Screening of sleep apnea syndrome using a wearable biosensor
Background: Home sleep monitors have evolved to complement the expensive laboratory based polysomnography (PSG) for screening of moderate-to-severe sleep apnea syndrome (SAS). However, the success rate of home tests has been found to be low due to many limitations including low compliance, obtrusive sensor attachments, and complex procedures.
HealthPatch® is a disposable biosensor worn on the chest that remotely monitors ECG, acceleration, vital signs and actigraphy measures, and has been found to be very useful in numerous clinical applications. U.S. Food and Drug Administration has cleared HealthPatch clinical-grade patient monitor for in hospital and home use. The study presents the efficacy of wireless patch sensor for screening of moderate-to-severe SAS.
Methods: An overnight PSG study recruited 85 volunteers (age: 21−80 years, female/male: 41/44 and Apnea-hypopnea Index (AHI): 0.1−87.7) of healthy and SAS patients, and was conducted at California Sleep Institute, Palo Alto, CA, USA. Participants were instrumented with a 22-channel PSG and HealthPatch sensors at the recommended chest locations. Standard protocol for an attended overnight PSG was followed, and data were wirelessly acquired. Sleep physicians performed sleep scorings with the PSG data per American Academy of Sleep Medicine (AASM) guidelines and calculated the AHI value in each subject as the number of apnea and hypopnea events per hour of sleep that is used to indicate the severity of SAS. Features were computed based on the overnight signal analyses of ECG, heart rate variability, respiratory signals, posture and movements. Support vector machine (SVM) classifiers were trained on the feature set, and optimized to classify subjects into two groups as control-to-mild apnea vs. moderate-to-severe apnea using two independent binary classification scenario with different AHI thresholds: (i) AHI<15 vs. AHI≥15 and (ii) (AHI<20 vs. AHI≥20). The SVM classifiers were trained on 53 subjects, tested on 32 subjects, and performance measures were quantified with 95% confidence intervals.
Results: The performance of SAS screening algorithm in 32 test subjects revealed the specificity and sensitivity of the classification problem (AHI<15 vs. AHI≥15) as 80.0% (69.3%−87.8%) and 88.5% (79.0%−94.0%), respectively. The other classification problem (AHI<20 vs. AHI≥20) offered the specificity of 83.7% (73.4%−90.5%) and sensitivity of 81.8% (71.3%−89.1%). The accuracy of SAS screening was found to be 83.1% (72.7%−90.1%) in both classification problems. The values given in parentheses represent 95% confidence intervals.
Conclusion: The wireless HealthPatch biosensor has been demonstrated to be highly accurate for screening of clinically significant moderate-to-severe SAS compared to the gold standard PSG. HealthPatch is an inexpensive biosensor that can continuously monitor vital signs as well as sleep of patients unobtrusively for multiple days. While mass screening of SAS risk using traditional laboratory based PSG is impractical, the disposable HealthPatch biosensor enables widespread screening of SAS risk in home settings.