Most RADV preparation improved chart retrieval. Extrapolated annual audits test the capture decision, which was made years before the sample arrives.
When CMS signaled that RADV was moving toward every contract every year, most organizations responded by getting faster at producing charts. Retrieval vendors, indexing projects, dedicated audit response teams, shortened turnaround times.
All of it useful, none of it addressing what an extrapolated audit actually measures. Retrieval determines how quickly you can hand over the chart. The audit outcome was determined years earlier, when a coder decided the condition was supported and something either did or did not attach the proof to that decision.
By the time the sample arrives, the error rate already exists. You are not preparing to pass an audit. You are preparing to discover a result that was set in a prior payment year.
Extrapolation Changes What Is Being Tested
The mechanic matters more than the frequency, and it is the part most audit readiness programs underweight.
Auditors sample members and apply findings across the contract. So the audit is not evaluating the captures in the sample. It is using them to estimate the reliability of the decision process that produced every capture you made. A sample finding is a measurement of your coding standard, not of those specific charts.
That reframes what readiness means. Improving how well you defend the sampled captures moves a small number. Improving the consistency of the decision that produced all of them moves the number that gets multiplied. An organization with a uniformly applied evidence standard and mediocre retrieval will outperform one with excellent retrieval and captures made under varying standards across vendors, years, and coder cohorts.
Which leads to the question worth asking internally, and it is not how fast can we produce charts. It is whether a capture made by a chart review vendor in one year would clear the same evidence bar as one made by your internal team last quarter. Where that answer is no, the exposure is already booked.
Where the Error Rate Is Actually Built
Three sources, none of which a retrieval program touches.
Variation in evidence standards across sources of capture. Internal coders, chart review vendors, prospective programs, and delegated groups rarely apply an identical threshold for what constitutes support. Each source contributes captures to the same population and the audit samples across all of them without distinguishing origin.
Captures with no durable link to their evidence. A condition confirmed against a specific page of a specific document in a system that did not record which page or which document is a condition you will re-adjudicate under time pressure years later, and sometimes fail to reproduce.
Conditions accepted into the score through paths nobody would defend deliberately. Codes that flowed through mapping logic, inherited from a prior year, or arrived on supplemental data without anyone examining the underlying support. These tend to be invisible in internal reporting precisely because no one decided to capture them.
Want to see what your capture lineage looks like today?
The Audit-Readiness Test Worth Running
Pull fifty captures at random from a closed payment year, weighted toward the highest-value HCCs, and try to produce for each one the document, the page, the clinician, the date, the coder who confirmed it, and the accepted encounter it rode out on.
Time it. The elapsed time is the finding, not the pass rate. If assembling that package takes hours of manual work across systems, an extrapolated audit will find whatever is weakest in your capture history and apply it to your contract, and you will be discovering your own record while defending it.
The second thing to look at is the variance. Captures that assemble instantly and captures that require investigation usually come from different sources, and that split is a map of where your evidence standard is inconsistent.
The Same Test Applies to Delegated Groups
Where documentation originates outside the plan, audit exposure is created outside the plan and lands on the contract holder.
A delegated group, IPA, or medical group generates the note and often the code. The plan submits, receives the response files, and carries the audit. If the evidence standard applied at the group differs from the one applied internally, the plan has inherited a capture population it never inspected and cannot easily characterize.
For groups, the exposure runs the other way as well. Documentation practices that produce captures a plan cannot defend become a contracting conversation after an audit result, which is the worst possible moment to discover that two organizations were working to different thresholds.
How Invent Health Handles Lineage
Invent Health runs risk adjustment as one connected loop rather than three tools that integrate, and it starts with analytics, not coding.
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, with conditions ranked by score impact and closure likelihood, and internal capture validated against CMS-recognized HCCs in the MOR.
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 never assigns a code, and every code stays linked to the chart for audit-defensible lineage. The Coder Workbench averages 95 percent coding accuracy, with NLP evidence detection at 85 percent or higher out of the box.
Encounter Submissions makes sure it counts, validating 837P, 837I, and DME for Medicare EDPS and Edge Server XML for ACA, flagging missing and non-risk-eligible diagnoses before a file leaves, then reconciling MAO-002 and MAO-004 for Medicare and the Edge Server reports for ACA.
