Abstract
Burroughs Wellcome Co. uses a slightly modified version of the Food and Drug Administration (FDA) COSTART terminology for encoding all adverse drug experiences (ADEs) reported to the company, including foreign reports. A hardcopy dictionary is available for translations between ADE verbatim text (raw terms) and preferred COSTART terms
A pilot project was initiated in July 1987 to evaluate the feasibility of automated translation of raw terms into COSTART. The pilot translation of 500 new raw terms using a “hashing” search algorithm, a word-to-synonym thesaurus, and historical Synonym-COSTART frequencies resulted in a coding accuracy of 73%, compared with an accuracy of 88% from manual coding
An extension of the pilot implementation retains the hashing search algorithm but uses logical groupings of synonym codes to isolate the preferred COSTART phrase. This extension provides an improvement in accuracy (overall coding accuracy of 98% for the previously described 500 raw terms), as well as a decrease in data processing resource requirements. The strengths, limitations, and potential applicability of these translation procedures will be discussed
Get full access to this article
View all access options for this article.