Johns Hopkins Hospital in Baltimore has adopted a new decision-support tool that enables nurses at triage in the ED to better differentiate patients based on acuity level. The electronic tool, dubbed e-triage, is able to quickly factor in a patient’s medical history, vital signs, and chief complaint, and compare this information to thousands of other patients. The tool delivers a recommended acuity score, ranging from 1 to 5. Triage nurses choose to accept the e-triage score or override it based on their observations and clinical judgment. The approach has replaced the emergency severity index (ESI) system, which is used widely in triage in EDs across the country.

  • Investigators at Johns Hopkins School of Medicine developed e-triage to mitigate the negative effects of crowding.
  • Although ESI is intended to differentiate patients based on acuity, investigators found that close to 70% of patients who present to EDs in the Johns Hopkins system were categorized as level 3, a group that typically includes very sick patients who will wind up in the ICU as well as patients who will be discharged within two hours.
  • The e-triage tool has reduced the level 3 group to 50% of patients. It has also increased the number of patients categorized as level 4 and level 5 by 10%.
  • In practice, 85% of the time the triage nurses agree with the score recommended by the e-triage algorithm, and 15% of the time they will override it.

The emergency severity index (ESI) is widely used in triage to determine the acuity level of patients so that care for the sickest patients, the level 1s and 2s, is expedited while the level 4s and 5s are deemed safe to wait. However, is there a better approach to triage that takes full advantage of IT without adding any time to the equation?

Investigators at Johns Hopkins University School of Medicine (JHUSM) think they have developed just such a method in e-triage, a tool that has been implemented fully at Johns Hopkins Hospital for more than a year and is in the process of being implemented at other sites, too.

The primary impetus for the tool was to mitigate the negative effects of crowding, a problem that presents implications for patients with time-sensitive needs, explains Scott Levin, PhD, an associate professor of emergency medicine at JHUSM. “When their care is delayed, that can result in a worse outcome,” he explains. “But [e-triage] was also developed to improve satisfaction, mainly for the lower acuity group that is being queued and waiting for resources that [these patients] may not need.”

Differentiate Patients

While ESI is intended to differentiate patients based on acuity, Levin notes that investigators found that nearly 70% of patients who present to EDs in the Johns Hopkins system were categorized as level 3.1

“When 70% of patients are in the same bucket, you are not really [differentiating],” he says. In fact, the group of level 3 patients typically includes a mix of people, some of whom are very sick and may end up in the ICU or even die during the encounter, as well as patients who are not as sick and will be discharged within two hours, Levin observes.

“That is where we started from [in developing e-triage],” Levin explains. “And what we did is develop a decision-support tool that looks at predicting patient outcomes. The predicted risk of these outcomes is what supports a triage nurse’s decision.”

When a patient arrives and is seen by a triage nurse, the nurse will gather all the same triage information that he or she would gather with ESI, including the chief complaint and all the patient’s vital signs. “Once the nurse enters all of that information into the EMR [electronic medical record], the e-triage machine learning algorithm will issue back a triage score on the screen,” Levin explains.

The score is derived from an analysis of hundreds of thousands of patient records to calculate the patient’s likelihood of dying, going to the operating room for an emergent, time-sensitive intervention, going to the ICU, or being admitted to the hospital. If the patient is at high risk for any of those outcomes, that will translate into the recommended triage level, Levin says. The triage nurse can either agree with the score that the algorithm is recommending or override it based on clinical judgment. For instance, while the algorithm can quickly factor in a patient’s full medical record and compare it to thousands of other patients, it cannot assess the patient’s appearance. In practice, 85% of the time the triage nurses agree with the score recommended by the e-triage algorithm, but 15% of the time they will override it. In those cases, triage nurses will record in the system why they overrode the recommended score.

“Through that feedback, over time we have developed and evolved the system,” Levin notes.

Expedite Triage Decisions

During the 12 months that e-triage has been in use at Johns Hopkins Hospital, the group of level 3 patients has been reduced from 70% of all patients to 50% of all patients.

“We have also been able to increase our level 4s and 5s,” Levin notes. “The patients that are fast-tracked are routed differently through the ED [because they do] not need a bed or as many resources or as much of a workup. We have increased those fast-track patients by about 10%.”

The level 1s and 2s are identified at a higher rate, too, expediting care to these patients. “We have gotten them out of the level 3 bucket ... which is good because [the level 1s and 2s] don’t really wait, and there is evidence that in these types of patients, if they wait they can deteriorate,” Levin says.

Of course, physicians can later upgrade or downgrade a patient’s acuity level once there is more information, but Levin maintains that e-triage represents a more outcomes-based way of distributing patients up front without adding any time to the process. In fact, triage now takes less time than it did before with ESI, Levin observes.

“Triage should take, at most, five minutes, and we’ve gotten it down to less than that,” he says, noting that this is the case even with the additional rigor that is applied to the process. “Nurses generally don’t have time to look through the patient’s medical history or anything like that. They may glance at the variables. But what this tool does is actually search through [a patient’s] entire medical history, so it searches their problem list and their medical/surgical history, and it [considers] the vital signs and the chief complaint. It compares that information with all of the patients that are in the same hospital.”

