Inductive plethysmography components analysis and improved non-invasive postoperative apnoea monitoring

Twenty-nine patients were monitored overnight for breathing distress patterns during postoperative analgesia. Nasal flow apnoea monitoring and pulse oximetry data were recorded at 50 Hz. Respiratory inductive plethysmography (RIP) tracked tidal volume (TV) thoracoabdominal motion, and supplemented the flow monitoring by identifying detected apnoea type. TVs were computed from linear combinations of the RIP signals, but calibrations showed that multiple regression approaches with fitting errors <1% had highly variable coefficients and estimate precisions. Simple least squares theory showed that unstable parameter calculation and coefficient variation with signal conditions were inherent in RIP calibration models. Principal components (PC) methods were well suited to mitigating these problems because the RIP signals were highly correlated. The two PCs tracked the relative changes in TVs and indicated the degree of signal asynchrony, enabling improved uncalibrated monitoring. For accurately measuring RIP phase differences, the cross-correlation function was calculated. A simple version of PC analysis is developed, avoiding matrices, to help clarify how RIP calibration problems can be addressed. The methods are illustrated for calibration in normal breathing, and for postoperative monitoring during Cheyne-Stokes breathing. Sum and difference combinations of the RIP signals could discriminate central from obstructive apnoeas to help improve flow monitoring efficacy on-line.