The AMIA 2013 conference was last month in Washington, DC. I had the honor of giving two talks on topics I often discuss on the Wolters Kluwer Health blog – medication safety screening and clinical decision support (CDS) standards.
My first talk was on the inter-rater agreement among physicians on the clinical significance of drug-drug interactions. Medication safety alerts can be conceptually divided into those that are clinically significant (signal) and those that are not clinically significant (noise). Alert fatigue occurs when the signal:noise ratio is too low and providers start to ignore the alerts. This problem is especially challenging to solve if physicians don’t even agree with each other on which drug-drug interaction alerts are clinically significant. I reported the results of a study we did looking at the this issue. We showed 16 physicians a set of 100 randomly selected interactions from our Medi-Span database. We asked each physician to indicate whether or not they considered each interaction to be clinically significant. The Fleiss Kappa, a chance-corrected measure of inter-rater agreement, was 0.21, which can be interpreted as poor agreement. Looking at it another way, there were only 3 interactions out of 100 where there was complete agreement among the 16 physicians on whether or not the interaction was clinically significant. These results underscore the need to provide user-level customization, including features such as, “Don’t show me this alert again” or “Don’t show me this alert again for this patient”.
My second talk was in the area of CDS standards, and in particular, I covered an example of a Health eDecisions Use Case 1 event-condition-action rule. This use case was balloted in January 2013 as an HL7 Draft Standard for Trial Use, but its publication is still pending because of the need to align it with evolving changes in the Virtual Medical Record model that it incorporates. The basic framework for a Use Case 1 artifact remains similar to what I described in a blog post about a year ago. The example I covered in the talk described how to represent a rule for a Practice-Changing UpDate from UpToDate that applies to adult women with breast cancer who are taking tamoxifen. The rule included the following components (each component being a section of XML):
- Metadata – e.g. title, identifiers, publisher, general applicability
- External data – e.g. definitions for breast cancer and tamoxifen using various standardized codes and where to look for the information in the Virtual Medical Record (VMR)
- Expressions – e.g. BreastCancerActive, TamoxifenActive, PatientAgeInYears
- Conditions – the conditions that must be true for the rule to evaluate to TRUE; in this case, BreastCancerActive AND TamoxifenActive and a few other criteria
- Actions – in this case, a message with the text of the Practice Changing UpDate
Also at AMIA, Dr. Jacob Reider, the US Acting National Coordinator for Health Information Technology, indicated that the US federal government has funded the creation of Use Case 1 artifacts in certain content areas, including immunizations, Hepatitis C, and management of HIV. He noted the importance of having some content artifacts available in this format in order to encourage electronic medical record (EMR) vendors to build support for this knowledge format.
As always, the AMIA Annual Symposium was an informative and intellectually stimulating meeting, as well as a great opportunity to catch up with my informatics colleagues from all around the country and all around the world.