
Risk Adjustment Analytics for Audit-Ready Risk Scores
Risk adjustment analytics turns encounter data into a clear view of risk scores, documentation gaps, and audit exposure. We cover the metrics that matter, the latest CMS and OIG changes, and five steps to build a program whose numbers you can defend.
Every diagnosis your organization submits for risk adjustment makes two claims. One is that the patient has the condition, and the other is that the medical record supports it.
Risk adjustment analytics checks both claims across every member, provider, and diagnosis source before an auditor does.
In this article, we cover what it measures, which metrics matter, how recent CMS and OIG actions change the work, and how to build a program whose numbers hold up under audit.
TL;DR: Risk Adjustment Analytics at a Glance
The essentials, in five points:
What it is: Analytics that shows risk scores, documentation gaps, and audit exposure across a population.
What to measure: Risk score trends, condition prevalence, recapture, provider outliers, diagnosis sources, and validation rates.
What changed: CMS no longer counts unlinked chart review or audio-only diagnoses toward Medicare Advantage risk scores, and RADV now covers every eligible contract.
Where accuracy starts: Every metric depends on the clinical note behind each diagnosis.
How to build it: Pair capture with validation, keep credentialed people on judgment calls, and keep an audit trail.
What Is Risk Adjustment Analytics?
Risk adjustment analytics measures how a population's documented diagnoses translate into risk scores, and how well the medical record supports each one. Health plans use it to forecast revenue and manage audit exposure, and provider organizations in risk-sharing and value-based contracts use it to see where care and documentation diverge.
The mechanics of turning one diagnosis into a score belong to HCC medical coding. Analytics works one level up, reading the patterns across thousands of those coding decisions.
Most programs draw on the same core data:
Claims and encounter data: The diagnoses submitted for each visit, which feed the risk score.
Clinical documentation: The notes in the electronic health record (EHR) that must support each submitted diagnosis.
Chart reviews and health risk assessments: Retrospective reviews and assessments that add or remove diagnoses, and that federal oversight now examines closely.
Model and payment data: The current CMS Hierarchical Condition Category (CMS-HCC) mappings and payment factors that convert diagnoses into dollars.
Audit history: Prior Risk Adjustment Data Validation (RADV) findings, internal audit results, and the codes your team has already deleted.

What to Measure in Risk Adjustment Analytics
A useful program tracks capture and accuracy together. A dashboard that reports risk score growth without validation results gives finance a number compliance cannot defend.
Six metrics carry most of the signal:
RAF Score Trends
Track average Risk Adjustment Factor (RAF) scores by contract, provider group, and month, then compare the trend with what CMS expects. CMS's latest Rate Announcement projects Medicare Advantage (MA) risk scores to rise about 2.5% on average from the underlying coding trend.
Growth well above that line deserves a closer look before anyone books the revenue.
Condition Prevalence Benchmarks
Compare how often each condition category appears in your population against prior years and peer groups. The HHS Office of Inspector General (OIG) recommends this benchmarking in its compliance program guidance for Medicare Advantage, followed by documentation reviews focused on the outliers.
Recapture and Open Documentation Gaps
CMS bases each year's payment partly on the prior year's health status, so chronic conditions need fresh documentation at a qualifying visit every year. Recapture rates show which conditions documented last year your clinicians still haven't addressed this year.
Treat an open gap as a reason to see the patient. A diagnosis belongs on the claim once a clinician evaluates and documents it.
Provider-Level Outliers
Compare diagnosis rates across providers with similar patient panels. The same OIG guidance recommends analyzing provider coding intensity to find providers who code at higher rates than peers, then following up with audits, education, and corrective action.
Providers far below their peers deserve attention too, since a low rate can point to thin documentation of conditions they treat.
Diagnosis Source Mix
Break down risk score contribution by source, including face-to-face encounters, linked chart reviews, unlinked chart reviews, health risk assessments (HRAs), and telehealth. OIG's review of health risk assessments found that diagnoses reported solely on HRAs and HRA-linked chart reviews generated an estimated $7.5 billion in MA risk-adjusted payments for 1.7 million enrollees.
