Skip to main content

Medical billing scrubber

A clean claim is not a correct claim.

A medical billing scrubber checks draft claim fields and supporting documentation against the applicable code, edit and payer-policy versions. The pattern Werkon would validate groups source-linked issues for qualified review and reruns checks after approved corrections. Clinicians, coders and billing owners resolve documentation and reporting questions; named authority releases the claim, with acceptance, adjudication and payment tracked separately.

Screening depends on source and effective version

Keep documentation, extracted facts, proposed and accepted codes, claim fields and payer results separate. Required fields, identifiers, code dates, line references, units, totals, duplicates and transaction constraints need deterministic checks. Applicable Procedure-to-Procedure, Medically Unlikely and add-on-code edits retain their exact scope and version; Medicare edits must not be assumed to govern every payer.

An edit hit is a review signal. A modifier indicator does not justify a modifier without documentation, and syntactic transaction validity does not establish accurate reporting. Missing, unsigned, copied, late or contradictory evidence needs review or a neutral clinician query, preserving original and amended records without supplying the answer.

Disposition, release and later payer evidence

Qualified owners decide sequencing, modifiers, units, present-on-admission status, attribution, medical necessity and payer interpretation. Record issue severity, owner, disposition and rationale, then rebuild approved corrections and rerun checks without erasing prior results. Named authority releases the claim. Transmission, clearinghouse acceptance, adjudication, denial, payment, correction, void and appeal remain distinct, linked back to the original evidence.

Scrub boundary

Flag the issue without deciding the code.

A rule can identify a condition worth review. Four boundaries keep that signal from becoming unsupported coding, documentation or payment authority.

01

Encounter, documentation, and claim identity

Bind the exact organization, patient, encounter, provider, facility, payer, coverage, claim type and service period; preserve authored records, signatures, amendments and prior submissions; and distinguish source facts from claim assertions.

Required evidence: Tenant and organization, patient and encounter identifiers, rendering and billing providers, facility and place of service, payer and coverage context, claim and frequency identifiers, service dates, document author and status, signature, amendment, source version and access history.

02

Date-effective coding and rule context

Resolve the code, transaction, edit and payer-policy releases that apply to the date and claim type; preserve licenses and source provenance; and classify each check by scope rather than treating one rule set as universal.

Required evidence: ICD-10-CM or PCS release, CPT or HCPCS context, modifier and place-of-service sources, transaction guide and companion rules, NCCI quarter and edit family where applicable, payer policy version, effective period, jurisdiction, claim type and configured exceptions.

03

Claim-line validation and issue candidates

Check structure, required fields, code dates, pointers, units, arithmetic, duplicates and configured edits deterministically; compare claim assertions with source evidence; and surface missing or conflicting facts as reviewable candidates.

Required evidence: Claim header and lines, diagnosis and procedure references, modifiers, units, charges, authorization and supporting fields, validation result, exact edit identifier, input values, source passage, confidence, missingness, contradiction, severity and rerun history.

04

Qualified disposition, release, and outcome

Assign each candidate to an authorized coder, biller, clinician, compliance or payer owner; preserve disposition and correction; release only after named approval; and reconcile transmission, adjudication, payment and later correction.

Required evidence: Reviewer role and evidence viewed, neutral query and response, accepted or corrected claim fields, rationale, approval and release time, payload digest, acknowledgment, rejection, adjudication, denial, remittance, payment, corrected or voided claim, appeal and variance cause.

Evidence-to-outcome path

Keep the claim draft, review decision, and payer result separate.

A claim changes as documentation is clarified, coding is approved and payers respond. Each stage retains what was known and who held authority at that time.

  1. 01

    Assemble the bounded claim packet

    Resolve patient, encounter, providers, facility, coverage, payer and claim type; collect the approved documentation and orders; record signatures and amendments; and hold mismatched or incomplete identities before screening.

    Owner
    Health-information, identity, privacy and billing-intake owners
    Evidence
    Identity resolution, encounter and service dates, provider and facility roles, payer and coverage context, claim type, source manifest, document status, access purpose, ambiguity and hold decision.
  2. 02

    Freeze the applicable rule context

    Select effective code and transaction releases, configured NCCI or payer edits and organization policy by date, claim type and jurisdiction; validate provenance and license boundaries; and record the exact version bundle.

