Show your work

From CMS’s file to this page

No hospital’s number is typed in by hand. A program reads CMS’s file, works out what the pages show, checks it, and only then writes the pages. These are the real numbers from the newest year’s run.

1. 1

   CMS publishes

   3,055 rows

   Provider Data Catalog dataset yq43-i98g, released August 13, 2026

2. 2

   This site pins it

   595,701 bytes

   Fingerprint (SHA-256) 5e152d9872e6… · if one byte changes, the tests fail

3. 3

   It reads every row

   6 measures

   CMS’s values, z‑scores, footnotes and decisions, kept exactly as CMS wrote them

4. 4

   It adds up every score

   2,929 scores

   Each hospital’s parts added up again by the year’s rule

5. 5

   It checks against CMS

   0.00008 largest miss

   2,886 decisions, 0 out of order; the build stops at a miss of 0.001 or any decision out of order

6. 6

   Then it writes the pages

   3,532 pages

   The program page, every hospital and every place

The formula

CMS takes the same steps for every hospital it scores.

limited value

the hospital’s value, held between CMS’s floor (the 5th percentile) and cap (the 95th percentile)

z‑score

(limited value − the measure’s average) ÷ the measure’s standard deviation

Total HAC Score

the sum of the hospital’s z‑scores ÷ the number of measures it was scored on

cut

a Total HAC Score above the 75th percentile of the scores of hospitals outside Maryland that year

CMS sets each floor and cap, average and standard deviation from the hospitals’ own results. Its public file doesn’t list them, so where this site needs them (to move a measure in a what-if) it fits them from the values and z‑scores CMS published, and says so.

This equal-weight formula started in fiscal year 2020. Before that, CMS grouped the measures into two domains and weighted them, and before fiscal year 2018 it gave each measure points by decile, one point for the best tenth of hospitals up to ten for the worst, instead of a z‑score. The check above uses each year’s own rule, which is also why these pages draw no Total HAC Score trend across those changes.

What comes from CMS, and what this site works out

Every figure says which it is. Where a page shows CMS’s own value, it shows CMS’s text unchanged.

The estimate tag

A recomputed z‑score can differ slightly from CMS’s. This site tags a measure “estimate” when its recompute misses CMS’s z‑score by more than 0.001 for any hospital; for fiscal year 2026 that is PSI 90 and CDI. The largest difference for a tagged measure is 0.0070. Every dollar figure carries the tag too: the dollars are this site’s estimate, explained under How this site estimates the dollars.

What “estimate” means across Paid & Penalized

How this site estimates the dollars

CMS’s HACRP file gives each hospital’s payment decision, not the dollars it moves. So this site rebuilds each hospital’s year of traditional Medicare inpatient payments, with the teaching payments Medicare makes on Medicare Advantage stays, from CMS’s FY2026 payment files, then applies Medicare’s three hospital quality programs: the readmissions and value-based adjustments to the base payment, and the HAC cut last, as the law sets. Every dollar figure this site estimates is tagged estimate.

The order is the point. The readmissions and value-based programs each adjust only the base payment; the HAC cut comes last and takes 1% of everything left.

Readmissions and value-based: each adjusts the base alone

Base payment $87.1BDRG rates × cases × case mix, plus a Medicare-dependent hospital’s add-on

Everything else $41.9Bteaching (Medicare Advantage stays too), low-income patients, outliers, capital, uncompensated care…

Then the HAC cut: 1% of all of it, last

Estimate · Source: this site’s arithmetic on CMS’s FY2026 IPPS final rule files (the impact file, the rate tables and the value-based factors), its FY2026 HRRP supplemental file and HACRP FY2026 (yq43-i98g), released August 13, 2026.

Show table

Part of the payment	This site’s estimate (whole dollars)
Base payment	$87,073,482,744
Everything else	$41,909,061,619

The totals on these pages add the 709 of CMS’s 718 cut hospitals with a usable row in CMS’s impact file; the other 9 have none, so they have no estimate. One more cut hospital, which CMS gave no Total HAC Score, shows its estimate on its own page and stays out of the totals, as it stays out of the site’s counts.

