Temporal mediation of relational databases for clinical decision support

Many health-care institutions would like to bring applicable clinical guidelines to the attention of providers via decision-support programs that examine patient data. Since the planning of medical therapies is highly time dependent, an essential step in predicating protocol-based advice is the detection of temporal conditions in a clinical database. A significant barrier to the task of integrating a decision-support application with a clinical database is the temporal model heterogeneity that exists between them. Current database methods cannot undertake the temporal-mismatch resolution and temporal-pattern matching that are necessary to resolve temporal model heterogeneity. Previous methods to overcome these limitations for relational databases have been neither general nor reusable for the task of temporal integration in protocol-based decision-support architectures.; In this dissertation, I present a novel method for temporal integration—which I call temporal mediation. This method consists of three parts: (1) a foundational model of time, (2) a temporal relational algebra, and (3) a software system. The model of time is based on a timeline representation that uses time-unit functions to map temporal values from one time-unit level to another and that uses temporal operations to match temporal patterns. I have incorporated the semantics of the model of time into a temporal relational algebra, which consists of a multisorted and a unisorted set of operators. The multisorted set applies the time-unit functions to heterogeneous data to transform them into a uniform interval-stamped representation at a single time-unit level. The unisorted set applies the temporal operations to verify temporal conditions among the uniform temporal data that result from the former set. I have implemented both sets of operators into the data-processing algorithm of a software module for temporal integration, which I call S YNCHRONUS.; I have undertaken a two-part process that formally evaluates the reusability of the temporal-integration module. I show that SYNCHRONUS satisfies the integration requirements within a test decision-support architecture, and I then demonstrate the transferability of the mediator to another existing software architecture. These results confirm the hypothesis that the temporal-mediation method can be reusable for the task of temporal integration among protocol-based applications and their relational data sources.