Our hypothesis is that clinical events that have no significant temporal duration, called point events, and events that have relevant temporal duration, called interval events, are essential elements in patient- specific clinical decision making. We describe a six-step research plan based on the premise that current clinical information systems cannot provide context-sensitive data retrieval for two reasons: (1) current data models do not explicitly store interval events, and (2) current query languages cannot specify temporal relationships among point and interval events. Our research plan will collect requests for patient information generated in actual clinical decision-making settings, analyze the temporal and contextual features of these requests, attempt to translate these requests into current database query languages, and study the sources of difficulty or failures in these translations. From this analysis, we propose to define a new data model for explicitly storing point and interval events and a new query language for specifying data retrieval based on relationships among instantaneous and interval events. 1. We will establish a library of context-sensitive medical queries obtained from recorded transcripts of teaching attending rounds in Internal Medicine and expert diabetologists analyzing timed insulin and glucose data in diabetic patient log books. 2. We will translate each query into an unambiguous representation to ensure that all implied or shared clinical events and states are made explicitly in an """"""""gold-standard"""""""" reference for each query. 3. We will translate each query into seven languages that have unique strengths in representing temporal and contextual relationships, analyze the source of difficulties when a translation is not possible, and generate a list of language primitives that must be present to specify time-and context-sensitive medical data retrieval. 4. We will develop a temporal semantic data model and query language that incorporates both point and interval events a primitive database objects. 5. We will implement our data model and query language by extending a commercial relational database system. 6. We will evaluate our temporal database system in a clinical environment.
Showing the most recent 10 out of 17 publications
