Prediction in annotation based guideline encoding.

The encoding of clinical practice guidelines into machine operable representations poses numerous challenges and will require considerable human intervention for the foreseeable future. To assist and potentially speed up this process, we have developed an incremental approach to guideline encoding which begins with the annotation of the original guideline text using markup techniques. A modular and flexible sequence of subtasks results in increasingly inter-operable representations while maintaining the connections to all prior source representations and supporting knowledge. To reduce the encoding bottleneck we also employ a number of machine-assisted learning and prediction techniques within a knowledge-based software environment. Promising results with a straightforward incremental learning algorithm illustrate the feasibility of such an approach.

Vocabulary requirements for implementing clinical guidelines in an electronic medical record: a case study.

As part of a larger effort to automate guidelines we determined the number and types of clinical variables required to implement two complex clinical guidelines and the adequacy of the electronic medical record (EMR) to capture them. 178 unique variables were required by both guidelines. Variables were classified as simple (existing observation terms in the EMR), calculated (transformations of simple variables), and complex (requiring multiple simple variables and logical rules for combining them). Many variables are unlikely to be instantiated in an EMR without focused efforts to collect them. In addition, many variables required knowledge that was neither provided in the guideline nor referenced. We conclude that, although the EMR contains the necessary variables to implement these guidelines, successful automated implementation requires unambiguous definition of required terms, incorporation of additional knowledge not provided in the guideline and modification of workflow to collect variables not normally captured in routine clinical care.