Statistics‐based alarms from sequential physiological measurements

Abstract

We have developed an anaesthesia alarm system that responds in a more clinically appropriate manner than current threshold alarms. A decrease in systolic arterial pressure of 10 mmHg from a previous value of 70 mmHg has a greater clinical significance than a decrease of 10 mmHg from 150 mmHg. However, it has been difficult to envisage a simple algorithm for the detection of these contextually adverse changes in physiological variables. We have processed systolic arterial pressure data to create a mathematically straightforward statistical tool for sampling intervals up to 5 min. Both the blood pressure and the change in blood pressure over a known time interval are plotted on x and y axes with the units in standard deviations. Some 10 824 measurements were obtained at 10-s intervals in 17 patients. The mean (SD) systolic arterial pressure for all observations in our patients was 118 (17.0) mmHg. The mean (SD) change in systolic arterial pressure over 5 min was - 0.35 (15.2) mmHg. Combining the value for the standard deviation of systolic arterial pressure and the standard deviation of the change in systolic arterial pressure using Pythagoras's theorem creates a value in standard deviations for this particular state. Instead of alarms being set in mmHg, they would be set in standard deviations. This technique was developed further using Principal Component Analysis to isolate uncommon deviations from normal, clinically unimportant physiological variations. These clinically unimportant changes occur in a predictable fashion only if the sampling interval is 90 s or less. This new alarm system is asymmetric - a small decrease in systolic arterial pressure from 90 mmHg may, appropriately, set off an alarm but it would require a much larger increase in systolic arterial pressure to do so.