Towards reliable and intelligent alarms by using median filters

In the last two decades there has been a rapid development of the equipment used for monitoring of critically ill patients at intensive care units. The monitoring has become more invasive, and concurrently new parameters have become available to routine bed-side monitoring. The automatic alarming systems employed in these monitors have not, however, developed at the same pace. Typically, upper and lower limits are used in generation of alarms for each monitored parameter. Several parameters are rarely combined to produce more reliable alarms. The current limit alarming methods can be considered unreliable for the monitoring of a critically ill patient. Typical reasons for this are momentary artifacts and short lasting insignificant changes in the monitored parameter values. The incidence of false alarms is high. In order to accomplish more reliable alarming the data from the monitored parameters should be combined, using clinical knowledge, with other data available concerning the patient. Such systems can be realised using knowledge based programming. The salient problem is then the reliability of the real time measurement. The measurement and processing of values of the intensively monitored physiological parameters were studied. The reliability of the existing limit alarms was assessed clinically. The use of median filters to accomplish more reliable limit alarming was suggested. Suitable filters were selected on the basis of clinical experimentation. A novel dual-limit alarmer was proposed and evaluated clinically. This alarmer allows reliable automatic monitoring of therapy and a rapid response to catastrophic failures of the physiology. On basis of the clinical tests, the proposed method was found very reliable compared with the existing methods. The use of median filters for detecting and quantifying trends in the monitored parameters was proposed. A method was suggested for quantitative measurement of trends relevant to therapy. Finally, reliable signal-to-symbol transformation (symbolisation) was studied, and a method was presented for symbolisation of both the value and the trend contents of the monitored data. These were actually employed in a knowledge based alarming system developed in the same research project. A brief description of the implemented knowledge based alarming system is followed by a clinical test of the system. There have been no publications concerning the application of median filters to intensive monitoring data in order to enhance the reliability of limit alarming. The experiences gained in the study would suggest that such study has great potential.