An adaptable architecture for patient cohort identification from diverse data sources

Objective: We define and validate an architecture for systems that identify patient cohorts for clinical trials from multiple heterogeneous data sources. This architecture has an explicit query model capable of supporting temporal reasoning and expressing eligibility criteria independently of the representation of the data used to evaluate them.
Method: The architecture has the key feature that queries defined according to the query model are both pre and post-processed and this is used to address both structural and semantic heterogeneity. The process of extracting the relevant clinical facts is separated from the process of reasoning about them. A specific instance of the query model is then defined and implemented.
Results: We show that the specific instance of the query model has wide applicability. We then describe how it is used to access three diverse data warehouses to determine patient counts.
Discussion: Although the proposed architecture requires greater effort to implement the query model than would be the case for using just SQL and accessing a data-based management system directly, this effort is justified because it supports both temporal reasoning and heterogeneous data sources. The query model only needs to be implemented once no matter how many data sources are accessed. Each additional source requires only the implementation of a lightweight adaptor.
Conclusions: The architecture has been used to implement a specific query model that can express complex eligibility criteria and access three diverse data warehouses thus demonstrating the feasibility of this approach in dealing with temporal reasoning and data heterogeneity.