The OIG said this pattern raises concerns about inaccurate diagnoses or enrollees going without necessary care.
Deletion and Validation Rates
Measure how many reviewed diagnoses your team removes or corrects, and how many pass internal audit. OIG's guidance lists failing to remove unsupported codes found during chart review among the practices federal investigations have revealed.
A program with no deletions to report has a gap worth explaining, since defensible risk adjustment coding removes unsupported diagnoses as routinely as it adds supported ones.
How Federal Oversight Moves Risk Adjustment Upstream
Two federal actions change what analytics can count on. Both reward diagnoses tied to a real encounter and a supporting note.
The same CMS announcement finalized the exclusion of diagnoses from unlinked chart review records, meaning reviews not tied to a specific encounter, from MA risk score calculation. A narrow exception covers members who switch between MA organizations.
CMS also excluded diagnoses from audio-only encounters and estimates the change in diagnosis sources lowers risk scores by about 1.5% on average. Plans may still submit unlinked chart review diagnoses, though they no longer count toward payment.
Separately, CMS has expanded RADV and plans to audit every eligible Medicare Advantage contract each payment year, up from about 60 plans a year. For analytics teams, a diagnosis now needs an encounter and a supporting note to count, and the odds that someone checks it have risen.

Why Analytics Accuracy Starts in the Clinical Note
Every metric above reads diagnoses that began as words in a clinical note. When the note is vague, no chart review can legitimately fill the gap, because the support was never written down.
A specific, complete note gives risk adjustment analytics three things:
A diagnosis that counts: Each condition traces to an encounter and the clinician's own assessment.
Detail for coding: Coders get the specificity they need to assign the most accurate supported code.
A ready audit response: The support an auditor asks for already sits in the record.
Our VP of Business Operations, Gautam Kumar, traces the problem to the note itself. "If the note is incomplete, or for whatever reason it is inconsistent, [...] every downstream process becomes reactive," he says.
Risk adjustment analytics inherits the revenue cycle's upstream problem.
InstaNote, our AI clinical documentation, is built around that first step:
Capture: Ambient conversation or dictation, in inpatient and outpatient care.
Fidelity: Speech recognition designed not to add anything that wasn't said.
Coding support: Coding suggestions drawn from the note help your providers hand off a more complete note, and your coders still approve the final codes.
Our Director of Product Implementation, Garrett Dunne, is direct about what the model will never do.
"There's not a situation where the engine is being essentially, the model is being asked to infer a diagnosis. We wouldn't allow it to do that."
Reliable analytics depends on notes that capture each condition clearly the first time, and that is where we focus.
At DeliverHealth, we provide AI-powered clinical documentation and coding solutions for small and mid-size practices and large health systems. Our InstaNote turns each encounter into a structured note with coding suggestions, and our InstaCode turns documentation into accurate, defensible codes with credentialed coders in the loop.
More than 40,000 healthcare providers trust InstaNote, and it runs on our eSOne platform, which is HITRUST i1 certified. Note formats and templates follow your organization's and each provider's requests.
Book an InstaNote demo to watch a sample encounter become a specific, structured note ready for your coding team.
How to Build a Defensible Risk Adjustment Analytics Program
Analytics earns trust when every number traces back to a note. The steps below apply whether you run analytics in-house, with a health plan partner, or through a vendor.
Five steps build that traceability:
1. Unify Claims, Encounter, and Clinical Data
Bring claims, encounter submissions, and EHR documentation into one view keyed to the member and the encounter. Linking each diagnosis to its encounter matters more now that unlinked chart review diagnoses no longer count toward MA risk scores in most cases.
2. Route Suspects to the Next Visit
Suspecting models flag members whose history, medications, or lab results suggest an undocumented condition. Send each suspect to the treating clinician as context for the next visit, where an evaluation can confirm or rule out the condition.
3. Pair Capture With Validation
For every workflow that adds diagnoses, run one that removes unsupported diagnoses. OIG's guidance recommends pairing completeness efforts with accuracy processes, along with audits of diagnosis data before and after submission.
