clinicians.build · interactive · September 20, 2026 · built on Thompson et al., Health Services Research 2016

Nothing Changed. The Grade Moved.

These are the 222 hospitals in Medicare’s worst decile for heart-failure readmissions over the three years ending June 2021. Press play and watch where they land in the three years that follow — a window that shares not one patient with the first.

Primary source: Thompson MP, Kaplan CM, Cao Y, Bazzoli GJ, Waters TM. “Reliability of 30‑Day Readmission Measures Used in the Hospital Readmission Reduction Program.” Health Services Research 2016;51(6):2095–2114 (doi:10.1111/1475-6773.12587, PMID 27766634) — readmission rates for medical conditions fall below the reliability benchmark at most hospitals, and “approximately 25 percent of payments for excess readmissions were tied to unreliable” rates.
Data: CMS Hospital Readmissions Reduction Program (Provider Data Catalog) via MIMI Labs. 2,223 hospitals with a published heart-failure ratio and denominator in both of two non-overlapping windows: Jul 2018–Jun 2021 and Jul 2021–Jun 2024.
Read the reliability paper → Or read today’s newsletter →
the number
1.136 → 1.044
Mean excess readmission ratio of the worst decile of heart-failure hospitals, 2018–21 → 2021–24.
The best decile moved the other way: 0.876 → 0.939. Both crowds walked toward 1.00 from opposite sides.

Medicare’s excess readmission ratio divides how often your heart-failure patients came back by how often CMS’s model says patients like yours should have. Above 1.00 is a finding against you, and it is worth up to 3% of your base Medicare pay.

Each dot below is one hospital, placed by its ratio and stacked to keep the crowd readable. Red is the worst decile of the first window — the 222 hospitals with the strongest case against them. Navy is the best decile. Everyone keeps the colour they were given in 2018–21, so the second frame shows where that cohort went, not who the new worst hospitals are.

window 1 of 2 · Jul 2018 – Jun 2021
worst decile, 2018–21 (222) best decile, 2018–21 (222) the other 1,779
worst decile, mean
best decile, mean
gap between them
worst decile now under 1.00
Hover any dot for the hospital, its discharge counts, and both ratios. Dots keep their colour from window 1, so the second frame shows where that cohort went — not who the new worst hospitals are.

The red crowd did not disperse evenly. It collapsed inward. None of the 222 was below 1.00 in the first window — that’s what put them in the decile. In the second, 63 of them are, on the side of the line where CMS has nothing to say to you. Eight are in the best decile outright.

Only 55 of the 222 are still in the worst decile. That is 25%, against 10% if the grade were pure noise — so there is something real in there, about two and a half times chance. It is just nowhere near a verdict, and the penalty is written as a verdict.

The objection that doesn’t work here

You’d reach for sample size. Small hospitals bounce; that’s the whole of it. It’s the right instinct and on this dataset it comes out backwards — sort these hospitals into five volume groups and the spread of the ratio grows with volume.

Standard deviation of the published heart-failure ratio within each fifth of the volume distribution, current window (2,316 hospitals with a published denominator). A textbook funnel plot slopes the other way.
the 80/20 lens Small hospitals aren’t steadier — CMS already knows they’re noisy and has dealt with it upstream. The ratio comes out of a hierarchical model that shrinks a low-volume hospital toward the national average before publication, which is why the smallest fifth is pinned near 1.00. So the number on the page has already been smoothed to make it trustworthy, and it still loses 83% of itself over the next three years (r = 0.407 across these 2,223 hospitals). Whatever is left after the smoothing is mostly not a property of the hospital.
what this is not

Why you already know how to think about this

M&M exists because medicine decided, formally, that the outcome is not allowed to grade the decision. You present the case that went fine and the room takes it apart anyway, because a good result can be built out of luck and a bad one out of nothing you did.

Then the same hospital gets a number like 1.14, and the number arrives with a dollar figure, and nobody runs it through the room.

The hospital at 1.14 that is now at 1.04 has a story about the discharge pathway it rebuilt. Maybe it did. The chart above cannot tell, and neither can the hospital. What the chart can tell you is that 222 hospitals moved an average of 0.09 in the same direction, together, and most of them had no pathway at all.