Risk adjustment analytics that surface genuinely supported conditions for clinician evaluation, assess whether documentation actually supports what was submitted, and identify codes that should be removed as readily as codes that should be added.
Most risk adjustment programmes are measured on score lift, which quietly makes them programmes for finding more codes. CaliberFocus builds them the other way round. We start from the population and the evidence, present suspected conditions to clinicians with the data behind them rather than as a prompt, assess whether existing documentation would survive a review, and report unsupported submissions for correction. The score that results is the one the record supports.
If your risk adjustment programme cannot produce a list of codes that should be deleted, it is not measuring accuracy. It is measuring revenue.
Risk adjustment rewards documenting the conditions a patient genuinely has. The difficulty is that every commercial incentive pushes one way only, while enforcement attention has moved toward unsupported diagnoses and the analytics that generated them.
A programme measured on score improvement will produce score improvement. Whether that improvement is defensible is a separate question, and it is the one that matters when charts are reviewed.Â
Recapture is an operational problem
Most of the apparent gap is patients with no qualifying encounter, or an encounter where the condition was never addressed.
A problem list is not evidence
Problem lists carry resolved conditions, duplicates, historical diagnoses and entries copied forward without current assessment.
A claim code is not the same as support
A submitted diagnosis without documentation evidencing evaluation or management is a liability rather than revenue.
Suspecting tools that prompt rather than inform
Bare code prompts invite rubber stamping or wholesale rejection. Neither produces accuracy.
No view of what the plan actually accepted
Submitted, accepted and acknowledged are three different populations.
Provider reporting that reads as policing
Clinician-level coding performance framed as a compliance scorecard destroys the engagement the programme depends on.
Under audit, a small error rate on a large population is not a small number.
Chart-review samples can be extrapolated. That arithmetic turns a modest documentation weakness into material financial exposure. We build for the sample, not for the score.
Chart-review samples can be extrapolated. That arithmetic turns a modest documentation weakness into material financial exposure. We build for the sample, not for the score.
Almost every engagement in this space begins by looking for what is missing. We begin by examining what is already there. That establishes credibility with clinicians and surfaces existing exposure before new opportunities are pursued.
Step 1
Establish population and attribution
Define which patients are in scope, under which arrangement, attributed to which providers, for which period.
Step 2
Build the condition picture
Assemble claims, problem lists, encounters, medications, labs, imaging and specialist correspondence.
Step 3
Audit what was already submitted
Test whether existing diagnoses are supported by current-period documentation.
Step 4
Identify recapture by cause
No qualifying encounter, encounter without assessment, or assessed and documented but not coded.
Step 5
Surface suspected conditions with evidence
Present the underlying clinical information and no code pre-selected.
Step 6
Deliver into the pre-visit workflow
Put the information into the workflow the care team already uses before the encounter.
Step 7
Support documentation quality
Help clinicians document what they assessed to the specificity the record supports.
Step 8
Reconcile the chain
Compare documented, coded, submitted, accepted and payer-acknowledged states.
Step 9
Retain evidence and monitor accuracy
Keep the logic, evidence and lineage required to explain the programme years later.
The technical work here is not condition detection. The defensible work is assembling the evidence, assessing whether documentation supports it, and delivering it in a form a clinician will engage with.
Build the Picture
Population and Attribution Management
Patients in scope, arrangements, providers and periods reconciled against the payer view.
Multi-Source Condition History
Claims, problem lists, encounters, medications, labs, imaging, specialist correspondence and external data.
Patient Identity as a Prerequisite
Unresolved matches enter stewardship rather than becoming confirmed clinical history.
Submission Chain Reconciliation
Documented, coded, submitted, accepted and payer-acknowledged compared as distinct states.
Find the Opportunity, Both Directions
Recapture Identification
Persistent conditions classified by the reason they have not been addressed.
Evidence-Based Suspecting
Objective clinical data attached to every candidate condition.
Documentation Sufficiency Assessment
Whether existing documentation evidences evaluation, assessment and management.
