The Anatomy of a Failed RADV Chart: What Every Retrospective HCC Program Gets Wrong

What a Failed Chart Looks Like From the Inside

The OIG’s March 2026 audit of BCBS Alabama tells the story in numbers: 91% error rate, $7.06 million in estimated overpayments, 100% failure rates for acute stroke and myocardial infarction categories. But the numbers don’t show what the failed charts actually looked like. Understanding the anatomy of a failed chart is what separates programs that repeat the same errors from programs that fix them.

A typical failed chart follows a pattern. The medical record contains a problem list that mentions a chronic condition, say diabetes mellitus type 2. The progress note for the relevant encounter says something like “DM2 stable, continue current medications.” A coder reviewing the chart sees diabetes mentioned, confirms the ICD-10 code maps to an HCC, and submits it. The code is technically present in the record. The submission looks reasonable.

Then the auditor reviews it. They don’t ask whether diabetes appears in the chart. They ask whether the documentation proves the condition was actively managed during the encounter. “DM2 stable, continue current medications” doesn’t reference a specific lab value (monitoring), doesn’t describe clinical findings (evaluation), doesn’t document the provider’s assessment of disease progression (assessment), and “continue current medications” is the thinnest possible treatment evidence. The code fails. Multiply this by hundreds of charts, and you get 91%.

The Three Layers of Failure

The first layer is provider documentation. The note wasn’t written to satisfy audit standards. It was written for clinical continuity. The provider knows they’re managing the patient’s diabetes. The note is a shorthand reminder, not a compliance document. Fixing this layer requires provider education and EHR template improvements that prompt more specific documentation without adding excessive burden.

The second layer is coder validation. The coder saw the diagnosis and submitted it without evaluating whether the documentation met MEAT criteria. This happens because most coding workflows don’t include a structured MEAT validation step. The coder’s job is defined as finding codes, not proving them. Under throughput pressure, coders accept what the chart appears to show rather than critically evaluating whether the evidence would survive audit scrutiny.

The third layer is technology. The tools the coder used identified the diagnosis mention but didn’t assess documentation quality. The system flagged “DM2” as a potential HCC and left everything else to the human. No MEAT mapping. No evidence scoring. No defensibility assessment. The technology accelerated the coder through a flawed process rather than catching the flaw.

Fixing All Three Layers Simultaneously

Programs that fix only one layer keep failing. Provider education without technology support produces better notes that coders still don’t validate systematically. Better coder training without AI assistance produces better judgment that throughput pressure still overrides. Better AI without workflow integration produces evidence assessments that coders don’t have time to use.

The fix requires all three layers working together. AI evaluates documentation against MEAT criteria and presents evidence-mapped recommendations. Coders validate those recommendations with the evidence visible, not hidden in 40 pages of chart. Provider feedback loops route documentation gaps back to clinicians with specific, actionable guidance rather than generic reminders to “document better.”

When all three layers function together, the chart that says “DM2 stable, continue current medications” gets flagged before submission, not during an audit. The system identifies the MEAT gap. The coder queries the provider or removes the code. The evidence trail documents the decision.

The Program Design Implication

Every failed RADV chart traces back to the same root cause: the coding process prioritized identification over validation. Retrospective Risk Adjustment HCC Coding programs that embed MEAT validation into the workflow, equip coders with evidence-first AI, and establish provider feedback loops are addressing all three failure layers simultaneously. Programs that fix one layer while leaving the others untouched will produce the same 80-91% error rates the March 2026 audits documented. The anatomy of failure is well understood. The question is whether the program’s design prevents it or enables it.