Maryland’s hospitals are left out of every dollar total: they are exempt from the cut.

The formula, with this year’s numbers

1. 1

   Base payment

   $87.1B

   DRG rates × each hospital’s cases and case mix, plus the Medicare-dependent add-on; 2,871 hospitals

2. 2

   Readmissions

   −$336M

   each hospital’s base × (its factor − 1)

3. 3

   Value-based

   −$3.2M

   each hospital’s base × (its factor − 1): bonuses and cuts

4. 4

   Everything else

   +$41.9B

   teaching (on Medicare Advantage stays too), low-income patients, outliers, sole community and low-volume payments, capital, uncompensated care

5. 5

   After the two

   $128.6B

   the payment the HAC cut is taken from

6. 6

   The HAC cut

   −$306M

   1% of the payment after the two, for 709 of the 718 hospitals CMS cut

Assumptions

Where the files leave a choice, this site makes one. Each choice and what it does to the estimate:

The national checks

Every build adds the estimate up nationally, checks the method against an earlier year’s cost reports, and holds the results to fixed ranges written into the build; the build stops if one fails. This build’s tape:

Checked against hospitals’ own reports

To learn how far its estimates can be trusted, this site checked its method against a year already on the record. It ran the same estimate on FY2024, with that year’s rates, the FY2022 claims volume CMS’s FY2024 impact file gives and CMS’s FY2024 decisions, and set each hospital’s figures beside its own Medicare cost report for that year.

A hospital’s cost report records the HAC cut on a line of its own, and the readmissions and value-based adjustments on two others.

The check takes only reports covering exactly the federal year, October through September, so each program’s line holds that one year’s adjustment: 498 hospitals, 134 of them cut by CMS. Of the 6,113 reports in CMS’s cost report file, it leaves out 5,249 covering another period, 358 without an inpatient payment worksheet, 2 from Puerto Rico, 3 for hospitals outside CMS’s FY2024 HACRP file, and 3 for hospitals the method can’t price.

Of the 134 cut hospitals, 3 can’t be compared, and sit outside every statistic of the HAC cut below: 1 whose cost report shows the HAC amount with the opposite sign and 2 whose cost report shows no HAC amount.

Each dot is a hospital CMS cut: its estimate over what its cost report recorded, so a dot right of the dotted line ran high. The band behind each group is its likely range, the middle 80% of the group’s misses; a group with fewer than 30 hospitals takes the range of a wider group, hatched.

Hatched: a group with too few hospitals of its own, taking a wider group’s range. Each group’s note: its hospitals, and how many of them fell inside the range it uses. SCH or MDH: a sole community or Medicare-dependent hospital. Hospitals’ names only in the tooltips.

Estimate · Source: this site’s check of its estimate, from CMS’s FY2024 IPPS final rule files and HACRP FY2024 (hacrp-fy2024), against hospitals’ Medicare cost reports in CMS’s HCRIS file for FY2024, retrieved October 23, 2026.

Show table

Group	Hospitals	Inside its range	Median, estimate ÷ cost report	Likely range, low end	Likely range, high end	Range from
Fewer than 100 beds, sole community or Medicare-dependent	26	21	1.15×	0.89×	1.4×	every hospital with fewer than 100 beds
Fewer than 100 beds, other hospitals	30	24	1.09×	0.8×	1.36×	its own
100 to 299 beds, sole community or Medicare-dependent	4	4	1.02×	0.82×	1.29×	every hospital with 100 to 299 beds
100 to 299 beds, other hospitals	40	32	1.03×	0.8×	1.31×	its own
300 beds or more, sole community or Medicare-dependent	3	2	1.09×	0.92×	1.23×	every hospital with 300 beds or more
300 beds or more, other hospitals	28	23	1×	0.92×	1.23×	every hospital with 300 beds or more

Each group’s note gives how many of its own hospitals fell inside the range it uses: from 67% to 100% of them, group by group. Those ranges were set on these same hospitals, so the count shows how well they fit, not how often another year’s figure will land inside.