Deletion and Correction Identification
Unsupported diagnoses identified and returned for correction as a standard output.
Deliver and Defend
Pre-Visit Workflow Delivery
Opportunities delivered into the pre-visit and encounter workflow.
Provider-Level Insight
Clinician-level patterns presented as education and adjusted for patient complexity.
Audit Evidence and Model Versioning
Supporting documentation, model version, logic version and lineage retained together.
What CaliberFocus does, and does not do?
We are not paid on score movement and we will not accept an engagement structured that way. We do not pre-populate diagnoses, we do not present codes without evidence, and we do not conclude that a patient has a condition. Analytics can surface a clinical question. Only the clinician can answer it, and only the record can support it.
We are not paid on score movement and we will not accept an engagement structured that way. We do not pre-populate diagnoses, we do not present codes without evidence, and we do not conclude that a patient has a condition. Analytics can surface a clinical question. Only the clinician can answer it, and only the record can support it.
Programmes tend to be sold on suspecting, which is the most interesting category and usually the smallest. The larger and safer opportunity is generally in conditions the organization already knows about, and in submissions it has already made.
Recapture
A persistent condition documented previously and not addressed this period. Lowest risk, highest volume.
Specificity
A condition documented less specifically than the record supports. Accuracy improvement with no new clinical assertion required.
Suspecting
A condition suggested by objective data and never documented. Genuine clinical value and highest scrutiny.
Correction and deletion
A submitted diagnosis the documentation does not support. Reduces exposure and is treated as an opportunity, not an embarrassment
Strong evidence
Objective results, condition-specific medication or specialist diagnosis, surfaced with source values and dates visible.
Moderate evidence
Indirect indicators, historical claims or related procedures, surfaced as a question with reasoning shown and clearly labelled as inference.
Weak evidence — Not surfaced
Statistical or population-level likelihood with no patient-specific clinical evidence. We do not put probability in front of a physician as if it were a finding.
What Sufficient Documentation Looks Like
A code is not support. Documentation sufficient to survive review generally shows that the condition was actually evaluated and managed in the period: an assessment, a status, and a plan or rationale for continuing current management.
A code is not support. Documentation sufficient to survive review generally shows that the condition was actually evaluated and managed in the period: an assessment, a status, and a plan or rationale for continuing current management.
A suspected condition is a question, not a diagnosis. Each signal is handled the same way: surface the evidence, frame the clinical question, route it for review, and stop.
| Signal | The Clinical Question | Action | What We Never Do |
|---|---|---|---|
| Prior-period condition not documented this period | Is the condition still present and clinically relevant? | Present history for provider review | Carry the diagnosis forward automatically |
| Problem-list entry with limited current documentation | Is it still active, and was it assessed or managed? | Route for clinical review | Treat problem-list presence as current-period support |
| Medication specific to a condition | Does the record support evaluating that condition? | Surface medication and context | Infer the diagnosis from the prescription |
| Laboratory or diagnostic result | Has the clinical significance been evaluated and documented? | Bring result and context into review | Convert an abnormal result directly into a diagnosis |
| Specialist documentation not reflected elsewhere | Is it current, supported and relevant here? | Make the documentation available | Assume it propagates to every encounter |
| Documentation supporting greater specificity | Does the record support a more precise representation? | Route for documentation or coding review | Code beyond what the record supports |
| Coded condition with weak or conflicting support | Can the diagnosis be substantiated? | Prioritize for validation and correction | Protect a code because removing it reduces capture |
Confirmed and documented
Already supported
Needs clarification
Not supported
Resolved or historical
Duplicate
Deferred
Risk stratification built purely for coding produces a list of patients worth a documentation visit. Risk stratification built for care produces a list of patients who need attention, most of whom also carry undocumented complexity.
Care management
Patients whose complexity or trajectory warrants proactive management.
Visit planning
Patients with unaddressed chronic conditions and no scheduled encounter.
Documentation completeness
Patients where the record does not reflect known complexity, addressed during the encounter.
Clinician-level coding reporting should be presented as education, adjusted for patient complexity, and designed with clinical leadership.
Adjust for the panel
Unadjusted comparison across physicians with different patient complexity is noise with a name attached.
Route through clinical leadership
Physician documentation feedback comes through clinical leadership, not finance.
Never link compensation to capture
Compensating clinicians on documented risk score creates the wrong incentive.
Organizations routinely manage risk adjustment against what they believe they submitted. The payer is working from what it accepted and acknowledged. The gap between those positions is invisible without deliberate reconciliation.
Documented but not coded
The note evidences evaluation and management, but the coder or logic cannot act on it.
Coded but not submitted
Present on the encounter, absent from the file actually transmitted.
Submitted but not accepted
Rejected for format, eligibility or timing and never resubmitted.
Accepted but not acknowledged
The payer accepted the record but the condition does not appear in its view of the member.
Built on the Governed Platform, Not Beside It
This work consumes the same certified data products as the rest of the organization. If the provider dashboard, coder worklist and finance report disagree about a patient’s risk profile, that is one unresolved data problem.
This work consumes the same certified data products as the rest of the organization. If the provider dashboard, coder worklist and finance report disagree about a patient’s risk profile, that is one unresolved data problem.
The governing question for every design decision is what happens when a reviewer pulls twenty charts. A programme comfortable with that scenario is built differently from one optimized for score.
Programme integrity
- Accuracy, not score movement, as the stated objective
- Defined deletion protocol
- No incentive paid per condition captured or score movement
- No clinician compensation linked to documented risk score
Documentation standards
- Written organizational standard for sufficient documentation
- Compliant and non-leading query practice
- Query templates reviewed by compliance
Audit readiness
- Supporting documentation, model version, logic version and lineage retained together
- Bidirectional audit trail
- Documented response runbook
Access and oversight
- Compliance visibility by design, including deletion-list access
- Clinician-identifiable performance data access controlled appropriately
The uncomfortable test.
Ask whoever runs your risk adjustment programme today for the list of codes they have recommended deleting in the last twelve months. If there is no list, the programme has only ever been pointed in one direction.
Ask whoever runs your risk adjustment programme today for the list of codes they have recommended deleting in the last twelve months. If there is no list, the programme has only ever been pointed in one direction.
Score is an output. Reporting a programme on it creates exactly the incentive the programme should not have. These are the measures we report.
| Category | What We Measure | Why It Matters |
|---|---|---|
| Documentation Sufficiency | Share of submitted conditions with documentation that survives internal review, on a rolling sample | The measure that determines exposure |
| Two-Directional Activity | Conditions added and codes identified for deletion, reported together every period | A programme with no deletions is not looking |
| Recapture Completeness | Persistent conditions addressed in the period, and reason for each that was not | The largest and lowest-risk part of the opportunity |
| Suspecting Quality | Clinician confirmation rate and rejection rate on surfaced conditions, by evidence tier | Very high confirmation may suggest rubber stamping; rejection is a healthy signal |
| Audit Position | Internal review pass rate, time to assemble an audit response, extrapolation exposure estimate | The scenario the whole programme should be designed for |
Expect the first year to be uncomfortable.
The first year may produce a lower net position than a capture-focused programme because unsupported submissions are coming off at the same time as supported conditions go on. Initial review may also identify exposure that needs correction. Those realities should be agreed with compliance and finance before the work starts.
The first year may produce a lower net position than a capture-focused programme because unsupported submissions are coming off at the same time as supported conditions go on. Initial review may also identify exposure that needs correction. Those realities should be agreed with compliance and finance before the work starts.
We will take a defined patient sample, assess whether the documentation supports what was submitted, identify the persistent conditions that were missed, and give you both lists with the evidence attached. That exercise takes weeks, it tells you your real position rather than your reported one, and it is the only honest basis for deciding what the programme should look like.
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
