[Regression analysis of transfusion requirements of severely injured patients in clinical emergency care-A feasibility study]
Related Experiment Videos
A Luckscheiter1,2, U Hoppe3, P A Grützner4
1Klinik für Anästhesie, Intensiv- und Schmerzmedizin/OP-Abteilung, BG Klinik Ludwigshafen, Ludwig-Guttmann-Str. 13, 67071, Ludwigshafen, Deutschland. andre.luckscheiter@medma.uni-heidelberg.de.
Background:
The acute management of severely injured patients with hemorrhagic shock and coagulopathy often requires the use of packed red blood cells (PRBC) and fresh frozen plasma (FFP). Using a single center dataset of the TraumaRegister DGU® of the German Trauma Society (Deutsche Gesellschaft für Unfallchirurgie, DGU) , this study aims to retrospectively estimate the required amounts of PRBCs and FFP from initial emergency room management to the end of any potential emergency surgery.
Methods:
A total of 721 primary treated patients from 2020-2023 at a level I trauma center were analyzed. A random forest model was used to regressively estimate the number of required PRBC and FFP units. Attributes included vital signs, laboratory findings, interventions and surgical indications from the prehospital care and emergency room phases.
Results:
In this study PRBCs were administered in 13.2% of cases (median 4, min-max 0-27 units PRBC), and FFP in 10.7% (median 4, min-max 0-28 units FFP). In the PRBC model the correlation coefficient (CC) was 0.70 (95% confidence interval, CI 0.70-0.71), with an R2 of 0.50 (0.49-0.50). The mean absolute error (MAE) and root mean squared error (RMSE) were 0.63 (0.63-0.63) and 1.81 units PRBC (1.81-1.82), respectively. The false negative rate was 0.03 (0.02-0.03), corresponding to a specificity of 0.82 (0.81-0.82) and an underprediction of 9.8%. For the FFP model, the CC was 0.74 (0.73-0.75), R2 was 0.55 (0.54-0.56), MAE was 0.58 (0.58-0.58) and RMSE was 1.71 units FFP (1.70-1.71). The false negative rate was 0.01 (0.00-0.01), resulting in a specificity of 0.82 (0.82-0.82) and an underprediction of 8.8%.
Conclusion:
The standard dataset of the TraumaRegister DGU® appears to enable regression-based estimation of the PRBC and FFP units required for severely injured patients in the time window of transfer from the trauma room to the operating room, showing a clinically acceptable model fit for the first time. To improve individual prediction accuracy, future models should be developed and validated using a multicenter cohort and the dataset should be expanded to include real-time parameters as well as attributes reflecting individual transfusion triggers.