Job 1: Linked corrections
- A documented, supported diagnosis coded wrong or missed on a submitted encounter can still be added through a chart review record linked to that encounter
Guide · HCC suspecting
HCC suspecting: prospective versus retrospective, and why 2027 picks a side
Short answer: both models look for the same missing conditions. Retrospective review produces a chart review record after the year closes; prospective suspecting produces a documented face-to-face visit. From payment year 2027 CMS counts only the second on its own [2], so the question is what retrospective review is now for.
PY2027 · unlinked CRRs excludedData year 2026 · DOS to Dec 31Modifier 93 / FQ · audio-only excludedSwitcher exception · kept
Focus
HCC suspecting
Audience
Risk-bearing groups
Framework
2024 CMS-HCC
Effective date
PY 2027
Metric
-1.53% payment
Length
7 min
Prospective suspecting versus retrospective chart review
Both start from prior-year diagnoses, active medications, lab values, specialist notes and discharge summaries. They differ in when the chart is read and where the diagnosis ends up; in 2027 that decides whether it counts at all [2]. Payment status rows follow the CY 2027 Rate Announcement.
| Dimension | Retrospective review | Prospective suspecting |
|---|---|---|
| When the chart is read | After the data year closes | Days before a scheduled visit |
| Who acts on the finding | A coder, submitting a chart review record | The clinician, documenting at the visit |
| Where the diagnosis lands | A chart review record, often unlinked | The encounter note and the encounter data record |
| Evidence created | None new; relies on the existing note | A current-year note showing the condition was addressed |
| Audit posture | Depends on the quality of an old note | Contemporaneous documentation from a face-to-face visit |
| 2027 payment status [2] | Counts only if linked to a submitted encounter | Counts, as every encounter diagnosis does |
The unlinked record, measured
A condition the clinician documents at a 2026 visit is on the encounter record by construction. A condition a coder finds in January 2027 has to be linked to a 2026 encounter that already exists; if no visit addressed it, there is nothing to link to [2].
Three jobs, none of them a standalone submission
CMS's own data rules support this: items and services must be reported on an encounter data record; a chart review record should not be the only record of a service; when additional diagnoses belong to an encounter, a CRR can be linked to it [2].
Job 2: Deletions
- Conditions submitted without support come off
- Protects the RADV sample as much as the score
Job 3: Prompt generation
- A retrospective read of the prior year is the raw material for next year's prospective list
- Where most of its value now sits
From evidence to encounter record
Example: this member carried chronic kidney disease stage 3 last year, the latest eGFR is on file, and the condition has not been evaluated this year; if present and addressed today, the note needs the stage. CMS reviewer guidance: undated problem lists are judged case by case, "DM, neuropathy" links while separate lines do not, and probable, suspected or rule-out outpatient diagnoses are not coded [5].
- 1EvidencePrior diagnosis, labs, medications, specialist notes, with dates
- 2PromptOne line, specificity named: stage, type, complication; a question, not a code
- 3VisitClinician evaluates; face-to-face or audio-video
- 4NoteCondition addressed and staged in the current-year note
- 5Encounter recordDiagnosis counts; coder confirms the code against the note
Short
- Three to five conditions, ranked by clinical relevance, not by coefficient
Specific
- Name the stage, type and complication, severity and episode, or treatment status the category needs
Evidenced
- Show the medication, the lab or the specialist note, with dates
Clinical
- Leave out RAF, dollars and coefficients; the prompt is for a clinician documenting a patient
Who does what in a prospective program
Reading a year of charts for every member before each visit is the volume problem. Our HCC risk adjustment AI reads de-identified charts, maps supported diagnoses through a V28 crosswalk of 7,903 ICD-10 codes to 115 payment HCCs audited at 1,500 of 1,500 codes against the CMS weights, and attaches the evidence to every suspect. Two fixed boundaries: charts are de-identified before any model reads them, zero identifier leaks in our verification run; and no code is submitted on the engine's word, a certified coder confirms every suggestion, per our risk adjustment coding services and HCC coding AI pages.
| Task | Engine | Coder | Clinician |
|---|---|---|---|
| Read prior-year charts, labs and medications for persistence evidence | Yes | Reviews | No |
| Rank and write the prompt | Drafts | Edits and approves | Reads |
| Evaluate and document the condition | No | No | Yes |
| Code the note | Suggests | Owns | No |
| Query a gap before the chart closes | Flags | Sends | Answers |
The suspect only counts if the visit does
Counts
- Face-to-face encounter; interactive audio and video telehealth qualifies [2]
Excluded from 2027
- Audio-only visits identified by modifier 93 or FQ, unless another line on the record is risk adjustment eligible [2]
RADV applies the same test
- Hospice, home health, lab-only and superbill records are invalid sources [4]
Takeaway
Prospective suspecting is the capture model the 2027 rules reward, because its output is a documented face-to-face encounter. Retrospective review becomes the research behind it, and the cleanup after it.
Five moves before December 31
Pull the 2025 chart review findingsSplit them into linked corrections, deletions, and prompts for visits still on the 2026 calendar.
Draft three-line prompts for November and December visitsShort, specific, evidenced, clinical.
Check every scheduled visit typeAudio-only with modifier 93 or FQ does not count on its own.
Route every suggestion through a certified coderThe engine drafts; the coder owns the code.
Read the operating planOur unlinked chart review playbook covers the rest of 2026; RAF score explained for 2027 shows the arithmetic each documented condition drives.
Frequently asked questions
What is HCC suspecting?
The process of identifying conditions a member likely has that have not been documented and coded in the current year, using prior diagnoses, medications, labs and specialist or hospital records, and then getting them evaluated and documented at an eligible encounter.
Is retrospective chart review still worth doing in 2027?
Yes, with a different output. Findings must be linked to a submitted encounter as corrections or deletions, or turned into prospective prompts for the next visit. A standalone unlinked chart review submission no longer counts toward the 2027 score, except for plan switchers.
Does an AI suspect list count as a diagnosis?
No. A suspect is a question for the clinician. The diagnosis exists only when the clinician documents the condition at a face-to-face or audio-video encounter and a coder confirms the code against that note.
Can audio-only visits support a suspect in 2027?
Not on their own. CMS excludes diagnoses from encounters where modifier 93 or FQ is the only service, unless another line on the record is risk adjustment eligible. Interactive audio and video telehealth meets the face-to-face requirement.
- CMS fact sheet: 2027 Medicare Advantage and Part D Rate Announcement (April 6, 2026)
- CMS: Announcement of CY 2027 Medicare Advantage Capitation Rates and Part C and Part D Payment Policies (April 6, 2026)
- CMS: Announcement of CY 2024 Medicare Advantage Capitation Rates and Part C and Part D Payment Policies (March 31, 2023), Attachment VIII, Tables VIII-1 and VIII-4
- CMS: Payment Year 2024 MA Contract-Specific RADV Audit Methods and Instructions (August 28, 2026)
- CMS: Contract-Level RADV Medical Record Reviewer Guidance, Version 2.0 (January 10, 2020)
Checked October 25, 2026. Rules change; confirm against the source before relying on them. CMS has said the RADV audit schedule is subject to change.
Want prospective prompts for your 2026 visits that are still on the calendar?
We read the prior year, draft a three-line prompt per member, and a certified coder approves every line before a clinician sees it. The output is a documented visit, which is the only thing 2027 pays for.