The same scenario plays out across doctor's offices and hospitals countless times across the world every day. A doctor, nurse or other medical practitioner quickly jots down a few notes into an EMR, providing sparse details and little context, and quickly moves on to the next patient. It's up to someone else to decipher the notes at a later point down the road.

A better way

Rather than asking people to change their ways and possibly disrupt their workflows, one large hospital network in the northeast US decided to use text analytics to classify, organize and then make searchable all of the important information contained within an EMR.

The hospital's CTO asked Repustate to help him solve this problem:

There's so much data buried in these records, I wish I could get it out. I wish could ask questions like "What was the result of prescribing a particular drug at various dosage levels?" And that's just one example, I have so many.

Taking that question as a good first problem to tackle, Repustate applied its text analytics API to each and every progress note in the EMR. Specifically, named entity recognition was applied to identify and annotate each of the following:

  • The primary reason the patient came (e.g. chest pain, shortness of breath)
  • Each and every medication prescribed and by any name it might go by (e.g. Tylenol, parecetemol, acetaminophen, Panadol)
  • The dosage it was prescribed at (e.g. 2 pills twice a day, 40mg daily)
  • The length of time between successive visits (e.g. patient returned two weeks later)
  • And the ensuing result (e.g. symptoms went away, symptoms persisted)

With this information tagged and stored, each health record now contains a tremendous amount of value. Reports can be run to monitor the effectiveness of certain medications and/or dosage combinations. The outcomes that patients have with certain medications and even specific medical practitioners also becomes self-evident. Sentiment analysis can be used to determine efficacy by looking out for positive or negative terms and phrases.

More work remains as the hospital administrators, having seen the success of the scenario above, have numerous insights they want to glean from this new found structured data. With the hard part of struturing and annotating progress notes out of the way, the fun part begins.