How we grade

What the grade measures

The CostGrade compares what a hospital charges against two things: what that same care is actually paid for, and what other hospitals charge for the same coded work.

It is not a quality rating, and it is not a claim that a hospital overcharged anyone. Medicare pays a formula rate that ignores the charge entirely, so for a Medicare patient a high charge changes nothing. The reason charges matter is narrower and real: the list price is where an uninsured bill, an out-of-network bill and most collection demands begin.

The grade is relative, on purpose

There is no absolute standard for what a hospital's charge-to-payment ratio ought to be. No law sets one, no regulator publishes one, and the national median moves every year CMS republishes. A threshold presented as absolute would be a number we made up.

So each component is scored as a percentile against the live population — every other U.S. hospital in the same federal files — and the ladders are rebuilt from scratch on every run. An A means a hospital is near the best end of the country, not that it passed a bar.

What we deliberately do not score

How many procedures a hospital publishes

It is tempting to read a long price list as transparency and a short one as hiding something. It is neither. These files report what a hospital actually billed Medicare, so the length of the list tracks how big the hospital is and what it treats. Scoring it would grade rural and specialty hospitals down for their size.

The CMS quality star rating

It is shown on every hospital page because readers want it, and left out of the score because this engine is about price. We also checked the direction before deciding: across every U.S. hospital where both are published, average markup runs 5.45x at one-star hospitals and 5.07x at five-star hospitals. A high markup is not a sign of better care, and a grade that quietly rewarded expensive charging would be worse than no grade.

Ownership

Ownership is not an input to the model — the engine reads prices and nothing else. It is worth stating because the output correlates with it hard: for-profit hospitals average 8.23x, private non-profits 4.74x, and locally-run government hospitals 3.77x. That is a finding in the data, not a thumb on the scale.

Rural status and hospital size

Neither is an input, and both were checked for fairness rather than assumed.

Rural hospitals average 3.96x against 5.80x in metropolitan areas, so the grade does not penalise them — if anything the data runs the other way.

Small hospitals also collect more A grades than large ones (23% against 8%), which looks like the grade rewarding thin data. It is not. Inside metropolitan areas alone the effect largely disappears (12% against 8%), and the small-hospital markup there is actually higher than the large-hospital one (6.36x against 5.76x). What the raw comparison was picking up is composition: 47% of the smallest hospitals are rural, against 9% of the largest. We have deliberately not "corrected" for size, because doing so would correct away a real difference in how rural hospitals price.

Missing data, and hospitals we refuse to grade

A component with no published figure scores exactly half its weight. Because every component is a percentile, the median hospital earns 50% of it by construction — so half marks are the only value that leaves a hospital with no outpatient department neither better nor worse off than one with a perfectly ordinary outpatient department. Anything above 50% would be a bonus for not publishing a figure, which is the last incentive this site should create.

A hospital with fewer than 10 priced procedures, or with no published payment figures at all, is left ungraded rather than given a flattering default. Ten is not an arbitrary floor: CMS suppresses any row covering fewer than eleven patients, so ten published prices means the grade rests on at least a hundred cases. At a threshold of five, the graded population included a hospital at 100x — a number that is one strange case, not a pricing policy. 2,891 hospitals are in that position. Their pages stay reachable and stay honest about it, and they are excluded from the sitemap — publishing a few hundred pages that carry a name and nothing else teaches search engines the site is thin.

When it updates

CMS republishes these files annually. The importer resolves the newest release from the CMS catalogue rather than a fixed URL, so a new data year is picked up rather than silently missed, and every percentile ladder is recomputed against the new population.