Taming a Toxic Threat

New AI tools send lifesaving alerts on sepsis.

Taming a Toxic Threat

New AI tools send lifesaving alerts on sepsis.

By Joe Kita

August 1, 2024

Gil Munoz was riding his mountain bike in Altadena, CA, when he took a spill. He got up, brushed himself off, and pedaled home, thinking nothing of it. But in the next few days he developed a foot wound and his left leg swelled. He went to the emergency department at Adventist Health White Memorial in Los Angeles, where doctors found multiple foot fractures, referred him to a podiatrist, and prescribed antibiotics. Everything was under control.

Two days later, Munoz returned to the ER. He'd developed a hand tremor. He didn't look sick, and his podiatrist had told him just that morning that his foot wound looked good. Fortunately, there was an experienced triage nurse on duty who consulted KATE, a sepsis specialist. KATE analyzed Munoz's situation. Based on his blood pressure and his medical history, which included diabetes, KATE warned that he might have sepsis, a life-threatening response to infection.

But KATE isn't a doctor. Or a nurse. Or any human. KATE is one of several new artificial intelligence (AI) tools that some U.S. hospitals are starting to use. They tap hundreds of data points to warn nurses and doctors of the potential for sepsis in hospitalized patients.

Sepsis, which draws its name from an ancient Greek word meaning putrefaction or rot, affects 1.7 million people in the U.S. each year. It has long plagued hospitals. It's their top cause of death: 1 in 3 people who die in a U.S. hospital had sepsis while in the hospital. Sepsis-related hospitalizations have been rising: up 20% from 2016 to 2019 and another 16.5% from 2019 to 2021, with 2.5 million hospital stays. And it's the most expensive condition that U.S. hospitals treat.

Will AI help hospitals save people like Munoz who come down with sepsis?

A Fast-Moving Threat

Sepsis stems from an infection — bacterial, fungal, or viral — that gains traction in the body. It happens when the body's response to the infection surges into overdrive. It can happen in any number of situations, whether you're an older adult with a vulnerable immune system, a pregnant person, or rarely even a healthy young adult who gets an infection from something like contaminated tattoo ink or a tampon. If it's not found and treated, sepsis can worsen very quickly through a series of chain reactions leading to tissue damage, organ failure, and death. And it's often hard to spot when it first takes root.

mobile infographic

desktop infographic

"The initial signs of sepsis can be almost undetectable and easy to miss," says Albert Wu, MD, MPH, an internist and patient care expert at Johns Hopkins. "It is one of the most difficult diagnoses to make because it is so common and so deadly."

Sepsis is so lethal because of its speed and how challenging it can be to diagnose early. There's no single test to detect it. Research shows that missed or delayed sepsis diagnoses in hospitals may range up to 20.8% of people who come in with other serious health problems, including heart attack or stroke, infection, and cancers.

"I couldn't believe what I was going through. It was all very surprising."

-- Gil Munoz

After KATE's alert, the medical staff began treating Munoz for sepsis, doing more tests and X-rays, which detected critically high lactic acid levels and the start of gas gangrene, an immediate, life-threatening infection of the soft tissue, in his ankle. He was admitted to the ICU and intubated.

While doctors couldn't save his leg (it was amputated below the knee), because of KATE's early vigilance and the human medical team's response to it, they saved his life.

"I couldn't believe what I was going through," says Munoz, 58, a retired carpenter. "It was all very surprising."

AI and machine learning may be unfamiliar to many patients and don't replace doctors and nurses. Still, as Stephen Liu, MD, medical director and chair of emergency medicine at Adventist Health White Memorial, who helped develop KATE, points out, "AI can do some wonderful things. In Mr. Munoz's case, it saw a pattern based on his medical history and vital signs and alerted the team there was something lurking that we weren't quite seeing. That enabled us to find what was going on much quicker than we probably would have otherwise."

Early Warning Systems From AI

KATE, which is being marketed by a company called Mednition, isn't the only AI-driven sepsis-detection system. Many others are under development, being field tested, or already running in hospitals nationwide.

One is Targeted Real-Time Early Warning System (TREWS), developed at Johns Hopkins by a team including Wu and Suchi Saria, PhD, a data scientist and AI learning expert. According to research done at five U.S. hospitals and published in Nature Medicine, TREWS caught 82% of sepsis cases, shaved 1.85 hours off the time to start antibiotic treatment in cases where doctors confirmed sepsis within 3 hours of the TREWS alert, and reduced death rates by nearly 19%.

