Multivariate analysis in clinical monitoring: detection of intraoperative hemorrhage and light anesthesia

The number of vital sign variables measured during a typical surgery is beyond the simultaneous surveillance capabilities of most experienced clinicians. Most intraoperative events cause trend changes in multiple variables, and many clinical events can only be detected by investigating the inter-relationship between the direction and amplitude of these trend changes in the whole measurement array. We have compared the techniques of principal component analysis (PCA) and factor analysis (FA) in extracting latent variables to represent the underlying physiological mechanism. The detection performance of each method was tested on three simulated cases of intraoperative hemorrhage and a case of variation in depth of anesthesia. The results show that although the detection schemes based on PCA and FA both reduce dimensionality and detect changes in the variance, the FA-based method performs better in detecting subtle changes in the correlation structure.