Validating a digital stethoscope for pediatric breath sounds, with Jennifer Nestor, MD

A recent study led by Jennifer Nestor, MD, fellow physician at Nemours Alfred I. duPont Hospital for Children, evaluated the effectiveness of digital stethoscope technology in recording pediatric breath sounds for clinical and educational use. The poster, titled “Validation of a Digital Stethoscope for Recording and Identification of Pediatric Breath Sounds to Aid in Development of a Machine Learning Algorithm,” was presented at the 2025 Pediatric Academic Societies meeting in Honolulu, Hawaii.

The project emerged from Nestor’s experience during her fellowship, which coincided with a national albuterol shortage amid a surge in pediatric respiratory illnesses. “We were working with a week’s worth of albuterol nebulized therapy for our patients, and trying to sort of be really judicious about where we were using the therapy,” Nestor said.

Faced with limited resources and high patient volume, Nestor and her colleagues saw an opportunity to explore how digital auscultation could optimize bedside decision-making. The prospective observational study recorded breath sounds from children younger than 18 years using EkoCore digital stethoscope technology. Recordings were captured from up to 10 positions and classified in real time by the bedside clinician. These same recordings were later assessed independently by five expert listeners, blinded to original classifications.

“There is no gold standard [for pulmonary auscultation],” Nestor explained. “The gold standard is the trained professional who’s listening to the patient at the bedside and hearing specific things.”

The study’s primary aim was to assess interrater and intrarater reliability using Cohen’s kappa and Fleiss' kappa to evaluate agreement between bedside and remote listeners. “My kappa values all fell within the fair to moderate agreement range,” Nestor said, citing background noise and extraneous patient movement as factors impacting classification accuracy.

Despite these challenges, the technology showed promise. Interrater agreement between the bedside recorder and majority classification of all listeners was moderate (k = 0.43), while all individual comparisons showed fair agreement.

“No difference in agreement based on age, gender, respiratory support or nebulizer use” was observed, and even conditions like bronchiolitis and respiratory failure showed only slight agreement, according to the study.

“My hope… is that we work on the technology for cleaning up some of the breath sounds and see if that can push us from that fair to moderate agreement into more of the substantial to almost perfect agreement classification,” Nestor said. Nearly all study participants consented to breath sound storage, which may contribute to the creation of the first pediatric database for machine learning development in pulmonary auscultation. “Ultimately… depending on what we’re able to do with the technology, [this] may lend itself to the development of that algorithm,” Nestor added.

Reference:

Nestor J, Slamon N, McMahon K, et al. Validation of a Digital Stethoscope for Recording and Identification of Pediatric Breath Sounds to Aid in Development of a Machine Learning Algorithm​. Poster. Presented at: Pediatric Academic Societies 2025 Meeting; April 24-28, 2025. Honolulu, Hawaii.