Risk Adjustment Accuracy
Capture risk accurately and prepare for audit.
AI agents prepare recapture and new suspect reviews with cited records. Clinicians decide in the EHR. Coders approve before submission.
How it works
From a suspected condition to a documented diagnosis.
The same four steps as every Fold workflow. Here the signal is a condition not yet documented this year.
- 1Signals inClaims · JanE11.22 coded last year, not yet this yearEHR · JunEcho: ejection fraction 35%Labs · Jul, OcteGFR 54 and 52, 113 days apartPharmacy · OctFurosemide filled monthly
- 2One patient contextMEMargaret Ellison, 74Medicare Advantage · visit WednesdayVisit with Dr. Hayes Wednesday2 suspected conditionsLast full exam in MarchLives aloneNot documented this yearVisit in 2 days
- 3Next best actionReview two suspected conditions before Wednesday’s visitReady for Dr. Hayes · WedAI agent pulled the supporting linesAI agent checked each against MEATAI agent drafted the pre-visit summaryRHDr. R. HayesOwner · makes the call
- 4Owner signs off, result trackedHeart failure confirmed in today’s noteKidney disease documented with eGFRAngela Brooks, CRC reviewed the codesEvidence kept for auditSource, decision and date kept with each diagnosis
Before the visit Prospective review
Review last year's conditions
and new suspects before the visit.
AI agents cite the source behind each item. Jordan Reyes, RN checks the list before Dr. R. Hayes sees Margaret Ellison.
One visit, two kinds of review
- Recapture E11.22Coded 2025, not yet assessed in 2026Preparing
- Suspect N18.31Possible CKD stage 3a · from 2 labsPreparing
- 54 on 07/08 and 52 on 10/29, 113 days apart
- E11.22 on file, no CKD stage coded
- Suspect opened: N18.31, CKD stage 3a
Check by Jordan Reyes, RNSent to Dr. R. HayesBefore the Wed 11/4 visitAt the visit Concurrent review
Inside the EHR, clinicians see
what the note still needs.
AI agents check each suggested diagnosis against MEAT and show the supporting line or the gap. Dr. R. Hayes decides, documents and signs.
MEAT check inside the EHR
- MMonitor Renal panel and uACR ordered 10/30
- EEvaluate eGFR 54 and 52, 113 days apart
- AAssess Not yet assessed in 2026
- TTreat No treatment documented
Creatinine 1.11 mg/dL (H). eGFR 52 mL/min/1.73m² (L), CKD EPI 2021.
Suggestions stay separate from the signed note. Not every MEAT element is required.
After the visit Retrospective review
Your coders approve every change
before anything is submitted.
Your coders check each proposed change against the source note, then approve, query or hold. Every decision is logged.
Angela Brooks, CRC · review queue
- To reviewAddN18.31HCC 329CKD stage 3aVisit note 11/04 · Dr. R. HayesAssessment: CKD stage 3a, eGFR 52. Recheck uACR.
- To reviewQuery providerF32.ANo HCCDepression, unspecifiedVisit note 11/04 · Dr. R. HayesAsks Dr. R. Hayes: type and severity, if clinically determinable.
- To reviewProposed deleteE66.01HCC 48Morbid obesityEncounter 02/18 · Dr. R. HayesNo BMI or plan in the note. Held for full record review.
Stage and MEAT support are in the signed 11/04 note.
- Note signed Dr. R. Hayes · 11/04 4:12 PM
- Evidence linked AI agent · 11/05 8:30 AM
- Approved Angela Brooks · 11/05 9:14 AM
At audit RADV readiness
Have the record ready before CMS asks for it.
Fold records the source, decision, owner and date behind each HCC as work happens. When CMS picks a member, export the records instead of rebuilding them.
One HCC, with its record for RADV
- Signed note, 11/04 visit
- Provider credential, Dr. R. Hayes, MD
- 2 lab reports, support only
- Audit log, every step
- Sources in one record.
- EHR notes, claims, labs and pharmacy data stay linked to the member.
- AI agents check at scale.
- They test each submitted HCC against the record and send weak support to a coder.
- Work by role.
- Clinicians, CDI nurses, coders and compliance each see their own queue.
- Coding beside the chart.
- Clinicians answer suggestions in the EHR they already use.
- APIs for write-back.
- Suggested diagnoses reach your EHR or coding system through APIs.
Why now
What CMS changed
for Medicare Advantage.
A diagnosis counts only when the signed note supports it.
- V28
CMS finished the V28 phase-in for payment year 2026. Stage and specificity can change the HCC, such as CKD 3a versus 3b.
CMS 2026 rate announcement - RADV
CMS says new RADV audits will cover every eligible Medicare Advantage contract, each payment year.
CMS audit announcement, May 2025 - PY2027
Diagnoses from unlinked chart reviews no longer count toward 2027 risk scores. A narrow exception covers members who were in a plan under a different parent organization the year before.
CMS 2027 rate announcement
Questions from coding
and compliance teams.
Do AI agents assign codes?
No. They suggest codes with cited evidence. Clinicians document and sign, and your certified coders assign the final codes from the signed note.
Will we have to rebuild the record for a RADV audit?
No. Fold records the source, decision, owner and date behind each HCC as work happens. When CMS picks a member, you export the records instead of rebuilding them.
How does Fold handle V28?
Suggestions map to V28 HCCs using the stage and specificity the evidence supports, such as CKD 3a versus 3b. Each one records the model and payment year.
Can Fold queries steer clinicians toward a diagnosis?
No. Queries cite clinical indicators and ask the clinician to clarify, without suggesting an answer. Your CDI and compliance teams approve the templates and monitor their use. Signed records change only through your amendment process.
Do clinicians have to leave the EHR?
No, where your EHR connection supports it. Fold opens in a panel beside the chart. Before go-live we confirm what your EHR accepts back, such as the note, problem list or claim.
Can our own AI agents use this?
Yes. Through APIs and events, they get the same evidence and decisions, under the same rules and audit log. People still decide.