An algorithm strategy for precise patient monitoring in a connected healthcare enterprise

Alarm fatigue continues to be one of the most important problems in health care. We describe our SuperAlarm framework, a strategy to not only solve the problem of alarm fatigue, but also enhance the utility of hospital monitoring systems.

“How was your night?” The answer from my post-procedural patients admitted for observation overnight in the hospital is almost invariably, “I didn’t sleep that well. The alarm kept going off overnight. But I’m sure I’ll sleep better once I go home tonight!” While this is but a single night’s bad experience for most of my patients, as a cardiologist who has worked closely with Dr. Xiao Hu on the SuperAlarm system over the past many years trying to address this very problem, these comments are important illustrations of a problem that remains a major problem in medicine.

As any clinician can attest, the barrage of alarms is nearly endless for hospitalized patients and for their caring physicians and nurses. Worse yet, nearly all of these alarms are unactionable; even those alarms meant to be the most critical of all such as “ventricular tachycardia” or “asystole” are frequently false. This has led to the appropriately termed condition of “alarm fatigue”: a desensitization of hospital staff to the system that constantly cries “wolf” and haphazard attempts to stop such alarms from repeating to the potential detriment of patient care.

As we discuss in this paper, the underlying problem is deep-rooted. Though we live in an ever more connected world, medical technology remains largely unconnected. Each medical device is developed with only itself in mind, running independently and with little purpose aside from performing its own narrow scope of functions, disregarding the patient’s condition in general. As such, the result is naturally a deluge of alarms which frequently do not assist in patient care, and may actually harm patients.

We believe that in order to fundamentally address the problem of “alarm fatigue”, monitoring systems must be transformed from an isolated series of devices to one that is patient-centered: a system that integrates data from a multitude of devices and other diagnostics to provide alarms that will actually help clinicians improve patient care.

Such alarms are 1) precise to specific patient conditions and thus actionable, 2) predictive of specific patient conditions with adequate lead time to allow early intervention to avert an adverse outcome, and 3) interpretable by clinicians such that further diagnostic workup and therapeutic management can be logically carried out. This goal requires a highly interdisciplinary approach and redesign of the fundamental approach to alarms using a data-driven and clinically validated approach, which we describe in this paper.

We reimagine the alarm system in a linguistic manner, where many alerts (what we know of as individual alarms in current day system) that frequently coexist are combined to form “words”; we call these SuperAlarms (In this manner, non-critical alerts that occur in isolation, are likely false and therefore suppressed). We then combine these “words” into “sentences” (sequences of SuperAlarms) through a data-driven and clinically validated approach to best describe (and warn of) specific patient deterioration events. These specific sequences of SuperAlarms can be further reviewed and verified by clinicians to justify further workup and treatment.

Using such an analogy, we can also see that to best describe a wide variety of patient deterioration events of interest to the clinician, our vocabulary of “words” must be expanded. This can be done by incorporating new sources of alerts into our SuperAlarms besides the traditional monitor alarms: data from imaging studies, laboratory tests, and clinical notes to name a few.

One source of alerts and SuperAlarms that is of particular interest to me, as a cardiac electrophysiologist, is the continuous electrocardiographic recordings. We have shown that many changes in electrocardiographic parameters can be detected preceding cardiac arrests; yet these and many other features of the electrocardiogram are completely ignored in today’s alarm systems. We believe that careful analysis of these signals, and integration of this information into monitoring systems will yield significant benefits in both predicting patient deterioration events and providing insights into the etiology of deterioration events to guide management in real time.

A significant amount of work remains to be done to adequately address the problem we currently face with alarm fatigue, but we believe that the framework presented in this paper will allow us to not only significantly reduce the burden of nuisance and unactionable alarms, but also greatly improve patient monitoring and patient care through the development of an interconnected, intelligent monitoring system.