A trend-based alarm system to improve patient monitoring in intensive care units

This paper presents an alarm system based on trend analysis. A trend is semi-quantitative information, extracted on-line, expressing the time evolution of a signal with several symbols such as {Increasing, Decreasing, Steady}. The trend description corresponds to an asynchronous abstraction of a digital signal. In this paper, trends extracted on-line from physiological data are used to generate alarms for medical staff. The quantitative information of the trend is used to set off an alarm, while the qualitative information is used to recognize specific situations on-line, such as device disconnections. The system was run on-line and tested on 36 h of data recorded on adult patients hospitalized in an Intensive Care Unit. Comparison with alarm periods set off by a traditional limit alarm system shows that the performance of the trend-based alarm system overtakes those of a traditional limit alarm system, with 33% of false alarms rejected without any clinically relevant alarms missed.