On the application of system identification and model validation methods for constructing multivariable anesthesia response models

This paper investigates the use of multivariable pharmacodynamic models for describing patient response to anesthesia. The main modeling paradigm considered is that of uncertain systems, with the eventual goal of developing robust feedback control methods for automated anesthesia delivery. We present an approach to modeling anesthetic response in which multi-input multi-output state-space models are considered as a means of describing the relations that exist between a variety of patient endpoints such as heart rate, mean arterial pressure and EEG signals, to anesthetic inputs and external stimuli. The main focus of the present discussion is on the construction of linear time-invariant models via standard system identification methods applied to data from a clinical study of volunteers. The use of recent model validation techniques for the construction of uncertain systems models and model validation purposes is also discussed.