According to a study published by the American Journal of Respiratory and Critical Care Medicine, an automated algorithm to analyze nocturnal blood oxygen levels of children at risk for obstructive sleep apnea (OSA) may increase diagnosis. Areas of the world that lack the resources to do overnight sleep studies would particularly benefit from the algorithm, researchers say.

The study reported on an automated neural network algorithm developed by researchers to detect OSA based on data from a pulse oximeter. They found that the algorithm’s ability to detect obstructive sleep apnea in children compared favorably to polysomnography, the diagnostic gold standard.

“Both in the U.S. and around the world, the vast majority of children with OSA are not diagnosed because of the scarcity of pediatric sleep laboratories and the costs associated with such tests,” said David Gozal, MD, MBA, of the University of Chicago and immediate past- president of the American Thoracic Society. “At the same time, many children with sleep apnea symptoms undergo surgical removal of tonsils and adenoids–the current standard of care treatment for pediatric OSA–even if they would not fulfill the criteria for OSA if they had been tested.”

The study authors wanted to address both issues—under diagnosis of OSA and unnecessary surgery—by developing a simple, accurate and inexpensive screening tool that could be used for all children who snore.

Previous studies using nocturnal oximetry as a screening tool for OSA found that the test had high specificity (it produced few false positives), but limited sensitivity (there were too many false negatives). The goal of the current study was to produce results with both high sensitivity and high specificity from 23 features of the oximetry recordings using an automated neural network algorithm. Neural networks are computing systems that can recognize patterns and reach conclusions much the way the human brain does.

Gozal and his colleagues compared data from both nocturnal oximetry studies and polysomnography performed on 4,191 children, ages 2 to 18, who were referred to 13 pediatric sleep laboratories around the world because they habitually snored or had other signs of OSA. The polysomnography study was considered definitive, and the researchers compared the apnea-hypopnea index (AHI), which measures the number of stopped or shallow breaths per hour, from those studies to the AHI obtained from the automated neural network algorithm.

The study found that the algorithm’s diagnostic ability increased with OSA severity. The algorithm’s diagnostic accuracy at:

AHI = 1 was 75.2%
AHI = 5 was 81.7% and
AHI = 10 was 90.2%.

The authors noted that the increased reliability of their automated methodology corresponded to the “more widely used and clinically relevant cutoff criteria of OSA,” an AHI ? 5, which would allow clinicians using the algorithm to “confidently both confirm and discard cases that would or would not fulfill OSA criteria” in most cases