Secondary Logo

Journal Logo

News: Sepsis Algorithm a Deadly Marker

Sorelle, Ruth MPH

doi: 10.1097/01.EEM.0000511941.27134.7e


An attempt by a Phoenix, AZ, hospital to develop a marker for deadly sepsis instead found that the algorithm identified patients at an increased risk of dying.

“What we found was astonishing,” said Hargobind Khurana, MD, the senior medical director of health management at Banner Health and the lead author of the report in The American Journal of Medicine. (2016;129[7]:688; “It was not just that these patients were sicker. What we stumbled upon is that this is a small cohort of patients [who would die in the hospital]. Not all of them were septic or those we thought would die,” he said.

Increasingly, algorithms govern daily life, playing an important part in monitoring the effects of drugs, treatments, and surgery on patients in the hospital. Dr. Khurana and his team used real-time automated sampling of electronic medical data to develop an alert that could tell physicians and nurses when a patient is at risk for dying. Continuous sampling identified patients with two of four systemic inflammatory response criteria along with at least one of 14 acute organ dysfunction parameters in those most likely to die.

The researchers implemented the plan in 312,214 patients and 24 hospitals. It was tested in the training and validation phases, with datasets from both periods analyzed. A total of 29,317 (18.8%) patients triggered the alert during the training phase of the study, and 5.2 percent of these patients died. Meanwhile, only 0.2 percent of patients for whom there was no such alert died. Only a quarter of the patients flagged by the alert had sepsis, but those patients were much sicker than the patient population in general.

Patients identified by the algorithm accounted for nearly 90 percent of all deaths in the hospital. Patients who triggered the alert for systemic inflammatory response syndrome (SIRS) and organ dysfunction were older, more likely to be male, have cancer, have undergone coronary artery bypass grafting, suffered trauma, or be labeled with sepsis by the Angus criteria. Patients who triggered the alert during their hospital stay were also more likely to suffer from chronic medical conditions.

Back to Top | Article Outline

Measured in Real Time

A total of 31.9 percent of all alerts (9,361) were triggered by patients in the emergency department during the training phase, and 35.7 percent (10,357) were triggered in the emergency department during the validation phase. Overall, 80.2 percent of alerts were noted in the first 48 hours of admission. The advantage of this alert, Dr. Khurana said, is that it is measured in real time and alerts the provider to the patient's status.

But he added that they are continuing to study and refine the alert, taking into account the phenomenon of alert burnout in physicians, nurses, and other health care providers. “We want to find the needle in the haystack that is important amid the other noise. Even though you use the right criteria, it's hard to make that criteria into an alert in the real world when blood work comes in the morning and vitals are measured on a regular basis. How do we put this in a patient scenario?

“Not all patients who triggered the alert were septic or would die,” Dr. Khurana said. “The mortality rate is five percent, [and] 95 percent of these patients could leave the hospital.” As a result, he said, they went back to explore the patient pool, wondering if they missed something, if the patient had the right diagnosis, or if the right treatment option was chosen.

“The most important part of the study is to realize that we can use the electronic medical record in new ways,” he said. “We are all making imperfect algorithms. We can learn from our mistakes and make them better for our clinicians.”

Back to Top | Article Outline

Will Doctors Use It?

This also opened their eyes to something else, Dr. Khurana said. “One in five patients who come into the hospital are this sick. If we can identify them in the first 48 hours, we have enough lead time to intervene. Those who died did so in five days after the alert. I hope people realize that's the power of the electronic medical record. I also hope they realize you might not strike gold the first time out.”

“This is a good first step and a good study,” said Hardeep Singh, MD, MPH, the chief of health policy, quality, and informatics program in the Center for Innovations in Quality, Effectiveness and Safety at Michael E. DeBakey VA Medical Center. “The question is who uses it next? Will physicians use it? Doctors don't like alerts, especially those that don't improve patient care. In the case of this one, who should use it next? Does it improve patient care?”

Dr. Singh, who specializes in the study of safety of electronic medical records and the risk of missed diagnoses in the ambulatory care setting, is also the director of the VA Center of Inquiry to Improve Outpatient Safety through Effective Electronic Communication and an associate professor of medicine at Baylor College of Medicine in Houston.

Share this article on Twitter and Facebook.

Access the links in EMN by reading this on our website or in our free iPad app, both available at

Comments? Write to us at

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.