pfb_fhir: A utility to extract clinical data systems into a portable format

BackgroundFast Healthcare Interoperability Resources (FHIR) is a server specification and data model that allows for EHR systems to represent clinical metadata using a consistent API. There is a critical mass of EHR and clinical trial data stored in FHIR based systems. Research analysts can take advantage of existing FHIR tooling for de-identification, pseudonymization, and anonymization. More recently the BiodataCatalyst consortium has proposed the Portable Format for Bioinformatics (PFB) which is a carrier format for describing raw data and the data model in which it is structured, based on an efficient binary format (AVRO). PFB allows an entire cohort of metadata to be loaded into a research data system. Here, we describe an open source utility that will scan FHIR based systems and create PFB based archives. Resultspfb_fhir scans data from FHIR based clinical data systems and converts the data into a self contained PFB file. This utility identifies types, customizations (extensions), and element connections. It then converts all of these components into a graph model compatible for storage in the PFB specification. The structure of the original FHIR system is faithfully reproduced using the PFB schema description system. All records from the system are downloaded, converted and stored as vertices in a graph described by the PFB file. This system has been tested against a number of different FHIR installations, including ones hosted by dbGAP, The Kids First Data Resource and AnVIL. Conclusionspfb_fhir helps to unlock the potential of EHR and clinical trial data. pfb_fhir allows researchers to easily scan and store FHIR resources and create self contained PFB archives, called FHIR in PFB. These archive files can easily be moved to new data systems, allowing the clinical data to be connected to more complex genomic analysis and data science platforms. The FHIR in PFB archives generated by pfb_fhir have been loaded into data platforms including the Broads Terra system, Gen3 based data system, custom graph query engines and Jupyter notebooks. This flexibility will enable genomics investigators to do more integrated genotype to phenotype association analysis using whichever tools suit their line of research.