clinicians.build · interactive · September 19, 2026 · built on HHS‑OIG, Sep 17 2026

Paperwork, Not Medicine

OIG pulled 100 claims at one hospital, found 27 wrong, and demanded $12.4 million. Fourteen years of Medicare’s own claim reviews say the thing an audit catches is almost never a judgment call. It’s a missing page.

Primary source: “Methodist Hospital Received at Least $12.4 Million in Medicare Overpayments,” HHS Office of Inspector General, report A‑09‑23‑03001, issued Sep 14 2026, posted Sep 17 2026 — 100 claims worth $1,426,020 reviewed; 73 compliant; 27 not; $256,926 in net overpayments found; $12.4M of $62M extrapolated. OIG: the errors occurred “primarily because the Hospital did not always follow its written policies and procedures.”
Data: CMS Comprehensive Error Rate Testing (CERT) claim-level review files via MIMI Labs — every claim CERT medically reviewed in reporting years 2011 through 2024: 2,685,339 claims, 553,163 flagged. Reporting year 2024 covers claims submitted Jul 1 2022 – Jun 30 2023.

Read the OIG audit → Or read today’s newsletter →

the number

64.8%

of the 553,163 Medicare claim errors found in 14 years of CERT review were documentation — insufficient or absent.
Medical necessity, the part that is actually a clinical argument, was 15.2%.

Extrapolation is what turns a sample into a bill. It works on one assumption: that a hospital which got 27 of 100 claims wrong has a habit, and the habit runs through the other 43,000 claims too.

So it is worth knowing what kind of habit auditors actually find. Not in one hospital — across every Medicare claim the government has pulled and read since 2011.

Scrub the year 2024

Bottom panel

Documentation — insufficient or none Medical necessity Incorrect coding Other

year—

claims reviewed—

flagged—

documentation share—

The shape of it

The top panel is the error rate CERT found each year. It peaked at 30.97% in reporting year 2014 and has roughly halved, to 15.80% in 2024. That is a real improvement, and it is mostly one category shrinking.

The bottom panel is what those errors were. The red band — documentation — is 75.4% of every error found in 2014 and 56.3% in 2024. It is the band that fell, and it fell a long way: from 42,761 claims to 16,476. Switch the panel to claim counts and you can see that medical necessity and incorrect coding barely move at all across the whole series. One category shrank. The others just stood still.

Which means the thing Medicare’s reviewers overwhelmingly catch is not a doctor being wrong. It is a chart that does not prove the doctor was right. Those are different failures with the same price tag.

the 80/20 lens Twenty percent of the categories carry eighty percent of the exposure, and it is the boring twenty. If you are buying a tool to protect your book against an audit, the question is not can it reason about the case. It is can it tell, before the claim goes out, whether the note supports the code. Those are not the same product, and only one of them is what 64.8% of findings are about.

Why 2012 looks strange

One year breaks the pattern. In reporting year 2012, medical necessity jumped to 32.9% of all errors — 19,338 claims, more than double any other year in the series — while documentation dropped to 44.8%. Then it snapped back.

A spike that size in one year, in one category, across a 264,003-claim sample, is what a change in review policy looks like, not a change in medicine. The dataset does not say which change. It is a good reminder that an error rate is partly a measurement of the auditor.

what this chart is not

These are claim counts, not dollars. Counting flagged claims across 14 years gives 20.60%. CMS’s own headline Medicare fee-for-service improper payment rate for FY2025 is 6.55% — because CMS weights by dollars and by sampling stratum, and CERT deliberately over-samples claim types it considers high risk. Same reviews, very different number. Neither is wrong; they answer different questions. Ours is “how often does a reviewed claim fail,” not “what share of Medicare’s money moved improperly.”
“Insufficient documentation” is not a finding that the care was unnecessary. CMS classifies a claim this way when the record submitted does not support the service billed. Some of those are genuinely unsupported services. Some are a signature, a date, or an order that exists but never made it into the packet a reviewer got 45 days to receive.
An error is not fraud. CERT is an accuracy measurement program, not an enforcement one. It makes no intent determination.
The categories are CMS’s, and they overlap in practice. A claim downcoded because the note was thin could be filed as documentation or as incorrect coding depending on the reviewer’s read.
14 years is not one continuous instrument. Sample sizes range from 158,250 to 264,003 a year, review contractors and edit policy changed repeatedly, and the 2020–2021 years sit inside a public health emergency with wide documentation flexibilities. Read the trend as a trend, not a measurement.

The practical version

Auditors do not sample your judgment. They sample your file room. That is good news and bad news: good, because the failure mode is mechanical and therefore checkable before the claim leaves; bad, because mechanical failures are exactly the kind software reproduces perfectly, on every claim, at three in the morning, without variance.

A hospital with a habit gets extrapolated. A system with a defect gets extrapolated harder.

Next: pull your own hundred → — the same dataset, 89 provider types, and what a sample of 100 can and cannot tell you about your own book.