Recognition of Obstructive Sleep Apnea: An Exploratory Bayesian Modeling Analysis

Area of Science:

• Sleep Medicine
• Medical Informatics

Background:

• Two main diagnostic methods for sleep studies exist: in-laboratory polysomnography (PSG) and home sleep apnea testing (HSAT).
• While HSAT is convenient and cost-effective, criteria for its use are complex and not solely based on apnea-hypopnea index (AHI) or oxygen desaturation index (ODI).

Purpose of the Study:

• To investigate associations between common demographic, clinical, and symptom data and objective sleep apnea severity indices (AHI and ODI).
• To generate hypotheses for future patient-level model development for diagnostic approach selection.

Main Methods:

• Retrospective analysis of 1100 individuals who underwent PSG.
• Included demographic, clinical, and symptom variables analyzed against AHI and ODI.
• Used variable selection, backward elimination, and Bayesian multivariable linear regression with Hamiltonian Monte Carlo methods.

Main Results:

• Male gender, BMI, Epworth Sleepiness Scale (ESS) score, reported sleep breathing interruptions, and COPD predicted AHI.
• Male gender, BMI, ESS score, sleep breathing interruptions, daytime sleepiness, obesity, and COPD predicted ODI.
• COPD showed an inverse association with both AHI and ODI.

Conclusions:

• Routinely available clinical data can be integrated into a Bayesian framework to estimate pre-test probability of sleep apnea severity.
• This exploratory study provides a hypothesis-generating foundation for future validated tools to guide HSAT versus PSG selection.