Enabling Early Obstructive Sleep Apnea Diagnosis With Machine Learning: Systematic Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Clinical prediction algorithms are recommended by the American Academy of Sleep Medicine for screening obstructive sleep apnea (OSA).
• These algorithms aim to assist in patient screening but do not replace polysomnography, the established gold standard for OSA diagnosis.

Purpose of the Study:

• To systematically identify, gather, and analyze existing machine learning (ML) approaches for disease screening in adult patients with suspected OSA.
• To evaluate the validity and performance of various ML techniques in OSA screening.

Main Methods:

• A comprehensive literature search was conducted across MEDLINE, Scopus, and ISI Web of Knowledge databases.
• Studies were evaluated for validity using the Prediction Model Risk of Bias Assessment Tool, with polysomnography as the gold standard outcome measure.
• Included studies were assessed for risk of bias and applicability.

Main Results:

• Out of 5479 retrieved articles, 63 were included in the analysis.
• Logistic regression was the most common ML technique (35 studies), followed by linear regression (16 studies).
• The highest reported area under the receiver operating curve (AUC) was 0.98, achieved using age, waist circumference, Epworth Somnolence Scale score, and oxygen saturation in a logistic regression model.

Conclusions:

• Machine learning models demonstrate potential for effective screening of obstructive sleep apnea.
• Current ML models for OSA screening require further external validation in large cohorts.
• Standardization of OSA criteria definition is necessary for improved reliability and generalizability of ML screening tools.