Over the last week, I have spent a great deal of time in hospital rooms listening to monitoring systems in alarm and observing the problems of poorly designed sensors. I was reminded of the development of large network Operational Support Systems (OSSs) from the 1980‘s and 1990‘s. These systems were designed to improve the availability of network services through rapid fault detection, isolation and repair. In those days, we talked about “alarm storms”... single or correlated events that would generate hundreds or thousands of alarms that were more than operators were capable of analyzing and taking action upon. An example of an event that would generate a large number of alarms is the failure of DC power feeds to a rack of transmission equipment. Hundreds or thousands of circuits passing through this rack would go into alarm. These alarms would be reported from equipment all over the network. The result of alarm storms were overwhelmed network control center personnel, who either couldn’t figure out what the root cause of all the alarms were or who grew fatigued from constant assault of systems in alarm. Overwhelmed operators resulted in poor networks service availability despite the promise of intelligent systems to improve service availability.
So to deal with this problem, alarm categorization, alarm filtering and alarm correlation capabilities were build into these systems. Alarms could be categorized. Minor alarms could be ignored. Circuit failures that could be mapped into the failure of a single board or system were rolled up into one alarm. These capabilities are normal now, however, they had to be learned through experience as networks of intelligent systems scaled up.
Back to the hospital intensive care unit... a single patient has multiple sensors and multiple machines providing services to the patient. Each sensor (heart rate, blood pressure, respiration rate, blood oxygen levels, etc.) generates an alarm when parameters get out of range. Ranges can be customized for each patient, however, some fundamentals do apply. For example, blood oxygen levels below 90% is bad for any patient. Machines providing services, such as an IV machine, go into alarm under a variety of circumstances... for example, if the supply bag is empty, or if the IV tube is blocked. Alarms can be suppressed, however, these machines go back into alarm after a time if the trouble is not resolved.
All alarms are not equally life threatening, however, each monitor or machine behaves as if they are. Some sensors are very unreliable and are constantly in alarm. Sometimes, different systems interact with each other to cause alarms. For example, when a Blood Pressure cuff is measuring, it cuts off blood supply to the oxygen sensor on the finger of the same arm, causing an alarm. Nursing staff, like operators at a network control center, learn to ignore unreliable alarms and work around some monitoring system problems, however, it is easy for them to be overwhelmed by alarm overload and miss critical alarms. Integrated alarm monitoring and management systems for intensive care seem to be emerging, however, they still have a long way to go. I found that there was no substitute for a patient advocate, providing an escalation path for the critical problems that get lost in the noise.
What’s your experience with monitoring, alarm management and intelligent systems?
No comments:
Post a Comment