The point of running all three on one platform is that the evidence trail is a byproduct of the workflow rather than a project run after the sample arrives. A single applied standard, a named coder, a cited document, and a traceable encounter for each capture, rolled up to the delegated group level where the documentation originated.
Why the Window Is Closing
Two changes are converging on the same exposure.
The CY 2027 Rate Announcement ends unlinked chart reviews beginning with the 2027 payment year, so a chart review diagnosis has to tie back to an encounter CMS accepted, and the July HPMS memo on CY 2027 risk adjustment implementation sets out what has to be in place. Chart review has been the mechanism organizations lean on for both capture and correction, and it now requires the encounter link it previously did not.
V28 removed a large share of diagnosis codes from mapping and re-based values, so fewer conditions carry payment and each surviving capture bears more weight in both directions: more revenue if it holds, more repayment if it does not. Combined with extrapolation across annual audits, the cost of an inconsistent evidence standard compounds rather than accumulates.
Retrieval speed will still matter on the day the sample lands. It just will not change the result.
Run the Fifty-Capture Test
Take fifty captures from a closed year and assemble the full evidence package for each. What that takes, and how much it varies by source, is a more honest measure of audit readiness than any retrieval metric.
Frequently asked questions
How should health plans prepare for annual RADV audits?
By making the evidence standard consistent across every source of capture and ensuring each capture retains a durable link to its supporting document, coder, and accepted encounter. Retrieval speed determines how fast charts are produced, not whether the captures inside them hold.
What does RADV extrapolation actually test?
The reliability of the decision process that produced your captures. Because findings from a sample are applied across the contract, the audit uses sampled captures to estimate the quality of every capture made, which makes consistency more consequential than performance on any individual chart.
Why is chart retrieval speed not the same as audit readiness?
Retrieval determines how quickly documentation is produced after a sample arrives. The audit result was set years earlier, when captures were made under whatever evidence standards were in force. Faster retrieval surfaces the existing result sooner without changing it.
Where does RADV error rate come from?
Mainly from three sources: differing evidence standards across internal coders, chart review vendors, and delegated groups; captures with no durable link to the document that supported them; and conditions that entered the score through mapping, inheritance, or supplemental data without deliberate review.
How can an organization test its own audit readiness?
Pull roughly fifty captures from a closed payment year, weighted toward high-value HCCs, and assemble the document, page, clinician, date, confirming coder, and accepted encounter for each. The time required, and how much it varies by capture source, is the finding.
Does coder-in-the-loop coding reduce RADV exposure?
It reduces the exposure created by captures that cannot be traced to a human decision and a documented source. Because extrapolation multiplies indefensible captures across a contract, a capture carrying a named coder and a cited chart line is materially cheaper to defend.
How do delegated groups affect a plan’s RADV exposure?
Documentation and often coding originate at the group while the plan holds the contract and carries the audit. If the group applies a different evidence threshold than the plan does internally, the plan has inherited a capture population it never inspected.
What should a plan ask a chart review vendor about audit exposure?
What evidence standard was applied, whether each capture retains a link to the specific document and page supporting it, and whether that standard matches the one used internally. Differences between sources are where extrapolated error rates originate.
Does ending unlinked chart reviews change audit risk?
It changes where captures can come from. Starting with the 2027 payment year, chart review diagnoses must tie to an accepted encounter, which removes the mechanism organizations have used to both capture and correct late, and pushes both upstream to the visit.
Why does V28 raise the cost of an audit finding?
Because it removed a large share of diagnosis codes from mapping and re-based values, so fewer conditions carry payment and each surviving capture represents more revenue. A removed capture under extrapolation therefore costs more than the same finding would have under the prior model.
What evidence should support a risk-adjusted diagnosis?
Documentation showing the condition was monitored, evaluated, assessed, or treated during a face-to-face encounter with an acceptable provider type, recorded in a note that can be located, dated, and attributed, and linked to an encounter CMS accepted.
How does Invent Health support audit defensibility?
Every code in the Coder Workbench is confirmed by a certified coder against cited chart evidence and stays linked to that document. Because analytics, coding, and submission run on one platform, the trail from chart to CMS outcome is a byproduct of the workflow rather than a project assembled after a sample arrives.
