OR WAIT null SECS
Big data fueled the real-time sepsis forecasting algorithm.
This bot sees what doctors sometimes cannot. Emory University researchers have created a “Sepsis Expert” algorithm that works in real time to predict the onset of sepsis, the deadly condition that often takes hold in healthcare settings. Banking on information from 31,000 patients admitted to 2 hospitals and data on 52,000 intensive care unit (ICU) patients from a public database, the researchers used machine learning to build an artificial intelligence (AI) technology that they hope will save lives.
According to the CDC, more than 1.5 million people in the United Sates contract sepsis every year, resulting in 250,000 deaths. But, until now, that knowledge has not translated to insights for the individual. “What we lack is ‘situation awareness,’ which is perceiving data, comprehending data, and projecting those data into the future to see whether there is an evolving threat to the patient,” says Timothy George Buchman, PhD, MD, director of Emory’s critical care center and co-author of a study on the tech.
He says the Sepsis Expert is sensitive to pattern detection over time, using layers of data that can project into the future. It acts like a smartphone pop-up notification, which happens to appear when you are craving a donut—and when you are “vulnerable” and conveniently close to a coffee shop.
The AI-guided Sepsis Expert predicted the onset of patients in the ICU between “4 and 12 hours before clinical recognition,” the researchers found. Time is crucial, as for patients who have severe sepsis—an infection along with dropping blood pressure—each delayed hour of treatment raises the chance of mortality by about 7%, according to experts. The algorithm is designed to identify those people and begin their treatments early on.
“Recognizing and initiating treatment early not only saves lives, but also tends to mitigate those longer-term problems that plague older folks who don’t have much reserve and yet survive the sepsis episode,” Buchman says. “Finding and fixing a problem before it becomes catastrophic seems to make a difference. That said, it is the comorbidities that exist prior to sepsis that have the greatest influence on function among sepsis survivors.”
So far, Sepsis Expert is only compatible with adults, even though children are also vulnerable to sepsis. The AI must be tuned to kids, as they respond to sepsis differently than adults.
With all the bedside alarms already in place in hospital settings, researchers were concerned that Sepsis Expert might be perceived as another nuisance to ignore. Buchman says healthcare professionals are bombarded with device alarms, drug interaction alerts, and critical lab value notifications, but the team at Emory has spent considerable time and thought on how the alerts will be adjudicated, who does the adjudication, and what professionals at the bedside want to and need to know. He sees an increased trust in AI, though a future study may examine the clinical utility of the Sepsis Predictor model.
The study, “An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU,” was published in the journal Critical Care Medicine.