Open source relational storage for FHIR targeting real production.

Here is an interactive demo, which is build with FHIRPlace & FHIRFace by-products.

While crafting Health IT systems wey understand a value of properly chosen domain model. FHIR is an open source new generation lightweight standard for health data interoperability, which (we hope) could be used as a foundationan for Health IT systems. FHIR is based on a concept of resource.

FHIR® is a next generation standards framework created by HL7. FHIR combines the best features of HL7 Version 2, Version 3 and CDA® product lines while leveraging the latest web standards and applying a tight focus on implementability.

Also we learned that data is a heart of any information system, and should be reliably managed. PostgreSQL is battle proved open source database, which supports structured documents (jsonb), while preserving ACID guaranties and richness of SQL query language.

PostgreSQL is a powerful, open source object-relational database system. It has more than 15 years of active development and a proven architecture that has earned it a strong reputation for reliability, data integrity, and correctness.

Here is list of PostgreSQL features we use:

• jsonb
• gin & gist
• inheritance

We actively collaborate with PostgreSQL lead developers to craft production ready storage for FHIR.

Why we are doing this inside database?

We decided to implement most of FHIR specification inside database for scalability reason (all data operations is done efficiently in databse).

This approach also gives you possibility use FHIRBase from your prefered lang/platform (.NET, java, ruby, nodejs etc). We implemented FHIR compliant server in clojure, with small amount of code - FHIRPlace.

And there is option to break FHIR specification abstraction (if required) and go into database by generic SQL interface and complete your business task.

• resources versioning
• CRUD operations on resources
• search operations
• optional indexing (to speedup search)
• history
• tags operations
• transaction

Please follow FHIRPlace installation instructions.

Requirements:

• PostgreSQL 9.4 (http://www.postgresql.org/about/news/1522/)
• pgcrypto
• pg_trgm
• btree_gin
• btree_gist

You can download postgresql 9.4 pre-release or build Postgresql from source on debian/ubuntu and create local user cluster with:

source cfg/local_cfg.sh && ./install-postgres
psql postgresql

By default postgresql will be built in .build directory. source cfg/local_cfg.sh in fresh shell to export required variables

NOTE: you can tune configuration in local_cfg.sh.

You can install FHIRBase:

source local_cfg.sh
echo 'CREATE DATABASE mydb' | psql postgres
export DB=mydb

./runme integrate

Fhirbase could be installed using docker

#run database container
docker run --name=fhirbase -d fhirbase/fhirbase

docker inspect fhirbase
# read ip of started container

docker run --rm -i -t fhirbase/fhirbase psql -h <container-ip> -U fhirbase -p 5432

Here we used image from dockerhub - fhirbase

You could build image by yourself:

git clone https://github.com/fhirbase/fhirbase/
cd fhirbase
docker build -t fhirbase:latest .
#run database container
docker run --name=fhirbase -d fhirbase

docker inspect fhirbase
# read ip of started container

docker run --rm -i -t fhirbase psql -h <container-ip> -U fhirbase -p 5432

To start quickly read Getting Started Guide.

We heavily use PostgreSQL inheritance feature, for polymorphic operations.

Here are base tables:

CREATE TABLE resource (
  version_id uuid,
  logical_id uuid,
  resource_type varchar,
  updated TIMESTAMP WITH TIME ZONE,
  published  TIMESTAMP WITH TIME ZONE,
  category jsonb,
  content jsonb
);

CREATE TABLE resource_history (
  version_id uuid,
  logical_id uuid,
  resource_type varchar,
  updated TIMESTAMP WITH TIME ZONE,
  published  TIMESTAMP WITH TIME ZONE,
  category jsonb,
  content jsonb
);

For each resource type FHIRbase generate two tables (which inherit from base tables) - one for actual resources and one for history of resource.

This is done, to separate dataspaces for each resource, so they are not messed up and can guarantee performance proportional to amount of data for particular type of resource.

Note: Same trick is used by PostgreSQL for partitioning.

