How Comparison Works

Two hospital prices are comparable only if they describe the same service and payment context. This page lists the rules enforced and tested in the open-source data layer that supplies the site.

The comparison key

Prices are compared across hospitals only within an exact service context, which is the combination of all of these:

billing code system + code
care setting (inpatient / outpatient)
billing type (facility / professional)
billing modifier set
drug unit context (for drug codes)
price type (list / cash / negotiated)
payment methodology (for negotiated rates: fee schedule / case rate / per diem — a per-diem is a DAILY amount and never mixes with episode or item prices)

If any part differs, the rows land in different contexts and are never mixed. Every exact context has a stable address of its own (/compare/context/…), so links from hospital and insurer pages always land on the precise comparison they came from. Three consequences worth knowing:

The same code appears multiple times. A knee MRI billed in an outpatient facility context is a different row from the same code billed professionally. Service pages show all contexts side by side.
"Not labeled by hospital" is a real context. When a hospital omits the billing type, we keep those rows in their own "unspecified" bucket rather than guessing — they only ever compare against other unlabeled rows.
Modifier rows stay separate. Modifiers like professional-only (26) or technical-only (TC) change what is being priced; we never blend them with unmodified rows.

Which codes can compare at all

Standard code systems (MS-DRG, CPT, HCPCS, APC, …) mean the same thing at every hospital, so they can anchor comparison. Internal chargemaster codes and local codes cannot — they stay visible but blocked. Codes must also be specific enough to identify one service; broad catch-all codes are blocked from ranking.

A note on descriptions: MS-DRG descriptions are public domain and shown. CPT/CDT descriptions are licensed by the AMA and cannot be republished here — those services show "Description not available" with the code still fully comparable. This is a licensing constraint, not a hospital failure, and the site says so wherever it appears.

One hospital, one vote

Every market statistic is computed hierarchically so that repetition in a published file can never add statistical weight:

Rows repeating one amount under one insurer contract collapse to a single contract amount. (One hospital published a single per-diem rate against 56 revenue-code variants of one MS-DRG; it counts once.)
A hospital's representative price for a context is the median of its contract amounts.
Market medians and percentiles are computed over those hospital representatives — one per hospital.

A contract whose rows carry several different amounts for the exact same context hides a distinction the file does not label. We exclude it from statistics rather than average it, and service pages show how many hospitals were excluded this way.

The 3-hospital floor

No market statistic — median, percentile, spread, ranking, or "vs. market" position — is computed unless at least 3 hospitals have a safely representable price for the exact same service context.

With 2 hospitals, a median is only the halfway point between two prices, and a range simply reports those two values. Even at n=3, a 10th or 90th percentile is close to the minimum or maximum. Below the floor, we show the individual prices and the hospital count and label the context "Too few hospitals."

The floor applies separately to every peer group: the all-corpus market, the same-state group, the same-hospital-type group, the same-health-system group, and each insurer's cross-hospital market.

Insurer (payer) matching

Hospitals write insurer names free-form — "BCBS", "Blue Cross TN", "BCBST" can all be one insurer. The pipeline matches published names to canonical insurer identities using documented rules. Only matched rates enter insurer views; unmatched names are counted openly as the "Insurer name couldn't be identified" blocker. Matching coverage per hospital is published as the insurer-name matching score.

Currentness

Only each hospital's current file feeds comparisons. Every hospital page shows the file's own published date, its recency bucket, and a link to the source file. Superseded files are blocked with the "outdated file version" reason.

The 12 blocker reasons

Every exclusion from a stricter comparison is one of 12 named, stable reasons — never an undocumented filter. Ten apply per row; two apply at a wider grain (the 3-hospital floor per service context, and mixed-amount contracts per insurer contract). The full plain-language catalog, with counts for the current corpus, lives on the data quality page.

What we deliberately do not do

No fuzzy matching of service descriptions across hospitals — descriptions can look alike while covering different things.
No cross-unit drug conversions.
No mixing of price types, ever.
No mixing of payment methodologies, ever — and no converting a per-diem into an episode price.
No derived-dollar rankings.
No observation-weighted statistics — one hospital, one vote.
No claims beyond the loaded corpus.

These rules reduce coverage in exchange for more consistent comparisons. The data quality page reports their effect on the published rows.