    Owner
    Coding, billing, compliance, payer and integration owners
    Evidence
    Code-set release and dates, transaction version, edit family and quarter, payer companion rule, organization policy, jurisdiction, claim-type applicability, source URL or licensed reference, configuration digest and approval.
  3. 03

    Generate source-linked issue candidates

    Run deterministic structure, date, arithmetic and edit checks; compare claim assertions with signed evidence; identify missing and contradictory facts; and group candidates without selecting a final code or changing documentation.

    Owner
    Billing-technology, data, coding and audit owners
    Evidence
    Input claim digest, rule and input trace, pass or hit result, source citation, extracted candidate, missingness and contradiction, confidence, affected line, severity, duplicate grouping and model or parser version.
  4. 04

    Resolve issues under qualified authority

    Route each issue to the right coder, biller, clinician, compliance or payer owner; use neutral clinical queries when needed; accept, correct, hold, escalate or reject with rationale; and preserve the original and every rerun.

    Owner
    Qualified coding, billing, clinical-documentation and compliance owners
    Evidence
    Assignment, evidence viewed, query wording and recipient, response and amendment, disposition, rationale, approved field correction, rule exception, reviewer identity, timestamp, unresolved item and rerun comparison.
  5. 05

    Release, observe, and reconcile

    Require final release approval, generate the transaction from approved fields, preserve payload and acknowledgment, track rejection and adjudication separately and connect denial, remittance, payment, correction, audit and appeal to the scrub record.

    Owner
    Billing release, clearinghouse, payer-relations, finance and quality owners
    Evidence
    Released claim and approver, transaction digest, send attempt, acknowledgment, accepted or rejected state, payer claim identifier, adjudication and denial details, remittance, payment, correction, void, appeal, post-payment finding and learning action.

Authority map

Separate exact validation, uncertain assistance, and final coding judgment.

Automation can prove that a field conforms to a configured rule. It cannot prove that clinical documentation supports a code or that a payer must pay it.

01

Deterministic claim controls

Software owns tenant and patient boundaries, identifiers, effective dates, required fields, code existence, transaction structure, line references, units, arithmetic, duplicate checks, configured edit execution, issue state, access and immutable receipts.

  • Patient, encounter, provider, payer, claim and line identifiers
  • Code-set, transaction, edit, policy and configuration versions
  • Required-field, effective-date, pointer, arithmetic, duplicate and schema checks
  • Issue, review, rerun, release, transmission and response receipts
02

Bounded AI assistance

Models can extract documented candidates, link claim fields to source passages, identify likely contradiction or missingness, cluster related flags, explain configured rules and draft neutral review questions while preserving uncertainty.

  • Documentation and claim-field extraction candidates
  • Source-to-line comparison and contradiction candidates
  • Issue grouping, prioritization and plain-language explanations
  • Neutral clinician-query and reviewer-packet drafts
03

Qualified human authority

Qualified people own documentation clarification, diagnosis and procedure coding, sequencing, modifiers, units, medical necessity, payer interpretation, compliance, claim correction, release, appeal and post-payment response.

  • Clinical fact and documentation clarification
  • Final code, sequence, modifier, unit and diagnosis-pointer decisions
  • Coverage, payer-policy, compliance and release judgments
  • Denial, appeal, correction, refund and complaint decisions

Scrub components

Build an evidence ledger, not a green checkmark.

The same claim can pass one rule set and still be unsupported, inapplicable or unpaid. Four components preserve the scopes that a single clean badge would hide.

01

Encounter and source-evidence registry

Bind organization, patient, encounter, providers, facility, payer, coverage, claim type, service dates, signed notes, orders, reports, amendments, prior claims, access purpose and correction history.

Operating contract: A claim field is not source documentation, a copied note is not proof of a distinct service, an unsigned draft is not a signed record, an amendment must not erase the original and a code must not be used to infer that care occurred.

02

Versioned code and rule registry

Record effective ICD-10-CM or PCS, CPT or HCPCS context, modifiers, place-of-service and other applicable code sources, transaction and companion rules, NCCI edit families, payer policies, jurisdiction, claim type and configuration approvals.

Operating contract: Current is date-specific, one release must not validate another service period, NCCI scope must not be generalized to every payer, an edit indicator is not modifier permission and a technical standard does not decide clinical support, coverage or payment.

03

Issue and disposition ledger

Preserve rule inputs and outputs, source-linked extraction, missing and conflicting evidence, affected claim lines, candidate severity, owner, neutral query, response, disposition, correction, rationale, approval and rerun result.

