A knowledge-based clinical decision support (CDS) system that was built upon software tools developed by BMIR remains in use after nearly a quarter century at VA Palo Alto Health Care System. Stanford Health Policy's Mary K. Goldstein, MD, and the Office of Geriatrics and Extended Care in the Veterans Health Administration, said the system was developed to support primary-care staff in providing evidence-based treatments for patients with hypertension, one of the most prevalent diagnoses in adult medicine.

In the late 1990s, Goldstein collaborated with fellow Stanford faculty member Brian B. Hoffman, MD, at VA Palo Alto and Mark Musen, MD, PhD, and his group in BMIR. They initially encoded a computable knowledge base of hypertension clinical knowledge using Protégé, the knowledge-acquisition software that BMIR developed. There was further work in knowledge representation with BMIR academic staff, including the internationally recognized expertise of Samson Tu, MS, and with VA Palo Alto, whose health informatics staff understood the importance of providing evidence-based CDS for clinical providers.

Those innovative approaches to implementing technologies led to development of an architecture for extracting patient data, processing it with the knowledge base, generating recommendations, and displaying recommendations to primary care providers (PCPs) in a computer window that displayed on top of the patient’s electronic health record and could be viewed on the screen in real time.

 

The original system was created to help PCPs in managing hypertension. Over time, the VA Palo Alto-based team has built on that system so that today it has evolved into a new CDS system that can be displayed within a clinical dashboard for additional diseases including diabetes. The current program, known as Med Safe CDS, still uses Protégé and a lot of the concepts—and program code—from the original work.

“Many decision support systems are based on simple rules with branching logic that reach a limit of how much complexity they can handle," Goldstein said in an interview for the newsletter of the Stanford Center for Biomedical Informatics Research. "At the VA, we wanted to be able to take account of more patient complexity so that multiple patient clinical characteristics would be processed at the same time. The underlying system for doing this has allowed us to generate much more complex recommendations than could be done with a rule-based branching logic system."

Read More