Emory University issued the following announcement on Feb. 8.
Emory researchers are investigating the use of artificial intelligence to forecast therapeutic effectiveness and outcomes for patients with sepsis, a syndrome that claims one in five lives around the world and has until now been a black box for rapid diagnosis and treatment.
In the United States, more than a quarter of a million people each year succumb to sepsis, which occurs when the immune system responds to an existing infection such as COVID-19 by turning on itself instead of fighting the germs.
The research team, led by Rishikesan Kamaleswaran, PhD, assistant professor in Emory’s biomedical informatics department, was recently awarded a $2.6 million award from the National Institute of General Medical Sciences of the National Institutes of Health to mine sensor-generated data streams for physio-markers that may be able to predict the onset of sepsis, inform treatment options, and support the discovery of sub-types of sepsis.
“The hope is to use AI in a new way so we can better see in an area where traditionally we have flown blind,” says Kamaleswaran. “By using existing and routinely collected physiological data in intensive care units (ICU), we hope to generate robust machine learning algorithms than can more 1. More accurately phenotype sepsis patients and 2. Improve outcomes with more personalized treatments – therapies that would elicit the best response.”
Approaches using machine learning have focused largely for predicting sepsis from electronic medical records which Kamaleswaran says suffer from a host of problems. “The data are not timely, significant portions are missing or wrong because of the manual process of entry, and the information often reflects individual and institutional biases, which all make it difficult to devise a treatment plan that can be replicated someplace else.”
The five-year study will tap into expertise from different disciplines at Emory including mathematics, computer science, and medicine to develop sophisticated tools that can analyze the data, identify patterns, and prescribe a course of action. “We will contribute significant knowledge about the role and utility of complex physiological interactions that are abundantly available in clinical practice but seldom used for clinical decision making.”
Kamaleswaran points out that sepsis results in many deaths from COVID-19 and other diseases. “The use of AI and machine learning here are powerful mathematical constructs that when placed in the hands of a capable clinician, can become an efficient resource for improving patient care.”
The tools the research team is developing rely on information from patients who developed secondary infections after admission to the ICU, a departure from existing algorithms for sepsis that use data from patients in general wards.
Apart from Kamaleswaran, the research team for the study comprises physician Craig Coopersmith, biomedical informatics scientists Gari Clifford and Qiao Li, all at Emory; scientist Omer Inan at the Georgia Institute of Technology; and physicians Michael R. Pinsky and Gilles Clermont at the University of Pittsburgh.
Infections that lead to sepsis most often start in the lung, urinary tract, skin, or gastrointestinal tract. Without timely treatment, sepsis can rapidly lead to tissue damage, multi-organ failure, and death.
The Centers of Disease Control and Prevention says one in three patients who die in a hospital in the U.S has sepsis, and in 87 percent of those cases, the infection begins outside the health care facility.
The World Health Organization has said that lowering the death and disability count from sepsis is particularly challenging because of “serious gaps in knowledge” particularly in low- and middle-income countries. Around 11 million people die each year of sepsis, many of them children, and even those who survive can sometimes have lifelong disabilities.
"The goal is to find a novel way to improve patient care and move the needle in a space that is plagued with complexity,” says Kamaleswaran.
The NIH grant supporting this research is R01GM139967.
Original source can be found here.
Source: Emory University
Alerts Sign-up