Application of machine learning to predict and identify factors associated with the need for surgery in traumatic
Iran Chanideh1, Masoud Ghadiri1, Tahereh Mohammadi Majd1
1Clinical Research Development Center, Taleghani and Imam Ali Hospital, Kermanshah University of Medical Sciences, Kermanshah, Iran.
Background:
Timely identification of the need for surgical intervention in traumatic epidural hematoma (tEDH) is critical to optimizing outcomes. This retrospective study aimed to identify predictive factors for surgical intervention in tEDH using machine learning and develop a nomogram to support clinical decision-making.
Methods:
In this retrospective study, data from 147 patients with tEDH at a major trauma center in western Iran (2023-2024) were analyzed. Demographic, Clinical, and CT scan data were extracted from medical records. Four machine learning models (Logistic Regression (LR)/ Support Vector Machine (SVM)/ Naive Bayes (NB)/Neural Network (NN)), were developed to predict surgical need. A Random Forest (RF) algorithm identified key predictors, and a nomogram was constructed from the LR model to facilitate individualized risk assessment. Statistical analyses were conducted using R software (version 4.3.2).
Results:
In this study, 131 (89.1%) of 147 patients with tEDH were male. Of these, 72 (49%) underwent surgery. The cause of brain trauma was a Motor Vehicle Accident (MVA) in 76 (51.7%) of patients and a fall in 50 (34%) of patients. The mean (±Standard Deviation) age of the patients was 31.47 (±18.27). The initial hematoma volume demonstrated the highest discriminatory power, with an AUC of 0.92 (95% CI: 0.83-1.00) and an accuracy of 0.89 (95% CI: 0.76-0.96). The Glasgow Coma Scale (GCS) score also exhibited strong predictive performance, with an AUC of 0.76 (95% CI: 0.62-0.89) and an accuracy of 0.71 (95% CI: 0.56-0.84). The SVM model demonstrated the highest AUC of 0.96 (95% CI: 0.91-1.00), with sensitivity and specificity values above 90%.
Conclusions:
In this study, the novel integration of machine learning with a nomogram offers clinicians a precise, user-friendly tool for rapid decision-making, potentially reducing complications. These findings help surgeons to make more informed clinical decisions by accurately assessing these parameters in the early stages and to identify patients at higher risk for surgical intervention more quickly.
