Neural Network Analysis of Factors Differentiating Recurrent Croup from Asthma in Children

Sunday, October 12, 2025

10:06 AM - 10:12 AM EDT

Location: 233-234

Disclosure(s):

Benjamin A. Aunins, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Introduction: Recurrent croup of childhood is an enigmatic, poorly-understood disorder that accounts for a significant amount of unscheduled health care. We examined structural, biological, pathogen-related, racial/ethnic, environmental, and socio-cultural variables differentiating recurrent croup from asthma in children examined by an aerodigestive (AD) team.

Methods:

Methods: Children underwent direct laryngoscopy, bronchoscopy with bronchoalveolar lavage (BAL), and esophagoduodenoscopy (EGD) under anesthesia. BAL was evaluated for pathogens and granulocytes, and blood for total eosinophils, IgE, and allergen-specific IgE. The sample was stratified post-exam into children with recurrent croup (n=180) and asthma (n=243). Monovariate tests identified independent variables associated with recurrent croup for inclusion into multivariable binary and deep learning neural models.

Results: Factors which differentiated recurrent croup from asthma in the binary model were laryngeal anomalies, school-age, male sex, and enrollment post-COVID quarantine. Net classification accuracy was 73%, and the area under the receiver operating characteristic (ROC) curve was 0.80. The neural network performed better, the testing data set had comparable classification accuracy of 72%, but a greater ROC area of 0.83. In contrast to the binary model, the neural network ranked lower BAL eosinophils and older symptom onset as important predictors of recurrent croup.

Conclusions: In children examined by an AD team, factors distinguishing recurrent croup from asthma included an assortment of laryngeal anomalies, lower BAL eosinophils, and older age. Whereas the presenting symptoms of recurrent croup and asthma overlap, an AD evaluation facilitated precision management of both disorders by the discovery of treatable patterns of immune dysregulation and asymptomatic pathogens.

Presenting Author(s)

Benjamin A. Aunins, MD

Resident
University of Virginia
Charlottesville, VA, United States

Disclosure(s):
faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Co-Author(s)

WB

William Brand, MD

University of Virginia Health, United States

Disclosure information not submitted.
WT

William Gerald G. Teague, MD

Professor of Pediatrics
University of Virginia Health
Charlottesville, Virginia, United States

Disclosure information not submitted.
DJ

Denise L. Jackson, PA-C, MA, CCC-SLP

Physician Assistant
University of Virginia Health
Charlottesville, Virginia, United States

Disclosure information not submitted.
CH

Cheryl Hubbard, RN

University of Virginia Health, United States

Disclosure information not submitted.
JM

Jeremy P. Middleton, MD

University of Virginia Health
Charlottesville, Virginia, United States

Disclosure information not submitted.
CM

Craig McKinney, MD

University of Virginia Health, United States

Disclosure information not submitted.
BW

Brittany Wall, MD

University of Virginia Health, United States

Disclosure information not submitted.

Senior Author(s)

Ariana G. Greenwell, MD

Assistant Professor
University of Virginia Department of Otolaryngology - Head and Neck Surgery
Charlottesville, VA, United States

Disclosure information not submitted.