Making ICU alarms meaningful: a comparison of traditional vs. trend-based algorithms.

Much of the work in the ICU revolves around information that is recorded by electronic devices. Such devices typically incorporate simple alarm functions that trigger when a value exceeds predefined limits. Depending on the parameter followed, these "boundary based" alarms tend to produce vast numbers of false alarms. Some are the result of false reading and some the result of true but clinically insignificant readings. We present a computerized module that analyzes real-time data from multiple monitoring devices using a customizable logic engine. The module was tested on 6 intensive care unit patients over 5 days, running alarm algorithms for heart rate, systolic and diastolic blood pressure as well as arterial oxygen saturation. Results show a ten-fold increase in positive predictive value of alarms from 3% using monitor alarms to 32% using the module. The module's overall sensitivity was 82%, failing to detect 18% of significant alarms as defined by the ICU staff. The results suggests that implementation of such methodology may assist in filtering false and insignificant alarms in the ICU setting.