The ranges aren’t centred on the estimates: each carries its group’s tilt, and the estimates are left as the method makes them. Of the groups with at least 10 hospitals, hospitals with fewer than 100 beds that are sole community or Medicare-dependent ran highest: by 15.3% at the median, across 26 hospitals.

Nationally, the cut hospitals’ cost reports added up to 99.9% of what the method estimated for them; resampling the hospitals puts the middle 80% of that between 97.8% and 101.9%. The program page’s national range is its estimate times those two: one year’s check, not a forecast.

A state’s or metro area’s range adds up its cut hospitals’ ranges, as if every estimate there erred the same way, so it is wider than the nation’s.

The same check for five of the figures the method makes, over the hospitals each applies to (the HAC cut over the hospitals CMS cut, the rest over all of them):

Its limits: it is one year’s check, and hospitals whose fiscal year doesn’t start in October go unchecked. And it is not an independent test of the teaching payments on Medicare Advantage stays: this site added them to the method after the check first ran, on these same hospitals, whose cost reports showed the cut taken on those payments too.

The infection baseline

An infection score compares a hospital’s infections with the number CDC’s model predicted, built on national data from 2015.

CDC is moving to a 2022 baseline, which CMS says reaches this program’s scores in fiscal year 2028. Care Compare is moving its infection results to the new baseline, and it shows one year at a time where this program uses two, so its numbers can differ from the ones here.

How to read a penalty

1. A penalty reflects how CMS’s formula ranked a hospital against others; it is not a verdict on the quality of its care.
2. The rule cuts about a quarter of hospitals in each year it runs, the ones with the highest scores, even if every hospital improves.
3. A hospital scored on few measures depends on those few. Scored on one, that one sets its whole score.
4. A CMS-assigned worst score means data were not submitted or not accepted. It is not a count of infections.
5. The patients counted differ by measure: PSI 90 counts traditional Medicare patients only; infections count patients of every insurer. The cut applies to Medicare’s own payments: traditional Medicare’s, and at teaching hospitals the teaching payments Medicare makes on Medicare Advantage stays.
6. The cut is 1% of the hospital’s Medicare payments for its inpatient stays, for one fiscal year.
7. Not included, among others: critical access, children’s, psychiatric, rehabilitation and long-term care hospitals, the cancer hospitals Medicare pays outside its inpatient system, VA hospitals and hospitals in U.S. territories. Maryland’s are scored but not cut.

Sources

Every input is a public file or an extract of one, pinned byte for byte: the build reads only these copies and what this site computes from them, and a test fails if any byte changes.

The code that builds these pages, the tests that check them and these pinned files are public in the site’s repository on GitHub (opens in a new tab).

What changed

Each change to how this site computes its numbers, and each correction, with its date.

1. November 2, 2026

   • The what-if’s ramp marks CMS’s line in earlier years scored by the same formula as faint ticks, and a line under it
   • The program page’s “Three programs” figure says whose dollars its totals add up (the hospitals this program scored
   • The states map’s keyboard note says what Enter does: it opens the state’s page, as a click does.
   • Methods links the current CMS page for the FY2024 hospital cost report file.
   • The program page’s script, which bundles parts of the d3 charting packages, carries their license notices.
   • The charts’ drawings are written in shorter path data, the same shapes: the program page is 18% lighter compressed

2. October 24, 2026

   • Every page links the family’s Terms of use, privacy, corrections, About and accessibility pages, and its feedback
   • Pages fit a phone’s width without scrolling sideways, the lightest gray text is a shade darker so it meets contrast

3. October 23, 2026

   • This site estimates what the HAC cut, the readmissions adjustment and the value-based adjustment come to in dollars,

4. October 22, 2026

   • Methods notes the one year whose count of hospitals cut differs from CMS’s results fact sheet (fiscal year 2016:
   • An accuracy pass: every sentence that states a fact about CMS’s program now cites its source in the claims register,
   • The Methods page takes CMS’s line as the 75th percentile of the scores of hospitals outside Maryland, as CMS’s FAQ
   • Each hospital’s record by year, back to the program’s first year, from CMS’s archived files, and the program’s history.
   • This Methods page, with the self-check drawn for every year.

