Beep, beeep, beeeep, beeeeeep

See allHide authors and affiliations

Science Translational Medicine  21 Oct 2015:
Vol. 7, Issue 310, pp. 310ec181
DOI: 10.1126/scitranslmed.aad4451

Modern intensive care units contain an impressive array of medical devices to monitor or manipulate a patient’s physiological state. Each device also provides visual and audible warnings to the health care providers when a measured physiological variable goes out of range or the device detects a malfunction. The design of the associated alarm algorithms has traditionally focused on analyzing individual physiological variables and on maximizing the alarm sensitivity so as not to miss a truly life-threatening event. This approach comes at the expense of alarm specificity and has led to a proliferation of monitoring alarms, often exceeding hundreds of alarms per patient per day, the vast majority of which are either technically false or clinically irrelevant. Health care providers respond to this cacophony of low-specificity alarms with desensitization (“alarm fatigue”), which can result in patient deaths, as truly life-threatening events are missed or the alarm features are simply turned off altogether.

Borges and Brusamarello focus on two types of critical heart-rate alarms—extreme bradycardia and extreme tachycardia—and address the false-alarm problem through sensor fusion and machine learning. First, the authors extracted a patient’s heart rate as well as measures of heart-rate variability and signal noise from simultaneously acquired cardiovascular signal streams. This information was then fed into machine learning algorithms that are trained to improve the alarm specificity. Using a three-layer neural network, the authors achieved a false-alarm suppression rate of 92.5% while encountering a true-alarm suppression rate of only 0.08%. Thus, more than 90% of false alarms were rejected while preserving essentially all true alarms.

Although the methodology can be extended to other critical cardiac alarms such as ventricular tachycardia, ventricular fibrillation, or asystole, the approach needs to be independently validated on a much larger sample size than the 1283 alarms from 90 patients used in the study. Nevertheless, the results demonstrate that a substantial reduction in false alarms is eminently feasible by leveraging the redundancies in physiological information among various monitoring signals. Reducing the source of alarm fatigue might also help to make an ICU the serene and safe place it ought to be.

G. Borges, V. Brusamarello, Sensor fusion methods for reducing false alarms in heart rate monitoring. J. Clin. Monitor. Comput. 10.1007/s10877-015-9786-4 (2015). [Abstract]

Stay Connected to Science Translational Medicine

Navigate This Article