The Case for Continuous Clinical Surveillance

by John Zaleski, Ph.D., CAP, CPHIMS
Chief Analytics Officer

How to prevent injury or death due to unrecognized patient deterioration ^[1] was the subject of debate at the AAMI Foundation Annual Forum on patient safety.

Best practices for preventing said events focused on two ideas. Those include:

• Widening single-parameter alarm limits to balance essential data delivery and reducing false alarms; and,
• Continuous clinical surveillance, which captures multiple parameters from bedside medical devices and filters and delivers clinically-actionable alarms in real time.

Both ideas share the same goals and accept current surveillance challenges. Those include:

• Reduce false alarms without weakening patient safety; and,
• Current best practices, like visual spot checks, leave patients exposed to preventable adverse events.

That said, these ideas diverge in notable ways. Noting these differences will help hospital leadership make informed decisions regarding what is best for the organization, the clinicians, and the patients.

Alarm Reduction

Reducing false alarms has been the subject of countless meetings, lectures and peer-reviewed studies. Many approaches have been identified for reducing alarms in high-acuity settings. For example, Görges ^[2] showed that a 14-second delay before alarm presentation would reduce non-actionable alarms by 50%. A 19-second delay would reduce this further to 67%.

Discerning the root causes of false alarms and standardizing management policies are exceedingly difficult.^[3] Respected organizations, including the AAMI Foundation, have contributed useful resources to help hospitals manage alarms ^[4]. The Joint Commission has issued a new National Patient Safety Goal (NPSG) on clinical alarm safety. It mandates the creation of clinical alarm management policies and education ^[5].

Adjusting default thresholds for devices such as pulse oximeters and capnographs can reduce the quantity of alarms issued. There is little evidence showing that widening thresholds increases risk to the patient.

However, even adjustments based on hospital policies and protocols may not capture at-risk patients. The simplistic threshold limits in pulse oximeters and capnographs are highly susceptible to false alarms and more so, do not capture underlying dynamic changes.

Continuous Clinical Surveillance

A patient who succumbs to an undetected adverse event can be compared to a plane crash. Aviation accidents are usually the result of a combination of separate events that cascade into a larger and more significant problem ^[6]. Similarly is the process of patient deterioration.

With this in mind, continuous clinical surveillance employs multi-variate data from multiple sources. Sources include EKGs, vital signs, laboratory tests, and EHR data. Identifying clinically relevant trends, sustained conditions, reoccurrences, and combinatorial indications may indicate a degraded patient condition. This helps clinicians to recognize and respond to signs of distress before the patient’s health declines.

An objective of predictive analytics is to uncover relationships among seemingly unrelated measurements. This process begins with the underlying physiological, biological, and chemical processes that describes expected behavior and then validating or correcting those assumptions.

This ensures that analytical models accurately represent measured behavior. An example of this follows.

Continuous Clinical Surveillance at Work

Our team recently published a study in the Journal of Biomedical Instrumentation & Technology. A collaborator for the article was an East Coast hospital.

The study measured pulse (HR), oxygen saturation (SpO₂), respiratory rate (RR), and end-tidal carbon dioxide (ETCO₂) continuously and compared alarms received through the bedside monitoring device with remote alerts triggered after a delay. The goal was to reduce the total number of alarms without increasing risk to patients at risk for respiratory depression.

The results showed that using only sustained alarms as the filter, alerts were reduced from 22,812 to 13,000; a number still high enough to risk alarm fatigue. However, when passing multiple data points through a multi-variable rules engine that monitored the values of HR, SPO₂, RR, and ETCO₂ in order to determine which alarms to send to the nurse-call phone system reduced respiratory depression alerts to 209—a 99% reduction. ^[7]

Conclusion

Healthcare organizations face big challenges in balancing alarm management and patient safety. Rescuing patients is costly in terms of resource utilization, morbidity, and mortality. Logic dictates that interventions that precede crisis events will better reduce the strain on the patients and clinical staff.

Hospitals need to take a system-wide inventory of alarms. Further, they must use project teams to guide the migration from reactive alarm response to a more proactive strategy. Frontline clinicians must be involved. Impacts to workflows must be considered. Other considerations include proper infrastructure: computer and monitor networks, a centralized data center, the computer capacity to run the algorithms, and the ability to distribute the data to clinicians carrying mobile devices or via a dashboard.

Hospitals with critical care units or ICUs already have a continuous monitoring infrastructure in place. Progressing to continuous clinical surveillance throughout the hospital or health system requires a careful investment of time and money in addition to workflow considerations to ensure the highest level of patient safety. That said, the health systems that empower their clinicians to make more intelligently informed decisions utilizing the data that are already available at the bedside will be the long-term winners in terms of predictive monitoring and intervention.