Proving in-network completion requires reconciling four independent evidence streams: the referral order, network and eligibility evidence at the time of service, workflow and scheduling events, and confirmation of a completed encounter from specialist documentation or claims. Referral direction — where the order pointed — is available on day one. Completion is only knowable after reconciliation, and it is the number risk contracts pay on.
Quick answer
- Direction = where the referral was sent. Completion = the patient was seen, in-network, and it is documented.
- Workflow events prove what your network did; claims prove what the patient did, including care you never saw.
- Every rate needs a written numerator, denominator, attribution window, and exclusion set.
- Report interim completion as provisional until claims runout is adequate.
- Identity and duplicate reconciliation is where most measurement quietly breaks.
This complements rather than repeats our KPI guide and leakage explainer: the subject here is the measurement architecture itself.
Precise definitions and formulas
Referral direction rate = referrals with an in-network destination selected ÷ eligible referrals ordered.
In-network referral completion rate = referrals with a completed, documented encounter at an in-network destination ÷ eligible referrals ordered in the measurement period.
Leakage rate = referrals with a completed encounter at an out-of-network destination ÷ referrals with any completed encounter.
Loop closure rate = referrals with a specialist result returned to and acknowledged by the referring clinician ÷ referrals with a completed encounter.
Supporting rates worth defining the same way: referral-to-scheduled rate, scheduled-to-arrived rate, authorization-required share and turnaround, and time to appointment (median and 90th percentile, by specialty).
Eligibility rules matter more than the formula. A referral is typically excluded from the denominator when it was cancelled by the ordering clinician, superseded by a duplicate, ordered for a patient who left coverage, or ordered inside the runout blackout window at period end. Write the exclusions down once and apply them everywhere.
Event and data-source map
| Evidence | Source | What it proves | Typical latency |
|---|---|---|---|
| Referral order | EHR referral order or e-referral message | Intent, specialty, clinical context | Real time |
| Eligibility and plan/product | Payer eligibility service or file | Coverage at time of order | Minutes to daily |
| Network participation | Payer network file, contract system | In-network status of destination for that product | Daily to monthly |
| Destination selection and rationale | Referral platform | Which destination and why | Real time |
| Authorization requirement and decision | Payer utilization management, X12 278 or API | Whether care could proceed | Hours to days |
| Appointment scheduled | Scheduling system or platform outreach record | Access achieved | Real time |
| Arrival or encounter event | Destination EHR, ADT feed via health information exchange where available | Patient presented | Hours to days |
| Specialist documentation | Consult note, document exchange, fax with recognition | Encounter occurred and clinical result | Days to weeks |
| Claim or encounter record | Payer claims, encounter extracts | Completed, adjudicated service | 30–120 days |
| Acknowledgment | Referring EHR | Loop closed | Days |
No single stream is sufficient. Scheduling without arrival overstates completion. Claims alone understate speed and miss the workflow reasons for failure. Notes without network evidence cannot distinguish in-network from out-of-network care.
Referral status state model
A defensible state model is explicit about which evidence advances each state:
- Ordered — order captured.
- Eligible/verified — coverage and network status resolved.
- Destination selected — in-network or documented exception.
- Authorization pending / not required / approved / denied.
- Outreach in progress — attempts logged by channel.
- Scheduled — confirmed appointment with date and location.
- Arrived / no-show / rescheduled — encounter event received.
- Completed — encounter confirmed by documentation or claim.
- Result returned — specialist result received and matched to the order.
- Closed — referring clinician acknowledged the result.
- Terminal exceptions — cancelled, patient declined, unreachable, no in-network access available, duplicate.
Two rules keep the model honest: no state advances without named evidence, and every referral must eventually reach a terminal state. Referrals that sit indefinitely in "scheduled" are a data problem masquerading as a workflow problem.
