7].
In addition, most MBT scores were created from welldeveloped countries with mature trauma systems and advanced prehospital care. However, some parameters, such as radiographic imaging, cannot be obtained on a timely basis in developing countries as in ailand. e majority of prehospital care providers in ailand cannot deliver advanced prehospital resuscitation which may lead to delayed resuscitation and deterioration of the hemodynamics by the time the patients arrive at the hospital. erefore, the cut points of the parameters used in well-developed countries may not be suitable in developing countries due to the limitation of prehospital management. is study aimed to create an MBTscore to predict MBT in a setting that does not have advanced prehospital care.

Design and Setting.
is study was a diagnostic prediction study conducted in Songklanagarind Hospital which is a university-based hospital and a level 1 trauma center in Southern ailand. Annually, Songklanagarind Hospital has about 10,000 trauma patients who visit the emergency department with about 1600 trauma admissions. Only 50% of the patients were transported to the emergency department by the basic life support team [8]. is study was approved by the ethics committee of the Faculty of Medicine, Prince of Songkla University.

Population.
e enrolled subjects were adult trauma patients aged ≥15 years who met at least one of the criteria for activating the trauma team of Songklanagarind Hospital within 1 hour after admission from January 2012 to December 2018. e criteria of the trauma activation protocol are (1) systolic blood pressure (SBP) ≤90 mmHg; (2) gunshot wound (GSW) at the chest, abdomen, or back; (3) stab wound at the chest or abdomen or both; (4) respiratory rate <12/min or >30/min; (5) heart rate (HR) >120/min; (6) Glasgow Coma Scale (GCS) score ≤8; and (7) a positive Focused Assessment Sonography in Trauma (FAST) exam. Patients were excluded if they were dead on arrival, injured from hanging or burns, or had missing primary outcome data due to incomplete medical records. Patients who died from exsanguination but did not receive MBT were also excluded from the study.

Predictive Parameters.
Only parameters that could be obtained within a few minutes after arrival were included in the study. Individual parameters such as age and gender were also collected. Physiological parameters consisted of vital signs and GCS score. Injury parameters consisted of the mechanism of injury (i.e., road traffic injury, GSW, stab wound, and fall), pelvic and femur fractures that were diagnosed by physical examination, and the presence of free fluid from the FAST examination. Laboratory results were blood lactate and base excess. Hematocrit results were not included in the study because in the local setting, hematocrit results take longer to obtain.

Data Collection Process.
e data were collected and were retrieved from the trauma registry and the hospital electronic medical records of Songklanagarind Hospital.

Sample
Size. According to the TRIPOD guideline [11] for a predictive model, at least 10 events are required for one candidate variable. Eight variables were estimated to be included in the final model. A previous study at our institute demonstrated that the incidence of MBT was 9% in patients activated by the activation protocol [7]. erefore, the sample size was estimated to be about 889 patients for 80 events.

Statistical Methods.
Continuous variables are presented as mean and standard deviation or median and range according to the distribution of the data. Categorical variables are expressed as frequency and percentage. Univariable analysis was performed, and the candidate predictors identified as variables with significant values (p < 0.20) in the univariable analysis, as well as clinically important variables, were selected for a multivariable logistic regression analysis [12]. Backward elimination was then conducted to identify a parsimonious model. e score was then derived from complete data only. Two-tailed p values <0.05 indicated a significant difference. e coefficients of the significant variables from multivariable analysis were used to generate risk scores. Sensitivity, specificity, and positive likelihood ratio (LR+) were also demonstrated. e cut point of the score was calibrated from LR+. Discrimination was demonstrated by AuROC. e Hosmer-Lemeshow calibration test was used to determine the concordance between prediction and observations. Internal validation was done with bootstrap replications.

Results
During the seven years of the cohort period from 2012 to 2018, 1023 patients met the eligibility criteria. One hundred and fifty-six patients were excluded because of death on arrival, hanging, burn, and missing medical records, and 11 patients did not receive MBT but died from exsanguination. erefore, the remaining 867 patients were included in the study.
e patients who were entered into the study are shown in Figure 1.
Among the 867 patients, 102 patients (11.8%) were in the group that received MBT, and the remaining 765 (88.2%) patients were in the non-MBT group. Most of the patients in both groups sustained motorcycle crash injuries (54.9% vs. 44.6%). e group that received MBT had a higher median injury severity score (ISS) (32 vs. 14, p < 0.001). e median number of transfusion units of PRBC in the MBT group was 17 units (IQR: 13-25), while the non-MBT group had a median number of transfusion units of PRBC of 0 (IQR: 0-2). e results of the univariable analysis are shown in Table 1. From the univariable analysis, 11 predictors were included in the multivariable model. ose predictors were age ≥60 years, GCS score ≤8, SBP ≤90 mmHg, GSW, HR ≥ 105/min, presence of pelvic fracture from physical examination, presence of femur fracture from physical examination, need for nasal packing for hemostasis, BE ≤ -10 mEq/L, lactate >4 mmol/L, and presence of free fluid from FAST.