Report both sets of results through your medical coding compliance program so leadership sees capture and accuracy on the same page.
4. Benchmark Providers and Close the Loop
Share provider-level benchmarks with clinical leaders, along with specific chart examples. Education built from a provider's own notes gives clinicians something concrete to act on.
5. Keep an Audit Trail From Note to Submission
Record who coded each diagnosis, what changed during review, and why. When a RADV request arrives, that trail turns record retrieval into a lookup.
Our AI-powered medical coding, InstaCode, builds that record into the coding workflow. Every coded encounter is auditable, with the AI's output, each human change and the reasoning behind it, and any later auditor edits kept together.
How to Use AI in Risk Adjustment Analytics Responsibly
AI improves analytics by sharpening the signal for the people who make decisions. Models can scan claims, lab results, and notes across a whole population and rank where review time should go.
OIG's guidance also sets a boundary for that work. Among the practices federal investigations have revealed, the guidance lists EHR prompts, including AI-generated ones, that push physicians to add risk-adjusting diagnoses patients did not have or that did not affect their care.
Keep AI in a supporting role. Models rank and flag; clinicians evaluate the patient, and credentialed coders confirm what the record supports.
Before you rely on an analytics model, ask the vendor about four areas:
Inputs: Ask which data sources feed each flag and whether the model reads the full note.
Evidence: Ask whether every flag shows the documentation or data behind it.
Direction: Ask whether the model flags unsupported diagnoses for removal as readily as missed ones for review.
Human review: Ask who confirms each flag before it reaches a clinician or a submission.
Accuracy and capture figures deserve the same scrutiny, and knowing how to interpret AI medical coding accuracy rates helps your team compare vendors on equal terms.

Frequently Asked Questions (FAQs)
These are the questions revenue cycle and compliance leaders raise most often about risk adjustment analytics:
What Is the Difference Between Risk Adjustment Analytics and Risk Adjustment Coding?
Risk adjustment coding assigns ICD-10-CM diagnosis codes to each encounter from the documentation. Risk adjustment analytics looks across all those encounters to show trends, gaps, outliers, and audit exposure.
Who Uses Risk Adjustment Analytics?
Medicare Advantage plans, health systems, and physician groups in risk-sharing or value-based contracts, and the vendors that support them all use it. Each applies it differently, and all depend on the same encounter-level diagnoses.
What Is a Suspect in Risk Adjustment?
A suspect is a member whose data suggest a condition that has not been documented this year, such as a medication with no matching diagnosis. The suspect prompts a clinical evaluation at the next visit, and the diagnosis counts once a clinician documents it.
Do Unlinked Chart Reviews Still Count Toward Medicare Advantage Risk Scores?
In most cases, they no longer do. CMS has finalized the exclusion of diagnoses from chart review records that are not tied to a specific encounter, with an exception for members who switch between MA organizations.
Plans can still submit those diagnoses, and CMS leaves them out of risk score calculation.
How Often Should You Review Risk Adjustment Analytics?
Review operational trends monthly, and bring capture and validation results to your compliance committee at least quarterly. OIG's guidance encourages compliance committees to meet regularly and oversee internal and external audits.
Final Thoughts on Risk Adjustment Analytics
Strong risk adjustment programs look calm from the outside. Their scores move with the population, their deletions are documented, and their charts answer an auditor's request without a scramble.
Reaching that point takes the note and the code working together.
At DeliverHealth, we provide AI-powered clinical documentation and coding solutions that connect the encounter to the claim. InstaNote captures each visit as a structured note with coding suggestions, and InstaCode codes from InstaNote, traditional dictation, operative reports, PowerScribe, or existing EHR notes, with no change to how documentation happens.
InstaCode provides 95% coding accuracy, delivered with human-in-the-loop review; our coders, credentialed through AAPC or AHIMA, are DeliverHealth employees in the U.S. and India, and they review every complex or flagged encounter. Billable codes are typically delivered within about 24 hours.
Schedule an InstaCode demo to walk through sample notes from your specialty and trace each code back to the documentation behind it.
Stay Updated
Subscribe to our newsletter for the latest healthcare AI insights and company updates.
