Refining continuous capnography monitoring by using a new algorithmic alarm filter in a medical-surgical unit dramatically reduced the number of clinically insignificant alarms, according to a study of 25 patients with obstructive sleep apnea (OSA).
The study, which was carried out at Virtua Memorial Hospital, in Mount Holly, N.J., found that the alarm filter, developed by Bernoulli Health, in combination with a 30-second notification delay, achieved a 98% reduction in the frequency of issued alerts, without compromising patient care.
“As an anesthesiologist, I see a number of opioid-related side effects during the postoperative period, including respiratory depression,” said co-author Leah Baron, MD, chair of the Department of Anesthesiology at the hospital and an anesthesiologist at Burlington Anesthesia Associates, in Mount Holly, part of MEDNAX Inc. “In our institution, a significant number of our surgical patients are diagnosed with OSA, or are suspected to have OSA. These patients are known to be more susceptible to opioid-induced respiratory depression.”
By observing this subset of patients in the PACU and on medical-surgical units, Dr. Baron and her colleagues recognized that better monitoring modalities were needed for early detection of respiratory depression. The goal of the study was to create a safer environment for their high-risk surgical population.
Previous investigations showed that hypoxia is a relatively late sign of opioid-induced respiratory depression, whereas monitoring breathing via capnography is an earlier sign of pending deterioration.
The current clinical study, which was published in Biomedical Instrumentation & Technology (2017;51:236-251), is a follow-up to a 2013 initial pilot study.
Actionable Alarms Versus Noise
The investigators identified sustained combinatorial alarms that principally combined respiration rate and end-tidal carbon dioxide (CO2), and that correlated with clinical significance. “The most significant drop in hypopnea/hypoventilation combinatorial alarm occurred at 18-second alarm delay, reducing the number of alarms from more than 4,500 to 209,” Dr. Baron said. The investigators used Bernoulli One analytics software in conjunction with the Medtronic Capnostream 20 bedside monitoring system.
The investigators defined the alarm threshold as five consecutive readings at six-second intervals over a 30-second period of time for CO2, respiratory rate, oxygen saturation or heart rate. Overlaying individual alarms that coincided with clinically significant events created an actionable alarm.
“Some of our parameters, though, were more important than others,” Dr. Baron said. “For example, oxygen saturation never reached low levels. This confirmed the findings of previous studies that oxygen saturation is a fairly late sign of respiratory depression.”
The reduced number of alarms was mostly confined to low CO2, an indicator of obstructive airways.
“Overall, there was a delay in notification for the clinical nurses to respond,” Dr. Baron said. “By implementing a 30-second delay, we had no patients that were undetected for respiratory depression. We were also able to intervene [in a] timely [fashion].”
Of the 25 study patients, seven required some degree of intervention. “You want to identify actionable alarms versus background noise,” Dr. Baron said. “According to ECRI Institute, addressing alarm fatigue is one of the 10 top priorities in hospitals. Finding ways to safely monitor the high-risk surgical population on med-surg units and decreasing the use of more expensive beds by this high-risk population provides significant cost savings for an institution.”
Dr. Baron said follow-up studies are needed to ascertain the clinical relevance of even longer delayed combinatorial alarms, for example, waiting 48 seconds for notification instead of 30 or 42 seconds.
Study co-author John Zaleski, PhD, is chief analytics officer of Bernoulli, which focuses on alarm management and medical device integration. Dr. Zaleski created the Bernoulli One analytics software. “I realized that the use of real-time, clinically actionable data requires more than filtering based on alarms and vital signs thresholds,” Dr. Zaleski said. “Combining analysis with real-time data at the point of collection creates a powerful tool for prediction and decision support.”
Dr. Zaleski’s original idea was driven by an investigation into the area of real-time patient care management and identifying markers of respiratory states that could herald the onset of an adverse event, particularly related to weaning from postoperative mechanical ventilation and respiratory depression. The application of this approach to respiratory depression in patients receiving postoperative opioids for pain management was a logical extension of the methodology.
“Early investigations into viewing a threshold-based alarm signal proved that an excessive number of false alarms was evident,” Dr. Zaleski said. “A closer review of the retrospective data revealed that certain correlations among parameters, together with delayed action on those alarms that sounded, could reduce the overall quantity of alarm being issued.”
The technical obstacles in developing the algorithm were both physical and analytical. “The analytical hurdle was identifying which parameters, when measured continuously, were indicative of the onset of an adverse event, whereas the chief physical hurdle was managing logistics associated with assigning and de-assigning patients to the capnography equipment,” Dr. Zaleski said.