Somewhere in your organization, three people are reporting risk adjustment performance this quarter, and none of their numbers agree.
Coding reports capture against suspects worked. Encounters reports acceptance against files submitted. Finance reports RAF against the MOR and the payment cycles. Every number is correctly calculated. Every number is defended by someone who can show their work. And the three of them describe a different year.
The usual response is to reconcile the reports. That is the wrong move, because the disagreement is not a reporting artifact. It is the most honest performance data your organization produces, and the size of the disagreement is roughly the size of what you are losing.
Every Vendor Scorecard Ends at the Handoff
Risk adjustment performance is measured in segments because it is bought in segments.
Your coding vendor reports coding accuracy and conditions captured. That number stops at the signed chart, because that is where their scope stops. Your submissions vendor reports first-pass acceptance and rejection rates. That number starts at the file, and it has no opinion about whether the diagnoses inside it were the right ones. Your analytics vendor reports RAF and gap closure, calculated from data that was already several handoffs old when it arrived.
Each of them is measuring their own segment honestly. What none of them measures is the seam, and the seam is where risk adjustment revenue actually goes.
A coder confirms a condition against solid evidence. It never lands on a risk-eligible encounter, or it lands on one CMS rejects, or a later void removes it. The coding scorecard still counts it, correctly, because the coder did their job. The submissions scorecard never sees it, because a diagnosis that failed on one encounter is not a rejection story, it is an absence. And finance sees a RAF that came in under projection with no line item explaining why.
Nobody is wrong. The condition is just gone, and no single scorecard is built to notice.
The One Number Almost Nobody Produces
There is a single measure of risk adjustment performance that survives all of this, and most organizations cannot produce it: of the conditions your organization identified, documented, and coded this year, what percentage did CMS ultimately recognize?
Accepted capture. Not capture rate, which counts work performed. Not acceptance rate, which counts files processed. The percentage of the clinical work you paid for that ended up in the risk score.
The reason it is rare is structural rather than analytical. Producing it requires joining the coded chart to the encounter it rode out on, that encounter to the MAO-002 and MAO-004 responses, and those responses back to the MOR at the member level. Organizations that buy coding from one vendor, submissions from another, and analytics from a third are asking three companies to jointly produce a number that would expose which of them is losing conditions. That number does not get built.
When it does get built, the finding is usually not a rejection problem. Rejections are visible and teams work them. The bigger share sits in conditions that were captured, submitted, technically accepted, and then filtered out as not risk-eligible, or removed by a void nobody traced back to the chart it came from. Those never appear on any rejection report, which is exactly why they persist year after year, resurfacing each cycle as fresh suspects and getting worked again.
Want your accepted capture rate for a closed year?
Delegated Arrangements Split the Number in Half
This gets worse the further the clinical work sits from the CMS contract, which is where a growing share of risk adjustment now happens. Health systems own plans. IPAs and medical groups carry risk under delegated and value-based arrangements. In those structures the two halves of the number live in different organizations.
The group holds the documentation, often does its own coding, and is measured on capture. The plan holds the encounter submission and receives the response files. The group can tell you what it captured. Only the plan can tell it what counted, and that feedback usually arrives, if it arrives, as an aggregate months later with no path back to a specific chart.
So a delegated group can run an excellent documentation program, hit every internal target, and never learn that a meaningful share of its conditions did not reach the risk score. The plan sees a RAF gap and reads it as provider documentation behavior. The group sees its own capture numbers and reads the plan as unresponsive. Both are looking at real data. Neither is looking at the same year.
For a health system that owns its plan, this is the most solvable version of the problem and the most commonly unsolved one, because the documentation and the response files are both in the building, just not in the same system.
What It Takes to Produce the Number
Invent Health was built around this specific problem, which is why analytics, coding, and encounter submission run on one platform rather than three that integrate.
Risk Analytics decides what is worth working. Suspect logic runs across historical, lab, pharmacy, comorbidity, and chart-derived signals, drawing on EMR data through CCD and FHIR. Conditions are ranked by how much they move the score and how likely they are to close, and internal capture is validated against CMS-recognized HCCs in the MOR with RAF progression tracked across the Initial, Mid-Year, and Final cycles in the MMR.
The Coder Workbench confirms the condition is real. AI-assisted coding reads the clinical documentation, shows the supporting evidence, and recommends ICD-10-CM codes for a certified coder to confirm and sign. It is a two-pass, coder-in-the-loop model. The AI surfaces what matters, a person makes the call, and every code stays linked to the chart for audit-defensible lineage.
Encounter Submissions makes sure it counts. One engine generates and validates clean submissions, 837P, 837I, and DME for Medicare EDPS and Edge Server XML for ACA, flags missing diagnoses and non-risk-eligible claims carrying risk diagnoses before a file leaves, then reconciles MAO-002 and MAO-004 for Medicare and the Edge Server reports for ACA.
