JMIR Publications

ABSTRACT

Background:

Harmonizing and standardizing the wealth of digital medical information has been pursued for many years. The research data repositories commonly created with research in mind typically require a large amount of harmonization and transformation efforts on the clinical data. The FHIR format on the other hand was designed to document clinical processes and is therefore closer to the clinical data model and more available across modern EHRs. However, all common standardized data formats do not directly lend themselves well to statistical analysis and therefore preprocessing of the data for statistical analysis is needed.

Objective:

This study aims to show how this preprocessing can be achieved on the FHIR format directly.

Methods:

We propose that the binary JSON (jsonb) format of the PostgreSQL open source database is suitable not only for storing FHIR data, but also for building preprocessing and filtering services on top of, which directly transform data stored in FHIR into prepared subsets of data for statistical analysis. We specified an interface for and implemented said preprocessor, deployed it at the University Hospital Erlangen-Nürnberg and created three example datasets and analyses on the available data.

Results:

We loaded patient data from 2016 to 2018 into a standard PostgreSQL database containing around 35.5 Mio. FHIR resources including: Patient, Encounter, Condition (ICD-10), Procedure (OPS) and Observation (laboratory results). We then integrated our pre-processing service with the PSQL database and the locally installed web-based KETOS analysis platform. We created three exemplary subsets and analyses to demonstrate the feasibility of the preprocessor.

Conclusions:

The study demonstrates how a standard open source tool like PSQL can be used to store FHIR data, but also the feasibility of developing further preprocessing on top to enable advanced filtering and prepared dataset creation for further statistical analysis. The web-based preprocessing could be deployed locally at a particular site, protecting a patient’s privacy and integrate well with existing open source data analysis tools currently being developed across Germany.