Using Data to Create New Sepsis Diagnosis Criteria

How is an international effort to redefine sepsis diagnosis criteria bringing better care to more kids and laying the foundation for future improvements?

Nearly 20 years ago, a few dozen people sat in a room and made a decision that informed the way doctors diagnose and treat children with sepsis, a host response to an infection that can lead to organ dysfunction. The criteria they developed for a sepsis diagnosis have been used around the world to help providers make decisions about suspected cases of the condition, which affects 25 million children each year. But a lot has changed in 20 years and researchers now have a better understanding of sepsis, data and care across global contexts. For Tell Bennett, MD, an intensivist and informaticist at Children’s Hospital Colorado, and his colleagues, the time was right for an update.

There is no single test that can definitively say someone has sepsis. Instead, providers have relied on a set of criteria that add up to a diagnosis. In 2016, a task force updated the adult sepsis diagnosis criteria. In 2019, a similar endeavor geared toward the pediatric population began. Not only were the pediatric criteria inconsistent with the ones used in adults, but they also didn’t function well in the clinical space, resulting in diagnostic criteria that were too sensitive to be effective.

“It led to a lot of confusion, and so they were not universally used because of that lack of performance,” Dr. Bennett says.

That confusion wasn’t just inconvenient for providers. It was dangerous to patients.

“We need criteria that aren’t too sensitive so we don’t, for example, give kids lots of broad-spectrum antibiotics they don’t need, because then they get other infections that can happen when your microbiome is degraded,” Dr. Bennett explains. “We also don’t want to give them too much resuscitation fluid, which is a cornerstone of sepsis therapy. We don’t want to subject kids to that if they don’t need it.”

Dr. Bennett was invited to join the Society of Critical Care Medicine’s international sepsis task force representing 12 countries and various technical domains, from nursing and emergency department care to intensive care, infectious diseases and informatics. The team, which also included Halden Scott, MD, a pediatric emergency medicine physician at Children’s Colorado, was charged with bringing the pediatric sepsis diagnosis criteria into the modern era. To kick off the work, the group held its first meeting in 2019 in Europe before the COVID-19 pandemic forced it to move to a virtual format.

Over the next five years, the participants worked to hash out a new understanding of sepsis and develop easy-to-follow criteria that could serve people in all different contexts, from big cities to rural towns, and from highly resourced countries to those with less developed medical systems. The fruits of their labor, the Phoenix Sepsis Score and the Phoenix Sepsis Criteria, were published in January 2024. Additionally, the team recently published a four-part series of papers designed to add context to some of the task force’s key decisions.

The importance of informatics

Because of his experience as a pediatric intensive care unit physician and his focus on informatics, Dr. Bennett was perfectly positioned to make a significant contribution to the task force as it worked to bring the Phoenix Score to life. His clinical work put him close to pediatric cases of sepsis on a regular basis, while his informatics specialty helped the team ensure its final product would be data-driven and evidence-based.

Along with his colleague Nelson Sanchez-Pinto, MD, at Lurie Children’s Hospital in Chicago, Dr. Bennett obtained funding from the National Institutes of Health for this work and designed and executed the computing effort for the task force. The team’s first role was to determine the best way to measure organ dysfunction, one of the key indicators of a true sepsis case. Using machine learning, Dr. Bennett took all possible measures of organ dysfunction for each organ system (cardiovascular, respiratory, etc.) and identified which were most accurate.

“Then, we took the best ones and put them in a variety of different machine learning models to see how they could best come together to form diagnostic criteria,” Dr. Bennett says. “Then we translated those models to an integer score, making sure that we didn’t lose accuracy along the way.”

His findings were then brought to the larger group, who spent grueling hours weighing different models against each other to determine which would function best in a real-life scenario. Together, they landed on the now-published criteria and decided that if a patient scores a two out of 13 possible points, an official sepsis diagnosis is confirmed.

Sepsis diagnosis and care across contexts

With those decisions in place, Dr. Bennett’s next step was to ideate ways to put the score into practice. Children’s Colorado is serving as one test environment for this endeavor. His team partnered with the Children’s Colorado Information Technology Division to code the criteria into the background of the hospital’s electronic health record (EHR) and is now testing its ability to accurately diagnosis sepsis. In the next six months, he expects to roll this approach out to other hospital systems, with the understanding that different providers might use the score in different ways.

“Once we’ve confirmed that it’s doing the right thing inside the EHR, we will be able to partner with the clinical and operational teams to figure out how best to change any workflows or surface this information to teams to best impact kids,” Dr. Bennett says. “And we anticipate that process being locally calibrated at each institution that uses this.”

But not all hospital systems have the same technology or the same access to testing. With that in mind, Dr. Bennett’s team also developed a mobile app for low-resource environments that may not have an electronic health record. Testing for the app is underway, and the hope is to roll it out for free use early in 2025.

“They enter whatever elements of the Phoenix Sepsis Score they have information about, and the application would give them the score and then also whether they meet criteria for sepsis or septic shock,” Dr. Bennett says. To ensure accurate diagnosis regardless of environment, the team was careful to build redundancies into its scoring system.

“What we saw is that even in low-resource environments, it was still accurate because they had enough of the elements of the sepsis score that they could get to two points for those kids who actually did have sepsis,” he adds.

Beyond sepsis diagnosis

With the Phoenix Sepsis Score officially out in the world, Dr. Bennett and Dr. Scott are beginning to work toward developing sepsis screening criteria.

The goal is to have a better understanding of the different elements that increase sepsis risk and to develop technology that allows providers to intervene before the condition sets in. So, when kids seek care for something like a simple fever, doctors may get a flag from their electronic health record indicating heightened risk.

Though Dr. Bennett knows there’s more work to be done to dramatically decrease the number of children who die of sepsis each year, the Phoenix Sepsis Score will not only improve care, but create a foundation for future improvements.

“I am very happy and proud to have been a part of the work and very proud of the work that our team here and around the world has done,” he says. “It’s an important step forward, and I hope that it improves the outcomes of kids, but it’s also a part of a process. I think we can have even more impact going forward.”

Citations

1. Schlapbach, Luregn J. et al. “International Consensus Criteria for Pediatric Sepsis and Septic Shock.” JAMA vol. 331,8 (2024): 665-674. doi:10.1001/ jama.2024.0179.
2. Sanchez-Pinto, L Nelson et al. “Development and Validation of the Phoenix Criteria for Pediatric Sepsis and Septic Shock.” JAMA vol. 331,8 (2024): 675-686. doi:10.1001/jama.2024.0196.