Prognostic value of admission ROTEM in trauma: enhancing 30-day all-cause mortality prediction using machine learning

Haemorrhage is a leading cause of trauma death, yet early coagulation markers are rarely used to predict long-term outcomes. This study assessed whether a single admission rotational thromboelastometry (ROTEM) test could independently predict 30-day all-cause mortality and improve existing trauma scores.

We conducted a retrospective cohort study of 1,498 adult trauma patients admitted to a Level 1 trauma centre, with ROTEM (EXTEM, INTEM, FIBTEM) acquired on admission. Machine learning models were developed to predict 30-day mortality using ROTEM alone, using conventional trauma scores (RTS, NISS, GAP, MGAP, TRISS), and their combination. Model performance was assessed through cross validation using AUROC, AUPRC, and specificity at 90% sensitivity. SHAP was used for explainability.

ROTEM alone predicted 30-day mortality with an AUROC of 0.80, comparable to RTS and NISS (both 0.79), and superior to PT-INR (0.63) and base excess (0.58). When combined with ROTEM, specificity significantly improved across all trauma scores, with the greatest gains observed in RTS (0.23 to 0.62) and NISS (0.36 to 0.69) (all p < 0.001). Key predictive ROTEM variables included clotting time, clot firmness time, and fibrinolysis indices. Model performance was notably lower in female patients.

A single admission ROTEM test predicted 30-day all-cause mortality with accuracy comparable to traditional trauma scores and outperformed conventional coagulation markers. Integrating ROTEM into established scores significantly enhanced predictive performance. Viscoelastic data appears to hold prognostic information capable of improving long-term trauma outcome assessments.