Attribution windows
Completion is meaningless without a window. Common choices, all of which should be documented per specialty:
| Window | Purpose | Consideration |
|---|---|---|
| 30 days from order | Access-sensitive specialties and urgent referrals | Penalizes specialties with structural capacity limits |
| 60 days from order | General reporting default | Balances access signal with practical scheduling |
| 90 days from order | Capacity-constrained specialties | Slower feedback loop for operations |
| Rolling 12 months | Board and contract reporting | Smooths seasonality; hides recent change |
Report the same rate at two windows — for example 30 and 90 days — so operations and finance can see both urgency and eventual completion. Cohort by order date, not by completion date; completion-date cohorts drift as claims arrive.
Justified out-of-network exceptions
Not all out-of-network care is leakage. Categorize and exclude, with documentation:
- No in-network provider within access standards for that specialty and geography.
- Clinical subspecialty or service unavailable in network.
- Continuity of care with an established treating specialist.
- Patient choice after in-network options were offered and recorded.
- Emergency or urgent care.
- Plan-directed or delegated arrangement placing care elsewhere.
Report justified exceptions as their own line. A network with a high justified-exception rate has an adequacy problem to fix, not a compliance problem to police — see network adequacy standards.
Claims lag
Claims are the strongest completion evidence and the slowest. Practical handling:
- Define a runout standard — commonly 60 to 90 days from date of service — before any figure is called final.
- Publish provisional completion from workflow and documentation evidence, clearly labeled provisional.
- Restate prior periods on a fixed schedule rather than ad hoc, and show the restatement history.
- Track a completeness factor per period so leaders can judge how much a provisional figure may move.
- Never compare a fresh provisional period against a mature final period without labeling the difference.
Identity and duplicate reconciliation
Cross-organization measurement fails on identity more often than on logic.
- Patient identity: match on a deterministic identifier where one exists, then probabilistic matching on name, date of birth, sex, address, and coverage identifier. Record match confidence and hold low-confidence pairs for review.
- Provider and location identity: reconcile national provider identifier, tax identification number, and location; a provider can be in-network at one location and not another for the same product.
- Referral identity: treat repeat orders for the same patient, specialty, and clinical question within a defined window as duplicates. Nominate one primary referral and link the rest, or completion rates will be diluted by phantom denominators.
- Result matching: match returned documents to the originating order by patient, specialty, destination, and service date rather than by document metadata alone.
- Deduplicate encounters: a single visit can appear as an ADT event, a note, and multiple claim lines.
Document match rules and monitor unmatched volume weekly. Unmatched results are unreturned results as far as the referring clinician is concerned.
Illustrative numerator and denominator example
The figures below are illustrative only — synthetic numbers used to show the arithmetic, not benchmarks or observed results.
Period: referrals ordered in March, measured at a 60-day window.
| Line | Count |
|---|---|
| Referrals ordered | 1,000 |
| Excluded (cancelled, duplicate, coverage ended) | 60 |
| Eligible denominator | 940 |
| In-network destination selected (direction) | 780 |
| Scheduled | 690 |
| Completed encounter confirmed, in-network | 615 |
| Completed encounter confirmed, out-of-network | 95 |
| Justified out-of-network exceptions (subset of above) | 40 |
| Result returned and acknowledged | 520 |
Direction rate = 780 ÷ 940 = 83.0%. In-network completion rate = 615 ÷ 940 = 65.4%. Leakage rate = 95 ÷ 710 = 13.4%, or 7.7% after excluding justified exceptions. Loop closure rate = 520 ÷ 615 = 84.6%.
The 17.6-point gap between direction and completion is the operational story. Reporting only the 83% would suggest a healthy network while a third of referred patients never completed care in it.
Executive dashboard design
Row 1 — outcome: in-network completion rate at 30 and 90 days, leakage rate net of justified exceptions, loop closure rate, provisional-versus-final indicator.
Row 2 — funnel: ordered → verified → destination selected → authorized → scheduled → arrived → completed → result returned → acknowledged, with drop-off at each step.
