Towards improved information retrieval from medical sources

We review a HEALNet health informatics project directed at information extraction from medical text resources in support of evidence based practice (EBP). The motivation for the approach is summarized. A brief review of pertinent methodology is given and the rationale for exploring approaches to statistical concept representation (SCR) derived. Data on the qualitative comparison of five related systems are presented. Installation and tests of three of these systems revealed that systems designed to deal with literature are not able to handle medical record text adequately because of various idiosyncrasies of the latter. To overcome these, it is necessary to control the system's text manipulation algorithms. This prompted us to build our own system. The principles of this development and its results are summarized. They include the construction of a core information retrieval (IR) system, which is now operational, initiation of work on the collection of test data bases and development of a conceptual framework for comprehensive system evaluation. A brief outline of further steps concludes the presentation.