We make one calculation of our own: the national median for each measure, used for the "better than" and "below" labels. Everything else on this site is a CMS value reproduced unchanged. This page explains both, including where they fall short.
1. Where the data comes from
We import 8 datasets from the CMS provider-data API, which is the same source behind Medicare.gov Care Compare. We read the dataset URL from the CMS metadata registry on every run rather than hardcoding it, so we always take the file CMS currently considers authoritative.
| Dataset | CMS ID | Hospitals with data | CMS release |
|---|---|---|---|
| Hospital General Information | xubh-q36u |
5,419 | Aug 13, 2026 |
| Patient Survey (HCAHPS) | dgck-syfz |
3,949 | Aug 13, 2026 |
| Timely and Effective Care | yv7e-xc69 |
4,658 | Aug 13, 2026 |
| Complications and Deaths | ynj2-r877 |
4,790 | Sep 30, 2026 |
| Unplanned Hospital Visits | 632h-zaca |
4,790 | Sep 30, 2026 |
| Healthcare Associated Infections | 77hc-ibv8 |
4,790 | Aug 13, 2026 |
| Medicare Spending Per Beneficiary | rrqw-56er |
4,626 | Aug 13, 2026 |
| Outpatient Imaging Efficiency | wkfw-kthe |
4,626 | Aug 13, 2026 |
"Hospitals with data" is the number of hospitals that reported at least one measure in that dataset — always fewer than the 5,419 hospitals in the directory, because not every hospital reports every programme.
2. What we change
Three things, all presentational:
- Capitalisation. CMS stores names in upper case ("MAYO CLINIC HOSPITAL"). We convert to title case, preserving abbreviations that belong in capitals (LLC, VA, UCLA, ICU).
- Labels. We replace measure codes with plain-English names.
HAI_1_SIRbecomes "Bloodstream infections (CLABSI)". The underlying value is untouched. - Empty values. CMS uses several strings for missing data ("Not Available", "Not Applicable", "Number of cases too small"). We normalise them to a single "Not reported" state and show the footnote explaining why.
We do not round, rescale, re-weight, average across measures, or adjust any published value.
3. How the national comparison works
Each measure on a hospital profile carries a comparison label. It comes from one of two places, and we always say which.
When CMS has already judged it
For mortality, readmission and infection measures, CMS publishes its own conclusion: "Better than the National Benchmark", "No different than the National Benchmark", or "Worse than the National Benchmark". We use that verdict whenever it exists.
This matters because CMS's judgement accounts for statistical uncertainty. A hospital that treated 20 patients can land far from the national figure purely by chance, and CMS's confidence intervals absorb that. Our own comparison cannot.
When it does not
For the remaining measures — patient survey scores, timeliness, imaging — CMS publishes no verdict. There we compare against the national median, and label the result "compared with the national median" so you know it is ours, not CMS's.
We calculate it like this:
- Take every hospital that reported a numeric value for that measure in the current release.
- Require at least 30 reporting hospitals. Below that we publish no comparison at all, because a median from a handful of hospitals is not a national anything.
- Take the median, not the mean.
- Treat a difference of under 5% as "about the national median" rather than better or worse.
We currently publish 146 measure medians calculated this way.
Why median rather than average
CMS measures are skewed. Infection ratios cluster between 0.5 and 1.0 with a long tail above 3.0; ER wait times have a few extreme outliers. An average gets dragged toward the tail, after which more than half of all hospitals appear "better than average" — a statement that tells the reader nothing.
The median answers the question people are actually asking: how does this hospital compare with a typical one.
Why a 5% threshold
Without it, a hospital 0.3% from the median would be labelled "better" — manufacturing confidence where the data supports none. Real confidence intervals for these measures are usually far wider than 5%, so even this threshold is generous. Treat small differences as noise.
4. Which measures we show, and which we hide
A hospital profile could show every row CMS publishes. We filter, in three ways:
- Technical rows are hidden. Infection datasets include confidence limits, device-day counts and eligible-case counts for every measure — inputs to the ratio, not findings. Showing them turns six meaningful infection measures into thirty-six rows of noise.
- Patient survey rows are consolidated. Each HCAHPS question has three rows (always / usually / sometimes-or-never) that sum to 100%. We show the "always" figure, which is the threshold CMS itself uses in star ratings, and hide the other two.
- Reported measures sort first. Within a category, measures with data appear above those without. A profile that opens with five "Not reported" rows looks broken even when the data below is complete.
Nothing is hidden because it is unflattering. Every measure CMS publishes for a hospital, with a value, appears on that hospital's page.
5. Known limitations
Read this before acting on any number here
The data describes the past. CMS refreshes quarterly, and each measure covers a reporting period that closed months earlier — mortality measures often span three years. A hospital that replaced its leadership, opened a new unit or lost half its nursing staff last month looks exactly the same here.
Risk adjustment is incomplete. CMS adjusts for age, sex and documented conditions, but not for poverty, housing, whether a patient has help at home, or whether they can afford their prescriptions. Hospitals serving poorer communities tend to look worse on readmissions for reasons that begin outside the building.
Small hospitals are systematically under-measured. Low patient volumes mean suppressed results, which means fewer measures, which means no star rating. Roughly 41% of hospitals in this directory have no overall rating. That is a limitation of the measurement, not a finding about the care.
Medicare patients only. Most measures are built from Medicare claims, so they describe care for patients aged 65 and over. A children's hospital or a maternity service is largely invisible to this data.
Averages hide specialties. A hospital with an outstanding cardiac programme and a weak orthopaedic one shows up as average. If you know what care you need, read the specific measure and ignore the summary.
6. How often we update
We re-import when CMS publishes a new release, normally in January, April, July and October. Each page shows the CMS release date rather than our import date, because what matters is how current the underlying data is, not when our server last ran.
Current release: August 13, 2026. Imported by us on November 2, 2026.
7. Corrections
If a number here does not match Care Compare, that is our bug and we want to know — report it and we will fix it and say so.
If a number matches Care Compare but a hospital believes CMS calculated it wrongly, that correction has to happen at CMS. We publish what the federal record says.