Exploration of OMOP and FHIR

Base project for exploration of OMOP to FHIR® conversion.

Overview

This document describes the creation of a test environment using synthetic clinical data to explore the GT-FHIR2 OMOP on FHIR project with two different versions of the OMOP CDM. Included here are terraform scripts to set up resources in AWS and instructions to perform the following:

1. Configure a PostgreSQL database for mulitple versions of the OMOP CDM
2. Use the SYNTHEA™ software package to generate sample data
3. Use the ETL-Synthea tools to create the CDM schemas, load vocabularies, and load synthetic data
4. Configure multiple versions of the GT-FHIR2 mapper and run via Docker

One can certainly just work with a single version of the OMOP CDM. In that case you can ignore the redundant steps.

Base infrastructure

For this project we use a PostgreSQL instance in AWS RDS and a single EC2 instance.

You can use the terraform scripts provided here (tf/) to create a user account with appropriate permissions and with this user account create the EC2 instance and RDS database. The scripts make use of AWS Secrets Manager for the database credentials.

Setup

You'll need to install the following on your test host

• git
• Docker
• docker-compose
• psql client
• java
• R, including devtools

Create database and schemas

psql -h <RDS endpoint> -U <username> -P <password> postgres
postgres=> CREATE DATABASE omopfhir;
postgres=> \connect omopfhir
omopfhir=> CREATE SCHEMA cdm60;
omopfhir=> CREATE SCHEMA cdm53;
omopfhir=> CREATE SCHEMA native60;
omopfhir=> CREATE SCHEMA native53;

The cdm{53,60} schemas will store the OMOP CDM (versions 5.3.1 and 6.0, respectively); the native{53,60} schemas will store the untransformed synthetic data.

Synthetic data

For the purposes of this project we will generate two synthetic datasets using the Synthea system.

Install Synthea

git clone https://github.com/synthetichealth/synthea
cd synthea
./gradlew build check test

Configure to generate CSV output

We will make use of the OHDSI ETL-Synthea loader. For this we need Synthea to create CSV output. We will be making two distict datasets to compare across populations so we also configure the system to create separate output directories. Edit src/main/resources/synthea.properties

exporter.csv.export = true
exporter.csv.folder_per_run = true

You may want to disable FHIR output to save disk space.

Create test datasets

./run_synthea -s 2020 -p 10000 Massachusetts
./run_synthea -s 2020 -p 10000 Illinois

This will create output files in timestamped subdirectories in synthea/output/csv

Get OMOP vocabulary files

Note that the ELT-Synthea concept loader expects lowercase filenames.

OHDSI Synthea data loader

Have a look at https://github.com/OHDSI/ETL-Synthea. Once you have successfully installed R and devtools, you can follow directions there to load the native and cdm schemas.

Now, we're going to run this twice because we are going to use different schemas with different versions of the CDM for our test datasets.

For the first pass, set cdmSchema to "cdm53" and syntheaSchema to "native53", and point sytheaFileLoc to one of the CSV output directories. For the second pass, use "cdm60" and "native60".

As an optimization, you can use the vocabulary files from the first schema to populate the second. For this you would use the LoadVocabFromSchema.r script instead of LoadVocabFromCsv.r, specifying "vocabSchema". This is much faster.

GT-FHIR2 server

Instructions for setting up multiple versions of the GT-FHIR2 system as used in this study can be found in the OMOPonFHIR directory.

Accessing data via the GT-FHIR2 server

The default configuration for either the v6 to R4 or v5 to R4 server creates the FHIR API at http://<server>:8080/omoponfhir4/fhir; the configuration used here specifies port 8080 for the CDM v6 schema and port 8081 for the CDM v5.3 schema. The server can be accessed easily from the command line for testing purposes, viz.,

curl -H "Authorization: Basic <auth>"  'http://<server>:8080/omoponfhir4/fhir/Patient/123/'

Where <auth> is the (base64-encoded) AUTH_BASIC string defined in the relevant Dockerfile.

Client examples

FHIR client libraries have been developed for both Python and R. The fhircrackr library for R supports basic authentication as is the default for the OMOPonFHIR server. Here is a simple example to identify all Condition resources that contain a code corresponding to infection by the SARS-CoV-2 virus:

library(fhircrackr)
fhir <- 'http://<server>:8080/omoponfhir4/fhir'
req <- paste0(fhir,"/Condition?code=840539006")
bundle <- fhir_search(req, username=user, password=pwd, verbose=0)

condition_design <- list(

    Conditions = list(
        
        resource = "//Condition",
        
        cols = list(
            CID       = "id",
            onset     = "onsetDateTime",
            abate     = "abatementDateTime",
            patient   = "subject/reference",
            encounter = "encounter/reference"
        ),
        
        style = list(
            sep = "|",
            brackets = c("[","]"),
            rm_empty_cols = FALSE
        )
    )
)
cracked <- fhir_crack(bundles=bundle, design=condition_design)

# Extract patients from bundle by “melting” the cracked bundle
cols <- ("patient")
melt <- fhir_melt(cracked$Conditions, cols=cols, 
                     brackets = c("[","]"),sep=" | ", all_columns = FALSE)

patients <- fhir_rm_indices(melt, brackets = c("[","]")

# patient[,1] contains the Patient resource identifiers associated 
# with the Conditions