THE VALUE OF A RANDOM FOREST MACHINE LEARNING MODEL IN PREDICTING MORTALITY IN PATIENTS WITH POLYTRAUMA
Main Article Content
Abstract
Objectives: To evaluate the performance of a Random Forest model in predicting mortality in patients with polytrauma and to compare its performance with ISS and blood lactate level at admission. Methods: A retrospective, descriptive study was conducted on 196 patients with polytrauma treated at the Surgical Intensive Care Unit, Military Hospital 103, from June 2020 to June 2023. Clinical variables, laboratory parameters, and mortality outcomes were collected from medical records. The dataset was divided into training and test sets at a 60:40 ratio. Boruta was used for feature selection. Model performance was assessed using AUC, sensitivity, specificity, PPV, NPV, Brier score, and decision curve analysis. Results: There were 80 deaths, accounting for 40.8% of all patients. After feature selection, 8 variables were included in the model. On the test set, the Random Forest model achieved an AUC of 0.94, sensitivity of 90.62%, specificity of 95.74%, PPV of 93.55%, and NPV of 93.75%, which were higher than those of ISS and blood lactate alone (p < 0.01, DeLong test). Conclusion: The Random Forest model showed high performance in predicting mortality in patients with polytrauma and may help physicians in early risk stratification in clinical practice.
Keywords
Polytrauma, Random Forest model, Mortality prediction
Article Details
References
2. Ngô Đình Trung, Nguyễn Chí Tâm, Nguyễn Thái Cường và CS. Giá trị tiên lượng tử vong của một số yếu tố lâm sàng, cận lâm sàng trong đa chấn thương. Tạp chí Y Dược Lâm sàng 108. 2024;19(1):1-6.
3. Çınar E, Usul E, Demirtaş E, Gökçe A. The role of trauma scoring systems and serum lactate level in predicting prognosis in thoracic trauma. Turkish Journal of Trauma and Emergency Surgery. 2021;27(6):619-23.
4. He W, Fu X, Chen S. Advancing polytrauma care: developing and validating machine learning models for early mortality prediction. Journal of Translational Medicine. 2023;21(1):664.
5. Zhou P, Ling L, Xia X, et al. Independent predictors of mortality for critically ill patients with polytrauma: A single center, retrospective study. Heliyon. 2024;10(5):e25163.
6. Pape H-C, Lefering R, Butcher N, et al. The definition of polytrauma revisited: An international consensus process and proposal of the new ‘Berlin definition’. Journal of Trauma and Acute Care Surgery. 2014;77(5):780-6.
7. Kursa MB, Rudnicki WR. Feature selection with the Boruta package. Journal of Statistical Software. 2010;36:1-13.
8. Pape HC, Moore EE, McKinley T, Sauaia A. Pathophysiology in patients with polytrauma. Injury. 2022;53(7): 2400-12.
9. Islam MM. Development and validation of two prediction models for 72-hour mortality in high-risk trauma patients using a benchmark dataset: a comparative study of logistic regression and neural networks models. Cureus. 2023;15(6):e40773.
10. Özel A, İlbeği EN, Yüce S. Predictive value of initial lactate levels for mortality and morbidity in critically ill pediatric trauma patients: a retrospective study from a Turkish pediatric intensive care unit. Acute and Critical Care. 2025;40(1):87-94.