Web services for data warehouses: OMOP and PCORnet on i2b2

Abstract

Objective: Healthcare organizations use research data models supported by projects and tools that interest them, which often means organizations must support the same data in multiple models. The healthcare research ecosystem would benefit if tools and projects could be adopted independently from the underlying data model. Here, we introduce the concept of a reusable application programming interface (API) for healthcare and show that the i2b2 API can be adapted to support diverse patient-centric data models.

Materials and methods: We develop methodology for extending i2b2's pre-existing API to query additional data models, using i2b2's recent "multi-fact-table querying" feature. Our method involves developing data-model-specific i2b2 ontologies and mapping these to query non-standard table structure.

Results: We implement this methodology to query OMOP and PCORnet models, which we validate with the i2b2 query tool. We implement the entire PCORnet data model and a five-domain subset of the OMOP model. We also demonstrate that additional, ancillary data model columns can be modeled and queried as i2b2 "modifiers."

Discussion: i2b2's REST API can be used to query multiple healthcare data models, enabling shared tooling to have a choice of backend data stores. This enables separation between data model and software tooling for some of the more popular open analytic data models in healthcare.

Conclusion: This methodology immediately allows querying OMOP and PCORnet using the i2b2 API. It is released as an open-source set of Docker images, and also on the i2b2 community wiki.

Figures

Figure 1.

Web services on data warehouses. Shown here: the i2b2 database uses i2b2 XML REST services to communicate with the query tool. The bottom two arrows on the left show hypothetical connections to PCORnet and OMOP.

Figure 2.

Showing the linkage between i2b2 ontology and non-i2b2 data models.

Figure 3.

A simplified query XML for the query in Figure 4, showing a query for all black, female patients with diabetes mellitus. The “item key” provides the unique ontology link that identifies each data element in the query.

Figure 4.

i2b2 querying OMOP’s synPUF 1000-patient dataset. This query took 2.2 seconds.