Everyone who works in U.S. hospitals knows that sepsis occurrence is a huge issue in healthcare, with the National Institutes of Health’s National Institute of General Medical Sciences estimating that over one million people experience sepsis in the U.S. every year, and between 28 and 50 percent of them dying. Not all of those cases occur inside hospital facilities, of course, but sepsis remains a major cause of hospital mortality.
One organization whose leaders have been attacking this problem strategically and systemically has been the UF Health Jacksonville Health System, the teaching hospital and medical system for the University of Florida in Jacksonville. Leaders at the 695-bed academic medical center have been working hard to reduce sepsis. And for their strategic initiative that has leveraged IT to improve patient outcomes in this critical area, the editors of Healthcare Informatics have named UF Health Jacksonville one of the semi-finalist organizations in the Healthcare Informatics Innovator Awards Program.
As UF Health Jacksonville’s leaders noted in their Innovator Awards submission, “The goal of this project was to provide a customizable, smart, reliable, surveillance and alert program that accurately notified specific groups of providers when early signs of sepsis are detected without increasing the likelihood of alarm fatigue. Early detection of sepsis leads to early management which has been shown to decrease mortality. Azrael is designed to be smart. The technology decides when it is reasonable to send an alert. The program is set up to send one alert per patient for a 24-hour period. However, since it is monitoring every patient, every hour, if it sees a worsening trend within a four-hour period on any of the parameters it is tracking, Azrael will send another targeted alert. Alerts are sent to identified responders through a page system. The tool is also built to be smart enough to decide whether it is appropriate to send an alert based on the patient’s location (ED, ICU, OR). The location based logic gave each area of the hospital the ability to customize the alert firing threshold depending on preference and ability to respond. The tool also identifies post-op status (immediate post-op, post-op day 1, etc). This solves the challenges of sepsis detection in surgical patients. Azrael tracks all the data it collects and it creates a daily summary of all the alerts it has sent, the abnormal values that triggered the alert and whether or not the alert was confirmed by the frontline providers. Currently, an add-on module is being developed to provide Azrael with the ability to track the implementation of the various CMS Sepsis Core Measure (SEP) components. Once in production, Azrael will be able to send out reminders to physicians about the timeliness of the various components needed in the SEP Core measure. This will also give Azrael the ability to do automated abstraction for the sepsis core measure. Azrael is customizable and can be used with any electronic health record. It was built by a physician with input from stakeholders, frontline staff and physicians.”
Importantly, UF Health Jacksonville’s leaders noted in their submission, “This project has had a significant impact on patient outcome. Early detection of sepsis is critical for early management. While some systems signal a provider through the electronic record that a patient is deteriorating, these systems only work if someone who can effect change in clinical management sees the signal. Azrael actually sends a page to someone that can go to the bedside and manage the symptoms. This of course can lead to alarm fatigue so Azrael is programed to only alert providers when a substantial and clinically relevant change occurs. This insures that clinicians pay attention to the alert and respond by going to the bedside. Results were measured by comparing our sepsis mortality index and sepsis length of stay index pre and post implementation. The sepsis mortality index (observed/expected ratio) in 2014 was 0.93. Azrael was rolled out in November 2014. In 2015 the sepsis mortality index dropped to 0.84. Similarly, the sepsis length of stay index was 1.25 in 2014 and decreased to 1.11 in 2015. In addition to the index the trend in actual sepsis mortality rate is also steadily decreasing.”
Recently, Susan Hendrickson, division director for quality at the UF Jacksonville Health System, and one of the leaders of the initiative, spoke with HCI Editor-in-Chief Mark Hagland about it. Below are excerpts from that interview.
Can you share with us the origins of this initiative?
The origin really came about when we started looking at opportunities for improvement around mortality. So we drilled down into the data, and saw sepsis as an opportunity. And the thing is, once we know you have sepsis, we do a pretty good job of saving your life. But as we started drilling down into the data, we found that we weren’t doing a good job of early identification of patients with sepsis. So we tried to figure out how we could identify the patients and where they were, and that really was the genesis of the project; that was back in 2014.
Let’s talk about process. It sounds as though a lot of people needed to come together collaboratively from different disciplines and areas to make this happen, correct?
Yes, absolutely. It really started with the physicians looking at this. And I have abstractors who are very talented, and they met with the physicians to figure out how to identify patients. So, what are the clinical indicators of pending sepsis? And they came up with four classic measures—temperature, heart rate, respiratory rate, and white blood count, plus bans (part of your blood count), and blood pressure, and other clinical indicators. But then again, the clinical indicators are different. And what makes this particular project particularly innovative. This is not a one-size-fits-all situation. The criteria in different areas like medical-surgical and ICU and other areas, can be different. The criteria are different for different patient populations in different clinical situations.
So this tool that we call Azreal, alerts somebody through a pager system. So when a patient shows certain criteria, when certain things fire, that sends a page to a certain person to get them to go to the bedside.
Is the key clinician who is alerted a physician? A nurse?
It’s different depending on the clinical area. We do have a rapid response team staffed by nurses, and sometimes also at the same time to a physician. So they’re paged and they go to the bedside. And once they go to the bedside, they make that decision. If it is a patient with sepsis, an order set fires.
And that involves clinical judgment.
Yes, and this tool looks at every patient every hour, and detects changes every hour. And the other thing is, we fired it to a pager, because some systems send up alerts in the EMR, but if you’re not inside the EMR at that moment, you won’t know. And if you’re not with Mrs. Smith at 8 PM, you’re probably not there anymore. So this tool looks at every patient every hour, including in the emergency department, and alerts the team.
How often does the alert fire?
It runs 24/7, every single hour. Every day, we actually get a report to see that. So let’s look at yesterday. About nine or ten times yesterday. We run an average daily census of about 500 (550 beds). So, six to 12 patients a day have an alert. I’d have to look up how many are confirmed and not confirmed. But we tweaked the tool—we didn’t want a lot of false alarms. We had this great team of physicians working with us who helped us to refine the tool.
Were there any particular challenges in architecting this solution?
Oh yes, there were lots of challenges. The biggest challenge was tweaking the system so that when you get an alert, it doesn’t turn out to be one that you ignore. And also, to tweak it for specific patient populations. So there was a lot of really good clinical work refining the tool.
Did it evolve forward about at the pace you had expected?
Yes. It certainly didn’t go faster than we’d expected. I don’t think anybody had any idea that it could be created instantly. And when we started, we started with a smaller population, and we added the ED later. So I don’t know that we had expectations. And we’re an academic medical center, and we have a lot of stuff going, all the time.
What should CIOs and CMIOs know about this kind of initiative?
That this is possible. The fact that we’ve done this is a success. And we did it for sepsis; but you could do it for anything. It’s just a matter of dialing down on criteria. If you survey your patients for certain criteria and then refine those criteria and apply them, it has a lot of potential as a process.
What have the biggest lessons learned been in all this?
I think the ability to alert somebody in real time to changing clinical conditions, can significantly impact outcomes.
And of course, being successful requires teamwork, and really bringing everyone together collaboratively?
Absolutely. This is not something anyone can do in isolation. We needed the developer, obviously; physician input and participation, and great work from our rapid response team, which is nurses; and among the physicians, from many specialties.
