NewMissing Records Detection: flags every visit, provider, and date missing from the file. See how →
Home›Solutions›RADV Audit Response
CMS RADV AUDIT RECORD RESPONSE

One best record per sampled condition. Thousands of charts to find it in.

CMS RADV audit response services read every candidate chart for each sampled HCC and identify the 1 record that best supports it under CMS reviewer guidance. Each selection is page-cited to the signature, the date of service, and the diagnostic statement, so the submission is checkable before it goes.

By Ahmed Jemaa, Co-Founder & CEO of Medrecords AI · Published 22 Aug 2026 · Updated 30 Aug 2026

Click any row → the source page it cites
Record selection · Case #IME-4812 documented
Adams, Timothy · RADV sample chart set
342 pp / 2 packets received logged
11 duplicate pages removed free
p.140 — wrong patient quarantined
Pages 342 Documents 27 Cited 100%

The reviewer guidance is published. So is the clock.

We do not build for a case type until we can name the standard that defines a correct output and the person who grades the file against it. In RADV audits, both are already published.

The standard and the referee
The rulebook
  • The CMS RADV medical record reviewer guidance, which defines an acceptable record: valid provider type, face-to-face encounter, date of service in the data year, and a legible signature and credential
  • 42 CFR Part 422 Subpart O and the RADV audit provisions, including the extrapolation methodology finalized for audits from payment year 2018 forward
  • The ICD-10-CM Official Guidelines, which govern whether the documented condition supports the submitted diagnosis code
  • The CMS-HCC risk adjustment model mapping the diagnosis to the HCC under audit
The referee
  • The CMS RADV medical record reviewer, who accepts or rejects each submitted record
  • CMS itself, which extrapolates the confirmed error rate across the contract
  • The HHS administrative appeal process, and the courts where an extrapolation methodology is challenged
Coders reading every candidate chart by hand against a 5-month submission window From 10¢ to as low as 5¢/page here, duplicates free
Record selection · Case #IME-4812 cited 100%
Packet342 pp / 2 packets · 11 deduplicatedlogged
Sampled HCCsBest record identified for eachcited
Signature & credentialLocated and page-citedcited
Unsupported samplesNo qualifying record foundflagged
Selection delivered · cited 100% · no submission on your behalf
How it works

3 steps between the sample and the submission.

01

Send the sample and the charts

The CMS sample list plus every candidate chart for each enrollee-HCC pair, in any format and any order.

02

We return the selection, page-cited

For each sampled HCC, the record that best meets the reviewer guidance, with the date of service, provider type, signature and diagnostic statement each cited to its page.

03

Your team submits

Coding leadership reviews the selection and makes the submission. We never submit to CMS, code a diagnosis, or certify a record as compliant.

The boundary, in writing

A selection from your charts, not a submission on your behalf.

We read the candidate charts and identify, with citations, the record that best matches the CMS reviewer guidance for each sampled HCC. We do not code diagnoses, certify a record as audit-compliant, submit anything to CMS, or estimate an error rate. Those stay with your coding and compliance leadership.

Where no candidate chart meets the guidance, we say so rather than submitting the least-bad option. A sample with no qualifying record is information your compliance team needs before the window closes, not after.

Audit trail · Case #IME-4812 exportable
08:12Packet received · 342 pp / 2 packetssystem
08:3111 pages deduplicatedsystem
08:44p.140 flagged — wrong patientsystem
08:54Record selection complete · selection complete · every sample cited · cited 100%system
09:20Record selection reviewed · citations verifiedreviewer
09:26Record selection exported · CMS submission made off-platformreviewer
Every access logged · deleted on request
Why Medrecords AI

The rules the platform never breaks.

Medrecords AI EVERY LINE CITED
CASE #IME-4812 · ADAMS, T.342 pp
2/14 — ER visit, right knee p.4
4/18 — arthroscopic surgery p.61
p.140 — wrong patient quarantined
Medrecords AI
Read every page · cite every line

HIPAA, under a signed BAA

Every file is handled under our Business Associate Agreement, from the first byte.

Never trains a model

Your records are never used to train any AI model — ours or anyone else's.

Every line cited

If we can't cite it, we don't say it. Every sentence links to its source page.

Deleted on request

Retention is configurable and deletion is on request, with a full audit log of every access.

Powered by the platform

4 capabilities behind every record selection.

The record selection is one deliverable of the same platform that reads, sorts, and cites the whole record.

Medical Records OCR

Every candidate chart read in full — EHR exports, scanned encounter forms, handwritten signatures and credentials — nothing skipped, nothing guessed.

IN ACTION · 342 pp / 2 packets → 27 documents, read in full

Medical Chronology

Each enrollee's encounters build into 1 timeline, so the qualifying date of service is found rather than hunted.

IN ACTION · data year start → year end, one continuous timeline

Verifiable AI Citations

If we can't cite it, we don't say it. Every selected record links to the page carrying the signature and the diagnostic statement.

IN ACTION · every selection → cited 100%

Medical Summary Reports

The selection formats straight into the submission workbook — your template, your letterhead.

IN ACTION · record selection → submission-ready

Medical Record Deduplication

Thousands of candidate charts for one sampled HCC often carry the same encounter more than once, resent across EHR exports or repeated inside a provider's own production. Every duplicate is confidence-scored and collapsed to one canonical copy before the selection starts.

IN ACTION · 342 pp / 2 packets → 11 duplicates collapsed, free

Missing Records Identification

When no candidate chart in the set actually supports a sampled HCC, that's a finding in itself. The platform flags the encounters a condition's own record implies should exist but that never showed up in the pull, before the submission window closes.

IN ACTION · candidate set checked → gaps named, cited to page
FAQ

RADV audit response, answered.

No. We identify and cite the best candidate record for each sampled HCC against the CMS reviewer guidance. Reviewing that selection and making the submission is your coding and compliance leadership's decision, on your own systems.

We flag it as unsupported rather than picking the least-bad record. A sample with no qualifying record is something your compliance team needs to know before the window closes, and submitting a record that fails the guidance doesn't improve the outcome.

Yes. Provider type, face-to-face encounter, date of service in the data year, and a legible signature with credential are each located and page-cited, because those are the elements a CMS reviewer rejects records on.

No. Medrecords AI does not retrieve records from providers or facilities. You bring the records you already have — review starts in minutes from upload. Retrieval vendors take days; you can keep yours and still cut the review to minutes.

Send 1 file. We'll tell you what we can read.

No obligation. If the case type is buildable we'll scope it; if it isn't, we'll say so.

Sources

Every rule, regulation, statute and published standard named on this page is linked below to the text the body that issues it puts out, read on the date shown. Where a standard is sold rather than published, the link goes to its publisher's page for it. A few rules are named without a link, because no source we could read and confirm publishes them.

Ranked lists