R027Clinical decision support for severe trauma patients : Machine Learning based definition of a bundle of care for Hemorrhagic Shock and Traumatic Brain Injury

Traumatologie / Urgences
E. Lang 1, A. Neuschwander 1, G. Favé 1, P. Esnault 2, T. Gauss 3, T. Geeraerts 4, A. Harrois 5, J.L. Hanouz 6, E. Kipnis 7, M. Leone 8, V. Legros 9, N. Mellati 10, J. Pottecher 11, S. Hamada 1, R. Pirracchio 12.
1Hôpital Européen Georges Pompidou - Paris (France), 2Hia Sainte Anne - Toulon (France), 3Hôpital Beaujon - Clichy (France), 4Chu De Toulouse - Toulouse (France), 5Chu De Bicêtre - Le Kremlin-Bicêtre (France), 6Chu De Caen - Caen (France), 7Chu De Lille - Lille (France), 8Hopital Nord - Marseille (France), 9Chu De Reims - Reims (France), 10Chr Metz Thionville - Metz (France), 11Chu De Strasbourg - Strasbourg (France), 12Zuckerberg San Francisco General Hospital And Trauma Center - San Francisco (États-Unis)

Conflits d'intérêt

No conflict of interest. 

Position du problème et objectif(s) de l’étude

Deviation from guidelines is frequent in emergency situations and may lead to increased mortality. In trauma patients, literature suggests that the maximal compliance rate to current guidelines is 55%. However, due to time constraints, this may be the maximum achievable rate when managing severe trauma patients. This study aimed to identify among all available guidelines, a minimum bundle of recommendations that should be followed to optimize the outcome of hemorrhagic shocks (HS) and severe traumatic brain injuries (TBI).​

Matériel et méthodes

A retrospective multicenter study of prospectively collected data from the TraumaBase registry (Traumabase®) was performed. We first estimated the compliance with the most recent French and European guidelines. Then, we used a fully automated procedure based on LASSO penalized regression to reduce the number of recommendations to a minimal set of items (bundle) to be followed in the early phase of a trauma to minimize 7-day mortality (training cohort). Finally, we evaluated the proposed bundles using an external validation cohort.​

Résultats & Discussion

This study included 5,924 trauma patients (1,414 HS and 4,955 TBI). Overall compliance rate to the 36 evaluable recommendation items for HS was 71.6% and of 66.9% for TBI. In HS patients, compliance to recommendations was significantly associated with 7-day mortality in univariate analysis (OR= 0.88, 95% confidence interval (CI)= 0.79-0.98, p=0.02) but not in multivariate  analysis  (OR =0.90,  95%CI=0.70-1.02,  p=0.1). In patients with TBI, compliance was significantly associated with mortality in univariate and multivariate analysis (OR=0.83, 95%CI=0.74-0.92, p=0.01). For HS, the bundle identified using LASSO included 13 recommendation items. In the validation cohort, patients following the bundle identified by the LASSO procedure had a lower 7-day mortality rate in multivariate analysis (OR=0.41, 95%CI=0.26-0.78, p=0.01).  In TBI, the bundle identified using LASSO included 7 recommendation items. In the validation cohort, the patients following the bundle identified by the LASSO procedure also had a lower 7-day mortality rate in multivariate analysis (OR = 0.39, 95%CI = 0.20-0.75, p=0.01).​


Using a machine learning procedure, we were able to identify a subset of recommendations that minimizes 7-day mortality following traumatic hemorrhagic shock and traumatic brain injury. These two bundles of care will have to be evaluated in a prospective manner.​


1. Crit Care. 2015, 19:423-431. 

2. Br J Anaesth. 2016, 117(5):592‑600.

3. Crit Care Med. 2012, 40(3), 778‑86.


Traumabase® group

Fig. 1