Clinically-relevant tracheostomy prediction model in neonatal bronchopulmonary dysplasia via respiratory MRI

Background: There are few prognostic indicators of need for tracheostomy in preterm infants with chronic lung disease (bronchopulmonary dysplasia, BPD). Aims: To develop a clinically-useful tracheostomy prediction model in neonatal BPD using quantitative biomarkers from respiratory MRI. Methods: Infants with and without BPD (N=61) underwent 3D lung and airway MRI (0.7 mm3) near term-age. Lung disease scores (0-14 points) were used to create a binomial logistic regression model (⅔*N) to determine likelihood of tracheostomy (yes/no tracheostomy outcome assigned by 75% probability threshold), with validation in a separate cohort (⅓*N). A sub-cohort model also included MRI-quantified tracheomalacia severity (n=36). Results: The model correctly classified 95% of the validation cohort. The full-cohort model had 89% accuracy, 100% positive predictive value, and 85% negative predictive value. The lung+airway model values were 83%, 92%, and 78%, respectively. Conclusions: Quantitative respiratory MRI can predict need for tracheostomy in neonatal BPD with high sensitivity and accuracy, providing an objective tool for clinical decision-making.