Presented by

Hopkins looks to code to identify a 'major and underappreciated' health problem

Suchi Saria, right, an assistant professor of computer science and health policy at Johns Hopkins’ Whiting School of Engineering  and Katharine Henry, a Ph.D student in Saria's lab analyze information on critically ill patients.

Johns Hopkins University researchers have taken to their computers to tackle one of the top killers in the nation's hospitals: sepsis.

Sepsis is a complication of infection. Chemicals released in the bloodstream trigger body-wide inflamation, which can weaken the heart, impede the flow of blood and oxygen, and cause organs to fail. It's difficult to predict, diagnose and treat.


But the Hopkins researchers have developed a model that has shown promise in identifying which patients are likely to develop the condition — often, before the damage begins and when treatments are more likely to work.

If their code is proved effective during a pilot program to begin in the next year at Howard County General Hospital, it could be rapidly deployed to any facility that uses electronic medical records.


"When sepsis treatment is delayed, mortality increases," said Suchi Saria, an assistant professor of computer science and health policy at Johns Hopkins' Whiting School of Engineering, who led a study of the model. "Normally we can't predict sepsis before there is organ dysfunction."

The computer model pores over data taken regularly from patients, such as blood pressure and heart rate, and determines the risk of sepsis. The condition is not only devastating; it's also costly, because it prolongs hospital stays.

In their study, published this month in the journal Science Translational Medicine, the researchers used thousands of health records from patients in the intensive care unit at Boston's Beth Israel Deaconess Medical Center over six years to predict the disease 85 percent of the time.

Two-thirds of the time, they predicted sepsis before it inflicted any damage.

The infection can come from E. coli, MRSA or other bacteria. Sometimes doctors know there is an infection, such as pneumonia or a urinary tract infection; sometimes germs invade during surgery or through a catheter and go undiscovered.

Clinicians most often detect sepsis in the beginning stages, when body temperature changes, heart and respiration rates rise, and they know or think there is an infection.

Without IV fluids, antibiotics and other treatments, the condition can cascade into septic shock, causing lasting damage or death.

There are more than a million cases of sepsis a year across the country. Twenty percent to 40 percent are fatal.


Dr. Peter J. Pronovost, a study co-author and senior vice president for patient safety and quality at Johns Hopkins Medicine, said doctors at Hopkins continue to work with engineers to develop the computer model.

There have been technical challenges in making the code work in Hopkins' electronic medical system, which is being upgraded. Once it's embedded and working, researchers will need to determine the best way to notify clinicians of potential sepsis cases and develop protocols to respond.

Another problem is that patients outside the ICU aren't hooked up to as many machines taking measurements that can be fed into the algorithm. The sepsis code uses more than 25. Hopkins is working on wireless technology that could increase the number of patients under surveillance in other departments, such as the emergency room.

There could also be "alarm fatigue," in which beeps and notifications are ignored because so many signals turn out to be false.

If the system is used, researchers will study whether it actually helps reduce sepsis cases and improves patient outcomes.

"We still have to prove it works in the real world," said Pronovost, who directs Johns Hopkins' Armstrong Institute for Patient Safety and Quality. "In theory, it would. It did in the study. Clinicians won't have to guess, they'd have a visual display."


The notice of potential sepsis could come more than 24 hours before caregivers would normally suspect the condition. That's critical time, Pronovost said: studies show that every hour patients' sepsis goes untreated increases their risk of death by 8 percent.

The protocols for caregivers would likely be the same as they are now, he said, just implemented sooner. Those come from the Surviving Sepsis Campaign in 2004, and were developed by the Society of Critical Care Medicine and the European Society of Intensive Care Medicine.

The protocols were created in response to a "massive public health problem" that most people don't understand, said Dr. Craig Coopersmith, president of the U.S.-based society and a professor of surgery in the Emory University School of Medicine.

Despite its relative anonymity, he said, sepsis is likely the No. 2 killer worldwide after heart disease.

While the group's advice has brought the rate of mortality down, he said, more people are dying because the number of cases has been rising. He said the reason for the increase could be an aging population, more drug-resistant bacteria or more medical interventions.

Coopersmith said compliance with the protocols is far from universal — a recent study found only 19 percent of hospitals in 62 countries were following them within three hours of suspecting sepsis. Compliance rose to 36 percent at six hours.


If the Hopkins algorithm is proved and adopted by other hospitals, he said, the alerts could spur clinicians to act when treatments can make the biggest difference.

"If we catch sepsis in early stages and nip it in the bud, we prevent all downstream bad effects," he said. "It's so devastating for families, and nurses and doctors, to watch patients deteriorate. … I'm extraordinarily excited about the potential."

Hospitals in the United States are under increasing pressure from federal regulators to reduce "preventable harm" from serious infections, among other causes.

Pronovost and Saria are working on other "big data" projects to predict disease. Already, some hospitals use computer code to determine who might be readmitted to the hospital in 30 days. Readmissions are one of the measures that federal regulators use to penalize facilities financially.

The Maryland Hospital Association and a patient safety center it supports have an initiative to reduce sepsis and serious infections. They reported a 32 percent drop in cases from 2013 to 2014.

Hospitals are trying to determine how to sustain and expand the drop, said Nicole Stallings, the association's vice president of policy and data analytics.


Critical care doctors have been hunting for ways to reduce sepsis. They are rooting for the Hopkins algorithm.

Dr. Allan Walkey, an assistant professor of medicine in Boston University's School of Medicine who studies the disease, said it's a "major and underappreciated" health problem.

He said it's unclear whether the Hopkins code will have the same accuracy when it's applied in real-world situations and in different hospital settings, and whether the alerts actually prompt change in clinical practice and patient outcomes.

But he said the researchers appear to be taking "a first step toward answering some of these questions."