Statistical Analysis of Distribution-based Spectral Features of Low-Sampled Snoring Vibrations in Predicting Treatment Outcomes in Obstructive Sleep Apnea

Abstract

This study delves into the statistical significance of spectral characteristics of snoring vibrations in the prediction of the efficacy of oral appliance devices in the treatment of obstructive sleep apnea. By analyzing data from 20 participants who underwent at-home sleep apnea testing both before and after a 5-month utilization of mandibular advancement devices, we established that specific distribution-based descriptive spectral features can predict the efficacy of oral appliances. Our analysis revealed that among 20 highly correlated features from an initial set of 192 features, only two features are significantly different between the group of responders and non-responders. Using these two features and a linear regression model, a predictive accuracy of 75%, coupled with a sensitivity of 67% and specificity of 82% was achieved. Our findings are also aligned with previous clinical outcomes on the snoring sounds, which share a lot of similarities with the snoring vibration signals.