Hidden in plain sight in electronic health records lies information that could help improve outcomes and lower costs. But sifting through a ton of clinical data, often held in free-text notes, to find the useful nuggets is a slog, and providers and payers alike are already short on time.
Extracting meaning from providers' EHR notes
One way to get around that is outsourcing that analysis—not to another person, but to a computer. Instead of having a person sort through all the unstructured data in an EHR, Linguamatics has its machine learning, natural language processing platform do the work.
The result is insight that can help providers and payers address gaps in care and that can point them to patterns they might not otherwise see. That's especially important in the move to value-based care, said Simon Beulah, senior director of healthcare strategy for Linguamatics. "This is going to give you insight into patients' wellness and their lifestyle," Beulah said. "That's really the ground on which population health is going to be successful."
Both providers and payers can use the Linguamatics platform to try to stratify patients by risk. "The health systems have more of their data structured already, but they haven't necessarily been looking at population stratification the way the payer has," Beulah said. On the other hand, "the payers suffer from not having as much access to structured data."
On the payer side, a large plan might use the platform to extract data about patients with a certain condition, like congestive heart failure, from unstructured notes. It could then analyze that information and create more detailed risk stratifications.
On the provider side, clinicians might use the platform to improve clinical care. At Atrius Health, data analysts use the Linguamatics platform to help clinicians improve care for advanced heart failure patients. A subset of these patients would benefit from certain interventions, but figuring out which patients those are can be tough, especially since it's difficult to capture in a structured format in the EHR how well their hearts function. "We wanted to make sure we were capturing what this population looked like," said Dr. Craig Monsen, medical director of analytics and reporting at Atrius Health.
So Atrius used a specific query for the Linguamatics platform that identifies the particular patients who would benefit from interventions. Providers, in turn, check to make sure those patients are undergoing the right interventions.
For something like this to work, Beulah said, the health system has to have reached "a level of digital maturity." In this case, that means having had the EHR system in place for awhile, so they can focus on natural language processing rather than implementing the EHR.
"The vocabulary in EHRs is very different from the vocabulary that's used in a lot of standard natural language processing domains," said Jeremy Weiss, a Carnegie Mellon University assistant professor of health informatics, in explaining one of the challenges of applying natural language processing to healthcare notes. But, he points out, extracting meaning from notes is so potentially useful because the notes contain exactly what providers want to communicate to each other.
Other institutions have dabbled in similar projects. Health Catalyst and the Regenstrief Institute partnered in 2017 to use Regenstrief's text analytics on unstructured EHR data, for example.
Over the next year, Linguamatics developers will focus on expanding what Beulah called the "360-degree view of the patient." The company already does work on social determinants of health, for instance. Next, Beulah and his team will dive deeper into behavioral health, he said. "There are really interesting clues and insights in there into the potential for an individual to be readmitted."
Send us a letter
Have an opinion about this story? Click here to submit a Letter to the Editor, and we may publish it in print.