Risk adjustment analytics measured on whether the documentation supports what was submitted, with unsupported diagnoses identified for deletion as a standard output rather than an optional module nobody buys.
A plan submits diagnoses it did not generate, using documentation it did not write, frequently supported by vendors paid in ways that reward finding conditions. The plan carries the audit exposure for all of it. Any programme measured on score movement is structurally pointed in one direction, and that direction is the one enforcement attention has been following.
If your risk adjustment programme has never produced a list of diagnoses to delete, it has not been looking for one
.
The diagnoses a plan submits originate in provider documentation the plan does not control, arrive through encounters and supplemental submissions of variable completeness, and are frequently supplemented by chart review and assessment programmes the plan commissioned. The plan is answerable for every one of them, and the practical question at audit is not who produced the code but whether the record supports it.
The commercial structure makes this harder. Programmes are measured on score improvement, vendors are frequently compensated in ways that reward volume of conditions identified, and nobody in that chain is incentivized to find a diagnosis that should come off.
Every incentive points one way
Internal targets, vendor compensation and programme reporting all reward conditions found. Nothing in the structure rewards a diagnosis identified as unsupported.
Vendors paid per condition are a design choice
Chart review, in-home assessment and coding vendors compensated on findings will produce findings. That arrangement is visible in an audit and difficult to defend afterwards.
Supplemental sources attract scrutiny
Conditions appearing only from a chart review or assessment, never addressed anywhere else in the member care, are exactly the pattern reviewers look for.
Submitted is not accepted, and accepted is not supported
Three different populations. Plans commonly manage the first, reconcile the second imperfectly and rarely test the third.
Extrapolation turns a small rate into a large number
Sample-based review can be extrapolated across a contract. That arithmetic converts a modest documentation weakness into material exposure.
Suspect models generate noise and call it opportunity
The question is not how many suspects were generated but how many clinically meaningful reviewable opportunities exist.
Look at how your vendors are paid before you look at your score.
Compensation tied to conditions identified creates an incentive the plan owns, whatever the contract says about compliance. Reviewing vendor economics costs one meeting and it is one of the highest-value checks a plan can perform.
Almost every engagement in this space starts with the gap. We start with the submission. Testing whether existing diagnoses are supported establishes the plan real position, frequently finds exposure nobody had quantified, and sets the credibility of everything that follows with compliance, with providers and with an examiner.
Step 1
Establish the population and the submission chain: which members, which contracts, and what was submitted, accepted and reflected in the final risk score.
Step 2
Validate what was submitted against a defined documentation standard.
Step 3
Assess the sources: encounters, chart review, assessments and vendors.
Step 4
Review vendor economics and quality: compensation, measures and independent validation.
Step 5
Identify recapture accurately, separating persistent supported conditions from conditions not addressed.
Step 6
Surface suspected conditions with evidence for clinical evaluation, never as a code recommendation and never on statistical likelihood alone.
Step 7
Reconcile the full chain to acceptance
Step 8
Prioritize by evidence strength rather than potential score impact.
Step 9
Address provider documentation with targeted data.
Step 10
Retain supporting documentation, model version and lineage per submitted position.
Both lists, every time.
Every engagement produces conditions that should have been submitted and conditions that should come off. A partner who only ever brings the first list is not measuring accuracy.
Condition detection is not the hard part and plenty of tools do it. What determines whether a programme is defensible is validating what was already submitted, assessing the sources it came from, and presenting suspected conditions in a form a clinician evaluates rather than confirms.
Validate
Submitted Diagnosis Validation
Testing whether documentation supports existing submitted conditions, with the unsupported list produced as a standard output.
Source Concentration Analysis
Identify which conditions derive from encounters, chart review, assessments or a specific vendor.
Vendor Output Validation
Independent testing of conditions produced by chart review, assessment and coding partners.
Documentation Sufficiency Standards
A written organizational standard agreed with clinical leadership and compliance.
Complete the Picture
Multi-Source Condition History
Conditions assembled from encounters, claims, pharmacy, laboratory, clinical data and available external sources.
Recapture Identification
Persistent supported conditions not submitted in the current period, classified by cause.
Evidence-Based Suspecting
Conditions suggested by objective clinical data, presented with the supporting data attached for clinician evaluation.
Submission Chain Reconciliation
Submitted, accepted and reflected compared as distinct states.
Defend
Deletion and Correction Workflow
Unsupported diagnoses identified, reviewed, corrected and recorded, with the decision and basis retained.
Provider Documentation Analytics
Identify providers, conditions and documentation gaps driving missed capture and unsupported submissions.
