Since January 1, 2026, hospitals that publish payer-specific negotiated charges as a percentage or an algorithm must also publish three percentile values per item or service: the 10th percentile, the median, and the 90th percentile of the allowed amounts they historically received, plus the count of remittances behind them. This page explains what 45 CFR 180.20 and 180.50 require, who must act and by when, and the one computational choice the regulation leaves open: which observed value a percentile formula lands on when the calculated position falls between two observations.
This guide is written for hospital revenue-cycle and compliance teams preparing machine-readable files, for payer and employer analysts who read those files, for health-data journalists checking a quoted percentile, and for developers building tools that parse or recompute hospital price transparency data. It quotes the governing clauses so every claim on the page can be checked against the primary text.
The Hospital Price Transparency rule, 45 CFR Part 180, requires every hospital to establish, update, and make public a list of all standard charges for all items and services online. Under 45 CFR 180.50(a)(1), that list lives in a machine-readable file on a publicly available website, in the format CMS specifies.
The percentile requirement is new. Under 45 CFR 180.50(b)(2)(ii)(C)( 2 ), where the payer-specific negotiated charge is based on a percentage or algorithm, the hospital must, beginning January 1, 2026, calculate and encode three dollar values for that item or service:
The 10th percentile of the total allowed amounts the hospital has historically received from a third party payer for the item or service.
The median of the same allowed-amount history, over the same lookback window.
The 90th percentile of the same history, over the same lookback window.
The total number of allowed-amount remittances used to calculate all three values, encoded in the same file.
The lookback window is fixed by the definitions at 45 CFR 180.20: a time period of no less than 12 months and no longer than 15 months prior to posting the machine-readable file. CMS's CSV technical guidance adds the operational details: the allowed amounts are computed from EDI 835 ERA transaction data or an equivalent remittance source, $0 remittances are excluded, and the count is encoded as 1 through 10 when fewer than eleven remittances exist, or as a whole number of eleven or more without a thousands separator (encode 13, never 2,025).
Around the percentiles sit the rest of the 180.50 requirements: the file must conform to a CMS template layout and data dictionary (180.50(c)(2)); it must be free to access, without an account, without submitting personal identifying information, and open to automated searches and direct downloads (180.50(d)(3)); and the standard charge information must be updated at least once annually (180.50(e)).
The duty falls on hospitals as 45 CFR 180.20 defines them: institutions licensed as hospitals under State or applicable local law. Each hospital location operating under a single hospital license that has a different set of standard charges than the other locations must separately make public the standard charges applicable to that location (180.50(a)(2)), which in practice means one machine-readable file set per differing location, each carrying its own allowed-amount percentiles.
The dates matter and they differ, so they are worth separating. The compliance date for the percentile and count requirements was January 1, 2026: from that date the encoding obligation applies. Enforcement of these new requirements, by contrast, began April 1, 2026. CMS's CY 2026 OPPS/ASC final rule fact sheet states the delay in one sentence: the 3-month enforcement delay applies solely to enforcement actions based on the new requirements at revised § 180.50. Older requirements, such as publishing a machine-readable file at all, have been enforced since their own earlier dates. Updates continue at least once per year under 180.50(e). This page reflects the text current on the eCFR as of September 2026.
“The effective date of the revisions at § 180.50, including removal of the estimated allowed amount, disclosure of the median, 10th, and 90th percentile allowed amounts and the count of allowed amounts, the attestation requirement, and the requirement that hospitals encode their organizational, or Type 2, NPIs in the MRFs will be January 1, 2026. However, CMS will delay enforcement of these finalized revisions until April 1, 2026.” CMS, CY 2026 OPPS/ASC Final Rule fact sheet: Hospital Price Transparency Policy Changes, retrieved 2026-09-25
Each of the three values is defined at 45 CFR 180.20 with the same structure: a percentile of the total allowed amounts historically received from a third party payer, over the 12-to-15-month lookback, plus a rule for a specific edge case. That edge-case rule is the tie rule, and it is quoted here from the 90th percentile definition:
“Should the calculated percentile fall between two observed allowed amounts, the 90th percentile allowed amount is the next highest observed value.” 45 CFR 180.20, definition of “Ninetieth (90th) percentile allowed amount”, eCFR current as of 2026-09-23
The median and 10th percentile definitions carry the same sentence with their own value substituted. So the regulation pins down what to do when a calculated percentile lands between two observations: take the next highest one. What it does not pin down is the position formula: how a percentile maps to a position in the sorted list in the first place. That gap is small in the text and large in practice, because two defensible readings of “the calculated percentile” produce different positions, and at the 90th percentile they produce different dollar values for most sample sizes a real file will contain.
