Deterministic structure in multichannel physiological data

We have found numerical evidence of deterministic properties in a multichannel physiological time series, a segment of the Santa Fe data set B which was recorded from a patient with sleep apnea. We used recently developed methods for finding good time-delay embeddings for multichannel data, together with nonlinear deterministic prediction. We show that determining good embeddings for this multivariate time series gives good short-term predictions, and convincing free-run (simulation) behavior, using only a simple local-linear approximation method. We also discovered relations between the different channels, which are heart-rate, respiration and blood-oxygen saturation.