"TREWS does not make the diagnosis of sepsis itself," Saria says. "It's more of a partner or a consultant. So far, we've deployed it in health care systems ranging from the Cleveland Clinic to small community hospitals."

COMPOSER is another AI-driven early-warning system. Gabriel Wardi, MD, an emergency medicine doctor at UC San Diego Health, started developing it in 2018 with colleagues. It's integrated into the existing hospital workflow system, so when someone checks into the emergency department, COMPOSER continuously monitors more than 150 patient variables -- such as vital signs, lab results, medications, demographics, and medical history -- and then signals the care team if sepsis is likely. Since COMPOSER went live in the ER on Dec. 7, 2022, sepsis death rates at his hospital have dropped by 17%, according to Wardi.

IntelliSep combines deep machine learning and a serum blood test to detect sepsis. It's in use at Our Lady of the Lake Hospital in Baton Rouge, which has the busiest emergency department in Louisiana. The blood test, which focuses on the shape and structure of white blood cells to estimate the risk of sepsis development, takes about 8 minutes and has been cleared by the FDA for marketing.

"IntelliSep has been a game-changer for us," says Mark Laperouse, MD, the medical director of emergency services at Our Lady of the Lake. "It is saving lives, money, and time. Since we started using it, the length of stay in the hospital has gone way down and the mortality rate for septic shock is in the single digits, which is unheard of in a facility our size," which has about 80,000 visits a year.

Overall, these AI systems are helping evolve sepsis care from reactive to proactive. Most of these models haven't been in hospitals very long, so there aren't a lot of studies on them. They're likely to become more common and eventually become widespread in hospitals. But there's no one-size-fits-all system. "The model that works best is the one that fits best into the hospital's existing workflow," Wardi says.

For patients, "this isn't something, if you're feeling sick, that you should start shopping around for," Wu says. "By that time, it'll be too late." In fact, even when you're hospitalized, patients probably won't know if a hospital uses AI to help detect sepsis because these tools run in the background.

But you can still do a lot to make yourself more aware of sepsis and its symptoms.

'I Missed It in Myself'

Karin Molander, MD, is an emergency room doctor at Mills-Peninsula Medical Center in Burlingame, CA. She's been involved in sepsis education and research since 2007 and developed the condition herself while delivering her first son.

"Even as a physician, I missed it in myself. I thought I was way too young and healthy to have sepsis."

-- Karin Molander, MD

"When I give talks on sepsis, I tell the audience, 'Look to your right. Look to your left. One of you will be impacted by sepsis," Molander says. "So it's in everybody's best interest to understand the signs and symptoms."

Yet in a 2023 online survey commissioned by the nonprofit Sepsis Alliance, only 63% of the 2,512 people who took the survey were aware of the term sepsis, and just 15% could identify four of its common symptoms:

  • High or low temperature
  • Infection
  • Mental decline
  • Extremely ill with pain, discomfort and/shortness of breath

"Even as a classically trained physician, I missed it in myself," Molander says. "I thought I was way too young and healthy to have sepsis. Fortunately, a well-trained maternity nurse recognized the signs and said, 'We need to deliver your baby as quickly as possible and get you some antibiotics.'"

Adults 65 and older with chronic medical conditions and weakened immune systems are at the highest risk. But sepsis can affect anyone, even babies, children, and fit young adults. If you suspect you might have sepsis and aren't sure if your hospital is running one of the new AI systems, Molander recommends asking your doctor this one simple question: "I'm concerned about sepsis. Can you tell me if it has been ruled out?" This will at least get them thinking about it.

Despite such progress, some doctors and nurses have their doubts about AI. In a 2023 Medscape survey of 1,043 doctors, 65% were at least somewhat "concerned" about AI making diagnosis and treatment decisions, although 56% were enthusiastic about having it as a help.

"Perhaps the best analogy is to think of it as an air traffic controller. It's better for pilots, but you still want a human flying the plane."

-- Albert Wu, MD, MPH

"While these AI-driven systems may in fact do a better job of predicting sepsis than physicians," Wu says, "the system plus the physician does even better. Perhaps the best analogy is to think of it as an air traffic controller. It's better for pilots, but you still want a human flying the plane."

"An AI system provides a gentle suggestion rather than a mandate," Wardi says. "It's a second set of eyes watching patients. Some nurses and physicians, the early adopters, think it's super cool. ... But at the opposite extreme there are those who say, 'I'm not using it and I don't trust it.'"

But among the experts interviewed for this story, the collective opinion is that such an attitude will change - and quickly.

