Published in the February 2016 issue of Today’s Hospitalist
WHEN HOSPITALISTS at Connecticut’s Yale-New Haven Hospital took up the challenge of improving the care of patients who develop sepsis in the hospital, they made a conscious decision: They wouldn’t let the perfect be the enemy of the good.
That meant making some tradeoffs. First, the hospitalists decided they’d be OK if the early-detection alert system they were setting up through their EMR missed a lot of cases in the beginning— as long as there was “exceptionally high” physician buy-in for the ones the system did pick up.
Their system would fire many fewer alerts so the project wouldn’t flame out due to “alert fatigue,” with clinicians ignoring the bells and whistles around them. If the EHR fired off fewer sepsis alerts, the architects of the Yale program figured, hospitalists would be much more likely to drop everything and respond.
“We just wanted to make things a little bit better.”
Another key tradeoff: The hospitalist receiving the alert—the physician carrying the rapid response team (RRT) pager—would typically have to do much less than in hospitals where clinicians are required to rush to a bedside every time a sepsis alert fires. Again, the goal was 100% compliance: No one, project leaders figured, should avoid a page or delay a response because they assume it’s a false alarm.
“We didn’t want a pilot project that no one could actually do after the grant ran out or one so expensive that it can’t be sustained,” explains hospitalist Robert Fogerty, MD, MPH, the assistant professor at Yale School of Medicine who spearheaded the project. Instead of tackling the entire problem of delayed sepsis identification and intervention, “we just wanted to make things a little bit better. If it’s inexpensive and easily ingrained without a lot of extra effort, that’s a win.” From the start, he adds, the project was designed with cost containment, ease of implementation and sustainability in mind.
Many hospitals are trying to harness their EHRs to flag patients becoming septic, and Yale-New Haven’s EHR-based protocol was launched last year. The hospital’s Epic EHR automatically fires off a page to the attending RRT hospitalist when a medical inpatient registers at least four of six criteria. (By comparison, some alert systems across the country fire when patients meet only two criteria.)
Four of those in use at Yale-New Haven are the well-known systemic inflammatory response syndrome (SIRS) criteria (see “SIRS criteria,” below), but these criteria have notoriously low specificity.
To compensate, the project team added two other criteria to the alert system: falling systolic blood pressure and rising creatinine. Both measures were selected, he says, because they are already being checked regularly.
Three response options
When an alert is fired, the attending who receives it must open the patient’s chart, figure out what is going on and make one of three decisions. One, hospitalists can conclude that the patient’s providers are already addressing the problem and merely add a brief note to the chart.
Or they can call the patient’s primary team members and alert them to the situation. Or hospitalists can “self-activate” the rapid response team, which sends that physician, a critical care-trained nurse and a respiratory therapist to the patient’s bedside.
“We rarely do that, but when we have, our average response time is close to four minutes,” says Dr. Fogerty. “If you cut out alarms that don’t need to be intervened on, when it’s the real deal, everyone shows up fast.” Weeding out the alarms that don’t need an intervention helps reduce the project’s impact on front-line providers.
He offers this example from his recent turn carrying the RRT pager. He received one sepsis alert during his shift, and when he went to the chart, he learned that the patient’s primary team had recognized the problem and already broadened the patient’s antibiotics.
But while the chart suggested a need to repeat blood cultures, Dr. Fogerty couldn’t find that new order. When he called the resident, he got this response: “Oh wow, we forgot. Thanks.”
His initial model for an automated sepsis alert was a project done at the University of Pennsylvania and described in the January 2015 issue of the Journal of Hospital Medicine.
Theirs is a “beautiful system,” Dr. Fogerty says. But in the Penn project, a team of providers “must be at the patient’s bedside within 30 minutes after every sepsis alert fires,” he notes. “They have great outcomes, but there is no way we could institutionalize that here.”
Responding to every alert
In the first six months of the project, Dr. Fogerty says there were fewer than 100 sepsis alerts out of about 12,000 medical hospitalizations. A preliminary look at those data indicates that every alert was responded to, with an attending leaving a note in the chart. About one-third of the alerts resulted in an “intervention,” either a call to the patient’s primary team or an RRT activation.
Based on that preliminary review, he says, “We think this has been a worthwhile intervention. It identified some patients we didn’t know about, and the fact that the hospitalists didn’t ignore the alerts makes us think we are on to something.”
Until the data are analyzed, Dr. Fogerty says he won’t know if the alert system improved patient outcomes. What he does know is that the project took very little to launch, is under budget and has had great buy-in with little to no pushback.
“We have so many flags and sign-this-and-sign-thats that any time people hear the words ‘alarm’ or ‘flag’ or ‘page’ or ‘warning’ or ‘trigger,’ they bristle, and we have all of those in this project,” Dr. Fogerty says. “But we have designed this to minimize your alarm fatigue. You know that when it goes off, we really need you to do something right away, and we aren’t asking you to do much, either.”
Putting the system in place was also easier than expected. Dr. Fogerty worked with one IT expert on a simple tweak in the EHR so the alert goes to a pager, rather than being displayed on a computer screen. The next step will be to revisit the criteria set to see if some should be added (or subtracted) to improve alert specificity and sensitivity.
The team may also rethink another tradeoff it made to minimize alert fatigue: deciding that each patient could have only one sepsis alert fired during his or her hospitalization. “As we improve our criteria, we may change that,” Dr. Fogerty says.
The project team, however, plans to move cautiously. The most important component of the initiative was the “conscious decision that every alarm that goes off has to result in something.”
Deborah Gesensway is a freelance writer who covers U.S. health care from Toronto.