Article Summary: AI-assisted coding is helping hospitals process claims faster, but it may also increase the risk of inpatient coding errors that lead to costly overpayments. Learn how coding inaccuracies in bronchoscopic lung procedure claims impact DRG reimbursement, why these issues often go undetected, and how clinical chart review helps health plans improve payment accuracy and control medical costs.
‍

By: Marilee Knack, Manager of DRG Validation Auditing, Penstock

A patient comes into the hospital. The imaging shows something — a shadow, a nodule, a finding that raises a question.

In the chart, it may be called a solitary pulmonary nodule. To the physician, it’s a diagnosis waiting to be confirmed. To the coding team, it becomes something else entirely: a sequence of diagnoses and procedure codes that will determine how the claim is paid.

The patient undergoes a bronchoscopy. A scope is guided into the airway. A sample is taken. The purpose is to diagnose what the nodule is.

But by the time the claim reaches the health plan, the procedure may not be reported correctly. It may be coded as therapeutic, as if the work was performed to treat a condition rather than diagnose one.  

The difference can be worth thousands of dollars. Multiply that across high-dollar inpatient claims, and the issue becomes less about coding nuance and more about preventable medical spend.

This isn't about denying necessary care. It's about ensuring that what's paid accurately reflects what was documented and delivered.

This is the kind of issue clinical chart reviewers are watching more closely as hospitals adopt AI-assisted coding. Most of what facilities call “AI” today is computer-assisted coding: encoder tools that suggest diagnosis and procedure codes before a coder ever looks at the record. The coder is supposed to validate every suggestion. In practice, that doesn’t always happen. And in payment integrity, a plausible code that moves unchecked isn’t enough.

The numbers reflect it. A June 2026 PwC analysis found that health plans project a 9% rise in commercial medical costs in 2027 — the highest in nearly two decades — with 70% of health plans ranking provider AI tools as a top-three cost driver. A separate 2026 Blue Health Intelligence analysis tied AI-enabled coding to an estimated $2.3 billion in increased spending over three years, without a corresponding increase in the complexity of care delivered.  

At Penstock, our DRG validation reviewers are seeing it firsthand: in bronchoscopy cases alone, roughly 8 or 9 out of 10 claims reviewed are coded as therapeutic when the clinical record supports a diagnostic procedure.

Why Inpatient Claim Coding Errors Are Hard to Catch Before Payment

Inpatient coding isn’t a simple translation exercise. A reviewer may need to move through the admission diagnosis, operative report, pathology, discharge summary, and final diagnosis before the full picture comes into view. They have to understand what was known at the start of the stay, what was discovered during it, and what ultimately drove the admission.

A procedure title may say “bronchoscopy.” But bronchoscopy is only the route into the body. It tells you how the physician got there, not what they did once inside. One bronchoscopy may involve a biopsy. Another may control bleeding. Some are diagnostic. Some are therapeutic. The answer lives in the operative narrative: what was seen, what was sampled, and why.

Physicians don’t write operative reports in the structured language coding systems require, and they’re not expected to. ICD-10-PCS guidelines place that interpretive burden on the coder. A tool may recognize the procedure title and suggest a reasonable code. But the most important question isn’t what happened. It’s why — and that requires reading the full clinical story.

How Repeated Coding Errors Drive Inpatient Claim Overpayments

Payment integrity rarely breaks down in one dramatic moment. It erodes through repetition.

A diagnostic procedure is coded as therapeutic. A complication is sequenced behind the underlying disease. A suggested code carries more specificity than the documentation supports. The claim pays. Then another. Then another. By the time the issue is visible in aggregate, the money’s already moved.

The biggest risk isn’t a code that looks obviously wrong. It’s a code that looks almost right.

Why Health Plans Need Chart Review Both Before and After Payment

The practical response isn’t a broad rejection of AI. Computer-assisted coding tools may help provider teams move faster and reduce administrative burden. Used carefully, with human validation behind every suggestion, they can be part of a stronger workflow.

But health plans can’t assume that speed creates accuracy.

Pre-pay chart review can flag high-risk claims before dollars go out the door. Post-pay chart review catches what the process missed and turns those findings into smarter pre-pay logic going forward.

When the coding loses the story, the payment loses its accuracy. Health plans that invest in the clinical expertise to read that story — before and after payment — are the ones that stay ahead of it.

‍

‍