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
| Function | Why it matters |
|---|---|
| Connectivity | MLLP, HTTPS, SFTP, database polling, message queues |
| Transformation | HL7 v2 ⇄ FHIR ⇄ CSV ⇄ XML ⇄ JSON |
| Terminology mapping | Local codes to LOINC, ICD, SNOMED CT |
| Routing | Content-based delivery to one or many destinations |
| Error handling | Retry, dead-letter queues, alerting on failure |
| Audit | A record of every message, needed for ISO 27799 |
| Monitoring | Throughput, 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
- One canonical model. Map each system to a shared representation — normally FHIR — rather than writing n×n direct mappings.
- Idempotency. Assume messages will be delivered twice; make processing safe when they are.
- Never drop a message silently. Dead-letter and alert; silent loss is the failure mode nobody notices until reconciliation.
- Version your mappings and keep them in source control with tests.
- Separate transformation from business rules. Engines that accumulate clinical logic become undocumented applications.
- Instrument from day one. Throughput and error-rate dashboards are how you find out an interface stopped, rather than hearing it from a clinician.
- Protect PHI in transit and in logs. Message payload logging is the most common accidental data exposure in integration work.
Related
- OpenHIE — where the engine sits in a national architecture
- APIs — observability for API traffic
- HL7 v2 and HL7 FHIR — what flows through it
References
- OpenHIM — https://openhim.org/
- Mirth/NextGen Connect — https://github.com/nextgenhealthcare/connect
- Apache Camel — https://camel.apache.org/