The objective of this study was to evaluate whether the addition of ROTEM data could improve the ability of the prediction model. We constructed two logistic regression models (Model 1: without ROTEM data, Model 2: with ROTEM data) and identified that the added ROTEM data did not significantly improve predictive accuracy. Although EXTEM CT and EXTEM ML were significantly associated with MT and the AUROC of Model 2 was higher than that of Model 1, there was very little difference in AUROC between the two models.
Accurately identifying the needs of MT and activating the MTP are essential in order to lower mortality in severely injured trauma patients with massive bleeding1-5. Previous studies have suggested various scoring systems or models for MT prediction. These models used diverse variables, including demographic data, clinical findings, hemodynamic status, injury mechanism, FAST results, and laboratory data13-15. Yücel et al. introduced a scoring system called TASH score. TASH score utilized seven variables such as laboratory and physiologic data, including SBP, hemoglobin level, FAST result, presence of long bone fracture, HR, base excess, and sex13. The ABC scoring system was developed in 2009 and consists of four dichotomous components: the mechanism of injury, FAST result, SBP, and HR. The ABC scoring system is broadly used because it has the advantage that components of the scoring system can be acquired early in the assessment phase but are as effective as the prior TASH scoring system. The AUROC of the ABC and TASH scores were 0.859 and 0.842, respectively. However, the difference between the two scores was not statistically significant14.
Two prediction models were constructed. The first model (Model 1) consisted of six variables selected without ROTEM variables by stepwise logistic regression analyses with backward elimination. SBP, HR, GCS score, hemoglobin concentration, platelet count, and PT were chosen. Each selected variable was statistically significant with a p-value lower than 0.05. The AUROC of the model using these 6 variables was 0.8542. Our model has a predictive value that is similar to that of the previously introduced scoring system.
For the second prediction model (Model 2), we added ROTEM data. Among the various ROTEM variables, EXTEM CT, MCF, and ML were selected. CT is the time from the start of the test until a clot firmness amplitude of 2 mm, and MCF is the maximum amplitude of clot firmness. CT and MCF are basic and crucial indicators for identifying the initiation of coagulation and the firmness of the clot, respectively. ML indicates maximum lysis during the run time and is expressed as a percentage of MCF19. Among the parameters representing coagulation activation, clot firmness, and clot lysis, we selected the most important variables one by one and added them to the model. MCF and ML were selected through a stepwise backward elimination process, and platelet counts were excluded. The AUROC of the second model, using seven variables, including ROTEM data, was 0.8603. The AUROC of Model 2 showed a higher performance, but it was not statistically significant.
In our study, it is unclear why additional ROTEM data did not improve predictive power. One possible explanation is that despite ROTEM’s unique strengths, the result is correlated with conventional coagulation tests30,31. Haas et al. showed ROTEM’s correlation with conventional coagulation tests in pediatric patients, and Schöchl et al. suggested that ROTEM-guided administration of fibrinogen concentrate and prothrombin complex concentrate was fast and effective32,33. We believe that adding the corresponding variable could not provide a major contribution to improve the prediction model.
Another explanation is that our study did not investigate the relationship between all the ROTEM parameters. In this study, three variables were thought to represent the crucial points of the coagulation cascade and were chosen to assess the relationship with the needs of MT. In the ROTEM test, multiple chambers with various parameters are suggested. Further studies are required to delineate the relationship between MT and other ROTEM parameters to improve the prediction model.
Our study had several limitations. First, because this study was retrospective, bias could exist. Second, this was a single-center trauma study. Further validation and multicenter studies are needed to generalize the clinical relevance of these findings. Another limitation was that we did not separately classify patients with traumatic brain injury (TBI). Patients with TBI have different characteristics and prognoses. Fourth, the physician decided whether to perform the ROTEM test based on the hemodynamic state, FAST result, and mechanism of injury rather than a protocol. Finally, we use only three ROTEM variables. We considered these variables to be the most important; however, it is possible that the result could be altered if additional ROTEM variables were added.
In conclusion, our study identified that ROTEM variables were independently related to MT and in-hospital mortality in patients with trauma. However, adding them to the prediction model did not significantly enhance its predictability.