A distributed analytics platform to execute FHIR-based phenotyping algorithms

Despite the benefits of reusing health data collected in routine care, sharing datasets outside of the organizational boundaries is not always possible due to the legal and ethical restrictions. The Personal Health Train (PHT) is a novel privacy-preserving approach to execute analytics tasks at distributed data repositories, without sharing the data itself. In this work, we report a proof-of-concept implementation of the PHT by using FHIR data standards and Clinical Query Language (CQL). The Semantic Web and containerization technologies have been utilized to move computations to the data. We developed tools to design phenotyping algorithms on the data consumer side, implemented an infrastructure to transfer and execute Docker containers at the data centers, and to return results to the consumers. We experimented the evaluated PHT infrastructure and tools by designing a phenotyping algorithm for diabetes mellitus and prostate cancer risk case-control study and executed it at three distributed FHIR repositories.