Correction and disclosure, 2026-09-18. Two figures in this article are wrong and the rest re-derive exactly. Taking the good news first, because it is the larger part: every rank correlation published below — all ten pairs, physician lines and lab lines alike — reproduces to the digit from the analysis that produced them, on the same vintage and the same population, using an estimator that only ever returns a rate that actually appears in the payer's file. So do the provider counts. The finding that a practice's position travels across the services a physician's office delivers, and stops at the lab line, stands as published.
Two things do not. First, the 0.96 this article quoted twice is not ours to quote any more. It comes from our earlier imaging article, which now carries its own correction: that figure covered the individually-priced providers within imaging, and it does not re-derive. We have removed the number from the two sentences below rather than replacing it, because the replacement would be measured on a differently-built cohort from the one this notice defines, and swapping one for the other is the mistake this correction exists to stop. What survives is the direction: imaging rank correlations are high, they are lower than the 0.96 we published, and the comparison against 0.83 is approximate. The imaging article's own notice carries the measured figures.
Second, the 8% is the wrong population. This article declares that everything after the method note describes the individually-priced group. The 8% of providers holding the same quartile across all five lines is the figure for every provider in the file. Within the individually-priced group it is 12.5%. The point the sentence is making — that the lab line is what pulls providers off an otherwise-steady position — survives either number, and is in fact slightly stronger on the correct one.
And the thing neither article disclosed at all: what puts a provider in the "individually-priced" group. The rule, as implemented: for each service code we take every provider's median rate for that code, rounded to the penny, using only the global component. We then take the five rate values that the most providers share on that code — those are the insurer's standard schedules for it. A provider is "on schedule" for a code if its rate is one of those five, and "individually priced" if it is not. A provider is counted as individually priced overall when at least half of the codes it appears on are individually priced. Codes a provider does not bill are not counted either way.
That rule is the same in both articles and it does not describe the same group in both, which is the part worth flagging to anyone comparing them. This article applies it across five codes in five different service lines; the imaging article applies it across ten imaging codes. Only one code is common to both. A practice that is individually priced relative to an imaging basket need not be individually priced relative to a basket spanning imaging, office visits, therapy, injections and labs — and it is not: the group is about 61% of providers here and about 65% there. The two articles' cohort figures should not be compared directly, and we should not have published them under one name without saying so.
Two things we've published about one insurer's public rate file keep pointing at a bigger question.
The first: in BlueCross BlueShield of Tennessee's in-network file for the Nashville area, a provider's imaging price barely moves from scan to scan. Rank every provider on a head CT and you've essentially ranked them on their brain MRI and their lumbar X-ray too — a rank correlation that stays high even among the majority of providers who don't sit on a standard fee schedule. Within imaging, a practice isn't ten prices. It's one position.
The second: the same file prices different kinds of care with completely different machines. A blood test carries a couple dozen prices and packs the overwhelming majority of providers onto a handful of them — an assigned commodity. A routine office visit carries hundreds of distinct prices — individual negotiation.
Put those together and a CFO's real question falls out. If you're expensive on imaging, are you expensive on office visits, therapy, and injections too? Is your position one fact about your practice — a single place you sit with this payer across everything — or is it set separately in every service line, so being high on scans tells you nothing about where you land on a blood draw?
We hadn't measured it. So we did. The answer is more useful than a simple yes or no: your position travels across the services your practice actually performs and negotiates — and stops cold at the ones it doesn't.
What we measured
We took the same BCBS-TN home-network file and picked one representative code in five service lines, using the global component only so no comparison mixes the professional and technical halves of a service:
- Imaging — head CT (70450)
- Office visit — established-patient level 4 (99214)
- Physical therapy — therapeutic exercise (97110)
- Injection — major joint injection (20610)
- Lab — complete blood count (85025)
For each code we ranked every provider from cheapest to most expensive, converted that to a percentile, and then — for the roughly 46,000 providers who appear in more than one line — asked: does a provider's percentile in one line predict its percentile in another? The measure is a Spearman rank correlation, computed per pair of service lines over the providers present in both. 1.0 means the two rankings are identical; 0 means one tells you nothing about the other.
