Predicting tracheostomy after operative cervical spine trauma: development and internal validation of the...

Tracheostomy decisions after operative cervical spine trauma vary widely, and existing prediction tools were developed in heterogeneous populations that included nonoperative patients or rely on variables not consistently available early after injury. No validated score exists specifically for surgically managed cervical trauma patients. The authors sought to develop and internally validate a practical risk score to estimate tracheostomy risk after operative cervical spine trauma.

The authors performed a retrospective cohort study of adults (≥ 18 years) who underwent operative management for traumatic cervical spine injury at a single level I trauma center (2017-2021). The primary outcome was postoperative tracheostomy during the index hospitalization. Candidate predictors available at or shortly after admission were selected a priori based on clinical plausibility and prior literature, then evaluated using multivariable logistic regression with a parsimonious model-building strategy. Model discrimination, calibration, and accuracy were internally validated using repeated stratified 10-fold cross-validation (10 repeats, 100 total fits) and bootstrap validation with optimism correction. Regression coefficients were converted into an integer Tracheostomy Likelihood Score (TLS) for bedside risk stratification.

Of 515 patients (66 tracheostomies [12.8%]), 461 with assessable admission American Spinal Injury Association Impairment Scale (AIS) examinations comprised the primary modeling cohort (47 tracheostomies [10.2%]). The final 6-variable model included admission AIS grade, clinically significant traumatic brain injury, number of cervical fracture levels, thoracic injury, nonspine injuries requiring surgery, and intubation on admission. Admission AIS grade (OR 2.27 per grade, p < 0.001) and intubation on admission (OR 6.54, p < 0.001) were the strongest independent predictors. The model demonstrated excellent discrimination (apparent area under the receiver operating characteristic curve [AUC] 0.903 [95% CI 0.858-0.945]) with preserved performance after repeated cross-validation (AUC 0.883 [95% CI 0.830-0.933]) and minimal overfitting (bootstrap optimism 1.5%). The TLS preserved nearly all the model's performance (AUC 0.897 [95% CI 0.849-0.943]) and stratified observed tracheostomy risk from 1.7% (TLS 0-5) to 10.7% (TLS 6-10) and 61.2% (TLS > 10), representing an approximately 36-fold gradient. On sensitivity analyses reincorporating patients with untestable AIS examinations, discrimination remained excellent (AUC 0.883-0.891).

The 6-variable TLS accurately stratifies tracheostomy risk after operative cervical spine trauma using early clinical and imaging data, including in patients whose neurological status cannot be formally assessed. This simple bedside tool may support more consistent airway planning, facilitate multidisciplinary communication, and improve patient and family counseling regarding anticipated respiratory trajectories.