201809 FHIR Storage and Analytics

Return to Fall 2018 Proposals

• Zulip Analytics stream
• Zulip Storage stream

Track Name

FHIR Storage, Query & Analytics ([1])

Track Overview

More and more developers about to start design storage for FHIR data. We hope this track will share experience about FHIR storage implementation, as well as analytics on FHIR datasets.

Please fill out: TODO [TODO Track Registration Spreadsheet]

Submitting WG/Project/Implementer Group

Should we create a group?

Justification

Questions to be answered

• How to store FHIR data?
• What is database schema design?
• Which databases can be used?
• SQL on FHIR
• FHIR data representation for storage
• FHIRPath subset for databases
• What i have to do to be part of it?
• How to approach FHIR search?

Proposed Track Lead

• Nikolai Ryzhikov ?
• Ryan Brush ?
• Patrik Sundberg ?
• Kenny Blanchette ?

Expected participants

• Nikolai Ryzhikov
• Tim Zallin
• Alexander Zautke
• Ryan Brush
• Patrik Sundberg
• Kenny Blanchette

Roles

• Database Provider: provides access to database with FHIR data
• Database User: design queries

Discussions

SQL on FHIR

Initial notes for creating a standard representation of FHIR that is directly usable in scalable SQL-based systems.

Original document by Ryan Brush - https://docs.google.com/document/d/1IPyI-4GYlF400vmZvrtjDYL6jauHdjeLfjGJMo6yWtA/edit#heading=h.ch8y6lqurlaj

FHIR data representation in database

JSON and other JSON like formats like yaml, edn, avro, protobuf are very popular in avantе-garde of programmers, some modern databases now could understad JSON as first-class data-structures, that’s why we think, good design for it is strictly required for FHIR adoption.

Original github - https://github.com/fhir-fuel/fhir-fuel.github.io/issues

FHIRPath subset for databases

It worth to discuss subset of FHIRPath, which is implementable in databases

Issue to discuss - https://github.com/fhir-fuel/fhir-fuel.github.io/issues/10

Scenarios

Scenario 1: FHIR search

• Design or take an existing database schema to store Patient, Encounter & Practitioner resources
  • relational (consider schema generation)
  • document oriented
    • postgresql jsonb
    • mongodb
    • big query
  • tripple store (datomic, EAV)
  • xml database (?)
• Load sample data
• Implement FHIR search for
  • Patient by name, address
  • Encounter by date and location/practitioner
  • Encounter include patient/practitioner
  • Encounter chained params
• On fly convertion to FHIR if format is different

Scenario 2: Advanced FHIR search

• Design or take an existing database schema to store Patient & Observation
• Implement search by quantity with respect to system and units

Scenario 3: Complex Queries / CQL

• Implement CQL to SQL (or other query lang) translation (automatic or manual)
• Another analytic queries???

Scenario 4: Analytical databases replication

• Get `transaction log` / history of all CRUD/transaction operations from kafka topic
• Transform and load into analytical databases
  • Click House
  • Elastic Search
  • Vertica
  • Relational databases (MS SQL, Oracle, Postgresql, Mysql)
• Run analytical queries

Scenario 5: Graphql implementaton

• prototype efficient graphql => sql transpilation

Discussion

fhirpath implementation/subset for databases

Assets

• we will provide you with test datasets
• jupyter environment with examples (will be used for demo after track)
• access to existing databases
  • fhirbase
  • Biq Query
  • aidbox

Outcomes

• make participants familiar with different approaches
• report/guidelines for implementation of FHIR database
• discuss open questions in a group :)

Databases

Relational

• PostgreSQL
• Big Query

Document databases

• MongoDB

Analytical

• ElasitcSearch
• ClickHouse
• Vertica
• Spark / Hadoop?

Integration bus

• Kafka