Referral Data Portability: Test the Export Before You Need to Leave

Why portability belongs in evaluation, not in the exit

Inventory where each required record lives. A referral platform may hold identifiers, communications, status history and open tasks, while some underlying documents remain in another authorized system. Do not assume either that everything is stored in the platform or that the EHR already has every event. The AHRQ Workflow Assessment for Health IT Toolkit frames health IT as something that changes both clinical and administrative workflow and that should be planned for across design, implementation and use (AHRQ). Leaving a system is a workflow change too. Identify the approved source for any history the export does not contain. If that source will become unavailable, the transition has an unresolved dependency.

The Linear Health Operational AI Evaluation Kit treats this as one of seven decision areas. Its commercial and exit dimension asks buyers to confirm data ownership, export formats, access during transition and deletion evidence, lists a sample data export among the evidence to request, and names as a red flag the case where a buyer cannot retrieve its records or explain the transition path before signing (Linear Health evaluation kit). This article narrows that dimension into a concrete work product you can reuse across vendors. It is not legal advice on contract terms, and it is not the full migration runbook; see the referral management system migration guide for that.

Step 1: List the records, attachments and histories that must remain usable

Start with what you would need on day one in a replacement system, then work backward to what the current system holds. A practical inventory has five groups:

Core records. Referral header data: patient identifier, referring and receiving provider, specialty, urgency, order date, reason text, insurance fields as captured, and current status.

Histories. Status change log with timestamps and the user or automation that made the change. Communication history: outbound fax or portal submissions, inbound responses, call notes, patient messages. Identify which of these records you need to answer "what happened to this referral" after the switch, including any that remain in another approved source.

Attachments. Clinical documents, order forms, insurance cards, consent scans, received reports. Note the file formats and whether the export links each file to its parent referral.

Mappings. Provider directory, location list, payer list, specialty codes, status vocabulary, and any crosswalk the vendor built between its internal codes and your EHR or practice management identifiers. Inspect the mappings explicitly, including who maintains them and whether they are included in the proposed export.

Open work. Unfinished tasks, scheduled follow-ups, items in exception queues, and automations mid-flight. Check how open work will remain assigned after the transition. A complete set of referral headers does not by itself establish that their pending tasks were exported.

For each group, record the system fields involved, the approximate volume, and the owner who can judge whether an exported version is usable.

Step 2: Request an export specification and a representative sample

Ask two things in writing. First, an export specification: file formats, table or object list, field definitions, how relationships are expressed (foreign keys, nested objects, filenames), how attachments are packaged, time zone handling, character encoding, and whether the export is self-service, scheduled, or requires a vendor ticket. Second, a representative sample drawn from your own tenant if you are already a customer, or from a demonstration dataset with similar structure if you are evaluating.

Define representative yourself. A sample should include at least one referral with several status changes, one with multiple attachments, one that was cancelled or declined, one with an open task, and one that was matched to a provider through a mapping rather than entered manually. The evaluation kit makes the same point in general terms: run a real workflow with normal complexity and at least one exception, because a scripted happy path is not enough (Linear Health evaluation kit).

If you are issuing a formal procurement, these requests fit naturally into the data section of the referral management RFP template.

Export acceptance specification (reusable artifact)

Adapt this proposed specification to the intended use and agreed scope. Mark each applicable item pass, partial, fail or not provided, with evidence. It is an operational worksheet, not a statement of contractual entitlement.

Step 3: Reconcile counts, relationships and open work

Compare the export with the same source population at the same cutoff or documented snapshot. Two files taken on the same date can still represent different states if work changed between captures. Use the three checks below to inspect coverage; they are not a guarantee that every possible defect will be found.

Count reconciliation. Define filters, included statuses and the cutoff for each object type before comparing counts. For a sample export, compare the selected sample population, not the full live total. Explain differences and inspect identifiers: equal totals can still hide one missing record and one duplicate.

Relationship reconciliation. Select cases that cover the relationships in your specification. Confirm that exported history, attachments and tasks can be associated with the correct parent record. An unmatched reference is a defect to investigate; do not assume the relationship from a patient name or a matching total.

Open-work reconciliation. For the selected population and cutoff, reconcile open tasks, their current owners and due actions. Document whether active automation is represented as data, paused for a transition or resumed through another approved process. Exported task data is not proof that a destination will execute it.

Labelled synthetic example

The following numbers are invented for illustration. They are not from any Linear Health customer and are not a measured result.

In this invented case, inspection establishes the stated omission of automation-generated history. The header totals match, but that does not resolve the missing task records or the 225 unlinked files. Each finding needs its own explanation, owner and retest. The table illustrates reconciliation logic; it does not estimate how often these defects occur.

Step 4: Test whether another approved system can read the export

Choose an approved destination suited to the intended test, such as a candidate replacement platform or a controlled analysis environment. Load the selected sample and locate a referral by its identifier, follow its event history and open the linked attachments. Also inspect an unfinished task and its assigned next action. A spreadsheet may help inspect fields, but it does not establish that a replacement workflow can resume operations.

Keep the test inside approved data paths. The evaluation kit flags sensitive data entering a pilot before responsibilities and data flows are documented (Linear Health evaluation kit). Use synthetic data to design the exercise. Before using any patient-derived sample, follow the organization's approved data-handling and access process with its designated reviewers. Calling a file de-identified does not itself authorize a transfer.

A successful import into a test destination establishes only the tested format, data population and operations. Record transformations, excluded objects and manual steps needed to make it usable. Ask any proposed referral coordination software vendor for that evidence rather than inferring import coverage from an integration logo.

If you want to walk through how Linear Health approaches exports and integration evidence with your own inventory in hand, you can book a demo and bring the acceptance specification above.

Step 5: Document gaps before the exit becomes urgent

Turn every partial or failed criterion into a gap record with four fields: what is missing, why it matters operationally, who owns resolution (buyer or vendor), and a target date. Keep the conclusion limited to what was tested. Linear's evaluation kit distinguishes claims, demonstrations and verified production evidence (Linear Health evaluation kit). One successful import does not automatically verify every object, permission or operating workflow. Record the supported scope and unresolved cases alongside any score.

As editorial guidance, repeat a light version of this test at each renewal and after any major platform change, and whenever a relevant format, mapping or source-system change gives you a reason to retest. Keep the gap register with your contract file so whoever handles a future transition starts from evidence rather than from memory.

For teams that want the broader seven-dimension scorecard, the Operational AI Evaluation Kit is a Linear-authored tool that can organize evidence requests across vendors.

If you are comparing referral coordination platforms now and want to see export and integration evidence rather than a scripted demo, book a demo with your inventory and sample scenarios ready.

Frequently asked questions

Sources

Sources accessed October 25, 2026.

• AHRQ, Workflow Assessment for Health IT Toolkit.
• Linear Health, Operational AI Evaluation Kit.