# Development and validation of a nomogram for predicting tracheostomy risk in traumatic cervical spinal cord injury

## TL;DR

This study creates a tool to predict tracheostomy risk in cervical spinal cord injury patients using five factors, showing strong accuracy in a hospital setting.

## Contribution

A new nomogram with five admission predictors is developed and validated for tracheostomy risk in traumatic cervical spinal cord injury.

## Key findings

- The nomogram achieved strong discrimination with an AUC of 0.844 in training and 0.903 in validation.
- At a 0.30 threshold, the model showed high sensitivity and specificity in both training and validation sets.
- Decision curve analysis confirmed the model's net benefit over treating all or none patients.

## Abstract

Tracheostomy is common in traumatic cervical spinal cord injury (TCSCI) because of respiratory complications, yet objective tools to estimate individual risk remain limited.

In this single-center retrospective cohort at the Second Affiliated Hospital, Zhejiang University School of Medicine, we enrolled 308 consecutive ICU admissions with TCSCI (January 2018–March 2023) and randomly split the cohort 7:3 (outcome-stratified) into training (n = 215) and validation (n = 93) sets. Candidate admission predictors were screened with Least Absolute Shrinkage and Selection Operator and then entered into multivariable logistic regression to construct a nomogram. Model performance included discrimination (AUC with bootstrap 95% CIs, 2,000 resamples), calibration (intercept, slope, Brier), and decision curve analysis (DCA). A prespecified clinical threshold of 0.30 was used to summarize sensitivity and specificity.

Five independent predictors were retained—smoking history, thoracic injury, BMI ≥ 25 kg/m2, cervical dislocation, and ASIA grade (A vs. B-D). The model showed strong discrimination (AUC 0.844, 95% CI 0.788–0.896 in training; 0.903, 95% CI 0.823–0.966 in validation) and good calibration. At the 0.30 threshold, performance was Sensitivity 0.781/Specificity 0.725 (training) and Sensitivity 0.812/Specificity 0.852 (validation); DCA demonstrated greater net benefit than “treat all/none” across threshold 0.10–0.70.

A parsimonious, five-factor nomogram based on routine admission data provides accurate, clinically interpretable stratification of tracheostomy risk in TCSCI. Clear reporting of ASIA coding and a prespecified decision threshold enhance bedside usability. Prospective, multi-center external validation is warranted.