Skip to content Skip to navigation

Leveraging Free Text in the Patient Chart

May 13, 2011
by Mark Hagland
| Reprints
One physician informaticist is innovating in a key area of interest around semantic search

Karl Kochendorfer, M.D., is director of clinical informatics, and medical director of the Living Lab within the Tiger Institute, an entity within the University of Missouri Health System in Columbia, Mo. He and his colleagues at the University of Missouri Health System (UMHS) have been collaborating with their core EHR vendor, the Kansas City-based Cerner Corporation, in the development of a solution called ChartSearch, a solution that allows clinicians and clinical informaticists to identify areas of free text within patient charts, and makes it possible to perform semantic search—to look up and extract keywords from the patient chart. Kochendorfer spoke recently with HCI Editor-in-Chief Mark Hagland regarding this IT research and development work.

Can you explain for us what semantic search is, and how this tool accomplishes it?
When I’m attempting to look for key terms in the free text portion of the electronic patient record, I choose a term from SNOMED [the Systematized Nomenclature of Medicine], and the tool can map that to other related terms. So I could put in Lisinopril [a generic blood pressure medication], or I could put in ACE inhibitors and find Lisinopril.

Karl Kochendorfer, M.D.

So it’s a search engine designed to comb the patient record for terms?
Yes. I started piloting it with my patients last October; and then we expanded it to 12 other physicians early last November. And in February, we opened it up to our entire provider community, including attendings, residents, medical students, and advanced practice nurses, which we call our provider community. And in March, we opened it up to all users—medical records people, nurses, etc., whoever has appropriate rights to open the patient record, can use ChartSearch.

How is it being used?
Right now, it’s a separate section of the chart; we’re hoping to embed it into the opening page of the screen. Right now, they have to open a separate section of the chart, and then can do a search. So, for example, I might be struggling with trying to identify when a patient’s last colonoscopy was performed. And the patient may think it was a year ago or so, but they don’t sound very confident; and I look for it where most reports are, but I don’t see it. So I would go into ChartSearch, and type in the word colonoscopy, and it could find other, related, terms like sigmoidoscopy, and it will pull up the results within a second. So I would find the procedure, then.

And you might be looking for a word or a thought, right?
Sure. Another example is that one of my colleagues got a call asking for a refill on some narcotic prescriptions; and the patient reported that their medications were stolen. And he went into the search, and typed in stolen, and he ended up with a story. The doc was wondering, was this a recurring problem? And he could easily see that the same thing happened three years ago. And in that case, there had even been a police report, but the patient had continued to call different clinics, saying the same thing. That search quickly gave us more information than we would easily have found otherwise, because these were coming from phone messages, from clinic notes; and this search pulled those up immediately. Another time, I was in a hospital setting, and we needed to notify next of kin. And this particular daughter of a patient was frequently caring for this patient; and a quick search pulled up her home number, cell phone number, etc.; and that’s not something that typically is stored any particular place in the chart.

Is this a parallel to the need for automated data pulling for meaningful use?
Yes, it is. And one real potential here is that this kind of search could pull across multiple charts. Right now, you need to use report-writing tools to do this. But if you could appropriately use a search engine like this, if there’s a drug recall, I don’t need to go to our CCL report writer people; I could quickly type in Celebrex, for example, if there were a recall or something, and within a minute, I’d have a list of patients who might be on Celebrex. And for meaningful use, we have to do quality measure reporting, and we’ve got teams and teams of people doing that. So I could quickly search for Pneumovax, a pneumococcal vaccine, to see which high-risk, over-60 patients might have had that immunization, for example.