Table 1.
Summary of main studies included in the review (see Additional file 1: Table S2 for full study summaries)
| Authors | Year | Purpose | Methodology | Features | Dataset used | Dataset size | AUROC (or relevant performance metric) |
|---|---|---|---|---|---|---|---|
| Ahmed et al.[27] | 2020 | Mortality prediction model | DNN | Age, INR, PT, PTT, haemoglobin, hematocrit, WBC, platelets, creatinine, glucose, lactate | MIMIC III | 3041 | AUROC: 0.912 |
| Kilic et al. [28] | 2010 | Determining time period for calculation and evaluation of trauma severity and predicted mortality after a period of resuscitation | Fuzzy-logic inference system | SBP, GCS, changes after 1 h of resuscitation | Data from hospital/ER records | 150 | AUROC: 0.925 |
| Kuo et al. [29] | 2018 | Mortality prediction of motorcycle riders suffering traumatic injuries | SVM | Age, SBP, HR, RR, RBC, platelet, haemoglobin, hematocrit, GCS, AIS, ISS | Data from hospital/ER records | 946 | AUROC: 0.9532 |
| Maurer et al. [30] and El Hechi et al. [31] | 2021 | Trauma-outcome predictor (TOP) smartphone tool | TOP | Age, SBP, HR, RR, SpO2, Temperature, comorbidities, GCS, injury mechanism, AIS | ACS-TQIP | 934,053 | AUROC (penetrating trauma: 0.920, blunt trauma: 0.830) |
| Cardosi et al. [32] | 2021 | Predicting trauma patient mortality | XGBoost | Age, SpO2, PR, RR, Temperature, GCS, injury type | NTDB | 2,007,485 | AUROC (children data: 0.910, adult data: 0.890, all aged data: 0.900) |
| Lee et al. [33] | 2021 | Prognostic prediction for critical decision-making | XGBoost | Age, HR, RR, MAP, GCS, AIS | Data from hospital/ER records | 2232 | AUROC: 0.940 |
| Tran et al. [34] | 2021 | Mortality prediction model | XGBoost | Injury mechanism | NTDB | 1,611,063 | AUROC: 0.863 |
| Tsiklidis et al. [35] | 2020 | Outcome predictor for survival | Gradient Boost | Age, SBP, HR, RR, Temperature, SpO2, GCS | NTDB | 799,233 | AUROC: 0.924 |
| Becalick et al. [36] | 2001 | Assessing probability of survival after trauma | ANN | Age, RR, SBP, SpO2, HR, Injury type, AIS, ISS, GCS | UKTARN | 2042 | AUROC: 0.921 |
| Sefrioui et al. [37] | 2017 | Predicting patient survival using readily available variables | SVM | Age, injury type, BP, GCS, RR, | NTDB | 656,092 | AUROC: 0.931 |
| Batchinsky et al. [38] | 2009 | Predicting life-saving intervention based on EKG derived data | ANN | Heart rate complexity | USAISR Trauma | 262 | AUROC: 0.868 |
| Liu et al. [39] | 2017 | Predicting life-saving intervention | MLP | HR, SBP, DBP, MAP, RR, SpO2, SI, PR | WVSM trial | 79 | AUROC: 0.990 |
| Liu et al. [40] | 2018 | Predicting life-saving intervention | MLP | HR, SBP, DBP, MAP, RR, SpO2, SI, PR | WVSM trial | 104 | Correlation coefficient: 0.779 |
| Kim et al. [41, 42] | 2018, 2021 | Decision-making algorithm for remote triaging | DNN | Age, HR, SBP, SI, SCS | NTDB | 1,204,290 | AUROC: 0.890 |
| Scerbo et al. [43] | 2014 | ML model for triaging trauma patients | RF | Age, HR, SBP, DBP, SpO2, RR, GCS, injury type | Data from hospital/ER records | 1653 | Sensitivity: 0.890, Specificity: 0.420 |
| Nederpelt et al. [44] | 2021 | In-field triage tool for determining shock, MT, need for major surgery | Dirichlet DNN | Age, BMI, HR, SBP, RR, Temperature, GCS, injury location | ACS-TQIP | 29,816 | AUROC (shock: 0.890, MT: 0.860, need for major surgery: 0.820) |
