gms | German Medical Science

Background: One major goal of decentral research networks is to be able to query and retrieve the data after it has been harmonized across multiple sites. This data is highly structured: In medical research, the data of a case or patient usually resides in multiple interlinked entities, each containing specific parts of the information. A rising standard for this kind of data is FHIR (Fast Healthcare Interoperability Resources), which provides predefined but extensible resources for those entities.

Methods: FHIR brings its own easy way to query for data based on the RESTful paradigm [1] as part of the FHIR standard: the so-called FHIR search specification [2]. While built into every relevant FHIR server, this API neither supports queries with complex interrelated criteria nor can it be used for subqueries and complex aggregations. In order to determine whether FHIR Search is a valid way to enable complex feasibility queries, a collection of 32 inquiries gathered from researchers of the German Cancer Consortium and the German Biobank Alliance to evaluate the different ways to query for FHIR data. The data model used as a reference and for testing, consists of the Oncology Dataset from the German Cancer Consortium, which will be part of the Medical Informatics Initiative [3] and the Biobank-oriented dataset from the German Biobank Alliance.

Results: Even seemingly simple queries prove to be problematic in FHIR Search. Because of the use-cases this search has to accommodate for, it is not sufficient to have a predefined way to combine criteria as it is done in the FHIR Search: In FHIR Search the criteria are always used on disjoint entities. For example, if a search is looking for a biosample a) of type tissue, that b) has been stored in a freezer at -90°C it would also return a sample of tumor tissue and another sample that has been frozen. To formulate such criteria, the query language needs to allow defining how exactly search criteria should be connected to entities, and to return the result of a query in a flexible manner using aggregations to visualise the result. These requirements are met by Clinical Quality Language (CQL), a rich query language with built-in arithmetic, relational, comparison and logical operators.

Conclusion: As of April 2020, there is a lack of available implementations of the Clinical Quality Language. This led to the development of a new FHIR store called Blaze [4] within the German Biobank Alliance as part of the Bridgehead [5]. To build an active development community, Blaze has been made open-source as part of the Samply open source community. Even with the technical way to build very complex queries solved, an open problem remains, as how to enable the user of a search portal to use the richness and complexity provided by such a language. While power-users can formulate Clinical Quality Language directly, the need for a highly flexible, yet intuitive search interface for non-technical users is an important next step for distributed research networks.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.