Performance analysis and assessment of a tf-idf based archetype-SNOMED-CT binding algorithm

Term bindings in archetypes are at a boundary between health information models and health terminology for dual model-based electronic health-care record (EHR) systems. The development of archetypes and the population of archetypes with bound terms is in its infancy. Terminological binding is currently performed “manually” by the teams who create archetypes. This process could be made more efficient, if it was supported by automatic tools. This paper presents a method for evaluating the performance of automatic code search approaches. In order to assess the quality of the automatic search, the authors extracted all the unique bound codes from 1133 archetypes from an archetype repository. These “manually bound ”SNOMED-CT codes were compared against the codes suggested by the authors' automatic search and used for assessing the algorithm's performance in terms of accuracy and category matching. The result of this study shows a sensitivity analysis of a set of parameters relevant to the matching process.