For hospital patients, the incessant beeping and buzzing of alarms from monitors, pumps and other devices can make sleep almost impossible.
But all that noise isn’t just annoying; it’s also potentially dangerous. For nurses, the barrage of alarms can be overwhelming, making it easier to tune them out—including when they indicate actual patient deterioration.
Hospitals have tried many strategies to combat alarm fatigue over the years, with mixed results. Now researchers are broadening the focus beyond simply reducing the number of alarms, with a larger goal of improving patient safety. They are making the alarms smarter, more useful and less annoying, while developing early-warning systems that could help clinicians intervene sooner and potentially prevent some alarms from being triggered in the first place.
Here are some promising approaches to easing alarm fatigue:
Making alarms wait
At Stanford Medicine’s Lucile Packard Children’s Hospital, bedside monitors in the pediatric intensive-care and acute cardiac-care units were generating large numbers of alarms considered “nonactionable”—caused by brief, self-correcting or clinically unimportant changes. Nurses had to listen to all those alarms and assess them, adding to alarm fatigue and potentially making it harder to respond to actual patient deterioration.
“As we were adding new technology for more things we can do at the bedside, the alarm burden was only increasing,” says Dr. Felice Su , medical director of the pediatric intensive-care unit.
Building on earlier work, including a program at Cincinnati Children’s Hospital Medical Center, Stanford programmed the monitors with time delays and conditional triggers so that some monitor readings didn’t automatically set off an alarm. For example, a brief drop in a patient’s blood-oxygen reading could be given time to correct itself, while a more serious or sustained drop would trigger an alarm sooner.
According to a study published last year, the changes cut the typical number of alarms per patient each day by 75% in the pediatric intensive-care unit and by 82% in the cardiac unit, over the one-month study period. Between the two units, that meant a total drop to 26,000 alarms from 60,000, says study co-author Dr. Jeffrey K. Yang , clinical assistant professor of pediatrics-cardiology at Stanford Medicine Children’s Health. The researchers found no increase in emergencies requiring urgent intervention or transfer to a higher level of care, suggesting that reducing the alarms didn’t compromise patient safety.
After the changes, nurses reported an improved ability to respond to alarms appropriately. The effect on patient sleep was harder to assess. Before the new alarm settings were introduced, about 40% of families said monitor alarms disrupted their child’s sleep. While that decreased after the new settings, the change wasn’t statistically significant.
A sound experiment
Other researchers are focusing not on how often alarms sound, but on what they sound like. One approach comes from Dr. Joseph J. Schlesinger II , a professor of anesthesiology critical care medicine at Vanderbilt University Medical Center. The doctor—who is also a musician—teamed up with McMaster University music-cognition researcher Michael Schutz to apply principles of music and acoustics to medical alarms.
Their goal isn’t to make monitors play catchy tunes, Schlesinger says, but to make alarms easier to distinguish, less irritating and more informative.
For instance, the system uses recognizable sounds, known as auditory icons, to tell caregivers what the problem is, so they can distinguish it immediately. For blood pressure, experimental alarms use the familiar “lub-dub” of a heartbeat. The system also gives each sound a “pointer,” using changes in pitch, harmony and texture to signal whether blood pressure is high or low, as well as the severity of the problem.
In a 2023 study, participants recognized an urgent high-blood-pressure alarm about 10% faster when it used musical cues. In a later experiment , clinicians and nonclinicians more accurately identified a redesigned pointer with additional musical and acoustic cues 90% of the time, compared with 80% accuracy for an earlier version, without slower response times.
Changing the sound itself may also make alarms easier to live with. In one experiment, participants heard six short alarm melodies played with either conventional electronic tones or a more musical, xylophone-like sound. Participants rated the musical versions as significantly less annoying.
“It doesn’t need to be the New York Philharmonic in the operating room, but as a system for communicating information effectively, music has a lot to teach us,” says Schutz.
Much earlier warnings
One way to reduce the burden of alarms may be to prevent some from becoming necessary in the first place, by warning clinicians earlier and without noise. If clinicians can spot deterioration earlier, they may be able to intervene before a patient reaches the crisis point that triggers a conventional alarm.
For example, there’s a tool called Concern that uses machine learning to look for potentially troublesome patterns in the hospital’s electronic health record. Nurses routinely use the record throughout their shifts, entering and reviewing information such as patients’ vital signs, assessments and medications.
Concern researchers found that nurses can pick up subtle signs that a patient is deteriorating before conventional measures such as heart rate, blood pressure and oxygen levels signal serious trouble. That concern can leave clues in the electronic health record—for example, nurses may check vital signs more frequently.
Concern analyzes those patterns in nursing documentation and turns them into a risk score for the broader care team. The system recalculates the patient’s risk every hour, but clinicians told the researchers they didn’t want another alert adding to alarm fatigue, so Concern doesn’t page, buzz or send a mobile notification. Instead, it displays a green, yellow or red icon alongside the patient’s name in the electronic health record, signaling low, increased or high risk. Clinicians can click on the icon to see how the patient’s risk has changed and what factors are contributing to the score.
“Unlike many early warning systems that generate interruptive alerts, we designed Concern as a noninterruptive alert,” says Patricia Dykes , a nurse scientist who was part of the original research group in Boston and is now executive director for data science at Emory University’s nursing school and director of patient safety and innovation at Emory Healthcare in Atlanta. The aim is to provide “actionable and meaningful information” for nurses “so you don’t have a system that’s alarming and people are ignoring it.”
In a randomized trial involving more than 60,000 hospital stays, patients on units using the system had a significantly lower risk of dying in the hospital and shorter stays than those receiving usual care. They were also more likely to be transferred from a regular hospital unit to intensive care, which researchers say could reflect earlier recognition of deterioration and escalation of treatment.
