RESUMEN
OBJECTIVE: As more sensors are added to increasingly technology-dependent operating rooms (OR), physicians such as anesthesiologists must sift through an ever-increasing number of patient parameters every few seconds as part of their OR duties. To the extent these many parameters are correlated and redundant, manually monitoring all of them may not be an optimal physician strategy for assessing patient state or predicting future changes to guide their actions. METHODS: The method is illustrated by application to seventy-six anesthetized patients for which thirty-two fundamental and derived variables were recorded at 20-second intervals. The Iterative Order and Noise estimation algorithm (ION) estimated the noise on each parameter. The performance of principal components analysis (PCA) was improved by normalizing the noise estimated by ION to unity. A linear regression of the resulting seven high signal-to-noise ratio principal components (PC's) predicted tachycardia 140 seconds in advance. RESULTS: ION estimated the noise on each parameter with sufficient accuracy to increase the number of significant PC's from two to seven, all of which had identifiable physiological correlates. The resulting receiver operating characteristic (ROC) suggested that a 70 percent prediction rate with 5 percent false alarms could be achieved. CONCLUSIONS: This paper illustrates the use of ION to improve significantly the performance of PCA in the efficient representation of patient state and in improving the performance of linear predictors of clinically significant parameters.