A Better Way to Detect Sepsis in Kids in the Emergency Department

A New Sepsis Alert System

Fran Balamuth, MD, PhD, MSCE


October 09, 2017

My name is Fran Balamuth. I am an assistant professor of pediatrics at the University of Pennsylvania Perelman School of Medicine and an attending physician in the emergency department (ED) at the Children's Hospital of Philadelphia (CHOP), where I also do research on pediatric sepsis.

I am excited to have the opportunity today to tell you a little bit about our sepsis recognition program. As you likely know, sepsis recognition is a key challenge in the ED. Lots of children present to EDs across the United States every year with fevers and infectious complaints, the vast majority of which are not septic shock. One of our challenges—like trying to find the proverbial needle in the haystack—is figuring out which of those children from among those hundreds of thousands who come to the ED with fever are going to require intensive care for septic shock.

We have been interested in this question for a long time here at CHOP and have done several interventions over the past several years to try to improve our ability to recognize children with septic shock. When we started this, we were concerned about recognizing children using any sort of electronic alert. Most of the electronic alerts for sepsis depend on abnormal vital signs: tachycardia and hypotension.

As you know, low blood pressure is a rare and late finding in children. Children most commonly present with compensated septic shock, where they have tachycardia alone. However, tachycardia with fever is an incredibly common phenomenon and is present in many children who do not have sepsis.

When we started our sepsis treatment program here at CHOP, we purposefully did not institute an electronic alert because we were worried that it was going to fire all the time and that there would be significant alert fatigue. In the beginning, we were only trying to identify patients with sepsis based on our own clinical judgement, looking at each patient who presented with an infection and determining whether we were worried about their perfusion and mental status, and deciding whether they needed to be treated with our sepsis protocol.

We tracked our performance using that method and found that we were missing about 20% of patients who had septic shock. Then we started to wonder whether there was an electronic alert that might help us better define these patients.

Before we implemented a live electronic alert, we wanted to do a simulated alert. We developed an in silico study where we took an imaginary alert and applied it retrospectively to a year and a half's worth of patients with fever in ED. What we found in that study,[1] which was published a couple of years ago, was that the putative or possible electronic alert would have found the vast majority of patients who we were missing clinically. At the same time, we found that patients who were being treated with our sepsis protocol were doing better clinically compared with patients who were being missed with this protocol.[2] They were getting their antibiotics and fluids faster, and their organ dysfunction was resolving sooner. We were very motivated to try to improve the sepsis recognition process at the beginning of the ED stay.

After finding that the in silico alert detected patients who were being missed clinically, we decided to implement an electronic alert prospectively and evaluate whether it could improve our ability to recognize children with sepsis.

We built a two-stage alert in EPIC, our electronic health record. The first stage fires if the patient has either low blood pressure or tachycardia. This prompts the triage nurse to answer whether he or she is concerned about infection in this patient. We wanted to be clear that we were looking for a concern for infection, and not just fever, because many patients with septic shock do not actually have a fever at the time of ED presentation, either because they took antipyretics at home or because they just have not mounted a fever yet. We put in this "concern for infection" question instead.

If the patient has a high heart rate or a low blood pressure, and there is concern for infection, then the triage nurse is prompted to answer three more questions: one about underlying high-risk conditions, one about perfusion, and one about mental status. If the patient has both high heart rate and one of these three abnormalities (a high-risk condition, altered mental status, or altered profusion), this prompts the second phase of the alert—what we call a "sepsis huddle," which means that a senior ED attending or fellow comes to the bedside of the patient and decides whether the child needs to be treated on our sepsis protocol.

We analyzed carefully how often this alert would go off. The first stage goes off in about 10% of our patients, and about 10% of the patients who fire the first stage alert go on to have a sepsis huddle. We held a sepsis huddle for just over 1100 patients during this time period.

We found over the first year of implementation of the alert that it improved our ability to recognize sepsis. The alert has a sensitivity of 86% when combined with additional patients who are treated for sepsis based on the clinician judgement at the bedside. These efforts allowed us to improve our sepsis recognition—from 83% of patients with sepsis who needed treatment at the beginning of the study, to 96% at the end of the intervention period.[3]

It's important to note, however, that the electronic alert, although it did improve our ability to recognize patients with sepsis, still required the bedside evaluation and clinician judgement to get that sensitivity up into the 90% range. We continue to strive to improve these alerts. What we have now gets us part of the way there, but not the whole way there.

The other thing is that it's very important to have a process at your institution for recognizing patients with sepsis, and that process will likely include some sort of electronic alert, as well as educating physicians at the bedside. Those two things together are needed to improve recognition of children with sepsis.

This is in line with the new pediatric sepsis guidelines[4] which came out just over the summer, which recommend that every hospital have some sort of sepsis recognition plan in place and that they have a protocol to treat sepsis patients. We are hoping that this alert will be implemented in other hospitals. Because we only have data from CHOP, the generalizability at this point is limited, but we do have active studies going on in other sites to evaluate generalizability. We continue to study this process. This alert has now been in place for almost 3 years, and our missed sepsis detection rate is still below 5%, which we are very happy with, but we continue to strive to improve these alerts and we are hoping to use some more advanced methods to help us build smarter alerts in the future.