One hospital in Tennessee had 31 heart-failure discharges and 14 readmissions. That is 45.2%. CMS publishes 21.5% for the same hospital, same period. Watch all 2,316 make the trip.
Both numbers come from the same government file. One is division. The other is a hierarchical model that risk-adjusts and then shrinks small hospitals toward the national mean, because at 31 patients the arithmetic is mostly noise wearing a percentage sign.
Hospitals
2,316
Spread (SD)
—
Largest single move
—
Median move, <75 disch.
—
Horizontal: heart-failure discharges per hospital, log scale. Vertical: 30-day readmission rate. Dashed line is the national pooled rate, 19.69%. Red segments are the distance between the two published numbers.
Nobody asks how they'd know a number was wrong when it looks broken. Broken is obvious. You need the question for a number like this one — arithmetic, published by CMS, correct in every digit.
It is not just heart failure
HRRP publishes six conditions. Each panel shows the middle 80% of hospitals — the 10th to 90th percentile — at six discharge volumes. Red is the rate you can compute. Navy is the rate CMS publishes. Read left to right: the red band narrows toward the navy one as the denominators get big.
Two of the six don't behave, and that's the honest part. CABG and hip/knee replacement have the fewest hospitals in the file (363 and 253) and the lowest event rates, so their percentile bands wobble instead of narrowing — the widest CABG band is in the second-largest volume bucket, which has 23 hospitals in it. If a trend only holds in four of six panels, the honest reading is that it holds in four of six panels. The buckets are marked below each chart; watch the ones with n under 60.
What the benchmark cannot see
Interrogating the file, not just plotting it
A quarter of the hospitals were removed before you got here. In the HRRP file, 36% of rows carry no ratio and 55–56% carry no discharge count — suppressed under low-volume rules. The 2,316 on this page are the survivors of a reporting threshold, so the left edge of every chart is already trimmed.
The denominator is Medicare fee-for-service only. Medicare Advantage enrollees — now more than half of Medicare — are not counted. A hospital's published readmission rate describes a shrinking slice of its own heart-failure population.
Risk adjustment reads claims, not lives. It sees coded comorbidities. It does not see housing, transport, pharmacy access, or whether anyone was home at discharge. CMS conceded the gap in FY2019 by stratifying hospitals into peer groups by dual-eligible share — an admission the flat model had been scoring poverty.
The window runs July 2021 to June 2024. Three years averaged into one number, across the tail of a staffing collapse. The average looks like a steady state. It was not one.
The shrinkage is a choice, not a fact. Pulling small hospitals toward the mean is defensible and it also guarantees that a genuinely bad small hospital looks average. The model does not tell you which one you are looking at. Neither does the raw number. That is the actual state of knowledge, and neither published figure shows it.
Why this is on a page about a play
Today's newsletter is about a phrase George Washington had worn smooth — the calm light — that came out on its own in the worst room of his career, because he had said it enough times on ordinary days.
The clinical version is what's the worst thing this could be, asked on five hundred chest walls that turn out to be pulled muscles, so that at 3 AM it asks itself.
The builder's version is how would I know if this were wrong? — and you will not think of it in the moment. You will only reach for the one you already wore smooth. Which is why it has to be rehearsed on the days the number comes back clean.