Machine Learning (ML) Models to Enhance the Berlin Questionnaire (BQ) Detection of Obstructive Sleep Apnea (OSA) at-Risk Patients

• DOI: 10.20944/preprints202406.1155.v1
• Publication Date: 2024-06-18T10:26:42Z
• Authors: Luana Conte; Giorgio De Nunzio; Francesco Giombi; Roberto Lupo; Caterina Arigliani; Federico Leone; Fabrizio Salamanca; Cosimo Petrelli; Paola Angelelli; Luigi De Benedetto; Michele Arigliani
• Keywords: Hypopnea; Area under curve; Clinical Practice

Objective: With just ten questions, the Berlin questionnaire (BQ) stands out as one of simplest and most widely implemented non-invasive screening tools for detecting subjects at high risk Obstructive Sleep Apnea (OSA), a still underdiagnosed syndrome characterized by partial or complete obstruction upper airways during sleep. The main aim this study was to enhance diagnostic accuracy BQ through Machine Learning (ML) techniques. Methods: A ML classifier (hereafter, ML-10) trained using questions standard BQ. simplified variant BQ, BQ-2, which comprises only two total ten, also assessed in context. 10-fold cross validation scheme used. Ground truth provided Apnea-Hypopnea Index (AHI) measured Home Testing. Model performance determined comparing ML-10 BQ-2 with Receiver Operating Characteristic Curve (ROC), Area Under (AUC), sensitivity, specificity, accuracy. Results: demonstrated superior predicting OSA compared capable classifying different AHI thresholds (AHI>=15, AHI>=30), typically used clinical practice. Remarkably, better sensibility assess moderate severe (AHI≥ 15) ML-10. Conclusions: underscores importance integrating techniques early detection, suggesting direction future research improve processes patient outcomes sleep medicine.