Nonlinear multivariable modeling and analysis of sleep apnea time series

This paper investigates the modeling and analysis of physiological data rec orded from a 49-year-old male and are composed of three time series: blood oxygen saturation, heart rate and respiration. In particular, it is desired to verify if the models estimated from data can distinguish between the dy namics underlying two different breathing patterns (normal breathing and ap nea). The estimated models are nonlinear autoregressive, moving average wit h exogenous inputs (NARMAX) and the regressors used to compose such models are carefully chosen, among hundreds of candidates, by an automatic procedu re. The results discussed in this paper suggest that the dynamics underlyin g the data are nonlinear and basically deterministic. Using estimated model s it seems to be possible to quantify the stability of the fixed point in p hase space reconstructed using the blood oxygen time series. This, as discu ssed, could be the basis of an algorithmic monitoring system. (C) 1999 Else vier Science Ltd. All rights reserved.