Computer-Assisted Assessment of the Interaction Between Arousals, Breath-by-Breath Ventilation, and Chemical Drive During Cheyne-Stokes Respiration in Heart Failure Patients

Abstract

Transient increases in ventilation induced by arousal from sleep during Cheyne-Stokes respiration in heart failure patients are thought to contribute to sustaining and exacerbating the ventilatory oscillation. The only possibility to investigate the validity of this notion is to use observational data. This entails some significant challenges: (i) accurate identification of both arousal onset and offset; (ii) detection of short arousals (<3 s); (iii) breath-by-breath analysis of the interaction between arousals and ventilation; (iv) careful control for important confounding factors. In this paper we report how we have tackled these challenges by developing innovative computer-assisted methodologies. The identification of arousal onset and offset is performed by a hybrid approach that integrates visual scoring with computer-based automated analysis. We use a statistical detector to automatically discriminate between dominant theta–delta and dominant alpha activity at each instant of time. Moreover, a statistical detector is used to validate visual scoring of K complexes, delta waves or artifacts associated with an EEG frequency shift, as well as frequency shifts to beta activity. A high-resolution (250 ms) state-transition diagram providing continuous information on the sleep-wake state of the subject is finally obtained. Based on this information, arousals are automatically identified as any state change from sleep to wakefulness lasting ≥2 s. The assessment of the interaction between arousals and ventilation is performed using a breath-by-breath, case-control approach. The arousal-associated change in ventilation is measured as the normalized difference between minute ventilation in the case breath (i.e., with arousal) and that in the control breath (i.e., without arousal), controlling for sleep stage and chemical drive. The latter is estimated by using information from pulse oximetry at the finger. In the last part of the paper, we discuss main potential sources of error inherent in the described methodologies.