How developing an algorithm will help physicians prioritize care and save lives

According to the Center for Disease Control, over 38,000 US adults die annually due to liver disease.

Most of these cases are related to cirrhosis, a chronic and progressive form of liver dysfunction. Cirrhosis has many causes and can present in a multitude of ways. Consequently, patients with this disorder are particularly difficult to predict and care for.​

To address these issues, a team of Masters students at the University of Virginia’s Data Science Institute is using a number of data science tools and techniques to try to predict mortality rates among patients hospitalized with cirrhosis.

“As scientific as medicine already is, the field needs more data-driven methods,” said MD-MSDS dual degree student Elizabeth Harrison, “and if our model proves successful enough to be implemented as a component of the University's electronic medical record system, it could make a significant impact on patient outcomes.”

Yi Hao, Myron Chang, and Harrison are working with Prof. Abby Flower to develop an algorithm based on de-identified patient records from the University of Virginia’s Epic Patient Database. They aim to use the most effective model to create a user-friendly system in which healthcare providers can enter information and receive immediate feedback about a patient’s risk for adverse outcomes in the next 24 hours.

The algorithm was originally designed to help with triaging patients in emergency medical situations, but the researchers have since shifted their attention.

“We have turned our focus more toward analyzing and predicting cirrhosis patients' mortality risk,” said Chang. “We are planning to change direction a bit and focus on a neural network to see how it performs.”

As of now the team has analyzed the data and is working on narrowing down the predictive models to find the one with the best performance. The ultimate goal is to create a system better than the current baseline model the UVA health system uses for predicting mortality.

“I actually do not have a health or medical background,” Chang said, “[but] this cohort has a few currently running capstone projects related to public health issues, which demonstrates the importance of using the data we have to address these issues.”


E. Harrison, M. Chang, Y. Hao, and A. Flower. "Using Machine Learning to Predict Near-Term Mortality in Cirrhosis Patients Hospitalized at the University of Virginia Health System." 2018 Systems and Information Engineering Design Symposium (SIEDS), Charlottesville, VA, 2018.


Learn about other Capstone Projects

Completed in:
2018
Project type:
Partner:

UVA Health System