Skip to content Skip to main navigation Report an accessibility issue

Machine Learning for Better ICU Outcomes

Anahita KhojandiSepsis is estimated to kill around 250,000 Americans each year. In 2018, the state of Tennessee alone saw nearly 1,000 deaths as a result of sepsis. Anahita Khojandi’s JDRD work could lay the foundation for improving those outcomes.

Sepsis is a body’s extreme reaction to an infection. This life-threatening condition typically occurs as a result of a pre-existing infection and has a mortality rate of around 40 percent. Survivability often depends on a patient receiving early identification and timely treatment of the condition. For every hour sepsis is undiagnosed, a patient’s mortality rate increases by 8 percent. Intensive care unit, ICU, patients are especially at risk.

“It’s actually one of the major causes of death in hospitals,” said Khojandi. “It’s very important to detect sepsis early and because this is such an important problem, we need to start addressing or thinking about it.”

Khojandi’s work seeks to improve sepsis outcomes by improving methods of early detection. Her team is working to combine Bayesian frameworks and machine learning to create a holistic means of peaking into the future of potential sepsis patients.

“For every minute a patient is in a non-sepsis state, there is a probability they will end up in sepsis. It’s a very small probability, especially when you’re looking at one minute, but then think about that over two or three hours. It’s compounded,” said Khojandi.

Khojandi believes the incorporation of a dynamic Bayesian framework will help account for the ongoing changes patients’ bodies experience, allowing for a decrease in misdiagnoses. Such a holistic framework could also incorporate the human element, which could be especially relevant in ICUs.

“You have a limited number of nurses in ICUs caring for patients, so how should they prioritize based on risk factors? Who should they cautiously monitor, who should they closely attend to, who should they rank how in terms of these risk factors?” Khojandi asked.

Her JDRD team has partnered with hospitals to gain access to a data set for a pilot study. Additionally, they have generated several papers on the topic and submitted to proposals to the National Science Foundation. Khojandi hopes this project will have multiple medical applications and improve outcomes for many people, both in and beyond Tennessee.