EHR Referral Reconciliation: Find Records That Did Not Synchronize

This guide proposes an administrative method and a synthetic worksheet. It does not describe a tested Linear Health feature or prescribe clinical follow-up timing.

Define what the comparison is meant to prove

An EHR and a coordination platform may contain different parts of a referral workflow. A document visible in one tool might not be intended to appear in the other. Start with a field and event map that states which records should exist in each system, which values should agree and which side supplies each value.

The Operational AI Evaluation Kit asks buyers for field-level reads and writes, failure and retry evidence, and handling of matching and recovery. Those questions provide a starting point for defining the comparison. They do not establish the behavior of any particular integration.

AHRQ's workflow assessment toolkit addresses how health IT affects administrative work. The reconciliation worksheet here is an original proposed method within that scope. For the broader operational context, see referral tracking and electronic referral management systems.

Fix the cohort and cutoff before exporting

Choose the population from an agreed originating event, such as referrals received at covered locations during a defined period. Carry that list of referral identifiers into both comparisons. Do not independently filter one system by receipt time and the other by local record-creation time: a later-created copy could fall outside the second export.

For each run, record:

A receipt-only cohort will not catch every update to an older referral. Include a separate change check or carried-forward list for those records. If either export cannot represent the agreed cutoff, record that limitation and repeat the affected comparison. Do not classify an apparent mismatch during ongoing processing as a confirmed failure without investigation.

Use approved reporting access and keep the working extracts in the organization's authorized location. Limit the worksheet to the identifiers and administrative fields needed for reconciliation.

Match referrals without merging ambiguous records

Prefer a stable shared referral identifier. If systems assign different identifiers, use a reviewed crosswalk that links the two. A patient identifier alone is insufficient for the proposed method because one patient can have more than one referral.

When no crosswalk exists, additional fields such as destination, originating order reference and date can help identify candidates. Treat those as matching evidence, not permission to combine records automatically. Route duplicate or ambiguous candidates to an authorized reviewer. Do not use a name-and-date-of-birth match as proof that two referral requests are the same.

Check key uniqueness before comparing fields. A duplicate identifier can make a simple join produce misleading counts. Keep matching exceptions separate from value discrepancies until the record relationship is established.

Distinguish missing records, conflicting values and proven delays

Use separate findings rather than forcing each referral into exactly one bucket:

• Not found in one extract: the expected identifier is absent. Investigate filters, export timing, exclusions and matching before confirming a missing record.
• Conflicting value: a mapped field differs after accounting for the agreed cutoff and translation. Record both values and their sources.
• Propagation beyond the agreed interval: a corresponding source event and destination acknowledgement or applied version show that an expected update arrived late or remains unapplied.
• Insufficient evidence: the available exports or logs cannot establish what happened. Assign the missing-evidence request rather than inventing a root cause.

Two systems can have different last-updated times because they track different actions. Subtracting those timestamps is not a reliable measure of the same event's transit time. Ask the integration owner which event identifier, version or acknowledgement can link the source change to its destination result.

One referral may have several findings. Report counts of distinct affected referrals separately from counts of issues or failed events.

Use the reconciliation worksheet

Keep a run summary and an issue table. The summary describes the population; the issue table records the work needed to explain and correct exceptions.

For a synthetic example, the same agreed cohort yields 212 EHR records and 209 platform records with unique shared keys. There are 207 shared keys, five found only in the EHR and two found only in the platform. The union contains 214 keys: 207 + 5 + 2. Those counts identify records to investigate; they do not explain the cause.

Within the 207 matched records, suppose three scheduled-date values differ and four event traces show delivery beyond the locally agreed interval. These groups may overlap. Do not add them to the seven unmatched keys and call the sum a count of unique affected referrals without checking overlap.

Investigate one discrepancy and verify the correction

Continue the synthetic example with referral R-1058, found only in the EHR extract. First confirm it belongs in the covered cohort and that the platform export was complete. If an intake log then shows an unresolved match, route it through the approved matching process. A missing record does not, by itself, prove a matching failure or justify creating another record.

For a scheduled-date conflict, consult the designated source and scheduling owner. A newer timestamp alone does not prove which date is correct. Use the supported correction or retry procedure, record who authorized it and repeat the relevant comparison. Leave the item open if the destination result cannot yet be verified.

Bring the worksheet and a synthetic discrepancy to a Linear Health demo to inspect how a proposed workflow would expose matching, failed updates and correction evidence.

Assign ownership and choose the review cadence

Name one accountable owner for each run and a backup for open issues. Choose a cadence based on the workflow's expected propagation, volume and existing operational review process. This article does not set a universal weekly check or one-business-day tolerance.

Carry unresolved items into the next run without resetting their original age. Distinguish resolved, still investigating and accepted exclusion. An accepted exclusion remains documented; it is not a successful synchronization. Trend confirmed causes and recurring evidence gaps using consistent definitions.

The referral coordination software overview provides broader workflow context.

Frequently asked questions

Sources

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