(0839) Using Machine Learning Modules to Predict Outcomes for Partients With Laryngotracheal Stenosis

Introduction: Laryngotracheal stenosis (LTS) has many etiologies, including trauma, surgery (e.g. tracheotomy), autoimmune disease, or idiopathic. Machine learning models can be used to uncover patterns in large datasets and make predictions. We describe the use of machine learning models to predict treatment outcomes of patients undergoing surgery for LTS.

Methods: A retrospective review of medical records of patients who presented with airway stenosis between 2017 and 2022 was conducted. Demographics, co-morbidities, characteristics of stenosis, and treatment modalities were collected. Logistic regression, random forest and support vector machine models were developed for analysis and used to predict treatment outcomes.

Results: A total of 204 patients were included. Median age was 56 (range 17-89), 71% were female, and 70% were Caucasian. Six patients had supraglottic stenosis, 59 had glottic, 126 had subglottic, and 91 had tracheal stenosis. Three were congenital, 83 were idiopathic, 136 were acquired, and 5 were rheumatologic in etiology. Logistic regression models found significant predictors of open surgery included male sex, longer duration of stenosis and comorbidities (asthma, diabetes). Predictors of endoscopic surgery included idiopathic or congenital stenoses. Significant predictors of a new tracheotomy included history of prior tracheotomy, non-Caucasian ethnicity, and male sex.

Conclusions: Machine learning models can be used to predict outcomes for patients with LTS. Further research is warranted on a larger population of patients to improve model accuracy. Ultimately, these models could be used to predict patients at high risk for treatment failure or recurrence, to improve follow-up and outcomes.