Battling Sepsis with Big Data

Fighting sepsis is one of the most common and challenging problems facing hospitals. Members of the UNC Health Care ISD team have been working hard to ensure that Epic@UNC is equipped with the best tools possible to help providers identify and treat this deadly infection.

When responding to a potential case of sepsis, time is of the essence. Calling a code sepsis means that a team is mobilized that is equipped to respond aggressively to this deadly disease with a fluids and other interventions.

But behind the scenes, computer algorithms in Epic@UNC are constantly running, on the lookout for the signs that someone may be developing this deadly disease or otherwise deteriorating in a way that busy providers in the hospital may not immediately pick up on.

Since before the launch of Epic@UNC the ISD team has been working on digital tools that can provide early warnings when a patient’s vital signs begin to indicate that there may be a problem. As Epic@UNC gathers data, an algorithm is running in the background; when certain criteria are met, it triggers a Best Practice Advisory (BPA) to let providers know that they should consider sepsis as an emerging feature of the patient’s diagnosis.

“The BPA we are using now is based on a logic algorithm we first started developing in the ED a few years ago to identify patients with systemic inflammatory response, which is the first portion of sepsis,” said Daniel Licatese, an application systems analysts for UNC Health Care who helped to design the tool. “We were taking kind of a wide berth in capturing things because we didn’t want to miss anybody, but we also wanted to limit the number of false positives. As we’ve developed it, we’ve continued to narrow the criteria to eliminate as many of those false positives as we can.”

These tools are not a substitute to an active and engaged clinical reasoning. But if designed well, they can be an invaluable supplement to it.

“When the education for this went out, one of the things that we stressed was that it’s an alert, it’s a tool, you cannot avoid using critical thinking,” explained Licatese. “But we also recognize that nurses, physicians, pharmacists, have a 1,001 things more that they have to do in a day than what might have been the case 20 years ago. They might not have the time to focus on all the little nuances, so we try to use technology wherever we can to call attention to the things they might not otherwise notice. That’s what the goal of these tools is.”

This is not a feature that Epic came with out of the box and the Epic@UNC team has been meticulously refining the criteria by which this BPA identifies at risk patients. Too broad a criteria set, and the message will appear constantly, too narrow and it may not appear when it should. 

The development process has involved a slow refinement of the criteria that would trigger a BPA and has also taken into account feedback from clinicians.

“It’s been a very long process, and it’s a very agile process,” said Kara Shore, an IT project manager for UNC Health Care. “There is so much more to this than just build it and leave it alone. It’s build it and see how it’s doing – continuous improvement. So that’s where it’s been good to see things getting better and the response from the clinicians.”

Behind the scenes, how often the BPA is triggered is also measured against how often sepsis is actually diagnosed.

If the conditions that trigger it are too narrow, it won’t trigger often enough to be useful.

In order to further refine this tool, ISD has been running its newest version behind-the-scenes, to see how well it is able to accurately diagnose sepsis, and so that the tool will be as accurate as possible once it goes live.

Once triggered, the BPA directs providers to an order set designed to help them keep track of all the protocols necessary to help someone recover from sepsis. The order set is designed to work at hospitals across the system. Developing a unified order set for this purpose has been a process of gradual expansion, explains Amy Jeroloman, RN, MSN, an application systems analysts for UNC Health Care.

“We started by vetting it across disciplines and services at the Medical Center. From there we’ve had different phases of development. After the Medical Center was online, we went to Rex and Chatham, and from there we started to expand it to some of the other facilities.  And everybody is on a different formulary and they have different policies and procedures so we’ve had to accommodate their workflows in a single order set.”

According to Mauri Williams, RN, Chief Nursing Officer for UNC Health Care, the main obstacle to overcome is developing an order set that is specific enough to be effective wherever it is deployed but general enough that it can be operationalized across the system.

“We give these hospitals the basic recipe, but sometimes the ingredients change. One hospital might have all the nurses draw the blood, another might have the phlebotomists draw the blood. Every physician and every nurse – every provider type – has access to this order set so these hospitals are able to operationalize this order set in the way that is most effective for them.”

Part of the challenge of implementing an order set like this is ensuring that it complies with CMS regulations. These regulations are designed to provide clear guidelines for providers, but in some cases, as when the patient’s condition is particularly complex, following these guidelines can cause more harm than good.

“The regulations are written as yes or no, pass or fail. There is no in between,” explains Williams. “They are trying for all the right reasons, and basically it’s to get us all moving in the right direction and to recognize this horrible condition. But in certain cases you do want to fail. You have certain patient populations where if you do that you are more likely to kill them with the treatment, than sepsis. And this is one of those instances that we need our providers using their clinical decision-making skills to make the right choice for their patients.”

The other challenge of implementing a tool like this is getting buy in from the providers and other stakeholders from across the system. Williams argues that if providers don’t understand the facility of the tool, it can’t do the good it was designed to do.

“The order set is a great tool but we also need to make sure providers are putting it to good use. Unless people know to use it, how to use it, why to use it, it means nothing. We can build all this stuff, but there are so many other things going into understanding if it’s useful.”

In order to continue refining the tools so that it works as well as it can, the ISD team has worked hard to ensure that the reporting provided is able to offer an apples-to-apples comparison of the data that are being gathered.

“We are able to gather lots of numbers, but what do the data mean?” asked Shore. “It took a long time to get everyone reporting the same numbers and we still struggle but that is one of the things across the hospitals where it was important to have a system level look. Now that these hospitals have been reporting every month at the system level for over a year, we can start to get a better sense of the big picture.”

Gathering all of this data means more work for providers, but, Licatese argues, it also means better outcomes for patients. “Yes, this is a change in their practice, but we can guarantee this is going to help providers detect those patients earlier, which means better care.”

-by , UNC Health Care

Filed under: