Recognition of Obstructive Sleep Apnea: An Exploratory Bayesian Modeling Analysis

Background/Objectives: Two diagnostic approaches for sleep studies are commonly used worldwide: in-laboratory polysomnography [PSG] and home sleep apnea testing [HSAT]. Although HSAT has gained increasing acceptance due to its convenience and lower cost, clinical criteria for HSAT use remain complex and cannot be inferred directly from AHI/ODI severity indices alone. The aim of the present exploratory study was to examine associations between routinely collected demographic, clinical, and symptom-related variables and objective indices of disease severity, namely the apnea–hypopnea index [AHI] and oxygen desaturation index [ODI] as an initial, hypothesis-generating step toward future patient-level model development and validation. Methods: A retrospective observational analysis was conducted in 1100 individuals who previously underwent in lab-polysomnography [PSG] at the University Hospital of Thessaly, Greece, between 2006 and 2023. Specific demographic, clinical and symptom-related variables were included in this study [six continuous and fifteen categorical], which were analyzed in relation to AHI and ODI values. A three-step process was carried out: variable selection followed a screening and backward elimination process. Multivariable linear regression models were subsequently estimated within a Bayesian framework using Hamiltonian Monte Carlo methods. Results: Out of 1100 individuals, the mean age was 51.9 years with the predominant gender being male [76%]. Obesity [65.6%] and hypertension [40.5%] were the most common comorbidities. For AHI, male gender, body mass index [BMI], Epworth Sleepiness Scale [ESS] score, reported breathing interruptions during sleep, and chronic obstructive pulmonary disease [COPD] were significant predictors. For ODI, significant predictors included male gender, BMI, ESS score, breathing interruptions during sleep, daytime sleepiness, obesity, and COPD. COPD showed an inverse association with both indices. Conclusions: These findings support the feasibility of integrating routinely available clinical variables within a Bayesian probabilistic framework to estimate disease severity pre-test probability. The current analysis may not constitute a validated tool for HSAT versus PSG selection; however, it is an initial, hypothesis-generating step toward future model development.