Score
Performance. e MBTT score had an AuROC of 0.85 (95% CI: 0.78-0.91) which demonstrated good prediction ( Figure 2). A Hosmer-Lemeshow goodness-of-fit test had a p value of 0.81 which demonstrated good calibration ( Figure 3). Risk stratification was done according to its LH+ into low risk and high risk. In the high-risk group with a score of ≥4, LH+ was 6.72 (95% CI: 4.7-9.6, p < 0.001), and in the low-risk group of MBT in our study with a score of <4, LH+ was 0.4 (95% CI: 0.27-0.59). e risk stratification of prediction values of the scoring system is demonstrated in Table 3. Internal validation with bootstrap replications had an AuROC of 0.83 (95% CI: 0.75-0.91) which demonstrated good internal validation.

Comparison with Other Scoring Systems.
Other scoring systems, such as the ABC score, Prince of Wales Hospital (PWH) score, and Trauma-Associated Severe Hemorrhage (TASH) score, were validated with the same dataset as the MBTT score, and the results showed that the MBTT score had the highest AuROC. e second-best prediction was the PWH score with an AuROC of 0.71 (0.60-0.82), while the ABC score had the lowest AuROC of 0.51 (0.37-0.63). Comparisons between the MBTTscore and the other scoring systems are shown in Figure 4.

Discussion
Our study included four significant parameters in the MBTT scoring system: age ≥60 years, BE ≤ -10 mEq/L, lactate >4 mmol/L, and HR ≥ 105/min. ese parameters showed a good prediction performance with an AuROC of 0.845. All predictors in the model were reported to be associated with MBT in other studies (i.e., age [13,14], BE [15], lactate level [16], and HR [13,15,17,18]. e incidence of MBT in this study was 11.6%, while the incidences from previous studies were between 4.8% and 25% [6,13,17]. e incidence of MBT varied because those studies were conducted in different populations. Some studies reported the incidence in patients who met the trauma team activation criteria [6], while some studies reported patients who had high ISS scores [17]. A study that Received massive blood transfusion N = 102 Did not receive massive blood transfusion N = 765 Critical Care Research and Practice 3 used the same population as our study was reported by Nunez et al. that reported an incidence of 12.5% which was similar with our study [6]. Compared with the scores that were previously published, our MBTT score had an AuROC of 0.845 which was close to the ABC score [6] that reported an AuROC of 0.842 and the TASH score [15] that had an AuROC of 0.892. e parameters in our study and in the ABC and TASH scores were also obtained rapidly after arrival. e AuROC in our study was lower than the modified Traumatic Bleeding Severity Score (TBSS) [13] which was 0.915 because the modified TBSS used more sophisticated parameters, such as type of pelvic fracture which may need a result from pelvic radiography, which take longer to obtain. e ABC score and the TASH score had similar AuROC values in the original research. However, when those scores    were validated in our study, the previously published scores had poor prediction. e explanation might be that the TASH score as well as the PWH score has hemoglobin as one of the predictors [15,17]. In our study, hemoglobin was not included and was not significantly different between the MBT and non-MBT groups (12.6 mg/dL vs. 13.7 mg/dL), thus the decreased performance of the PWH and TASH scores. It must be noted that, in patients with acute blood loss, hemoglobin might not drop initially, especially in patients who do not receive a prehospital intravenous fluid infusion which is the case in ailand.
In addition, the ABC score had a penetrating mechanism as one of the predictors [6]. In our sample, the predominant mechanism of injury was blunt force trauma. Penetrating mechanisms consist of GSW and stab wound. A stab wound causes less damage compared with a GSW. Our data revealed that 18% of patients who had GSW received MBT, while only 5% of patients who had stab wound received MBT.    Critical Care Research and Practice erefore, when the ABC score included these two mechanisms of injury into one parameter, it decreased the power of discrimination. e strength of this research lies in the fact that the data were collected from a trauma registry that were recorded prospectively. erefore, the data were complete with only a few missing values. Since the parameters in this scoring system are easy to collect, this score is easy to apply in clinical practice. e limitation of the MBTT score is the availability of the lactate and BE measurements. Even though these two parameters are point-of-care tests, they are not available in all hospitals. Also, this study also needs an external validation in the future.

Conclusions and Clinical Applications
Our MBTTscore consisted of four initial parameters that can be obtained immediately after arrival. With a MBTT score of <4, LH+ was 0.4 and the posttest probability was 0.04 which indicated that MBT would be rarely required. With a score of ≥4, LH+ was 6.72, in which case we would recommend preparing blood components.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Disclosure
is research was presented at the European Society of Intensive Care Medicine Conference 2019.

Conflicts of Interest
e authors declare that there are no conflicts of interest related to this research.

Authors' Contributions
Osaree Akaraborworn was involved in literature search, study design, data collection, data analysis, data interpretation, and writing. Boonying Siribumrungwong was involved in study design and critical revision. Burapat Sangthong and Komet ongkhao were involved in critical revision.