Correction in progress, 2026-09-10. This article was produced by a pipeline that reduced each provider to a single rate using a median which, on an even number of rates, returns the midpoint of the two middle values — a number that need not appear anywhere in the payer's file. We have now measured what that did to every figure here, on the same vintage and the same population, so that the estimator is the only variable.
The headline is confirmed and moves in our favour, by a hair. The average rank correlation across all ten imaging codes is 0.978 on observed rates, against the published 0.977. The share of providers who move more than one quartile between codes is 2.09%, against the published 2.08%. Rank correlations turn out to be nearly immune to this defect — unlike a distinct-price count or a dollar level — because reordering providers by a midpoint rather than by an observed rate barely reorders them at all. The finding that a provider's imaging rate is one position expressed ten times stands.
Two figures do not re-derive, and they are the two in the middle table. The separate correlations for providers on a standard schedule (published 0.992) and off one (published 0.958) reproduce under neither estimator: we now measure 0.932 and 0.945 respectively. The gap is not the estimator — the estimator moves these by less than 0.005 — it is the definition of the on/off-schedule split itself, which is sensitive in a way the article does not disclose. Do not rely on 0.992 or 0.958 until they are re-derived. The same sensitivity moves the cohort sizes: on observed rates about 35% of providers sit on a standard schedule and about 65% do not, rather than the 31% and 69% (31,689 providers) stated below.
What this does not change. The off-schedule quartile-crossing figure (3.04% published, 3.02% measured) holds, and with it the article's actual argument: the individually-priced majority hold their rank too, which is the part that is not arithmetic. The correlation among them is lower than we published and still high.
The body below still carries the original numbers. Where the body conflicts with this notice, this notice is correct.
Take one insurer's public transparency file, pick ten common imaging procedures, and rank every in-network provider from cheapest to most expensive on each one. You now have ten separate lists. The obvious assumption is that they'd be ten different lists — that a practice might negotiate a strong rate on MRI and a weak one on plain film, that its position would move around depending on the procedure.
It doesn't. In BlueCross BlueShield of Tennessee's in-network home-network file, we ranked 46,195 providers across ten imaging codes — CT, MRI, ultrasound, X-ray, and mammography — using the global component only, so every comparison is like-for-like. The ten rankings are almost the same ranking. Where a provider sits on a head CT predicts, with startling precision, where it sits on a brain MRI and on a lumbar-spine X-ray. A provider's imaging rate is not ten independent numbers. It is one position, and the ten prices are that position expressed ten times.
The rankings barely move
The clean way to say "these ten lists are the same list" is a rank correlation. Across all forty-five pairs of codes, the average Spearman correlation is 0.977 — where 1.0 would mean the rankings are identical and 0 would mean they're unrelated. Kendall's W, which measures agreement across all ten rankings at once, is 0.977 as well. In plain terms: 60.7% of providers never leave their quartile across all ten codes, and only 2.08% are what you'd call genuine rank-breakers — providers who move more than one quartile between one scan and another.
Some of that stability is unsurprising, and I want to be honest about which part. We showed in an earlier piece that this insurer assigns most providers onto a handful of proportional fee schedules — the top standard schedule pays exactly 4.23× the bottom one on nearly every code. If you're one of the roughly 31% of providers sitting on those standard schedules, your rank has to be stable across codes, because your rate is the same schedule multiplied by the same factor every time. That's arithmetic, not a finding.
The finding is what happens to everyone else.
Even the providers who price individually hold their position
About 69% of providers — 31,689 of them — do not sit on one of the dominant shared schedules. Their rates are bespoke: not one of the handful of penny-identical values that thousands of providers share, but their own number. If rank stability were purely a fee-schedule artifact, this is where it would fall apart. These providers were priced one at a time; there's no proportional table forcing them to hold position.
They hold position anyway. For this individually-priced two-thirds, taken alone:
| Group | Share of providers | Rank correlation across the basket | Providers crossing >1 quartile |
|---|---|---|---|
| On a standard schedule | ~31% | 0.992 | 0.00% |
| Priced individually (off-schedule) | ~69% | 0.958 | 3.04% |
| Everyone | 100% | 0.977 | 2.08% |
Read the middle row, because it's the real story. A correlation of 0.958 among providers who each negotiated (or were handed) their own separate rate means the thing that makes a practice expensive isn't procedure-specific. A practice that's in the 90th percentile on its ultrasound rate is almost certainly up near the 90th percentile on its MRI and its X-ray too. Being expensive is a property of the provider, not of the scan.
The examples make it concrete. The most expensive providers sit at the 97th-to-99th percentile — top quartile — on all ten codes at once, and they're off-schedule, individually priced. The cheapest sit at the half-percentile, bottom quartile, on all ten. And the single largest rank-breaker in the entire 46,000-provider population — the one provider whose position moves the most from code to code — still holds the same quartile on nine of the ten codes, breaking rank only on screening mammography, the one procedure federal law requires plans to cover at no cost to the patient. Even the exception mostly proves the rule.
Why this is good news, and why it's a trap
The good news is that it makes your position knowable cheaply. If your imaging rank barely moves across ten procedures, you don't need a ten-code audit to understand where you stand with this payer. One well-chosen benchmark — where do I sit on a single common scan — tells you, to a close approximation, where you sit across your entire imaging book with that insurer. You are one number, so you only have to find one number.
The trap is that the public file will not hand you that number. You can download every one of those 46,195 rates. What you cannot see in the file is which row is you, what percentile that puts you in, and — the question that actually matters — whether the position you're in is one you could move. The transparency rule turned the prices into a public commodity. It did nothing to tell any single provider where it stands in them.
And there's a boundary worth stating plainly, because it's the honest limit of what this data shows. This is one payer, one state, one file. Within BlueCross BlueShield of Tennessee, a provider's position is essentially one number. Whether that same provider holds the same position with a different insurer — whether being top-quartile here means top-quartile everywhere — is a question this file cannot answer, and it's exactly the question a practice most needs answered. The number being public doesn't make your position across your whole book legible. If anything, ten million public rows make it easier to assume you know where you stand and be wrong.
That gap is the whole point. The rate stopped being the scarce thing the day the file went public. What's scarce now is the neutral read of your position — across codes, and across payers — and the specific contract action it implies: which of your rates are worth reopening, which are already as good as the schedule allows, and which aren't yours to move at all. That's an answer, not a data pull. It sits on neither side of the table.
Method: BlueCross BlueShield of Tennessee in-network home-network machine-readable file (raw _890 schema, July 2026 vintage), all segments. Ten imaging CPT codes; global component only, one median rate per provider-code pair. 46,195 providers present across all ten codes. Rank agreement measured by pairwise Spearman correlation and Kendall's W; "off-schedule" = providers whose rate is not among a code's five most common shared values on a majority of their codes. All figures anonymized by rank and percentile — no provider names, NPIs, or tax IDs. Single payer, single state; not a claim about any other insurer's file.