Operating contract: Flag is not error, absence is not negative evidence, model confidence is not coding confidence, pass is not correctness, accepted exception is not deleted history and automation must not write clinical facts or select financially favorable codes.

04

Claim and financial-outcome ledger

Link reviewed and released claim versions, transmitted payload, acknowledgment, payer claim identifier, rejection, adjudication, denial, remittance, payment, correction, void, appeal, post-payment review and variance cause.

Operating contract: Release is not transmission, transmission is not acceptance, acceptance is not adjudication, adjudication is not payment, payment is not permanent correctness and later outcomes must not rewrite the evidence or judgment available at release time.

Delivery path

Prove one claim family through remittance.

Start with one claim type and payer path whose documentation, current manual review and later payer outcomes can be observed without expanding coding authority.

  1. 01

    Choose one bounded claim cohort

    Select one organization, claim type, service family and payer path with stable source records, licensed code access, named qualified reviewers, meaningful current issues and enough adjudication evidence for later comparison.

  2. 02

    Map sources, versions, and authority

    Inventory encounter evidence, claim fields, code and edit releases, payer rules, current review steps, clinician queries, release permissions, submission route, acknowledgments, denial handling and correction paths.

  3. 03

    Build deterministic and source-linked checks

    Implement exact identity, date, required-field, code, reference, unit, arithmetic, duplicate and configured-edit checks; add source-linked extraction candidates; and preserve every input, version and result.

  4. 04

    Test difficult billing cases

    Exercise wrong patient or encounter, unsigned and amended notes, service-date release edges, obsolete codes, missing pointers, duplicate lines, unit outliers, edit hits, unsupported modifiers, conflicting facts, corrected claims and delayed payer responses.

  5. 05

    Release narrowly and reconcile outcomes

    Begin in advisory mode with full qualified review, compare supported detections and burden with the current process, monitor holds and inappropriate suggestions and join rejection, adjudication, denial, payment and correction before expanding.

Release controls

Six controls before a flag can influence a claim.

A plausible correction can create clinical, financial and compliance harm. These controls keep every proposed change subordinate to source evidence and qualified authority.

Patient, encounter, and source identity are exact
Resolve organization, patient, encounter, providers, facility, payer and claim type; preserve authored documents, status, signatures and amendments; enforce tenant and purpose boundaries; and hold mismatches before any comparison.
Rules are versioned and scoped
Select code, transaction, edit and payer-policy versions by service date, claim type and jurisdiction; record provenance and license constraints; test applicability; and reject silent fallbacks to the newest or broadest rule set.
Every issue exposes evidence and rule inputs
Show the exact claim line, source passage, missing or conflicting fact, rule identifier, version, evaluated values and uncertainty; distinguish structural, coding, documentation, coverage, payer and compliance categories.
No automated documentation or coding authority
Models never add clinical facts, alter notes, select final codes, choose modifiers to bypass edits, upcode, unbundle or release claims. Neutral queries, coding changes and exceptions require authorized people and attributable rationale.
Corrections are reversible and reviewable
Preserve original claim and issue state, apply only approved field changes, rebuild and rerun deterministically, retain disagreements and exceptions and provide independent hold, correction, void, appeal and incident paths.
Submission and financial states stay separate
Record release, transmission, acknowledgment, acceptance, rejection, adjudication, denial, remittance, payment and post-payment review independently; reconcile later variance without calling a clean or paid claim permanently correct.

Outcome evidence

Measure supported dispositions and downstream variance, not flags.

More edits and more alerts can increase work without improving claims. Evidence must show whether the right issue reached the right reviewer and how the released claim behaved later.

Baseline

  • Claim types, service families, payers, source systems, code and rule releases, qualified owners, review paths, release permissions, submission routes and correction processes
  • Current time and human effort from documentation and claim draft through coding, edits, clinician queries, billing review, release, rejection, denial and appeal
  • Current missing, conflicting and unsigned evidence; obsolete codes; pointer, unit, modifier and duplicate issues; edit hits; false alerts; holds; overrides and unresolved candidates
  • Current acknowledgments, rejections, adjudication outcomes, denials, remittance variance, payment corrections, post-payment findings, complaints and operational delays

