Prevent Duplicate Referral Work Across Coordinators and Systems

This article proposes administrative handling methods. It does not decide clinical equivalence, referral validity or payer requirements.

Distinguish a repeated document from a separate referral request

A fax arrives twice. Is it one referral or two? Either kind of mistake matters. Closing a genuine second request as a duplicate loses a patient from the queue. Opening a second record for a resent fax splits one patient's history across two places.

The practical test is to compare the request, not the document. A document is the carrier: a fax image, a portal attachment, an e-referral message. A request is the underlying ask: this patient, from this referring party, for this service, around this date. Two documents can carry one request. One document can also contain two requests, for example two services for one patient.

Proposed comparison fields, in order:

1. Patient identity, using the identifiers your organization already relies on for matching.
2. Referring provider or organization.
3. Requested service or specialty.
4. Referral date or order date on the request, not the fax timestamp.
5. Any reference number from the referring system.
6. Content differences: new diagnosis text, new attachments, changed insurance, a note saying "resend" or "corrected".

These fields identify candidates for review; they do not prove equivalence. Matching values can occur on distinct requests, while a corrected date or destination may belong to an existing request. Use the organization's approved matching and correction process, including clarification from the sender when needed. Preserve an uncertain classification and assign its next action instead of automatically linking or closing a record.

The AHRQ Workflow Assessment for Health IT Toolkit addresses health IT impact on administrative workflow and the use of information to improve it (AHRQ). The point for your team is that the comparison rule needs to exist on paper before any system applies it.

Identify when two staff members are about to repeat the same action

A repeated action can occur even when the referral record itself is correct. Two coordinators may both prepare a call, or a staff member may repeat a send whose result is unknown. Investigate how each action was claimed and what the second person could see before assuming the cause.

Use an action-level ownership method:

• Claim and confirm: staff confirm that their claim succeeded before acting. Test simultaneous claims; a name entered in an ordinary text field may not exclude another worker.
• Separate actions: scheduling and a records request may have different owners on the same referral. Identify the action being claimed.
• Record current state: show preparation, execution, confirmed result or uncertainty, along with the last relevant event.
• Transfer with acceptance: keep an accountable owner until the recipient or covering lead accepts the action. Do not clear ownership first and leave a gap.
• Check uncertain results: before repeating an automated or manual send, inspect completion evidence and pending retries under the supported process.

These are design requirements to evaluate, not claims that a particular system enforces them.

Build the investigation record without silently merging requests

Use the worksheet below to preserve the investigation. Complete it before linking, closing or merging anything. Retain enough evidence for another authorized reviewer to understand the decision.

Duplicate-investigation worksheet

Two rules keep the worksheet honest. First, "Uncertain" on patient identity stops the process; you do not link on a probable match. Second, "Kept separate" is a valid outcome, and a record kept separate after investigation should carry a note so the next person does not repeat the investigation.

Resolve duplicate work while preserving source history

Use the approved procedure for each confirmed classification:

• Repeated document: link the receipt to the existing request where supported, retaining its source and receipt time.
• Correction or amendment: identify the authorized field changes, preserve prior information according to the record process and inspect any downstream action affected.
• Separate request: retain its own workflow and history. Cross-reference related requests where appropriate without treating that link as clinical equivalence.
• Repeated action: determine what already occurred. Cancel only pending duplicate work through the supported path, record completed effects and decide whether follow-up is needed.
• Uncertain: keep the question assigned and visible. Do not close it merely to reduce the queue.

Synthetic example

A second fax arrives for the same patient and apparent request with an updated administrative attachment. Similar dates and names make it a candidate amendment, not a confirmed duplicate. The authorized reviewer verifies its relationship to the existing request under local procedure, records the source and approved correction, and checks whether any queued action used the old information.

If that relationship cannot be established, both sources remain available and the unresolved question stays assigned. A changed service or date prompts further review; it does not, by itself, prove that the request is either new or duplicate. This example is illustrative, not a customer result.

Bring a synthetic version of the case to a Linear Health demo to inspect source history, uncertain matching and current action ownership.

Review the causes and test a prevention rule

Count confirmed classifications and unresolved investigations over a defined period, with referral and document volumes alongside them. A high repeat-document count may justify investigating receipt communication, but it does not prove that missing acknowledgements caused the resends. Likewise, repeated actions may reflect unclear ownership, retries or another cause. Check the event history.

Choose one supported cause and a bounded change. Record the hypothesis, affected workflow and measures before testing it. Compare like populations and note changes in volume or source mix. Keep the number of investigations separate from confirmed duplicate actions and unresolved cases.

The Operational AI Evaluation Kit asks vendors to demonstrate matching, duplicate handling, retries and recovery. Use a synthetic packet with a later amendment, an ambiguous second request and two users attempting the same action. Inspect the retained sources, classification evidence, accepted claim and failed-action recovery.

Ask what is written to the system of record and what remains only in a vendor worklist. A demonstrated lock in one tool does not establish that another tool cannot perform the same action. Include every execution path relevant to the proposed workflow.

For broader context, see referral intake automation, referral tracking and referral coordination software. To review the two-user and cross-system tests, book a demo with the worksheet and synthetic cases.

Frequently asked questions

Sources

Sources accessed November 9, 2026.

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