Audit Evidence Retention
Supporting documentation, model version, logic version and lineage retained per submitted position.
Evidence Chain Gap Analysis
Identify where the chain stopped rather than simply reporting the shortfall.
Accuracy Monitoring
Documentation sufficiency, deletion volume, suspect confirmation and rejection rates, and source concentration.
What CaliberFocus does, and does not do?
Analytics can identify evidence worth reviewing. It cannot manufacture documentation, and a signal is never converted into a diagnosis anywhere in our pipeline. Evidence and diagnosis remain separate objects. We are also not paid on score movement and will not accept an engagement structured that way. We do not pre-populate diagnoses, present codes without evidence, or conclude that a member has a condition. We will also review how your existing vendors are compensated, and where that structure is indefensible we will say so.
These programmes are all legitimate and they do not carry equal scrutiny. The third column reflects how a reviewer is likely to regard each, and it should influence sequencing and design more than the yield estimate does.
| Programme | What It Does | How It Is Regarded |
|---|---|---|
| Submission Validation | Tests whether existing submitted diagnoses are supported. | The safest and least common. Start here regardless of what the roadmap says. |
| Encounter Completeness | Diagnoses on paid claims that never reached an accepted encounter. | Low risk. Revenue already earned and not credited, and purely a data problem. |
| Recapture from Existing Documentation | Persistent conditions with support that were not submitted. | Low risk where the documentation genuinely exists in the record. |
| Provider Documentation Improvement | Education and feedback targeted by evidence. | Well regarded when data-driven and specific, poorly regarded when it reads as coding coaching. |
| Retrospective Chart Review | Reviewing records for conditions not submitted. | Legitimate and heavily scrutinized. Two-directional review is what makes it defensible. |
| In-Home Assessments | Assessments producing diagnoses outside routine care. | The most scrutinized programme in risk adjustment. Conditions never addressed elsewhere attract attention. |
| Vendor-Sourced Conditions | Conditions identified by a compensated partner. | Regarded according to how the partner is paid. Per-condition compensation is the issue. |
| Statistical or Model-Based Suspecting | Conditions inferred from population likelihood. | Never submitted on this basis. It may direct clinical review and it is not evidence. |
Two-Directional Chart Review Is What Makes Chart Review Defensible
Retrospective review that looks only for conditions to add is difficult to characterize as anything other than revenue work. The same programme, instructed and measured to identify both unsupported submitted conditions and unsubmitted supported ones, is a documentation integrity programme.
Programmes are sold on suspecting because it is the most interesting category. It is usually the smallest, the most scrutinized and the one requiring the most care. The larger and safer opportunity sits in conditions the plan already has evidence for, and in submissions it has already made.
| Category | What It Is | Risk Profile |
|---|---|---|
| Validation and Deletion | Submitted diagnoses the documentation does not support | Reduces exposure. The category with no downside and the one least often run. |
| Encounter Recovery | Supported diagnoses that never reached an accepted encounter | Purely a data and process problem. Revenue earned, not credited. |
| Recapture | Persistent conditions with support, not submitted this period | Low risk where documentation exists. Classify by cause before acting. |
| Suspecting | Conditions suggested by clinical evidence and never documented | Highest scrutiny. Requires evidence, clinician judgement and restraint. |
Strong evidence is surfaced with its source
Objective clinical results, a condition-specific medication or a specialist diagnosis, presented with values and dates visible for clinical confirmation.
Moderate evidence is labelled as inference
Indirect indicators, historical claims from another setting or a related procedure, presented as a question with the reasoning shown.
Weak evidence is not surfaced at all
Population or statistical likelihood with no patient-specific clinical evidence is not put in front of a clinician as though it were a finding.
A rejected suspect is evidence the control worked
Not a failed opportunity. A confirmation rate near total suggests confirmation rather than evaluation.
Gap Analysis Asks Where the Chain Broke
A supported condition can fail to reach the risk profile because the evidence was never surfaced, documentation was incomplete, coding did not capture it, the item remained unresolved, submission failed, or downstream status was never reconciled. Each has a different owner and fix.
Risk revenue follows accepted encounters, not paid claims. A plan that reconciles claims to submission and stops there is measuring effort rather than outcome, and the gap between submitted and accepted is where earned revenue is lost quietly while unsupported accepted diagnoses sit unexamined.
Encounters with acceptance state
Submission, acknowledgement, acceptance, rejection and correction tracked through to what was actually counted.
Claims with final action resolved
Diagnoses on paid claims compared against accepted encounters.
Clinical documentation
Records, notes, results and correspondence supporting or failing to support each submitted condition.
Pharmacy and laboratory
Objective evidence for suspecting and validation.
Provider data
Mastered provider identity and attribution for provider- and group-level documentation patterns.
Membership with enrollment periods
Eligibility, product and contract periods aligned to the service period.
Reconcile to acceptance, not transmission
The only state supporting risk revenue is accepted and counted. Everything upstream of that is process.
Retain the supporting document
The extraction is derivative. The record is the evidence a reviewer will examine, and it must be retrievable per condition.
Preserve provenance per condition
Which source produced each submitted diagnosis, retained, since source concentration is a risk indicator and vendor-sourced conditions warrant separate tracking.
Use effective-dated enrollment, provider relationships and model versions
Enrollment, provider relationship and model version as of the service period, since current-state joins produce findings that cannot be defended.
A signal is never converted into a diagnosis
Evidence and diagnosis remain separate objects throughout the pipeline. A record that merges them cannot later show what was inferred and what was documented.
Missing evidence stays visible as an unresolved state
Absence of documentation is not silently treated as confirmation or as rejection. It is an unresolved state with an owner.
The governing question for every design decision is what happens when a reviewer pulls a sample and extrapolates the result. A programme comfortable with that scenario is built differently from one optimized for score, and the differences are visible in the incentives, the evidence retention and whether anyone is looking for over-documentation at all.
Programme integrity
- Accuracy stated as the programme objective rather than score improvement
- No vendor, partner or internal incentive paid per condition identified
- Defined deletion protocol with compliance visibility
- Two-directional reporting as standard
Documentation and clinical standards
- Written organizational standard for sufficient documentation
- Coding performed or reviewed by appropriately credentialed staff with QA sampling
- Compliant, non-leading query practice
- No condition submitted on statistical likelihood
Audit readiness
- Supporting documentation, model version, logic version and provenance retained
- Every suspect retains reason, evidence, prioritization, review status and disposition
- Standing internal chart review on a rolling sample
- Error rate and extrapolation exposure understood
- Response runbook for external review
Oversight
- Compliance visibility by design
- Vendor and delegate activity held to the same standard as internal work
- Independent validation of vendor output distinct from the vendor's own quality process
Do not govern only the diagnoses you keep.
The audit trail should preserve the suspects rejected, the diagnoses removed, the opportunities that could not be supported and the items that expired unresolved. Those decisions are the evidence that the programme applies controls rather than accumulating risk-adjusted conditions.
Reporting a risk adjustment programme on score movement creates exactly the incentive the programme should not have. These are the measures we report, and the first two establish whether anything else can be trusted.
| Category | What We Measure | Why It Matters |
|---|---|---|
| Documentation Sufficiency | Share of submitted conditions surviving internal review on a rolling sample | Determines your exposure, and almost nobody reports it. |
| Two-Directional Activity | Conditions added and conditions deleted, reported together every period | A programme with no deletions is not looking. |
| Encounter Recovery | Supported diagnoses recovered through acceptance reconciliation | Revenue earned and not credited, recovered without any new clinical assertion. |
| Source Concentration | Share of conditions derived from chart review, assessment or a single vendor | The pattern a reviewer examines first, tracked before somebody else tracks it. |
| Suspecting Quality | Clinician confirmation and rejection rates by evidence tier | A confirmation rate near total suggests confirmation rather than evaluation. |
| Audit Position | Internal review pass rate, time to assemble a response, extrapolation exposure estimate | The scenario the programme should be designed for. |
Two things clients find uncomfortable and should hear first.
The initial validation may find exposure that has to be disclosed and corrected, which is a cost this year and a substantially larger avoided cost later. And the first year net position will usually be lower than a capture-focused programme would report, because unsupported conditions come off while supported ones go on. Both need agreeing with compliance and finance before the work starts.
We will take a defined sample, test whether documentation supports what was submitted, trace conditions to their source including vendor-derived ones, reconcile the submission chain through to acceptance, and give you both lists with the evidence attached. That exercise takes weeks and it tells you your real position rather than your reported one.
Bring us a specialty or clinical setting where clinicians are spending too much time creating notes. We will assess where ambient documentation fits, what must remain clinician controlled, how it should integrate with your EHR, and how to measure whether it is actually reducing burden.
One conversation with people who have run these deployments, and a written readiness view you can use with or without us.
- AI Agents and Workflow Automation
- Voice and Conversational AI
- Document AI and Intelligent Processing
- Generative AI and Enterprise Copilots
- AI Strategy and Governance
- HCC and Risk Adjustment Analytics
Security & Compliance
