The graph
Every source is a relationship with a spec
Diagnostic data moves between organizations that have no shared identity system, no shared code set, and no shared incentive to make it easy. The integration surface is not one connection, it is a graph of them, and the posture toward each edge is different.
| Connection | Direction | How | What to know |
|---|---|---|---|
| Laboratory information systems | Source | HL7 v2.5.1 ORU^R01 over TLS or SFTP, plus a FHIR R4 read where the LIS supports it. Orders return as ORM or OML. | GPS LIS is a first-class source. Every other LIS is treated as a stranger with a spec, which is the correct posture. |
| Electronic health records | Source and destination | SMART on FHIR app launch for in-workflow access, US Core profiles for read, and Bulk FHIR export for panel-level pulls. | Certified EHRs must expose a standards-based API. When one does not, that is a finding, not an integration problem. |
| Reference and send-out laboratories | Source | Result-only interfaces reconciled back to the originating accession, so a send-out returns to the case it left. | The failure mode is an orphaned molecular result nobody bills and nobody reads. Reconciliation is the whole job. |
| National networks | Source and destination | Query and retrieve through a designated QHIN under TEFCA, including individual access services for patient-initiated requests. | Useful for prior results and outside history. Not a substitute for a direct interface where volume is predictable. |
| Health data aggregators | Source | Normalized clinical and lab feeds from a network aggregator, used to fill in outside results in weeks rather than quarters. | Fastest path to coverage, thinnest control over data quality. Use it to start, replace it where the volume justifies a direct build. |
| Direct secure messaging | Destination | Direct addresses for practices with no interface and no API, because a surprising number of destinations are still an inbox. | Lowest common denominator, and the one that keeps a rollout on schedule. |
| Patient endpoints | Destination | Portal, patient-facing FHIR API with a registered third-party app, and a full electronic export on request. | Patient-directed transmission to a third party is a distinct path with its own record, not a shortcut around authorization. |
| Payers and clearinghouses | Destination | Claims and attachments out, remittance and adjudication back, reconciled against the charge that produced them. | This is where Metaclinic already lives. Diagnostic data access makes the claim defensible instead of merely submitted. |
How a new connection actually goes live
- Specification exchange and a signed agreement that names the data classes in scope. No connection is built against a verbal scope.
- Test messages with synthetic patients, validated against the mapping and against the release rules, not just against the parser.
- Parallel period where the existing path stays live. Cutovers with no fallback are how results get lost.
- Production with monitored acknowledgement rates, a named contact on both sides, and an alert when volume deviates from the established baseline.
Silence is the dangerous failure
An interface that breaks loudly gets fixed the same day. An interface that quietly stops sending one result type looks exactly like a slow week. Volume baselines per message type and per source are the only real defense, and they are monitored rather than reviewed monthly.
Identity across tenants
The same person is a patient at a practice, a subject at a laboratory, a claimant on a matter, and a member at a payer, under four identifiers that do not agree. Metaclinic maintains a link between those identities with a deliberate bias: a missed link is an inconvenience and a false link is a disclosure to the wrong person.
- Deterministic matching on strong identifiers first, probabilistic only as a candidate generator.
- Ambiguous candidates go to a human review queue and stay unlinked until resolved.
- Every link and unlink is recorded with who did it and on what evidence.
- Merges are reversible, because they are sometimes wrong.
Which stakeholder is currently asking you for data you cannot easily give them?
That is the useful first conversation, and it is a short one. Bring the laboratories, the practices, and the thing that breaks today.
service@meta.clinic Read the release model first