From 2027, diagnoses on unlinked chart reviews don't pay. Compare prospective, concurrent and retrospective review, and choose the right mix for your program.
The difference between prospective vs retrospective risk adjustment review is timing. Prospective review happens before the patient visit. Concurrent review happens during or right after the visit, before the claim goes out. Retrospective review happens after the claim, often months later.
For years, that difference was mostly operational. From 2027, it's financial. CMS no longer counts diagnoses submitted on unlinked chart reviews when it calculates Medicare Advantage risk scores. That change hits the most common form of retrospective review directly, and it changes which mix of the three approaches makes sense.
The three approaches, defined
Prospective review looks at a patient's history before a scheduled visit and flags conditions that may need to be evaluated: chronic conditions documented last year but not yet this year, or clinical indicators such as a medication or lab result with no matching diagnosis. The clinician sees the flags, evaluates the patient, and documents what they find. Nothing is coded from the flag itself.
Concurrent review codes and validates the encounter while it can still be corrected, before the claim is submitted. A coder or clinical documentation integrity (CDI) specialist reviews the note, checks that each diagnosis is supported, and sends a query to the clinician when documentation is unclear or incomplete.
Retrospective review goes back through medical records after claims have been submitted, looking for diagnoses that were documented but not coded, and for diagnoses that were coded but aren't supported. Findings are submitted to CMS as chart review records: adds for new diagnoses, deletes for unsupported ones.
Side-by-side comparison
| Prospective | Concurrent | Retrospective | |
|---|---|---|---|
| When | Before the visit | During or right after the visit, before claim submission | After the claim, often months later |
| Who | Analysts or coders prepare; clinician evaluates | Coders or CDI specialists, with clinician queries | Coders, often external vendors |
| Input | Prior-year diagnoses, claims, labs, medications | The current note | Submitted claims and the full medical record |
| Output | Flags for the clinician to evaluate | Corrected codes and documentation on the claim | Chart review records adding or deleting diagnoses |
| Submission path | The encounter record itself | The encounter record itself | Linked or unlinked chart review records |
| 2027 payment status | Unaffected | Unaffected | Unlinked adds excluded for non-PACE plans; linked adds and deletes still count |
| Main compliance risk | Leading prompts; diagnoses the clinician didn't independently evaluate | Leading queries; coder bandwidth | Add-only programs; late addenda; unlinked submissions |
| Best at | Annual recapture of real chronic conditions | Accuracy at the source | Finding unsupported codes to delete; feedback |
What changed for 2027
A chart review record (CRR) is how an MA organization adds or deletes diagnoses after an encounter has been submitted. A linked CRR is tied to a previously accepted encounter data record. An unlinked CRR isn't tied to any specific encounter.
In the CY 2027 Rate Announcement, CMS finalized excluding diagnoses from unlinked CRRs from risk score calculations. Plans can still submit them; they just don't affect payment. Two exceptions apply: members who switched between MA organizations in the prior year, and PACE organizations, which are exempt for 2027.
Linked add CRRs still count. Delete CRRs still work, in both directions.
Why CMS did it
The change answers a 2019 finding. The HHS Office of Inspector General estimated that diagnoses reported only on chart reviews, with no matching service record, drove $6.7 billion in 2017 MA payments. More than 99 percent of chart reviews added diagnoses rather than deleting them.
OIG put $2.7 billion of that on unlinked chart reviews. CMS disputed the figure, estimating the actual value closer to $675 million because OIG hadn't applied CMS's model blending. Either way, the direction was the problem: a review process that almost never deleted anything. OIG's recommendation to reassess unlinked chart reviews was closed as implemented in April 2026, ahead of the 2027 exclusion.
Retrospective review's new job
Retrospective review isn't dead. Its job changed.
Deletes come first. A retrospective program that finds and deletes unsupported diagnoses reduces audit exposure directly. Every unsupported HCC removed before a Risk Adjustment Data Validation (RADV) audit is one that can't drive an extrapolated payment error.
Adds must link. A diagnosis found on review still counts if it's documented in the record of an accepted encounter and submitted on a linked CRR. One documented only in a note with no accepted encounter record no longer pays.
Findings feed the front end. Every recurring miss on retrospective review is a documentation or coding pattern that prospective or concurrent review should catch next year. A retrospective program that doesn't send findings back to coders and clinicians pays twice for the same gap.
Late addenda stay off the table. Asking clinicians to amend notes months after a visit, so a diagnosis can be submitted, is the conduct at the center of the largest recent risk adjustment enforcement actions. Retrospective findings that need more documentation should become prospective flags for the next visit, not addenda.
