Hospital price transparency files: how to actually use one

Where the file actually is

There is a standard address, and it is the single most useful thing in this article: https://<the hospital’s domain>/cms-hpt.txt

That plain text file is not the prices. It is the pointer to them. Each entry names a hospital location, the page a human should read, and the direct URL of the machine-readable file itself. CMS provides hospitals a generator for it, and its stated purpose is to support automated searches for the price files, which is exactly what you want when the price page is buried four clicks deep under a menu nobody would guess.

Discovery is uneven in practice. We checked these six hospitals on 3 September 2026:

Scroll the table sideways on a narrow screen.

None of that is evidence that any of these hospitals is out of compliance, and we are not claiming it is. It is evidence that “the file is published” and “you can find the file” are different statements, and the gap between them is where most people give up.

One more thing worth knowing: the price file itself has a naming convention, so once you are looking at a URL you can usually tell what you have. Covenant Health Hobbs’s file is named for its employer identification number, the hospital name and the words standardcharges, ending in .json.

What is in the file, and what changed in 2026

The machine-readable file has to contain the hospital’s gross charges, its discounted cash price, the payer-specific negotiated charges, and the de-identified minimum and maximum negotiated charges, coded with whatever billing codes the hospital uses along with the code type: CPT, HCPCS, DRG, NDC or revenue center code, per 45 CFR 180.50. The hospital must update it at least once annually, and it must be reachable free of charge, without an account, without a password, and without submitting any personal information.

As of 1 January 2026 the file also carries what payers actually allowed, not just what the hospital charges. CMS removed the old estimated allowed amount and replaced it with the median allowed amount, the 10th percentile allowed amount and the 90th percentile allowed amount, in dollars. The regulation also requires the count of allowed amount remittances behind each of those figures, and the hospital’s organizational Type 2 NPIs. Those additions entered the rule in the CY2026 outpatient payment final rule, published at 90 FR 54087 on 25 November 2025.

CMS’s own guidance for hospitals goes further than the rule text on method. It tells hospitals to derive the allowed amount figures from EDI 835 electronic remittance advice, or an equivalent source of remittance data, over a lookback period of no less than 12 to 15 months, and it states that enforcement of the new and updated requirements began on 1 April 2026.

That change matters more than it sounds. A chargemaster number tells you what a hospital asks. An allowed amount tells you what somebody actually paid. For anyone arguing about what a charge should have been, the second number is worth far more than the first, and until this year it was not there.

Why six hospitals and not one

One hospital’s price is a data point, not a benchmark.

The six were chosen for comparability, which in this context means three things at once: roughly the same size, a similar service mix, and the same region. A rural 25-bed facility and an urban trauma center are not each other’s comparables, and neither is a hospital four states away. Get the comparison set wrong and every number computed from it is wrong in a way that is very easy for the other side to point at.

Six is also the practical floor for the statistic we needed. Computing a percentile off three numbers is not really computing a percentile. The more comparable hospitals you can find, the more the result behaves like a distribution instead of an accident.

What percentile, and why the file does not hand it to you

Clients ask for the 80th percentile. That is convention in this work rather than a rule we derived, and it is worth saying so plainly rather than dressing it up as methodology.

Here is the wrinkle. CMS now requires hospitals to publish the median, the 10th percentile and the 90th percentile of their allowed amounts. It does not require the 80th. So the file gives you three points on a curve and not the point you were asked for, which means you still assemble the comparable set yourself and compute across hospitals rather than reading a number out of one file.

That is the honest state of it. Price transparency data did not turn this into a lookup. It turned an unanswerable question into an answerable one that still takes work.

The hard part is deciding that two lines are the same thing

Finding the files is tedious. Matching across them is the actual difficulty, and it is where the judgment lives.

Hospitals describe the same procedure differently. The code may match while the description does not, or the description matches while the code sits at a different level of specificity, or one hospital bundles what another itemizes. So a match is not a string comparison. We work it from several directions at once: the revenue code, the billing code and its type, the text of the description, and the medical record itself, which says what was actually done to the patient.

When those agree, the lines are comparable. When they disagree, somebody has to decide, and that decision belongs in front of a person rather than inside an automated match. The same rule holds here as everywhere else in this work: a wrong comparable is worse than a missing one, because a wrong comparable produces a confident number with nothing visibly wrong about it. That principle, and what it costs to hold to it, is the subject of why no dollar figure in our output is produced by a model.

Hospital, facility and physician are three different prices

This is what outsiders get wrong, and it is the fastest way to produce a number that falls apart.

One episode of care can generate a hospital charge, a facility charge and a professional charge from the physician, and those are priced on different schedules by different rules. A hospital price transparency file speaks to the first two. It is not a source for what a physician should have been paid. Mixing them, or benchmarking a professional charge against a hospital’s published rate, produces an apples-to-oranges figure that an opposing expert will find immediately.

Keeping those three separate is unglamorous and it is most of what makes the output survive scrutiny.

What this approach does not give you

Not every hospital is covered. 45 CFR 180.30 deems federally owned facilities compliant, including Department of Veterans Affairs facilities and Department of Defense military treatment facilities, along with hospitals operated by an Indian Health Program and state forensic hospitals treating only people in the custody of penal authorities. If your comparable set would naturally include one of those, it will not be there.

Beyond the exemptions, the files vary in quality. Discovery is inconsistent, as our own check of six hospitals showed. Formats differ. Descriptions are written for the hospital’s internal use rather than for your comparison. CMS says it identifies hospitals to review using a web scraper and public complaints and performs at least 200 comprehensive reviews a month, which tells you both that enforcement is real and that it is sampling rather than universal.

None of that makes the data unusable. It makes it work rather than a lookup, which is the correct expectation to set before starting.