| Follin et al. [45] | 2016 | Predicting need for specialized trauma care | DT | Age, HR, SpO2, SBP, GCS, ISS, injury mechanism | Data from anonymized prospective trauma registry | 1160 | AUROC: 0.820 |
| Mina et al. [46] and Hodgman et al. [47] | 2013, 2018 | Smartphone app for predicting Massive Transfusion cases | LASSO regression | Mechanism of injury, HR, SBP, BD, ISS, RBC, resuscitation intensity | Data from hospital/ER records. Validation data from PROMMTT database | 10,900/1245 | AUROC (training: 0.956, validation: 0.711) |
| Feng et al. [48] | 2021 | Demand prediction for traumatic blood transfusion | XGBoost | Trauma location, Age, HR, RR, SI, SBP, DBP, SpO2, Temperature | Data from hospital/ER records | 1371 | AUROC: 0.940 |
| Lammers et al. [49] | 2022 | Predicting risk of requiring massive Transfusion | RF | HR, RR, DBP, SBP, SpO2, Temperature, INR, Hematocrit, Platelet, pH, mechanism of injury, GCS, AIS, ISS | DoDTR | 22,158 | AUROC: 0.984 |
| Chen et al. [50] | 2008 | Determining hypovolemia in patients | Linear ensemble classifiers | HR, RR, DBP, SBP, SpO2 | Data from hospital/ER records | 898 | Accuracy: 0.760 |
| Convertino et al. [51] | 2011 | Determining patients at greatest risk of ongoing hemorrhagic shock | undefined ML algorithm | SBP, DBP, RR, blood pH, base deficit | Data from subjects under LBNP | 190 | Accuracy: 0.965 |
| Rickards et al. [52] | 2015 | Determining hypovolemia in patients | undefined ML algorithm | HR, stroke volume, ECG, heat flux, skin temperature | Data from subjects through various exercises under LBNP | 24 | Accuracy: 0.926 |
| Davis et al. [53] | 2022 | Intracranial hemorrhage detection | NLP tool | CT scan images | Data from hospital/ER records | 200 scans (25,658 images) | Precision: 0.730 |
| Ginat et al. [54, 55] | 2020, 2021 | Intracranial hemorrhage detection | NN software | CT scan images | Data from hospital/ER records | 8723 scans | Accuracy: 0.965 |
| Davuluri et al. [56] | 2012 | Hemorrhage detection and image segmentation model | SVM | CT scan images | Data collected from hospital/ER records | 12 scans (515 images) | Accuracy: 0.943 |
| Perkins et al. [57] | 2021 | Prediction tool for detecting TIC | BN | HR, SBP, temperature, hemothorax, FAST scan, GCS, lactate, pH, mechanism of injury, fracture assessment | Data from hospital/ER records | 1091 | AUROC: 0.930 |
| Li et al. [58] | 2020 | Prediction model for acute traumatic coagulopathy | RF | RBC count, SI, base excess, lactate, DBP, pH | Emergency Rescue Database | 1014 | AUROC: 0.830 |
AIS abbreviated injury scale, ANN artificial neural network, BMI body mass index, BN bayesian network, DBP diastolic blood pressure, DNN deep neural network, DT decision tree, ECG electrocardiography signal, FAST Focused Assessment with Sonography for Trauma, GCS Glasgow Coma Score, HR heart rate, INR international normalized ratio, ISS injury severity score, LASSO least absolute shrinkage and selection operator, LBNP low body negative pressure, MAP mean arterial pressure, MLP multi-layer perceptron, NLP natural language processing, PR pulse rate, PT prothrombin time, PTT partial thromboplastin time, RBC red blood cell, RF random forest, RR respiratory rate, SBP systolic blood pressure, SCS Simplified Consciousness Score, SI shock index, SpO2 oxygen saturation, SVM support vector machine, TOP trauma outcome predictor, WBC white blood cell