Fusing multiple heart rate signals to reduce alarms in the adult intensive care unit;
The objective of the study was to compare the performance of two physiologic monitor algorithms in their ability to generate true heart rate alarms and to avoid producing false alarms. The “standard” algorithm, the algorithm currently used in GE/Marquette monitors installed at LDS Hospital, was compared with a “fusion” algorithm that combined heart rate data independently determined from electrocardiogram (ECG), intra-arterial pressure (ART), and pulse oximeter signals. Data were collected from patients admitted to the medical, surgical, and cardiac ICUs at LDS Hospital in Salt Lake City, Utah, from April through September, 2001. Data from a total of 109 patients were collected for periods of up to 24 hours for each patient, which resulted in a total of 1902.25 patients-hours of data. The physiological signals were then presented to both algorithms to allow a direct comparison of both alarm methods. A physician reviews the heart rate alarm results of each algorithm and determined whether each alarm generated was true or false. Since alarm conditions were only studies if they were detected by one of the algorithms, we were not aware of any condition that should have generated an alarm but did not. The following five alarm conditions were studied: low heart rate (LHR), high hear rate (HHR), asystole, ventricular tachycardia (VT), and ventricular fibrillation (VF).The “standard” algorithm generated 341 alarms; 118 (34.6%) were true and 223 (65.4%) were false. The fusion algorithm produced 184 alarms; 126 (68.5%) were true and 58 (31.1%) were false. There were 149 instances in which both algorithms produced the same alarms. Of these instance 111 (74.5%) were true alarms and 38 (25.5%) were false. Of the combined total of 525 alarms (341 standard + 184 fusion), 316 (60.2%) had durations of 10 seconds or less. The determine if a patient’s average heart rate was associated with low or high heart rate alarms, the average heart rate was calculated. Average heart rate was determined by summing all heart rate values and dividing the total number of heart rate of each patient’s data set. Of the 267 LHR alarms, 148 (55.4%) were from patients who had average heart rates of 80 beats per minute (BPM) or less. Of the 127 HHR alarms, 114 (89.8%) were from patients who had average heart rates of 90 BPM or greater. While there was no “gold standard” test available to calculate the actual sensitivity and false positive rate of each algorithm, we were able to compare the two algorithms to each other using relative sensitivity (RSN) and relative false positive rate (RFP). The RSN of the fusion algorithm compared to the standard algorithm was 1.09 (95% CI 1.01 to 1.17). The 95% confidence interval (CI) for the RSN indicated that there was a 95% chance that the true positive rate for the fusion algorithm was between 1% and 17% greater than the standard algorithm. The RFP of the fusion algorithm compared to the standard algorithm was 0.27 (95% CI 0.21 to 0.34). The 95% CI for the RFP indicated that there was a 95% chance that the false positive rate for the fusion algorithm was between 66% and 79% lower than the standard algorithm. Using the fusion algorithm to reduce false alarms did not result in a concomitant increase in the number of true alarms that were “missed” by the fusion algorithm. On the contrary, the fusion algorithm missed only 5 true alarms that the standard algorithm caught, while the standard algorithm missed 15 true alarms that the fusion algorithm caught. Combining redundant physiologic signal measurements through the use of a fusion algorithm was an effective way to eliminate false positive alarms. Because the fusion algorithm was more sensitive, had a higher positive predictive value, and missed fewer alarms that the standard algorithm, it was determined to be a superior method of generating alarms. In addition, we found that simple delaying alarms by 10 seconds could result in a 60.2% reduction in false alarms. Considering the patient’s average heart rate when setting low or high heart rate alarm threshold could also reduce false alarms. All of these methods should be integrated into new bedside monitors to reduce false alarms.
University of Utah;
Intensive Care; Intensive Care Units;
Heart Rate; Monitoring, Physiologic;
University of Utah;
Relation-Is Version Of
Digital reproduction of “Fusing multiple heart rate signals to reduce alarms in the adult intensive care unit”. Spencer S. Eccles Health Sciences Library. Print version of “Fusing multiple heart rate signals to reduce alarms in the adult intensive care unit” available at J. Willard Marriott Library Special Collection. RC39.5 2005 .P66.