Sepsis is a life-threatening condition caused by the body’s response to an infection. In the US alone, there are over 970,000 reported cases of sepsis each year accounting for between 6-30% of all Intensive Care Unit (ICU) admissions and over 50% of hospital deaths. It has been reported that in cases of septic shock, the risk of dying increases by approximately 10% for every hour of delay in receiving antibiotics. Early detection of sepsis events is essential in improving sepsis management and mortality rates in the ICU.
Since 2000, PhysioNet has hosted an annual challenge on clinically important problems involving data, whereby participants are invited to submit solutions that are run and scored on hidden test sets to give overall rankings. This year’s challenge was the “Early prediction of Sepsis from Clinical data.”
A team from Oxford Mathematics and Oxford Psychiatry which consisted of James Morrill, Andrey Kormilitzin, Alejo Nevado-Holgado, Sam Howison, and Terry Lyons ranked in first place out of 105 entries. The team built a method based on feature extraction using the Signature method. They showed how the model predictions could be used to provide an early warning system for high risk patients who can be given additional treatment or subject to closer monitoring.
Their work was made possible by support from the The Engineering and Physical Sciences Research Council (EPSRC) and the Alan Turing Institute.