ALARM-INDUCING VARIABILITY IN CARDIAC POSTOPERATIVE DATA AND THE EFFECTS OF PREALARM DELAY

Objective. Our objective was to study the distribution of invasively m easured hemodynamic data to enhance the reliability of patient monitor alarm systems. Methods. Monitoring data were collected, preprocessed off-line, and analyzed in 10 postcardiac surgery patients. The data we re studied statistically to estimate the probability distributions, th e probability of alarm at various probability levels in these distribu tions, the effect of the PREALARM DELAY to the alarm probability, and the effect of preprocessing the monitoring data using one- or multidim ensional MEDIAN FILTERING. Results. Fifteen percent of all registered values fell outside of commonly applied alarm limits. Doubling the pre alarm delay from 5 to 10 sec reduced the mean alarm rate by 26%. A fur ther decrease of 8% in the alarm rate was observed when a multidimensi onal vector median filter was used to remove the variable value interd ependencies. Conclusions. Brief excursions beyond clinically optimal a larm limits were frequent and can occur without leading to significant degradation of the patient's state. Preprocessing can decrease the al arm rate effectively. Multidimensional preprocessing may produce more reliable alarms than one-dimensional processing.