One honesty note, the same one we made about imaging. Roughly 39% of providers sit on the insurer's standard fee schedules; if those schedules coordinate across lines, correlation within that group is partly arithmetic rather than a discovered fact. So we split the population and report the individually-priced providers (about 61%, ~29,600 per pair) separately — that's the number that means something. Everything below is that off-schedule group unless stated. And to be sure the pattern isn't an artifact of the method, we shuffled each ranking at random and re-ran: the average correlation under the null is 0.0002. The real numbers are not that.
The physician-delivered lines move together
For providers who each negotiated (or were handed) their own separate rates, the four services a physician's practice actually delivers rank almost the same way across all of them:
| Pair (individually-priced providers) | Rank correlation | Providers compared |
|---|---|---|
| Imaging × Physical therapy | 0.905 | ~29,600 |
| Physical therapy × Injection | 0.903 | ~29,600 |
| Office visit × Injection | 0.848 | ~29,500 |
| Imaging × Injection | 0.815 | ~29,600 |
| Office visit × Physical therapy | 0.802 | ~29,500 |
| Imaging × Office visit | 0.691 | ~29,500 |
These average about 0.83 — well below what you see inside imaging, but nowhere near zero, and light-years past the 0.0002 null. In plain terms: a practice that sits in the 90th percentile on its scans is very likely up near the top on its office visits, its therapy, and its injections too. Being expensive is largely a property of the practice, carried across everything a physician's office bills — not a decision the payer makes fresh for each service.
The file even shows the wiring underneath it. The single most expensive standard schedule for imaging (a head CT at $346.05) and the most expensive standard schedule for physical therapy (therapeutic exercise at $82.37) are held by the same providers — of the ~3,160 practices on the top imaging tier, 97.8% are the identical set on the top PT tier. The premium isn't negotiated line by line. For a large block of providers it's one tier assignment that lifts several lines at once.
Labs don't move with anything
Then there's the blood test, and it breaks the pattern completely:
| Pair with the lab (individually-priced providers) | Rank correlation |
|---|---|
| Imaging × Lab | 0.217 |
| Physical therapy × Lab | 0.129 |
| Injection × Lab | 0.004 |
| Office visit × Lab | −0.225 |
A provider's lab rank tells you essentially nothing about where it sits on anything a physician does — and against office visits it's even slightly inverted. The reason is the same thing that made labs interesting in the first place: this insurer prices a complete blood count as a pure commodity. 93.7% of all providers sit on just the five most common lab rates, across only about twenty distinct prices in the whole market. When almost everyone is paid one of a few numbers, there is barely a "position" to hold — and whatever sliver of variation exists doesn't track the practice's negotiated position on physician work at all. Labs aren't part of your rate position. They're a separate, assigned fact.
That's why, when you force all five lines together, only 12.5% of these providers hold the same quartile across every one of them — not because their physician rates jump around, but because the lab line is a different machine that pulls almost every provider off its otherwise-steady position.
The number is public. Your position isn't — and neither is which one you're reading.
Here's the practical shape of it. A single well-chosen imaging benchmark is a decent proxy for your entire physician-services position with this payer — scans, visits, therapy, injections all sit near the same percentile, so one number approximates four. But that same benchmark says nothing about your commodity lines, which are assigned on a separate table you didn't negotiate. Your contract with this payer isn't one position. It's a small number of them, set by different processes, and they don't all move together.
Which is the quiet catch in "price transparency." Anyone can now download BlueCross BlueShield of Tennessee's entire fee file — every rate, every provider, every line — and soon they'll be able to ask a chatbot for a single price and get one back. The raw number stopped being scarce. What's still scarce is the neutral read of where a specific practice actually stands across all of its lines at once, and a clear answer to the only question that pays: which of these positions were negotiated, which were simply assigned, and which are worth trying to move. A benchmark hands you the market's number. It doesn't tell you which of your positions you're looking at, and it doesn't take a side on what to do about it. That assessment is the work. The public file is only where it starts.