We hit 97% accuracy in risk adjustment HCC coding last quarter. The champagne came out. Bonuses got approved. Then CMS audited us and failed 43% of those “accurate” codes.

How can both numbers be true? Because we’ve been measuring the wrong thing this entire time.

The accuracy illusion

When we measure coding accuracy, we check if the coder selected the right HCC for the documented condition. Diabetic neuropathy documented? E11.42 coded? Accurate! Except nobody checked if that documentation would survive an audit.

Here’s what our 97% accuracy actually meant: coders correctly identified diagnoses mentioned in charts. That’s like celebrating that your teenage kid correctly identified a stop sign while texting and driving through it.

The provider wrote “probable diabetic neuropathy” in his note. Our coder caught it, coded it, and our accuracy metrics counted it as a win. CMS saw “probable” and rejected it instantly. Probable isn’t definitive. Our accurate code was built on sand.

We audited 1,000 of our “accurately” coded HCCs last month. Want to guess how many had rock-solid documentation? 312. The rest had problems: missing signatures, unclear dates, absent MEAT criteria, or hedging language like “likely” and “possible.” All accurately coded. All audit failures waiting to happen.

The speed trap

Our fastest coder, Marcus, reviews 55 charts daily. Our most accurate coder, Linda, manages 22. Guess who gets praised at team meetings?

Marcus skims documentation, spots obvious diagnoses, codes them, moves on. His accuracy rate? 94%. Looks great on paper. His audit survival rate? 41%. He’s accurately coding conditions that aren’t properly documented.

Linda reads every note twice. She verifies MEAT criteria. She checks provider signatures. She confirms dates align. Her accuracy rate? 92%. Slightly lower because she refuses to code borderline cases. Her audit survival rate? 89%.

We’re incentivising Marcus to plant time bombs in our submissions while punishing Linda for protecting us. The metrics say Marcus is our star. The audit results say he’s our biggest liability.

The pattern nobody tracks

Every organization tracks coding accuracy by percentage. Nobody tracks accuracy by condition type, provider specialty, or documentation source. That’s where the real story lives.

Our accuracy for diabetes? 99%. Fantastic! Our accuracy for diabetes specifically from endocrinology notes? 61%. Our coders nail simple cases but struggle with specialist terminology. The aggregate metric hides the pattern.

Mental health HCCs show 95% coding accuracy overall. But drill down: psychiatrist notes have 98% accuracy, while primary care mental health diagnoses hit only 72%. We’re missing thousands in HCCs because coders don’t recognize depression when PCPs document it differently than psychiatrists.

The worst pattern? Hospital discharge summaries. Our accuracy rate is technically 93%, but we only code from them 30% of the time. The other 70% never get reviewed because they’re “too complex”. Can’t be inaccurate if you don’t code at all, right?

The documentation reality check

We started measuring something different: defensible accuracy. Not “did you code what was mentioned?” but “would this code survive aggressive audit scrutiny?”

The results were sobering. Our real accuracy, measured by audit-ready standards, was 53%. Almost half our “accurate” codes were wishes built on weak documentation.

Now every coded HCC gets two scores: coding accuracy (did you identify the right condition?) and documentation strength (can you defend it?). High coding accuracy with low documentation strength means the coder did their job but the provider didn’t. That’s actionable intelligence, not feel-good metrics.

We track rejection risk by provider. Dr Thompson has 94% coding accuracy but 38% documentation strength. Every HCC from her notes is a gamble. Dr. Patel shows 91% accuracy with 94% documentation strength. His notes are gold.

Your Monday morning audit

Pull 20 random HCCs your team coded last week as “accurate.” Now pretend you’re a hostile auditor. Can you find the exact words supporting each diagnosis? Not implied. Not suggested. Exact words.

Check for hedging language: possibly, probably, likely, suspected, rule out. If these words appear anywhere near your coded diagnosis, your “accurate” code is actually inaccurate for audit purposes.

Time how long finding supporting documentation takes. If it takes you more than 30 seconds to locate clear support for an HCC, an auditor won’t find it at all. That accurate code is functionally worthless.

Count how many of your 20 “accurate” codes you’d bet your own money would survive audit. If it’s less than 15, your accuracy metrics are vanity numbers that will evaporate under scrutiny.

The truth about risk adjustment HCC coding accuracy? Most organisations are accurately coding conditions that can’t be defended. We’re precisely wrong, which is worse than being vaguely right. Stop celebrating accuracy metrics that measure coding correctness while ignoring documentation reality. Start measuring what matters: defensible accuracy that survives when CMS comes calling.

Tim Williamson, a psychology graduate from the University of Hertfordshire, has a keen interest in the fields of mental health, wellness, and lifestyle.