Relational Database Design for Computer-Analyzed Medical Narrative.

Abstract A detailed analysis of the syntactic and semantic patterns of the narrative portions of patient records reveals surprising regularity in both regards. Building on these observed regularities, the NYU Linguistic String Project has programmed and tested a processor that accepts free text input and produces information-structured output, suitable for mapping (with some adjustments) into a database of the network or relational type. The first schema, a network design, was presented at the Fourth Annual Symposium on Computer Applications in Medical Care. This paper will describe the development of a relational data model which parallels the earlier CODASYL design. Experiments using SYSTEM R will be described and compared with similar experiments using DMS1100. The suitability of SQL for queries addressed to a database of natural language medical data is also discussed.