Information exchange between medical databases through automated identification of concept equivalence

The difficulty of exchanging information between heterogeneous medical databases remains one of the chief obstacles in achieving a unified patient medical record. Although methods have been developed to address differences in data formats, system software, and communication protocols, automated data exchange between disparate systems still remains an elusive goal. The Medical Information Acquisition and Transmission Enabler (MEDIATE) system identifies semantically equivalent concepts between databases to facilitate information exchange. MEDIATE employs a semantic network representation to model underlying native databases and to serve as an interface for database queries. This representation generates a semantic context for data concepts that can subsequently be exploited to perform automated concept matching between disparate databases. To test the feasibility of this system, medical laboratory databases from two different institutions were represented within MEDIATE and automated concept matching was performed. The experimental results show that concepts that existed in both laboratory databases were always correctly recognized as candidate matches. In addition, concepts which existed in only one database could often be matched with more “generalized” concepts in the other database that could still provide useful information. The architecture of MEDIATE offers advantages in system scalability and robustness. Since concept matching is performed automatically, the only work required to enable data exchange is construction of the semantic network representation. No pre-negotiation is required between institutions to identify data that is compatible for exchange, and there is no additional overhead to add more databases to the exchange network. Because the concept matching occurs dynamically at the time of information exchange, the system is robust to modifications in the underlying native databases as long as the semantic network representations are appropriately updated. Thesis Supervisor: Peter Szolovits Title: Professor of Computer Science and Engineering