Enhancing Patient Stratification in Suspected Obstructive Sleep Apnea: Introducing a Novel Intelligent Decision Support System

<bold>Introduction:</bold> Traditional methods of obstructive sleep apnea (OSA) diagnosis are overwhelmed by increasing patient volumes, leading to delays in treatment. <bold>Objective:</bold> To develop an Intelligent Decision Support System (IDSS) using a minimal set of variables to streamline the management of patients suspected of having OSA. Methodology: We started with a database (DB) from the Álvaro Cunqueiro Hospital in Vigo, containing over 5,000 patient records and more than 40 variables, reserving 5% for testing. The DB was improved by recoding, rescaling, and removing nearly constant variables. Using Lasso regression, we then identified the most impactful variables based on two Apnea-Hypopnea Index thresholds (AHI≥15 and AHI≥30). Following this, a LightGBM algorithm was trained for each AHI threshold. <bold>Results:</bold> Preliminary tests indicate that the predictive performance is comparable to that achieved with the full DB, exhibiting AUC-ROC scores above 0.8. Furthermore, we applied SHapley Additive exPlanations to highlight the significant influence of factors such as BMI, age, neck circumference, reported apneas, and the Epworth score. <bold>Conclusions:</bold> The IDSS shows considerable promise in simplifying the management of potential OSA patients by focusing on a concise set of variables. Future studies are essential to confirm its clinical utility and expand its application.