clinicians.build · interactive · 17 sep 2026
One Lab, One Code
Memorial Sloan Kettering just licensed OncoKB — the annotations its curators have hand-maintained for a decade — into OpenEvidence, for every clinician outside MSK. The interpretation layer is now everywhere. Here is the layer it interprets: 330 molecular diagnostic codes in Medicare Part B, $2.80 billion, and 117 of them billed by exactly one entity in the entire country.
Primary source: “Memorial Sloan Kettering Cancer Center and OpenEvidence Partner to Advance Precision Oncology at the Point of Care,” Newswise, 16 September 2026
Data: CMS Medicare Physician & Other Practitioners — by Provider and Service, CY 2024 vintage, via MIMI Labs
OncoKB is not a model and it is not patient data. It is a catalog of what each genomic alteration means — oncogenic effect, level of evidence, the therapies attached — written down by people, dated, revised when the evidence moves. That catalog is the asset that changed hands this week.
But an annotation needs something to annotate. Someone has to run the test. So the useful question about a knowledge base going national is: how national is the thing it reads?
The median molecular diagnostic code in Medicare has three billing entities in the entire United States.
Every molecular diagnostic code in Medicare Part B, 2024
Each dot is one HCPCS code. Horizontal position is how many distinct entities billed it anywhere in the country; vertical position is total Medicare allowed dollars; area is the number of tests. Both axes are logarithmic, because they have to be. Set the concentration dial and the codes at or below it turn red.
Billing entities vs. Medicare dollars — every molecular diagnostic code, CY 2024
at or below the dial
above it
area ∝ tests · hover any dot
117 codes were billed by exactly one entity in the whole country in 2024, and they carry $1.02 billion — 36.4% of all molecular diagnostic spending in Part B.
The ten largest codes at the current setting
| Code | Description | Entities | Tests | Allowed |
|---|
The dollar in five that has no description
Drag the dial to 40 and one dot never turns red: 81479, at 76 billing entities. Its full CMS description is “Molecular pathology procedure.” That is the entire description. It is the single largest molecular line in Medicare Part B — 232,621 tests, $572.1 million, one dollar in five of everything on this chart — and the claim does not say what was measured.
80/20 — the part that transfers
A knowledge base is only as reachable as its substrate. Before you build on top of any clinical data layer, ask who produces it and how many of them there are. If the answer is one vendor, you have not built a product on a data source — you have built a feature on someone else’s contract, and their renewal is your roadmap.
Where this is thin
- The billing entity is not the orderer. These counts are rendering NPIs — the laboratory that billed, not the oncologist who ordered. The chart shows who supplies molecular testing, not who uses it.
- Suppression hides the small. CMS drops any provider/code cell with fewer than 11 beneficiaries. Small labs are not zero here; they are absent entirely, with no flag. Low-volume codes are undercounted, and a code shown at “1 entity” may have others below the threshold. Read the dial as a floor on concentration for the big codes and as unreliable at the bottom of the chart — which is exactly what the “hide below” slider is for. Drag it up and watch how much of the pattern survives.
- Medicare only. Part B fee-for-service. Medicare Advantage, commercial and Medicaid volume is not here, and molecular testing skews younger in some indications than the Medicare population.
- Categories are keyword guesses. The chips group codes by matching words in the CMS long description — “tumor,” “hla,” “cytochrome.” It is a reading aid, not a CMS taxonomy, and it will miscategorise at the edges. The underlying dots do not move when you change it.
- One entity is not always a monopoly. Several of these are proprietary assays with a dedicated PLA code — the code is the product, so a single biller is the design, not a market failure. That is the honest reading. It is also exactly why the interpretation layer matters: when a test has one source, the annotation on top of it is the only part anyone can compete on.