Language is always evolving, which is a challenge for researchers, particularly if they want to turn a decade of doctors’ free-text notes in Epic into discrete data to improve cardiac care. Mercy in St. Louis is tackling this problem with natural language processing, using it to smooth out the linguistic variation between clinical narratives into discrete, analyzable units that will help make better devices to treat heart failure.
To get the data, Mercy’s team needed to account for evolution. There “were standard ways of saying things internally in 2012 that had changed a couple of years later,” said Kerry Bonmarito, manager of data science at Mercy. For example, by 2014, the cardiac measurement called a QRS had become known as a QRSd.
An Epic community member since 2008, Mercy can draw on a deep well of electronic notes. The data they’ve collected can improve treatment by giving them a more detailed picture of “the progression of heart failure, how symptoms change over time,” Bonmarito said. Understanding this can help Mercy’s cardiologists improve treatment, and sharing the data with a cardiac device manufacturer multiplies the benefit by helping them improve design.
Mark Dunham, director of data engineering and analytics at Mercy, sees great potential for NLP to improve the data that drives better patient care. Is it possible, he asks, to push back the edges of narrative notes “and color in around the edges where we could get better-quality data?” The results of this project suggest it is. “We’re just scratching the surface of what we can do with this.”
Read more about Mercy’s use of NLP in Healthcare IT News.
