December 20, 2018 | Written by: Claudio Abiuso and Dr. Walter Kessler
Categorized: AI/Watson | Healthcare
Share this post:
At the Centre Hospitalier Chrétien (CHC), we are committed to putting patients first. Quality of care and patient safety are our top priorities. And we care for many patients. In our hospital and clinics, we have more than 1,000 beds for acute care patients, 30 beds for psychiatric care and 675 beds for retired residents.
That means a large volume of patients being admitted and discharged on a daily basis—each of whom receive personal attention and care from our medical staff. Each patient is diagnosed, treated and released with instructions for follow-up visits, medications, physical therapy and more. To do so successfully, achieving positive outcomes, is our mission at the CHC.
However, there is more to the story than the care our doctors and nurses deliver every day. We do an enormous amount of work behind the scenes to make sure we get government financing for our services. Recently, we started using AI technology to help do this work more efficiently and effectively.
Diagnostic codes and the promise of AI
If you’re familiar with hospital operations in general, you might know about diagnostic and procedural codes—a kind of standard language for the medical community to request financing from government agencies.
At CHC, we employ a team of about 15 people to review each and every hospitalization case—reading discharge letters, medical records, lab results and more—and assign the proper codes from a list of 69,000 options. The group responsible for this crucial work is called the Résumé Clinique Minimum (RCM) team.
It’s true that the work is repetitive and tedious, but it also requires a great deal of experience and concentration. On an average day, the RCM team can work through 25 hospitalization cases. There is no way around it—it simply takes time to do the work right.
Or so we thought, until we started looking at AI technology.
Teaching IBM Watson the language of medical codes
We started by running a pilot project with IBM Global Business Services. With IBM’s help, we automated some of the more tedious parts of the coding process by building an AI-powered assistant based on IBM Watson Explorer and IBM Watson Knowledge Studio technology.
Over the course of 18 months, we were able to train Watson to understand the terminology in the World Health Organization’s ICD-10 classification system by tagging and annotating nearly 2,000 medical documents. This was how Watson started as the newest member of the RCM team.
Just as we would treat any new employee in training, we watched the AI solution’s performance and corrected it as necessary, teaching Watson to interpret medical documentation and translate it into the proper codes.
Within a few months, the solution was giving us codes with a 28 percent accuracy rate. That number jumped to 80 percent in a year. Watson had become a viable assistant so, in June 2018, we went live.
The benefits of a smart new team member
The project has been an instructive one. We’ve learned that AI technology can help us with the cognitively heavy tasks that are part of hospital administration. It can make our processes faster, less tedious and more rewarding for our coding experts.
Today, the RCM team works alongside Watson as it helps read medical documents and suggest proper codes for diagnosis and treatment. Employees simply review the codes it proposes and the rationale it offers for its suggestions. Every time they accept or reject the suggestions, they are helping the system learn.
Earlier, I mentioned the RCM team can handle around 25 cases per day. With the Watson solution, that number is up to 35 cases—a 25-30 percent gain in productivity.
We see a bright future for AI technology at the CHC. This project is a good first step.