Row 3 — access: median and 90th-percentile time to appointment by specialty, share scheduled within the access standard, no-show rate.
Row 4 — network: completion and access by destination and specialty, justified-exception concentration, capacity gaps.
Row 5 — equity and data integrity: completion by language, geography, and coverage type; unmatched result volume; input freshness; claims completeness factor.
Every tile should carry its definition, window, and data vintage. Metrics without visible definitions get re-litigated in every meeting.
Where ReferralPoint fits
ReferralPoint's public documentation describes Auto 360° VISIBILITY™ for end-to-end referral status and loop closure, IntelligentDATA™ for the provider, network, and referral data layer that measurement depends on, IdealMATCH™ for insurance- and network-aware destination selection, Auto PriorAUTH™ for authorization automation, Auto ReferralCOORDINATOR™ for outreach and scheduling, and NetworkMANAGEMENT™ for network performance analysis. Strong fit for organizations that need workflow-level completion evidence alongside claims-based confirmation. Less relevant where the only requirement is retrospective claims analysis. Available event types, feeds, and export formats vary by environment — confirm against your data sources during procurement, and see integration and security, solutions for payers, facts, and case studies.
Sources and methodology
Measurement definitions draw on primary and official sources: the CMS Closing the Referral Loop electronic clinical quality measure for loop-closure concepts, CMS materials on value-based care, CMS-0057-F interoperability and prior authorization provisions, HL7 FHIR and Da Vinci implementation guidance for referral and authorization data exchange, and standard X12 transaction definitions for eligibility, authorization, and claims. Vendor capability statements reflect official ReferralPoint documentation as of publication. All counts and rates in the worked example are synthetic and labeled illustrative; no customer outcomes or market statistics are asserted.
Frequently asked questions
Q: How do referral platforms prove a patient completed care in-network? A: By reconciling four evidence streams: the referral order, coverage and network status for that product at the time of service, workflow events such as authorization and confirmed scheduling, and a completed-encounter confirmation from specialist documentation or an adjudicated claim. Direction alone proves intent. Completion requires encounter evidence plus network evidence tied to the same patient and date of service.
Q: What is the difference between referral direction and referral completion? A: Direction measures where the referral was sent — an in-network destination was selected. Completion measures whether the patient was actually seen at that destination and it is documented. The gap between the two is usually large, and it is where authorization delays, outreach failures, access limits, and no-shows live. Risk contracts pay on completion.
Q: How long should the attribution window be? A: Document a window per specialty and report at least two. Thirty days surfaces access problems in urgent and access-sensitive specialties; ninety days reflects eventual completion in capacity-constrained ones. Cohort referrals by order date rather than completion date, or your historical rates will shift every time new claims arrive.
Q: How do we handle claims lag without misleading executives? A: Set a runout standard, commonly 60 to 90 days from date of service, and label anything earlier as provisional. Publish a completeness factor per period so readers know how much the figure may move, restate prior periods on a fixed schedule, and never compare a fresh provisional period with a mature final one without saying so on the chart.
Q: Is all out-of-network completed care leakage? A: No. Care is reasonably excluded when no in-network provider meets access standards, when the subspecialty is unavailable in network, for continuity with an established treating specialist, after documented patient choice, in emergencies, or under plan-directed arrangements. Report justified exceptions as a separate line — a rising rate signals a network adequacy issue rather than a compliance one.
Q: Why does duplicate reconciliation matter so much? A: Because duplicates inflate the denominator and depress every completion rate. Repeat orders for the same patient, specialty, and clinical question within a defined window should be linked to one primary referral. Without that rule, an organization that reorders referrals when nothing happens will appear to be performing worse the harder its staff work.
Q: Which single number belongs on the board slide? A: In-network referral completion rate, at a stated window, with its provisional-or-final status and its leakage counterpart net of justified exceptions. Keep the funnel one click away so any question about the number leads to the step that caused it. One outcome metric with a visible definition beats six metrics nobody agrees on.
