Predictive model of tracheostomy in acute traumatic cervical spinal cord injury

English Keywords:Cervical spinal cord injury  Tracheostomy  Prediction model

English Abstract:

  【Abstract】 Objectives: To develop a risk prediction model for tracheostomy in acute traumatic cervical spinal cord injury (SCI) patients by using accessible data obtained from the bedside. Methods: Clinical and radiological data of 345 patients with acute traumatic cervical spinal cord injury were retrospectively reviewed. The information about patients′ demographics, trachestomy placement, pre-existing medical comorbidities (except for lung diseases), pre-existing pulmonary diseases, smoking history, presence of respiratory complications during treatment, admission ASIA motor score(AAMS), neurological level of impairment(NLI), ASIA grade, associated injuries, presence of cervical fracture and/or dislocations was gathered. The preoperative magnetic resonance imaging(MRI) was checked to determine the highest signal change(HSC) level in the spinal cord, lesion length, maximunm spinal cord compression(MSCC), maximum canal compromise(MCC) and the presence of intramedullary haemorrhage. The multiple logistic regression(MLR) analysis and classification and regression tree(CART) analysis were used to develop the prediction models. Cross-validation was used to conduct an external validation in order to assess the prediction performance of both models, using parameters of sensitivity, specificity, overall correction rate and area under receiver operating characteristic curve. Results: Among 345 patients, 58 patients underwent tracheostomy. The CART model suggested that the incidences of tracheostomy for patients with AAMS ≤1 and patients with AAMS ≤22 who suffered from respiratory complications during hospitalization and patients with AAMS ≥23, suffering from incomplete SCI, whose HSC was at C3 or lower was 66.7%, 69.0% and 0.8% respectively. The MLR model suggested that the independent risk factors included AAMS ≤22, ASIA grade A or B and respiratory complications during treatment. The comparison on sensitivity, specificity, overall correction rate and area under receiver operating characteristic curve between the CART model and the MLR model was 73.7% vs 81.8%, 89.7% vs 86.4%, 87.3% vs 85.7% and 0.909 vs 0.889 respectively. Conclusions: The CART model is preferred in prediction. AAMS ≤22, ASIA grade A, respiratory complications during hospitalization and HSC at C2 or higher are considered independent risk factors of tracheostomy for patients with traumatic cervical SCI patients