"It's a breakthrough," Molander says. "I think machine learning may prove as vital to sepsis care as the EKG has proved to be in diagnosing a heart attack. AI-driven systems can change the nature of health care." Indeed, Saria points out that Bayesian Health, a company she founded, "is building TREWS to be not just for sepsis but for a variety of other complications and adverse events."

As for Munoz, although he hasn't returned to carpentry, he did get a prosthetic leg and, with the support of his family, is living on his own. "I'm very grateful that my doctors were able to respond so quickly," he says. "I wasn't sure what was going on or what was going to happen, but they saved my life, and now with my prosthetic, I'm walking again."

What Is Sepsis?

Reporter's note: During the writing of this article, my 90-year-old mother acquired an infection from a leg wound that led to sepsis and her eventual death. Doctors immediately diagnosed it in the emergency room without one of these new AI-driven systems. I witnessed firsthand what a terrible condition this is, how it progresses through the body, and how it can resist antibiotics and treatments. Do yourself a favor and familiarize yourself with the symptoms listed in this article. It may one day save your life. — Joe Kita

Reviewed by Melinda Ratini, DO, MS, on June 11, 2024

Slideshow Image Credits:

(1)    Photo Researchers / Getty Images

(2)    iStock

(3)    Taxi

(4)    Juergen Berger / Science Source

(5)    Thinkstock

(6)    Thinkstock

(7)    Science Source

(8)    Thinkstock

(9)    Brand X Pictures

(10)  Thinkstock

(11)  The Image Bank

SOURCES:

Gil Munoz, Altadena, CA.

CDC: "About Sepsis."

Agency for Healthcare Research and Quality: "Overview of Outcomes for Inpatient Stays Involving Sepsis, 2016-2021," "National Inpatient Hospital Costs: The Most Expensive Conditions by Payer, 2013."

Applied and Environmental Microbiology, "Detection of anaerobic and aerobic bacteria from commercial tattoo and permanent makeup inks."

Albert Wu, MD, MPH,  professor of health policy and management and medicine; director, Center for Health Services and Outcomes Research, Johns Hopkins Medicine.

Diagnosis: "An Estimate of Missed Pediatric Sepsis in the Emergency Department," "Rate of diagnostic errors and serious misdiagnosis-related harms for major vascular events, infections, and cancers: toward a national incidence estimate using the 'The Big Three.'"

Critical Care Explorations: "Diagnostic Delays in Sepsis: Lessons Learned From a Retrospective Study of Canadian Medico-Legal Claims."

National Library of Medicine: "Gas Gangrene."

Stephen Liu, MD, medical director; chair, emergency medicine, Adventist Health White Memorial, Los Angeles.

YouTube: "AI in Action: Surviving Sepsis," Mednition.

NPJ Digital Medicine: "Impact of a deep learning sepsis prediction model on quality of care and survival."

Nature Medicine: "Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis," "Factors driving provider adoption of the TREWS machine learning-based early warning system and its effects on sepsis treatment timing."

Suchi Saria, PhD, the John C. Malone Associate Professor of computer science, Whiting School of Engineering, and of statistics and health policy, Bloomberg School of Public Health; director, Machine Learning and Healthcare Lab; founding research director, Malone Center for Engineering in Healthcare, Johns Hopkins University.

Gabriel Wardi, MD, emergency medicine doctor; associate professor of clinical, emergency medical services, UC San Diego Health.

FDA letter to Cytovale Inc., Dec. 20, 2022.

Mark Laperouse, MD, chief of emergency medicine, Our Lady of the Lake Regional Medical Center, Baton Rouge, LA.

Karin Molander, MD, emergency room doctor, Mills-Peninsula Medical Center, Burlingame, CA; past chair, board of directors, Sepsis Alliance.

Sepsis Alliance: "2023 Sepsis Awareness Survey."

Medscape: "Medscape Physicians and AI Report 2023: A Source of Help or Concern?"

Angus, DC, Critical Care Medicine, July 2001.

CDC: "Sepsis," "Sepsis Questions and Answers."

Mayo Clinic: Diseases and Conditions Sepsis "Symptoms."

MRSA Survivors Network.

National Institutes of Health: "Sepsis Fact Sheet."

Sepsis Alliance: "Amputations," "Sepsis and MRSA," "Sepsis and Pregnancy."

Lieu, V. The Journal of the American Medical Association, May 18, 2014.

Donald M. Yealy, MD, professor, chair, University of Pittsburgh Department of Emergency Medicine, Pittsburgh, PA.