Outcome evidence

  • Correct patient, encounter, provider, payer, claim type, service date, source and effective-rule handling against approved reference evidence
  • Supported issue precision and recall by category, source-citation quality, qualified disposition time, query burden, rerun stability and unresolved-case ownership
  • Wrong-version, unsupported-code, modifier-bypass, duplicate, arithmetic, privacy, cross-tenant, unapproved-change and premature-release prevention
  • Rejection, adjudication, denial, remittance, payment, correction and appeal variance against released claim evidence with payer changes and other confounders visible

Guardrails

  • Patient or encounter mismatch, cross-tenant disclosure, unauthorized access, wrong claim type, excessive clinical data, missing document status and overwritten amendment history
  • Wrong effective code or edit release, unlicensed reference use, NCCI applied outside scope, payer rule generalized, blank treated as no issue and transaction conformance called correctness
  • Invented clinical fact, model-selected final code, leading query, unsupported modifier, upcoding, unbundling, hidden contradiction, alert fatigue and unowned hold
  • Release without authority, acknowledgment called adjudication, acceptance called payment, denial not reconciled, correction history erased and claim activity presented as financial or compliance outcome

Fit test

Use this pattern when issue decisions can be traced through payer outcomes.

Good reason to begin

  • One claim family has stable source records, date-effective rule access, named qualified reviewers, a known submission path and enough later adjudication evidence to test the screening result.
  • Documentation, extracted facts, code candidates, claim fields, rule hits, human dispositions, released payloads and payer results can remain linked but distinct.
  • The organization can preserve original records, exact versions, missingness, disagreements, neutral queries, approved corrections and reversible claim history.
  • False alerts, unsupported suggestions, payer outages, delayed responses, correction, appeal and manual fallback can be tested without delaying necessary care or claims.

Resolve before beginning

  • Patient and encounter matching, documentation status, code-set licensing, effective-date logic, claim type, qualified review, release authority or outcome evidence is undefined.
  • The process cannot preserve source records or distinguish extraction, validation, edit, coding, documentation, coverage, submission, adjudication and payment states.
  • Success is defined by flags or clean claims without reviewer burden, supported dispositions, false-alert analysis, denials, corrections and downstream variance.
  • The scrubber is expected to invent documentation, assign final codes, bypass edits, optimize reimbursement, decide medical necessity, release claims or guarantee acceptance and payment autonomously.

Source basis

Sources behind the control model.

  • 01

    National Center for Health Statistics

    ICD-10-CM files and effective releases

    CDC and NCHS publish ICD-10-CM files by fiscal release and service-date period. The April 1, 2026 FY26 release applies to services from April 1 through September 30, 2026 and replaces the October 1, 2025 FY26 release. A current code file does not prove that documentation supports a diagnosis, that sequencing or reporting is correct, that a claim is covered or that a payer will adjudicate it favorably.

  • 02

    Centers for Medicare and Medicaid Services

    Medicare NCCI Procedure-to-Procedure edits

    CMS states that NCCI PTP edits prevent inappropriate payment for code pairs that generally should not be reported together and publishes additions, deletions and modifier-indicator changes quarterly. The July 1, 2026 files are current for the reviewed date. An edit hit or modifier indicator does not establish documentation support, clinically appropriate modifier use, medical necessity, other payer policy, legal compliance or final payment.

  • 03

    Centers for Medicare and Medicaid Services

    Health Care Claim or Equivalent Encounter Information

    CMS identifies ASC X12N 837 Version 5010 as the adopted US HIPAA electronic claim standard for institutional, professional and dental claims. A transaction can conform structurally while containing unsupported, incomplete or payer-inapplicable information, and conformance does not prove documentation, coding, coverage, adjudication, payment or compliance with every applicable requirement.

  • 04

    Health Level Seven International

    HL7 FHIR Release 5 Claim resource

    The current published FHIR R5 version 5.0.0 trial-use Claim resource represents a provider request for adjudication and can carry patient, provider, insurer, diagnosis, procedure, supporting information and claim items. It is a data-exchange model, not validation that documentation supports a code, the patient or coverage was resolved, a payer accepts the request, adjudication is favorable, payment occurred or implementation complies with applicable rules.

[ WORKFLOW / SYSTEMS AUDIT ]
THE FIRST ENGAGEMENT

Start with one real workflow

A Systems Audit is the usual starting point. If the opportunity is already clear, we can move directly into a focused build.

Show Us the WorkflowStart with the free automation readiness checklist

OBSERVEQUANTIFYDECIDEBUILD