5. October 21, 2026

   • The program page rebuilt, and a page for every region, division, state and metro area.

6. October 20, 2026

   • First build, from CMS’s newest results at the time.

Corrections go in the family’s public, dated corrections log; this site cannot change CMS’s values.

Glossary

Hospital-acquired condition (HAC)

A condition a patient develops in the hospital while being treated for something else.

Healthcare-associated infection (HAI)

An infection a patient develops as a result of medical care.

PSI 90

CMS’s patient safety and adverse events composite: one number built from patient-safety measures found in traditional Medicare claims. It is CMS’s version of the measure; AHRQ’s version shares its component definitions. Higher is worse.

NHSN

The CDC’s National Healthcare Safety Network, the system hospitals report the program’s infection data to.

Standardized infection ratio (SIR)

Infections a hospital reported divided by the number NHSN’s model predicted. Above 1.0 means more than predicted; below 1.0, fewer.

Predicted count

The number of infections NHSN’s model expects for a hospital, adjusted for factors that predict infections there. With fewer than 1.0 predicted, there is no SIR.

Standard deviation

A measure of how spread out hospitals’ results are.

Z-score

How far a hospital’s limited result is from the average for that measure, in standard deviations. Positive is worse than average; negative is better.

Floor and cap (winsorizing)

Before scoring, CMS limits each measure’s results to the range from the 5th to the 95th percentile; a result past either end counts as that end.

Total HAC Score

The average of a hospital’s z‑scores on the measures it was scored on, each weighted equally.

CMS’s line

The score above which CMS cuts payments: the 75th percentile of that year’s Total HAC Scores of hospitals outside Maryland. CMS’s file doesn’t give it, so the pages show it as the gap between the highest score CMS marked No and the lowest it marked Yes.

Payment reduction

A 1% cut to the Medicare payments for inpatient stays at a hospital above the line, for one fiscal year.

Subject to the reduction

A hospital outside Maryland, which CMS marks Yes (cut) or No (not cut), whether or not it was scored. Maryland’s hospitals are scored but marked N/A.

Payment decision (in counts)

On these pages, a hospital CMS scored and marked Yes or No. CMS also marks some hospitals it did not score, nearly all No; the counts leave them out, and each one’s page shows CMS’s decision as CMS wrote it.

Fiscal year (FY)

Medicare’s year, which starts in October of the year before and ends in September.

Performance period

The stretch of care a measure covers, usually two years. It moves forward a year each program year.

CMS-assigned score

When a hospital did not submit an infection measure, or its data failed CMS’s validation, CMS gives it the worst possible z‑score on that measure. It is not a measured result.

Maryland

Medicare pays Maryland’s hospitals under the state’s own arrangement with Medicare, and under an agreement between CMS and the state, CMS exempts them from the cut. They are scored but not cut.

CCN

CMS Certification Number, the ID CMS’s files use for each hospital.

National median

For each measure, the middle value among hospitals with a published value. This site computes it from CMS’s file.

Domain (earlier years)

Until fiscal year 2020, CMS grouped the measures in two domains, patient safety and infections, and weighted the domains instead of the measures.

Feedback, privacy and AI

The same rules for feedback, privacy and AI tools cover every Paid & Penalized site: see what the site and its host can see and how AI tools are used.

Reuse

This site’s text, charts and derived data may be reused with credit to Paid & Penalized (CC BY). CMS’s data are U.S. government works. Every hospital’s CMS values are in one CSV file; keep the CCN as text so leading zeros survive.

Accessibility

Every chart has a text description, and nearly every one has a table of its numbers behind “Show table.” The what-if and the sortable tables need JavaScript; without it, the values and the CSV stay available. Pages respect reduced-motion and forced-colors (high-contrast) settings.