While experts seem to agree that natural language processing is an important technology in healthcare -- especially...
as the need to use unstructured data grows -- many healthcare organizations aren't utilizing NLP in healthcare to its full potential.
"We're finding new, different types of dictated documents we haven't used," Scott Evans, director of medical informatics at Intermountain Healthcare in Salt Lake City, Utah, said. Therefore, healthcare organizations need a way to make sense of all that data. And since the amount of dictated documents and unstructured data is growing, the need for NLP in healthcare is also growing, he said.
How Intermountain Healthcare is using NLP
Intermountain is using a self-developed, Java-based NLP program to identify what illness a patient is suffering from, Evans said. Intermountain has used NLP to identify stroke patients, cancer patients, heart failure patients and patients with venous thromboembolisms.
"By using NLP, we found that we could identify heart failure patients out of reading data from 25 different types of free text documents stored in the electronic health record," he said. "You never know where you're going to find these little so-called nuggets that tell you something that nobody really charted anyplace else."
Evans said that often heart failure patients will come into the hospital or healthcare organization for a reason other than their risk for heart failure.
"Heart failure patients oftentimes come to the hospital … not because of their heart failure problem, but they may come in for a total hip, total knee [replacement], respiratory issues, their diabetes, [etcetera]," Evans said. Because of this, heart failure patients are often not identified as heart failure patients during their visit for another problem.
Using NLP in healthcare, "we identify those [heart failure patients] so our cardiovascular people are aware that they are in the hospital, [and] make sure they continue on with their heart failure meds and care," Evans said.
How UPMC is using NLP
At the University of Pittsburgh Medical Center, one way NLP in healthcare is being used is for clinical decision support and managing risk around chronic diseases more intelligently, said Rasu Shrestha, chief innovation officer at UPMC.
"A lot of times we're inundated with data, right?" Shrestha said. "There's lots of data that's hidden away in these unstructured documents, we're entering notes and we're interacting with our electronic medical record systems."
UPMC is using MedCPU, which is an overlay layer that sits on top of UPMC's EHR systems and adds intelligence to a physician's workflow.
"It looks at context, it has a context engine that's really driven by natural language processing and it's mining these unstructured documents and the things that I'm inputting into the EMR and it's presenting more intelligent information to me around clinical decision support as well as around … things like stroke management or for sepsis management," Shrestha said.
UPMC is also using NLP to manage risk more intelligently when it comes to chronic diseases. Shrestha said UPMC is using Health Fidelity, a risk adjustment solution that automates HCC coding operations and more, to do this.
Shrestha explained that it can be difficult for physicians to look at the many Hierarchical Condition Category (HCC) codes to figure out whether a patient has diabetes without complications, for example, or chronic obstructive pulmonary disease or breast cancer.
Shrestha said the question is: "What level of disease burden is this that we're talking about in these unstructured documents?"
Shrestha explained that Health Fidelity leverages NLP capabilities to mine through documents and create the right sets of capabilities to help UPMC better manage risk.
"It's really interesting because as a clinician, myself and my other colleagues, right, what we're seeing is the capability of having, not just data at our fingertips but information and knowledge, right? Because I think, in many ways, we're data rich and information poor," Shrestha said. "But NLP … converts all that data into information and knowledge that's actually actionable at the point of decision making."
Possible future use cases for NLP in healthcare
Shrestha believes that the next step for NLP is deep learning and machine learning, "where you're able to not just mine for information and put it in the right context but also you're able to have levels of predictive analytics and artificial intelligence that further aids clinicians in the decision-making process that continues to be complex in this environment."
Evans believes genomics may also be a good use case for NLP because when physicians send genetic tests out they usually contain a lot of pages of dictated reports, Evans said.
"That sounds to me like that is a big place where NLP could help out. Because the docs read these reports, and they're not really sure what it means. Or even how now they're going to explain that to patients," Evans said.
How natural language processing, big data will affect patients
Interoperability to be driven by next-gen EHRs
The role of medical transcriptionist is changed by use of EHRs
Enterprises implement chatbot component pieces to add business value