Classification and regression tree model for predicting tracheostomy in patients with traumatic cervical spinal cord injury

Abstract

Purpose: In patients with cervical spinal cord injury (CSCI), respiratory compromise and the need for tracheostomy are common. The purpose of this study was to identify common risk factors for tracheostomy following traumatic CSCI and develop a decision tree for tracheostomy in traumatic CSCI patients without pulmonary function test. Methods: Data of 105 trauma patients with CSCI admitted in our institution from April, 2008 to February, 2014 were retrospectively analyzed. Patients who underwent tracheostomy were compared to those who did not. Stepwise logistic regression analysis and classification and regression tree model were used to predict the risk factors for tracheostomy. Results: Tracheostomy was performed in 20% of patients with traumatic CSCI on median hospital day 4. Patients who underwent tracheostomy tended to be more severely injured (higher Injury Severity Score, lower Glasgow Coma Score, and lower systolic blood pressure on admission) which required more frequent intubation in the emergency room (ER) with a higher rate of complete CSCI compared to those who did not. Upon multiple logistic analysis, Age ≥ 55 years (OR: 6.86, p = 0.037), Car accident (OR: 5.8, p = 0.049), injury above C5 (OR: 28.95, p = 0.009), ISS ≥ 16 (OR: 12.6, p = 0.004), intubation in the ER (OR: 23.87, p = 0.001), and complete CSCI (OR: 62.14, p < 0.001) were significant predictors for the need of tracheostomy after CSCI. These factors can predict whether a new patient needs future tracheostomy with 91.4% accuracy. Conclusions: Age ≥ 55 years, injury above C5, ISS ≥ 16, Car accident, intubation in the ER, and complete CSCI were independently associated with tracheostomy after CSCI. CART analysis may provide an intuitive decision tree for tracheostomy.