ASSOCIATE : the interpretation of ICU data using ASSOCIAtional and TEmporal knowledge
Intensive Care depends on sophisticated life support technology. Effective management of device-supported patients is complex, involving the interpretation of many variables, comparative evaluation of numerous therapy options, and control of various patient-management parameters. Raw data, when taken literally, can lead to the wrong interpretation of the patient. We propose a system which processes raw data in real-time for intelligent alarming and analyses historical data for summarisation and patient state assessment. This will utilise a temporal expert system which incorporates associational reasoning. Using continuous physiological data from monitors, patient history and times of therapy administration, our research consists of applying three consecutive processes: filtering which is used to remove noise in the physiological data; interval identification which generates temporal intervals from the filtered data points which have abstractions relating to their direction of change (i.e. increasing, decreasing and steady); and interpretation which performs summarisation and patient state-assessments from a historical point of view and intelligent alarming from a real-time point of view. Using the temporal intervals, interpretation involves differentiating between events which are clinically insignificant and events which are clinically significant. We need to identify and remove clinically insignificant events (e.g. line flushes, blood samples etc.). Similarly, we need to identify clinically significant events i.e. clinical conditions (e.g. hypovolaemia, pulmonary haemorrhage etc.) and the outcome of therapies - this will utilise the patient history and times of therapy administration. Inherent in this process is the trend template which is used to represent events. Trend templates support temporal reasoning, knowledge to differentiate between events and taxonomical knowledge. Algorithms which are analogous to the way clinicians identify events use these trend templates.