A federal watchdog went looking at insurers who billed Medicare for strokes. It pulled 97 patient records. Every single one failed review. Its estimate of the cost for one year, for that one condition: about $462 million.
That is public money. Same pool that funds everyone else’s coverage.
But the number is not the interesting part. The interesting part is that the watchdog had already published, in detail, exactly how it catches this. Two and a half years earlier. And it still found a 100 percent failure rate.
Follow the incentive, not the villain
Here is how the money moves. Medicare pays private insurers a monthly amount for each person they cover, and pays more for people who are sicker. That design is deliberate and it is basically sound, because without it insurers would quietly avoid anyone expensive.
How sick you are gets established by diagnosis codes pulled from your medical notes. Sicker codes, bigger payment. Which means somebody has to check whether the notes actually say what the codes claim.
Now think about what software got built in that environment. Insurers bought tools to review medical records and find diagnoses that had been missed, because every one found meant more money. That is what the market paid for, so that is what got engineered.
Nobody commissioned software to go looking for codes that should not be there. There was no revenue in it. So it never got built.
That is the whole story. Not a conspiracy. An incentive structure doing exactly what it was designed to do, with nobody on the other side of it until a regulator showed up.
They handed over the answer key
In December 2023 the watchdog published the method it uses to find these cases. Not a policy summary. The actual database queries.
The logic is almost embarrassingly simple. A real acute stroke leaves a trail: an ambulance, an emergency room, a scan, an admission. So the query asks whether there is a stroke on a doctor’s office record with no matching hospital record anywhere that year. When those two facts contradict each other, a human opens the chart.
The watchdog had also said plainly what it kept finding across its audits: “approximately 70 percent of those diagnosis codes were not supported in the associated medical records,” with some categories failing “over 90 percent of the time.”
Method published. Failure rates published. Thirty months of warning. Then 97 out of 97.
What was actually in those records
Worth being precise here, because the headline invites the wrong conclusion.
Sixty-eight of the 97 patients had really had a stroke. It was in the chart, documented, undeniable. The problem was that it had happened earlier and the insurer billed a code saying it was happening right then. An old stroke and an active one are different codes worth different money.
Twenty-two records had no stroke documentation at all. Four could not be found. One had been signed by a pharmacist instead of a doctor. One nobody could read.
Why a wrong record follows you
This is where it stops being an accounting story.
A chart calling a six-year-old stroke an active emergency is wrong in a way that travels. The next doctor reads it. The emergency room reads it. Decisions about your care get made from it, by people who have never met you and have nothing else to go on.
For communities already getting worse access and worse outcomes, a medical record that misrepresents you is not paperwork. It is one more thing stacked against you in a room where you already have the least say.
Some insurers are now moving to a risk adjustment solution built to remove unsupported diagnoses as well as add missing ones. It took a federal audit and a nine-figure estimate to make that worth building. The instructions had been sitting in public the entire time.