Filtering of ICU monitor data to reduce false alarms and enhance clinical decision support

The monitors in the Intensive Care Unit generate alarms whenever a signal passes beyond a preset limit. Such an approach for alarming generates many false alarms because, in general, clinically insignificant events cause signals to go beyond these limits e.g taking a blood sample. Here the alarm has been caused by a clinically insignificant event, not a disturbance in the patient's physiology. If these limits are set to the maximum allowable physiological deviation from the normal or expected value, the monitor will alarm when the patient is already in a serious condition. Likewise, if the limits are adjusted to increase sensitivity, the monitor will be more prone to giving false alarms. There is, therefore, a strong need to reduce the number of false alarms. Our approach to reducing false alarms is to use filtering techniques which will not only remove clinically insignificant events but also allow medical staff to view noise free data to enhance clinical decision support. Using real monitor data, in this paper we review a number of filtering techniques and present our findings to determine which is most suitable for the Intensive Care Unit monitors.