Skip to main content

Integration Engines

An integration engine sits between health systems and handles the work that neither end wants to own: transport, message transformation, terminology mapping, routing, retries, and the audit trail of what was sent where.

Without one, every pair of systems needs its own interface — and interfaces grow as n².


What an engine does​

FunctionWhy it matters
ConnectivityMLLP, HTTPS, SFTP, database polling, message queues
TransformationHL7 v2 ⇄ FHIR ⇄ CSV ⇄ XML ⇄ JSON
Terminology mappingLocal codes to LOINC, ICD, SNOMED CT
RoutingContent-based delivery to one or many destinations
Error handlingRetry, dead-letter queues, alerting on failure
AuditA record of every message, needed for ISO 27799
MonitoringThroughput, latency and failure dashboards

Common options​

OpenHIM — the interoperability layer of the OpenHIE architecture. Purpose-built for health exchange: mediators, transaction log, role-based routing. The default choice for national-scale health exchange.

Mirth Connect / NextGen Connect — the long-standing workhorse of hospital integration. Channel-based, strong HL7 v2 support, JavaScript transformers. Widely known by hospital IT teams, which lowers the staffing risk.

Apache Camel — a general-purpose integration framework with a large connector library and an HL7 component. Powerful, code-first, and it expects engineers rather than integration analysts.

Rhapsody / InterSystems Ensemble (IRIS) — commercial engines used at hospital and regional scale, with vendor support and clinical toolkits.

OpenFn — workflow automation used widely in the global development sector for connecting programme systems.


Design guidance​

  1. One canonical model. Map each system to a shared representation — normally FHIR — rather than writing n×n direct mappings.
  2. Idempotency. Assume messages will be delivered twice; make processing safe when they are.
  3. Never drop a message silently. Dead-letter and alert; silent loss is the failure mode nobody notices until reconciliation.
  4. Version your mappings and keep them in source control with tests.
  5. Separate transformation from business rules. Engines that accumulate clinical logic become undocumented applications.
  6. Instrument from day one. Throughput and error-rate dashboards are how you find out an interface stopped, rather than hearing it from a clinician.
  7. Protect PHI in transit and in logs. Message payload logging is the most common accidental data exposure in integration work.

  • OpenHIE — where the engine sits in a national architecture
  • APIs — observability for API traffic
  • HL7 v2 and HL7 FHIR — what flows through it

References​