The result of implementing e-triage has been a 40% shift in how patients are distributed, Levin observes. For example, a young patient who arrives complaining of abdominal pain might have been triaged to a level 3 under ESI, but with all the information that is considered by the algorithm, such a patient might be considered at higher risk for appendicitis and triaged to level 2 with e-triage, he observes.

In another example, a young patient with a headache who might have been triaged to level 3 and designated to wait several hours for a bed under ESI might be triaged instead to level 4 where the patient will not require a bed and can be expedited through the fast-track area.

One significant change that Levin has noticed since Johns Hopkins Hospital began using e-triage is that elderly patients tend to be triaged differently than when ESI was used. “Using this tool, they are triaged to a higher acuity level, we believe, appropriately because it is in line with the evidence,” Levin says.

Encourage Feedback

Long before implementing e-triage, Levin worked with a core group of triage nurses who reviewed the tool and offered feedback, explains Heather Gardner, MSN, the program manager of clinical informatics for the ED at Johns Hopkins Hospital. “After these five or six people used the tool for a while, Scott came to the triage meetings to explain what the tool is and how it works,” she says.

Nurses continued to provide feedback as the tool was refined further.

“We were familiar with the tool, but it was a slow implementation,” observes Sophia Henry, MS, RN, the adult ED triage coordinator at Johns Hopkins Hospital. “We were hesitant to start using e-triage because, for us at Hopkins, getting to be a triage nurse is indicative of [having] great clinical skills ... and we were concerned about whether or not this machine learning program would eliminate the need for triage nurses to even be present.”

However, the nurses realized early that the process worked best when the e-triage tool was paired with a skilled clinician.

“[The tool] will give you an acuity, but it doesn’t see the patient, so from the computer’s perspective a patient might be a rock-solid 4, but we are looking at the patient and see that the patient is sicker than [a level 4], so we have the ability to change the acuity,” Henry explains.

In fact, using the new tool has helped the triage nurses feel more comfortable designating patients level 5, something they rarely did when using ESI.

“Before we went live with e-triage, we had less than 1% of patients at level 5 acuity. Now, almost 10% of our patients are level 5,” Gardner notes. “It made a huge jump in the number of patients that are in that lower acuity bucket.”

It definitely took some time to make the switch from ESI to the e-triage tool, but the nurses have come to trust the new approach, Gardner observes. Henry agrees, noting that she was a late adopter.

“I was very resistant to starting this, and I love it now,” she says.

Customize to Fit Needs

While the algorithm in place at Johns Hopkins Hospital is working well, it likely would not translate directly to another ED because every ED is different, Levin observes. “They all have unique objectives and unique patient populations,” he says.

However, the e-triage tool is designed to adapt to such differences. “That happens through the machine learning aspects of this,” Levin adds. For instance, two other hospitals in the Johns Hopkins system that are in the process of implementing e-triage, Bayview Medical Center in Baltimore and Howard County General Hospital in Columbia, MD, are working with algorithms that are built and tuned to those facilities.

“At Howard County General Hospital, if you come to the ED by ambulance, you are a pretty sick patient relative to the rest of the patients they see,” Levin explains. “If you come to Johns Hopkins Hospital by ambulance, you may not be that sick. You may have used the ambulance as a mode of transportation to get to the ED to get some food to eat, so the meaning of the ambulance is pretty different to the physicians and the clinicians practicing at those facilities.”

The patient populations exhibit different characteristics as well. For example, Levin notes that Johns Hopkins Hospital sees a lot of patients with substance use problems as well as social and psychological issues. The conditions that people present with at Howard County General Hospital tend to be different, and the algorithm must adjust to those, he says.

Levin acknowledges that well-functioning EDs that have been able to minimize waiting have no need for a tool like e-triage because everybody gets the care they need right away, but he observes that for many EDs, problems like crowding and waiting have only gotten worse in recent years.

“The reality is that something like 80% of EDs will report that they are chronically crowded ... and these triage decisions have to be made.”


  1. Levin S, Toerper M, Hamrock E, et al. Machine-learning-based electronic triage more accurately differentiates patients with respect to clinical outcomes compared with the emergency severity index. Ann Emerg Med 2017 Sep 6. pii: S0196-0644(17)31442-7. doi: 10.1016/j.annemergmed.2017.08.005. [Epub ahead of print].


  • Heather Gardner, MSN, Program Manager, Clinical Informatics, Emergency Department, Johns Hopkins Hospital, Baltimore. Email: hmacpher@jhmi.edu.
  • Sophia Henry, MS, RN, Triage Coordinator, Adult ED, Johns Hopkins Hospital, Baltimore. Email: swalke16@jhmi.edu.
  • Scott Levin, PhD, Associate Professor, Department of Emergency Medicine, John Hopkins University School of Medicine, Baltimore. Email: slevin33@jhmi.edu.