Alarm Filtering in Intensive Care Units Using Multivariable Analysis of Physiological Parameters

Abstract: Several variables are usually recorded on line in Intensive Care Units (ICU) to detect changes in a patient’s state. Current alarm systems are based on fixed thresholds, generating many false alarms. This paper presents a filtering alarm system developed to reduce these false alarms, without missing any true alarm. A vital physiological variable, the oxygen saturation rate signal (SpO 2 ) is taken as an example. The filtering alarm system is started when the threshold based alarm system sets off an alarm. A multivariable analysis of the other monitored physiological variables is carried out on a fixed temporal window preceding the alarm, using change indices calculated from the physiological variables trend. These indices are aggregated using fuzzy decision making tools and allow classifying the alarm in “true” or “false”. The system was tested on 45 hours of data recorded every second on adult patients hospitalized in ICU.