grafhir

This is an attempt to load FHIR R4 json files into a graph data model for running graph analytics to build applications for chronic care management, Population health etc.

Introduction

Current set of artifacts, loads FHIR for multiple patients and runs similarity alogorithms on different dimensions.

Current data loading is simple per patient. Here we link patient with conditions, encounters, medications, procedures, observations.

but my aim is to build cross patient grapth with temporal,status labeled relationship with conditions, medications, procedures, encounters and observations.

System Requirements

Memory : 8G+

Java 11+

docker

Create directories in local file system for

log : Neo4j database logs

plugins : Neo4j add ons like APOC, gds jars are to be copied here

imports : FHIR data to be loaded as graph is stored here

Downlaod plugin jars from neo4j site for following and copy in plugins folder

1. neo4j-graph-data-science-1.8.6.jar
2. apoc-4.2.0.4-all.jar

Downlaod patient FHIR R4 data for your study and copy to "import" directory

Start NEO4J container locally using following command

docker run --name testneo4j -p 7474:7474 -p7687:7687 -d -v $[DATA]:/data -v $[LOG]:/logs -v $[IMPORT]:/var/lib/neo4j/import -v $[PLUGINS]:/plugins --env NEO4J_AUTH=neo4j/******* --env NEO4J_dbms_connector_https_advertised__address="localhost:7473" --env NEO4J_dbms_security_procedures_unrestricted="apoc.,cyfhir." --env NEO4J_dbms_connector_http_advertised__address="localhost:7474" --env NEO4J_dbms_connector_bolt_advertised__address="localhost:7687" --env NEO4J_dbms_security_procedures_unrestricted=apoc.,gds. --env NEO4J_dbms_memory_heap_max__size=8G --env NEO4J_apoc_import_file_enabled=true neo4j:4.2.19

Note : Replace values for DATA,LOGS,IMPORT,PLUGINS varibales as per your file system. and password for NEO4J_AUTH

Visualize patient FHIR in NEO4J browser

Open a neo4j browser using below

http://localhost:7474/browser/

Refer cypher-queries.cql file for loading dataset and executing analysis queries. Including GDS queries with cosine similarity