Preliminary clinical trials of a computer-based cardiac arrest alarm.

The use of a high reliability cardiac arrest alarm utilising the continuously monitored values of patient heart rate and mean arterial blood pressure is described, based on a sample of 167 patients monitored for a total of 5116 h. The analogue heart rate and mean blood pressure signals are sampled at 1 s intervals, and a smoothing algorithm is applied to each of the resulting series which rejects artefacts, and identifies slope and step changes in each. Certain combinations of events in the 2 series, occurring within a preset time window, determine whether a cardiac arrest alarm or warning signal should be activated by the system. A total of 30 acute events occurring in 14 patients during the course of the study were each identified within 10 s. No cardiac arrest event was misdiagnosed by the algorithm during the period of the study. The algorithm also generates warnings which may have predictive value, and which will be the subject of further research. A final false alarm rate of about 1/200 h of monitoring was observed in adults (1/50 h in children), with evidence that these rates could be substantially improved.