Dimension reduction for highdimensional online-monitoring data in intensive care

Nowadays high dimensional data in intensive care medicine can be captured, stored, and retrieved with the help of clinical information systems. Intelligent alarm systems are needed for an adequate bedside decision support, in the course of which the detection of qualitative patterns in physiologic monitoring data such as outliers, level changes, or trends aims at a proper classification of the patients state. Statistical time series techniques have already been applied successfully to the analysis of single physiological variables. The simultaneous online analysis of the multivariate patient curve yields further challenges. We describe methods for reducing the dimension and for keeping the computational efforts necessary for monitoring low. We present preliminary results of an ongoing study on monitoring critically ill patients.