Prospective review done right
Prospective review is the strongest way to capture real chronic conditions every year, because the documentation is created at a face-to-face encounter by the clinician who evaluated the patient. It's also where the line between helping and leading is thinnest.
| Compliant prospective flag | Leading prompt |
|---|---|
| "COPD documented in 2025; not yet addressed in 2026." | "Code J44.9 COPD." |
| "On furosemide; no heart failure diagnosis on file. Please evaluate." | "Add chronic systolic heart failure." |
| Clinician can dismiss the flag with no friction | Dismissing requires justification; accepting is one click |
| Acceptance and dismissal rates tracked | Only acceptances counted |
The test is simple. The clinician has to evaluate the patient and decide, and the note has to show that evaluation. A flag accepted without an assessment in the note fails the same documentation test as any other unsupported diagnosis; what that note needs to show is in MEAT documentation for chronic conditions.
In-home health risk assessments are a form of prospective review that has drawn particular scrutiny. OIG found that diagnoses reported only on HRAs and HRA-linked chart reviews generated an estimated $7.5 billion in 2023 MA payments, with 20 companies driving 80 percent of it. A prospective program built around in-home assessments with no follow-up care is the profile auditors look for.
Concurrent review: accuracy at the source
Concurrent review produces the cleanest data, because problems are fixed before the claim is submitted and the clinician can still clarify while the encounter is fresh. Coding is accurate on the encounter record itself, with no chart review needed.
Its limits are practical. It takes coder and CDI capacity at the speed of the clinic, and queries have to be non-leading: they present the clinical indicators and let the clinician decide, rather than suggesting a diagnosis. Programs usually run concurrent review on higher-risk encounters first, such as annual wellness visits, chronic care visits and new patients with long problem lists.
One member, three approaches
Illustrative, not a real patient. A 74-year-old MA member has COPD documented in 2025. In 2026 she sees her primary care physician twice and a pulmonologist once.
Prospective. Before her March primary care visit, a flag notes that COPD hasn't been addressed yet this year. The physician asks about symptoms, reviews her inhaler use, and documents "COPD, stable on tiotropium, no exacerbations since last visit." The diagnosis goes out on the encounter record, with the evaluation in the note.
Concurrent. At the same visit, the note also says "CHF" in the history with no assessment. A concurrent reviewer catches it before the claim is submitted and sends a non-leading query showing her furosemide prescription and the 2025 echo. The physician documents chronic diastolic heart failure with a current assessment, or replies that it has resolved. Either answer is accurate, and the claim goes out right the first time.
Retrospective. In early 2027, a reviewer reads all three 2026 encounters. The pulmonologist's note documents chronic respiratory failure with home oxygen, which wasn't coded. The pulmonologist's encounter was accepted, so the diagnosis can go out on a linked chart review and still counts. The same review finds "history of stroke" coded as acute stroke at the second primary care visit. That diagnosis is deleted.
Each approach did something the others couldn't. Prospective review produced the documentation. Concurrent review fixed it before submission. Retrospective review caught a specialist's diagnosis the plan would otherwise have missed, and, more importantly, removed an error that would have failed audit.
Choosing the mix
Most mature programs run all three, weighted to their role.
| Organization | Lead with | Support with | Use retrospective for |
|---|---|---|---|
| MA plan | Prospective flags to network providers | Concurrent review for owned or employed practices | Deletes, provider feedback, mock RADV |
| Provider group in MA risk contracts | Concurrent review | Prospective flags at scheduling | Internal audit and coder education |
| Small practice in a value-based contract | Prospective flags for annual visits | Point-of-care coding support | Periodic sample audits |
What to measure
Each approach needs metrics that reward accuracy, not volume:
- Prospective: flag acceptance and dismissal rates, and the share of accepted flags with supporting documentation on audit
- Concurrent: query response rate, query agreement rate, and pre-bill validation rate
- Retrospective: delete rate alongside add rate, validation rate on internal RADV-style samples, and time from finding to front-end fix
A program that reports only HCCs added or RAF lift is measuring the thing auditors look for. The delete rate is the single best evidence of good faith.
FAQ
What is the difference between prospective and retrospective risk adjustment?
Prospective risk adjustment review happens before a patient visit and flags conditions for the clinician to evaluate. Retrospective review happens after claims are submitted and searches records for missed or unsupported diagnoses, which are submitted to CMS as chart review records.
Do retrospective chart reviews still count in 2027?
Partly. Diagnoses on chart reviews linked to a specific accepted encounter still count. Diagnoses on unlinked chart reviews are excluded from 2027 risk scores for non-PACE plans, except for members who switched MA organizations. Delete chart reviews still work.
What is concurrent risk adjustment review?
Concurrent review checks and codes an encounter during or right after the visit, before the claim is submitted. Coders or CDI specialists validate each diagnosis and query the clinician when documentation is unclear, so the encounter record is accurate from the start.
Is prospective risk adjustment compliant?
Yes, when flags inform rather than direct. A compliant flag shows the clinician what to evaluate; the clinician decides what's present and documents the evaluation. Prompts that suggest a specific code, or make dismissal harder than acceptance, create compliance risk.
Which approach is best for annual recapture?
Prospective review, because it puts the right conditions in front of the clinician at a face-to-face visit, where the documentation is created. Concurrent review then checks that the documentation supports what was coded.
This article is general education on risk adjustment programs. It isn't legal or compliance advice for any specific organization.
