Summary
Objectives: To assess the relevance of various potential confounding factors (comorbidities,
obesity, body position, ECG lead, respiratory event type and sleep stage) on the detectability
of sleep-related breathing disorders from the ECG.
Methods: A set of 140 simultaneous recordings of polysomnograms and 8-channel Holter ECGs
taken from 121 patients with suspected sleep related breathing disorders is stratified
with respect to the named factors. Minute-by-minute apnea detection performance is
assessed using separate receiver operating characteristics curves for each of the
subgroups. The detection is based on parameters of heart rate, ECG amplitude and respiratory
myogram interference in the ECG. We consider spectral and correlation-based features.
Results: The results show that typical comorbidities and supine body position impede apnea
detection from the heart rate. Availability of multiple ECG-leads improves the robustness
of ECG amplitude based detection with respect to posture influence. But quite robust
apnea detection is achievable with even a single ECG channel – preferably lead I.
Sleep stages and respiratory event type have a significant and quite consistent effect
on apnea detection sensitivity with better results for light sleep stages, and worse
results for REM sleep. Mixed and obstructive events are better detected than central
apneas and hypopneas.
Conclusions: Various factors confound the detection of sleep apnea based on the ECG. These findings
should be taken into account when comparing results obtained from different data sets
and may help to understand limitations of current and to improve robustness of new
detection algorithms.
Keywords
Sleep-disordered breathing - confounding factors - screening - ECG - heart rate