Take a synthetic item or service with 13 allowed-amount remittances in the lookback window, sorted from lowest to highest (all inputs below are invented for illustration):
| Rank | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Allowed amount | $120 | $180 | $240 | $300 | $350 | $400 | $460 | $520 | $590 | $660 | $740 | $830 | $950 |
Both conventions below apply the tie rule the same way: land on a position, and if that position falls between two observations, take the next highest. They differ only in where they place the position. Reading A sets the position at p×n, so the 90th percentile sits at position 0.9×13 = 11.7, between the 11th value ($740) and the 12th ($830); the tie rule promotes it to the 12th value, $830. Reading B sets the position at p×(n+1), so it sits at 0.9×14 = 12.6, between the 12th ($830) and the 13th ($950); the tie rule promotes it to the 13th value, $950.
| Value | Position formula | Rank chosen | Result |
|---|---|---|---|
| 10th percentile | A: floor(0.10×13)+1 = 2 | B: ceil(0.10×14) = 2 | Both readings: $180 |
| Median | A: floor(0.50×13)+1 = 7 | B: ceil(0.50×14) = 7 | Both readings: $460 |
| 90th percentile | A: floor(0.90×13)+1 = 12 | B: ceil(0.90×14) = 13 | A: $830 · B: $950 |
Same remittances, same tie rule, same lookback: the 90th percentile differs by $120 because of the position formula alone. The count element for this example encodes 13, a whole number above ten with no thousands separator. Whoever republishes a file's percentile should therefore say which position convention produced it; the number alone is underspecified.
Reading A places the percentile at position p×n of the sorted list and, for a non-integer position, applies the tie rule to move up to the next highest observation; as a rank that is floor(p×n)+1. Reading B places it at position p×(n+1), which as a rank is ceil(p×(n+1)). Both are standard percentile definitions in general statistics, and both are compatible with the wording of 45 CFR 180.20. 45 CFR 180.20 defines the tie rule, not the position formula, so the rule text alone cannot settle which reading a hospital applied.
Where the two readings agree matters as much as where they split. At the 10th percentile and the median they choose the same rank for every sample size n up to 50: the n = 12, p = 0.10 case lands on the 2nd value under both, and the median of 80 observations lands on the 41st under both, which matches CMS's own gloss that an even-count median is the next highest of the middle two values. The 90th percentile is where they separate, at 32 of the 50 sample sizes up to 50. The complete divergence map follows.
| Percentile p | Sample size n | Reading A rank (floor(p×n)+1) | Reading B rank (ceil(p×(n+1))) |
|---|---|---|---|
| n = 11–18 | |||
| 0.9 | 11 | 10th of 11 | 11th of 11 |
| 0.9 | 12 | 11th of 12 | 12th of 12 |
| 0.9 | 13 | 12th of 13 | 13th of 13 |
| 0.9 | 14 | 13th of 14 | 14th of 14 |
| 0.9 | 15 | 14th of 15 | 15th of 15 |
| 0.9 | 16 | 15th of 16 | 16th of 16 |
| 0.9 | 17 | 16th of 17 | 17th of 17 |
| 0.9 | 18 | 17th of 18 | 18th of 18 |
| n = 21–28 | |||
| 0.9 | 21 | 19th of 21 | 20th of 21 |
| 0.9 | 22 | 20th of 22 | 21st of 22 |
| 0.9 | 23 | 21st of 23 | 22nd of 23 |
| 0.9 | 24 | 22nd of 24 | 23rd of 24 |
| 0.9 | 25 | 23rd of 25 | 24th of 25 |
| 0.9 | 26 | 24th of 26 | 25th of 26 |
| 0.9 | 27 | 25th of 27 | 26th of 27 |
| 0.9 | 28 | 26th of 28 | 27th of 28 |
| n = 31–38 | |||
| 0.9 | 31 | 28th of 31 | 29th of 31 |
| 0.9 | 32 | 29th of 32 | 30th of 32 |
| 0.9 | 33 | 30th of 33 | 31st of 33 |
| 0.9 | 34 | 31st of 34 | 32nd of 34 |
| 0.9 | 35 | 32nd of 35 | 33rd of 35 |
| 0.9 | 36 | 33rd of 36 | 34th of 36 |
| 0.9 | 37 | 34th of 37 | 35th of 37 |
| 0.9 | 38 | 35th of 38 | 36th of 38 |
| n = 41–48 | |||
