RETRAUCI is an observational, prospective, and multicentre nationwide registry that currently includes 52 ICUs in Spain. The RETRAUCI database only collects traumatic patients admitted to the ICU. It has the endorsement of the Neurointensive Care and Trauma Working Group of the Spanish Society of Intensive Care Medicine (SEMICYUC) and currently operates in a web-based electronic format . We include a five-year study period (2015–2019). Ethics Committee approval for the registry was obtained (Hospital Universitario 12 de Octubre, Madrid: 12/209). Due to the retrospective analysis of de-identified collected data, informed consent was not obtained. Hospital mortality was used as the outcome variable.
The variables collected were classified into several groups (Table 1).
First, we considered patient variables, such as Age and Sex. Variables were used that describe the importance of injuries by anatomical area according to the AIS model (2005 version) - severity levels ranging from 1(least severe) to 4- 6 (most severe) . The anatomical areas were head (AHEAD), neck (ANECK), face (AFACE), thorax (ATHORAX), abdomen (AABDOM), spine (ASPINE), upper extremity (AUPPEREXT), lower extremity (LOWEREXT) and external and thermal injuries (AEXTERNAL).
Also, we considered variables derived from the T-RTS, obtained from first medical attention before initiating resuscitation and/or mechanical ventilation, such as the Respiratory Rate (PointRF), Systolic Blood Pressure (PointSBP) and the Glasgow Coma Score (PointGCS), which range between 0 points (greater severity) and 4 points (normality) .
Next, patient treatment variables, such as the presence at the ICU of mechanical ventilation (MV) or the occurrence of a Massive Haemorrhage (MASSIVEHEM) requiring activation of the massive transfusion protocol, were also included .
Finally, variables that defined organic failures during the ICU stay: hemodynamic failure (HEMODINAM) indicated by the presentation of an SBP lower than 90 mmHg requiring the administration of volume, blood products, and vasoconstrictor support; respiratory failure (RESPIRATORY), indicated by the presence of PO2/FiO2 below 300; renal failure (AKIDNEY), indicated by an increase in creatinine > 1.5 times the initial, or 25% reduction in urine flow to less than 0.5 ml/kg/h for at least 6 h; and the presence of coagulopathy (COAGULOP), indicated by the prolongation of prothrombin and activated partial thromboplastin times in > 1.5 times the control or by levels of fibrinogen < 150 mg / dl or thrombocytopenia < 100,000 [13,15,16].
Variable distribution was tested with the Kolmogorov–Smirnov test. The variable AGE did not meet the criteria of normality (p > 0.05). Variables are described as median (interquartile range) or as a percentage. For the comparison of survivors (A-ALIVE) and non-survivors (D-DIED), the Mann-Whitney test was used for continuous variables, and the chi-square test or Fisher’s exact test for categorical variables. A p-value of < 0.05 was taken as significant.
Machine learning techniques
We used the WEKA Platform (version 3.8). We first use attribute selection methodology. Attribute selection is a technique used to extract the ranking of attributes and can help us by reducing the work of processing algorithms by discarding irrelevant variables. WEKA incorporates various attribute selection techniques. We use the Information Gain Attribute evaluation method. This method measures the significance of attribute by measure of information gain calculated with respect to target class and orders the variables according to their importance .
Second, we use WEKA's EXPLORER module to determine the optimal parameters for each algorithm used. The parameters chosen were those that achieved the best performance values (see Algorithm evaluation section). A ten-fold cross-validation process system was used in all algorithms .
And third, using WEKA's EXPERIMENTER module, run all algorithms 10 times, using repeated ten-fold cross-validation, to facilitate comparison of the predictive performance based on the different evaluation criteria that are available in WEKA 
Of the multiple algorithms included in WEKA, we selected nine supervised algorithms classified in traditional and ensemble methodology. The first six are traditional models based on logistic regression binary (LR) functions, a neural network according to multilayer perceptron (NN), sequential minimal optimization (SMO), classification rules (JRip), classification trees (CT) and Bayesian networks (BN), respectively. We also included three models that use ensemble classification algorithms: adaptive boosting (ADABOOST), bootstrap aggregating (BAGGING), and random forest (RFOREST) . With the WEKA EXPLORER module we select the optimal parameters of the different algorithms used.
For the LR model, we used a backward stepwise regression system with variable input with p < 0.05 and removal with p < 0.10. Odds ratios (OR) with a 95% confidence interval were calculated.
In the CT model, we used the J48 algorithm based on C4.5, obtaining a pruned tree . The JRip algorithm uses a rule learner: Repeated Incremental Pruning to Produce Error Reduction (RIPPER) . We limited tree growth (CT) and the number of rules (JRip), with a minimum of 20 instances.
For the BN, we used the TAN (Tree Augmented Network) variable relation search algorithm, which generates a graph that can be interpreted. This method does not assume the independence of the variables [21,22].
The SMO implements John Platt's sequential minimal optimization algorithm for training a support vector classifier . In NN, we used the automatic mode for selecting the number of nodes in the hidden layer, with a learning rate of 0.3 and a momentum of 0.2 . In RFOREST, we selected ten trees with the C4.5 algorithm . In the rest of the algorithms (ADABOOST and BAGGING), we used the parameters that WEKA incorporates by default [18,26].
To evaluate the performance of the algorithms, we used the calculation of accuracy, specificity, precision, recall, F-measure, and the area under curve ROC (AUC).
A patient who dies can be classified correctly (true positive-TP) or incorrectly (false negative-FN) and a patient who survives can be classified correctly (true negative-TN) or incorrectly (false positive-FP). We define the evaluation indices as:
Accuracy. The proportion of patients that are correctly labelled among the total number of patients. Accuracy=(TP+TN)/(TP+TN+FP+FN).
Specificity. The proportion of patients predicted as survivors and are correctly identified. Specificity=TN/(TN+FP).
Precision. The proportion of patients that are correctly predicted as dead among those labelled as dead. Precision=TP/(TP+FP)
Recall (Sensitivity). The proportion of dead patients that are correctly labelled. Recall=TP/(TP+FN)
F-measure. A measure that combine both Precision and Recall. F-measure= (2 x Precision x Recall)/ (Precision + Recall).”
WEKA's Experimenter module, with ten repetitions, allows one to establish whether there are statistical differences between the evaluated properties of the algorithms using the paired T-Test (corrected) .