Two-Stage Interpretation of ICU Data Based on Fuzzy Sets

Intelligent on-line monitoring of ICU patients requires mechanisms to detect and interpret significant patterns in the flood of recorded data. Many monitoring methods have been proposed, ranging from simple surveillance of thresholds to complex reasoning about and forecasting of the patient ́s behaviour based on computational models of the human organism. From the many components that constitute a comprehensive intelligent monitor we focus on two the combination of which results in a workable device capable of mapping collected data onto physiological and pathophysiological states. They are arranged as stages and form the kernel of a bigger, still evolving system called DIAMON-1, which is dedicated to the on-line interpretation of patients ́ status. Stage one, which has also been more generally described as preprocessing in [7], deals with the transformation of incoming data streams into a sequence of events. In this paper we focus on a special issue of temporal abstraction: the detection of fuzzy trends. The event of detection is reported to stage two, a fuzzified deterministic automaton capable of assigning diagnoses to sequences of fuzzy events. For the purpose of evaluation, ICU data of an eight month old female suffering from adult respiratory distress syndrome (ARDS) was distributed to the symposium participants. The data set contains continuous heart rates, blood pressures, O2 saturation values, and ventilator settings sampled over a 12-hour period. In addition, discontinuous data from laboratory tests and the flowsheet was provided. In applying DIAMON-1 to the presented case we chose to concentrate on the events centred around hand bagging sessions where the mode of ventilation was changed from continuous mandatory ventilation (CMV) to hand bagging, accompanied by a temporary increase in the fraction of inspired oxygen (FIO2) from 0.5 to 1 and the delivery of Ventolin.