Because those three share a platform, the lineage holds end to end. Chart, coder validation, encounter, CMS outcome, payment cycle. A condition that did not count returns to the worklist as a known condition needing a corrected encounter, ranked by RAF value, instead of reappearing next year as a fresh suspect. And for organizations working with delegated groups and IPAs, that lineage rolls up to the group level, so the feedback lands where the documentation happens.
The Seam Gets Expensive in 2027
For years the retrospective sweep absorbed most of this. Whatever leaked during the year got caught in a chart review and submitted as a correction, and the seam stayed survivable because there was a mechanism at the end that did not care how the condition got lost.
The CY 2027 Rate Announcement ends unlinked chart reviews beginning with the 2027 payment year. A chart review diagnosis now has to tie back to an encounter CMS accepted, which means the sweep can no longer rescue a condition whose encounter failed. The July HPMS memo on CY 2027 risk adjustment implementation sets out what has to be in place. V28 compounds it by removing a large share of diagnosis codes from mapping, so fewer conditions carry payment and each loss weighs more. And with RADV moving toward every contract every year and findings extrapolated across the contract, the same broken lineage that hides lost revenue is what makes an audit expensive, because a condition you cannot trace is a condition you cannot defend.
The organizations that come through this well will not be the ones capturing the most. They will be the ones who can name their accepted capture rate without a project to calculate it.
Find Out What Your Number Is
Take a closed payment year and answer one question: of everything your team identified, documented, and coded, how much did CMS recognize? If producing that answer requires pulling three systems together by hand, that is the finding.
Frequently asked questions
How should risk adjustment performance be measured?
By accepted capture, meaning the share of identified and coded conditions CMS ultimately recognized. Capture rate measures work performed and acceptance rate measures files processed. Only accepted capture connects the clinical work an organization paid for to the risk score it was paid on.
Why do coding, encounter, and finance risk adjustment numbers disagree?
Because each measures a different segment and stops at its handoff. Coding scorecards end at the signed chart, submission scorecards begin at the file, and finance sees RAF after everything settles. Conditions lost between segments are counted by the first team, never seen by the second, and appear to finance only as an unexplained shortfall.
What is accepted capture rate?
Accepted capture rate is the percentage of conditions an organization identified, documented, and coded that CMS recognized for risk adjustment. Producing it requires joining coded charts to encounters, encounters to MAO-002 and MAO-004 responses, and those responses to the MOR at the member level.
Why is accepted capture rate hard to produce?
It is a structural problem more than an analytical one. When coding, submissions, and analytics come from different vendors, no single system holds the lineage from chart to CMS outcome, and the number would reveal which vendor is losing conditions. So it usually does not get built.
Where do most risk adjustment conditions get lost?
Less often in outright rejections, which are visible and worked, than in diagnoses that were submitted, technically accepted, and then filtered as not risk-eligible or removed by a later void. Those never appear on a rejection report, so they persist across years and resurface as repeat suspects.
Can a rejected or voided encounter cause a condition to be lost?
Yes. A diagnosis on a rejected encounter, or one removed by a later void, never reaches the risk score even when the chart fully supports it. Internal reporting typically marks the condition as captured, so the loss stays invisible until someone reconciles the CMS response files back to the chart.
How do delegated arrangements affect risk adjustment performance measurement?
They split the measurement across organizations. The provider group holds documentation and coding and is measured on capture. The plan holds submission and receives the response files. Without lineage that crosses the boundary, the group cannot see what counted and the plan misreads the shortfall as documentation behavior.
Why do health systems that own plans struggle with this?
Because both halves of the number exist inside the organization but in different systems. Clinical documentation sits in the delivery side and CMS response files sit in the plan side, and connecting them at the member and chart level usually requires work no one owns.
What is the difference between capture rate and acceptance rate?
Capture rate counts conditions a coding operation captured. Acceptance rate counts files or encounters CMS processed without rejecting. Neither answers whether a specific coded condition reached the risk score, which is why an organization can post strong numbers on both and still lose revenue.
How do the MOR and MAO-004 fit into performance measurement?
MAO-004 shows which submitted diagnoses were treated as eligible for risk adjustment, and the MOR shows which HCCs CMS recognized per member. Read against internal capture, they convert a submission log into an accounted-for number and identify the specific conditions that went missing.
Does ending unlinked chart reviews change how performance should be measured?
Yes. Starting with the 2027 payment year, chart review diagnoses must tie to an accepted encounter, so the retrospective sweep can no longer rescue conditions whose encounters failed. That makes accepted capture a live operating metric rather than a year-end reconciliation exercise.
How does Invent Health measure risk adjustment performance end to end?
Analytics, coding, and encounter submission run on one platform, so each condition stays linked from chart through coder validation, encounter, and CMS response. Conditions that did not count return as corrections ranked by RAF value, and reporting rolls up to the delegated group level where documentation happens.
