| Abstract | Monitoring is the serial evaluation of time-stamped data, and the volume of such data in an intensive care unit is huge. Clinical and biochemical data may be available at hourly or more frequent intervals but physiological data are 'continuous'. Although sophisticated monitors display the physiological data in multiple and varied combinations, staff are challenged by the frequency of the false alarms and lack of knowledge of the patterns from which they could predict problems. All these data, together with large amounts of clinical data, lead to information overload. In this paper, the case is made for the development of automatic decision-support system based on statistical and probabilistic analysis of data patterns appropriate for the level of cognition of the user (nurses and juniors at the bedside rather than consultants). Such decision support could both reduce the false-positive alarms that frustrate clinical staff, and improve the early detection of pathophysiological events. We have used the development of a pneumothorax as our paradigm. Our data indicate that the clinical diagnosis of pneumothorax takes a median of 127 minutes, but using short decision algorithms based on routinely available monitoring data, most can be detected within 10-15 minutes of occurrence. |