Can machine learning detect sleep-disordered breathing events as well as experts?
Sleep-disordered breathing is a condition where breathing is repeatedly interrupted during sleep. Detecting these interruptions is traditionally done by experts reviewing overnight recordings. Recent research shows that computer algorithms can match human experts in identifying these events.
What the research says
A study using gradient-boosted decision tree models found that machine learning achieved an F1-score of 0.818 for detecting respiratory events like apnea and hypopnea. This performance was comparable to the agreement seen between different human experts scoring the same data 3. The models also correctly classified sleep stages with an accuracy of 0.840 3.
Other research focused on specific patient groups. In patients with hypertrophic cardiomyopathy, a machine learning model identified sleep-disordered breathing with a sensitivity of 0.91 and a specificity of 0.68 4. Another study using cardiac and respiratory signals achieved a Cohen's kappa of 0.70 for sleep-wake classification and estimated sleep time with a strong correlation to expert scoring 5.
Even in critical care settings, machine learning algorithms processed respiratory and oxygen signals to measure the apnea-hypopnea index in ICU patients 6. While predicting risk factors was difficult, the models successfully processed continuous signals to estimate breathing patterns and hypoxic burden 6.
What to ask your doctor
- How accurate are the machine learning tools used to score my sleep study?
- Does the computer algorithm used in my test match the accuracy of a human expert?
- What specific metrics, like the apnea-hypopnea index, were used to determine my diagnosis?
- Are there any limitations in the machine learning model that might affect my specific results?
This question is drawn from common patient questions about Pulmonology & Critical Care and answered using cited medical research. We do not provide individualized advice.