| 0.9 | 41 | 37th of 41 | 38th of 41 |
| 0.9 | 42 | 38th of 42 | 39th of 42 |
| 0.9 | 43 | 39th of 43 | 40th of 43 |
| 0.9 | 44 | 40th of 44 | 41st of 44 |
| 0.9 | 45 | 41st of 45 | 42nd of 45 |
| 0.9 | 46 | 42nd of 46 | 43rd of 46 |
| 0.9 | 47 | 43rd of 47 | 44th of 47 |
| 0.9 | 48 | 44th of 48 | 45th of 48 |
The 32 divergent (p, n) pairs for sample sizes up to 50, computed with integer arithmetic (percent-scaled, no floating-point noise), each reading capped to the valid rank range 1..n. Every divergence sits at p = 0.9, and in every divergent pair reading B sits exactly one rank above reading A. Between the shown ranges (n = 19–20, 29–30, 39–40, 49–50 and all n ≤ 10) the two readings agree.
Practical upshot for anyone publishing or auditing these values: record the convention next to the number. A file that says “90th percentile $950” and one that says “90th percentile $830 (position p×n)” carry different information, and only the second lets a reader recompute the figure. Until CMS pins the position formula in guidance or a future rule amendment, treat the convention as part of the data's provenance.
The claims-side data pipeline behind these calculations is covered by tools already live on this site. All of them run in your browser, and the same EDI 835 remittance records that feed the allowed-amount percentiles are the records these tools lint, price, and dispute-check.
Validate X12 837 claim and 835 remittance record structures in the browser. The 835 ERA transaction is the remittance source CMS names for computing the allowed-amount percentiles, so well-formed 835 data is the precondition for a defensible percentile.
Open Tool →Model the economics of straight-through claim processing: clean-claim rates, per-touch handling costs, and the savings curve as automated claim decisions rise. Useful context when allowed-amount data quality determines how much of the pipeline can run untouched.
Open Tool →Assemble a structured dispute bundle from claim and remittance evidence. Where a published allowed amount disagrees with the remittances behind it, the discrepancy surfaces here first.
Open Tool →For the wider insurance and payer context, see the Insurance & InsurTech Hub, or browse the full catalogue from the AINumbers.co index.
Every regulatory claim on this page traces to one of the following primary sources. Retrieval dates are stated so the currency of each citation is checkable.
| Source | Cited for | Currency |
|---|---|---|
| 45 CFR 180.20 (eCFR) | Definitions of median, 10th percentile, and 90th percentile allowed amounts; 12-to-15-month lookback; the next-highest-observed-value tie rule | eCFR issue 2026-09-23, retrieved 2026-09-25; last amended by 90 FR 54087, Nov. 25, 2025 |
| 45 CFR 180.50 (eCFR) | Encoding obligation at (b)(2)(ii)(C)( 2 ); location-level publication at (a)(2); CMS template format at (c)(2); accessibility at (d); annual updates at (e) | eCFR issue 2026-09-23, retrieved 2026-09-25; last amended by 90 FR 54087, Nov. 25, 2025 |
| CMS CY 2026 OPPS/ASC Final Rule fact sheet | January 1, 2026 compliance date and the April 1, 2026 enforcement start for the new § 180.50 requirements | Retrieved 2026-09-25 |
| CMS Hospital Price Transparency CSV technical guidance | EDI 835 ERA remittance source; $0 remittance exclusion; count encoding buckets and the rejected “2,025” example; even-count tie-rule gloss | CMSgov master commit 5333564a, retrieved 2026-09-25 |
This page is an explainer, prepared to publication standard as of September 2026. It is a summary for practitioners and does not constitute legal advice; verify current obligations against the eCFR and CMS guidance cited above.
v1.0 · September 2026 · Guide · CMS Hospital Price Transparency · 45 CFR Part 180