JMIR Publications

ABSTRACT

Background:

American Academy of Sleep Medicine guidelines suggests that clinical prediction algorithms can be used to screen obstructive sleep apnea (OSA) patients without replacing polysomnography (PSG) – the gold standard.

Objective:

We aimed to identify, gather, and analyze existing machine learning approaches that are being used for disease screening in adult patients suspected of OSA.

Methods:

We searched MEDLINE, Scopus and ISI Web of Knowledge databases for evaluating the validity of different machine learning techniques, with PSG as the gold standard outcome measures. This systematic review was registered in PROSPERO under reference CRD42021221339.

Results:

Our search retrieved 5479 articles, of which 63 articles were included. We found 23 studies performing diagnostic models’ development alone, 26 with added internal validation, and 14 applying the clinical prediction algorithm to an independent sample (although not all reporting the most common discrimination metrics - sensitivity and/or specificity). Logistic regression was applied in 35 studies, linear regression in 16, support vector machine in 9, neural networks in 8, decision trees in 6, and Bayesian networks in 4. Random forest, discriminant analysis, classification and regression tree, and nomogram were each performed in 2 studies, while Pearson correlation, adaptative neuro-fuzzy inference system, artificial immune recognition system, genetic algorithm, supersparse linear integer models, and k-nearest neighbors’ algorithm each in 1 study. The best AUC was .98 [.96-.99] for age, waist circumference, Epworth somnolence, and oxygen saturation as predictors in a logistic regression.

Conclusions:

Although high values were obtained, they still lack external validation results in large cohorts and a standard OSA criteria definition.