For example for resource Patient we are generating tables:

CREATE TABLE "patient" (
  logical_id uuid PRIMARY KEY default gen_random_uuid(),
  version_id uuid UNIQUE default gen_random_uuid(),
  resource_type varchar DEFAULT '{{resource_type}}',
  updated TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP,
  published TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP,
  content jsonb NOT NULL,
  category jsonb
) INHERITS (resource);

CREATE TABLE "patient_history" (
  version_id uuid PRIMARY KEY,
  logical_id uuid NOT NULL,
  resource_type varchar DEFAULT '{{resource_type}}',
  updated TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP,
  published TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP,
  content jsonb NOT NULL,
  category jsonb
) INHERITS (resource_history);

For more information see source code:

Most of FHIR complaint operations could be done with FHIRBase procedures, which guaranties data integrity and do heavy job for you. All procedures have first parameter _cfg::jsonb with configuration params, which required for url generation. Now there is only one paramenter [base] (Service Root URL): {"base":"http://myserver"}

CRUD operations on resource are represented as set of procedures:

Create a new resource with a server assigned id Return bundle with newly entry;

Example:

SELECT fhir_create(
  '{"base":"fhir/end/point"}',
  'Patient',
  '{"resourceType":"Patient"}');

Result will be:

Result:

{
  "id": "f27638e4-801b-4c5b-8d16-81912358f537",
  "entry":
   [{
     "id": "fhir/end/point/Patient/7043326d-f2eb-4d8b-81c5-13ace5a8a3d7",
     "link": [{"rel": "self", "href": "fhir/end/point/Patient/7043326d-f2eb-4d8b-81c5-13ace5a8a3d7/_history/7dd8adfc-0f3b-4b21-b44a-497dba699df9"}],
     "content": {
        "resourceType": "Patient"
     },
     "updated": "2014-12-19T22:54:16.926811+04:00",
     "category": [],
     "published": "2014-12-19T22:54:16.926811+04:00"
   }],
  "title": "Concrete resource by id 7043326d-f2eb-4d8b-81c5-13ace5a8a3d7",
  "updated": "2014-12-19T22:54:16.926811+04:00",
  "resourceType": "Bundle",
  "totalResults": 1
}

fhir_create has several forms:

• fhir_create(_cfg jsonb, _resource jsonb)
• fhir_create(_cfg jsonb, _type text, _id uuid, _resource jsonb, _tags jsonb)
• fhir_create(_cfg jsonb, _type text, _resource jsonb)
• fhir_create(_cfg jsonb, _type text, _resource jsonb, _tags jsonb)

You can skip _type parameter (it will be taken from resourceType attribute of resource).

Also you can pass tags as jsonb array: '[{"label":???, "system":???, "value":???}]'::jsonb

Read the current state of the resource by logicalId (TODO: or by reference/url) Return bundle with only one entry for uniformity;

SELECT fhir_read('{"base":"fhir/end/point"}', 'Patient', uuid);

-- or you can search resource by id with ordinary SELECT

SELECT * FROM patient WHERE logical_id = 'uuid'

Read specific version of resource with type Returns bundle with one entry;

Update resource, creating new version Returns bundle with one entry;

DELETE resource by its id AND return deleted version Return bundle with one deleted version entry ;

Retrieve the changes history for a particular resource with logical id (id) Return bundle with entries representing versions;

Search in resources with type by params Returns bundle with entries;

Helpful functions:

• search('Patient', 'name=smith') -- returns rows from resoure table
• build_search_query('Patient', 'name=smith') -- shows generated search sql
• analyze_search('Patient', 'name=smith') -- shows execution plan for search query

Indexes are optional.

Index for specific search parameter could be created using index_search_param() function:

SELECT index_search_param('Patient', 'name');
-- Time: 96168.725 ms

SELECT count(*) FROM patient;
--  count
---------
--  2262107
-- (1 row)

SELECT content->'name'
FROM search('Patient', 'name=john');
-- Time with index: 20.859 ms
-- Time without index: 1186.767 ms

SELECT content->'name'
FROM search('Patient', 'name=nonexisting');
-- Time with index: 20.01 ms
-- Time without index (i.e. full scan): 74429.122 ms

SELECT fhir_search('{}'::jsonb, 'Patient', 'name=john');
-- Time: 35.785 ms

-- But they are not for free

SELECT admin_disk_usage_top(10);

-- public.patient,                      "454 MB"
-- public.patient_name_name_string_idx, "140 MB"

You can drop indexes:

SELECT drop_index_search_param('Patient','name');
SELECT drop_resource_indexes('Patient');
SELECT drop_all_resource_indexes();

Returns conformance resource jsonb;

• Star us on GitHub
• If you encountered a bug, please make an Issue
• Contribute to FHIRBase − see dev/README.md

Powered by Health Samurai

Sponsored by:

See development details in dev/README.md

Copyright © 2014 health samurai.

FHIRbase is released under the terms of the MIT License.