My eye tagged a new article in NEJM about Congress granting coverage privilege to a class of interventions, like those which have FDA breakthrough status. That's Kadakia et al. 2026.
But I'll devote 3/4 of this blog to discussing their citation #2, Kang et al., 2025, on CMS NGS claim denials.
Along the way, I noticed several interesting recent titles which I'll highlight in fine print:
- Grossman et al. (January 7) discuss "FDA law used to threaten medical practice." While it's not news that the Administration opposes transsexal healthcare especially in teens, in areas like Medicaid payments, this article says that DOJ sent "more than 20 subpoenas" to physicians or clinics focusing on the doctor's "off label' use of drugs [hormones.]
- Grossman et al here.
- Harvard's Jerry Avorn has an article, "25th Anniversry of a Nearly Unknown Health Policy Turning Point," December 17, here.
- This was a Massachusetts law for (near) universl coverage, which he sees as a direct precedent to the Affordable Care Act (ACA) in 2010. For Avorn, Massachusetts events circa 2000 (25 years) led to "Romneycare' in 2006 (20 years) and ACA (15 years).
- Sklar and Richman discuss "Financing Telehealth," January 10, here. Lots of ins and outs.
Technology & Medicare Coverage: Kadakia 2026 and its link to Kang 2025
Kadakia et al. write, "Automatic Medicare Coverage for New Medical Technologies — Here We Go Again." January 10, here. They oppose two specific situations ,which they treat as examples of a larger paradigm.
First, Kadakia et al. discuss proposed legislation to give temporary coverage to approved FDA breakthrough devices. This proposal recreates a Trump I policy, MCIT, which gave coverage by regulation, and was replaced by a much weaker Biden policy (via individual NCDs) called TCET. It's been proposed a couple times and is not currently headed to any vote. The Trump CMS could do this by regulation.
Second, they discuss proposed legislation that allows CMS to write NCD's for multi cancer early detection (MCED) after FDA approval but before USPSTF approval. It does not create coverage, only the possibility of an NCD that creates coverage. This has actually been included in the current (late January) government financing bill; we'll see if it holds in the final legislation hopefully passing by January 31 (to avoid a shutdown). Read about MCED legislation at the American Cancer Society here.
##
The first legislation (about BT devices) would create coverage by law, but it isn't in play for any vote. The second legislation (about MCED) would only create the right to later write an NCD, but it IS up for a vote within days.
##
The Kadakia article in NEJM has an interesting Reference 2, which I hadn't seen when it came out in 2025.
- This is Kang, Odouard, Gresenz, open access, Claim Denials for Cancer-Related Next-Generation Sequencing in Medicare,
- JAMA Netw Open April 25, 2025, here.
They state that, under the 2017 NCD for NGS in cancer patients (NCD 90.2), the claims denial rate runs about 20%. I think they may not understand claims processing as well as some of us do (see below!), but it's still worth reading. They write, "The findings suggest the continued existence of uncertainty regarding the boundaries of coverage for NGS despite the implementation of the National Coverage Determination."
SCITE says Kang has been cited 3 times in 10 months (one being Kadakia). To me, there are some ins and outs - for example, before an FDA test under the NCD gets a PLA code, CMS variably instructed MACs to process early claims with 81479 versus 81455. That means, nearly zero tests would have been paid by this FDA CDx NCD for <50 gene FDA CDx tests using NGS - yet they make significant discussion of 81445 vs 81455 denials (<50 and >50 genes).
###
Do Kang et al. drop the ball? Experts will be startled to find their analysis is based on generic CPT codes 81445, 18450, 81455, and yet they assert the time period means they are learning about claims processing under the NGS NCD. This is flat wrong; Almost no claims under the NCD's coverage for FDA CDx tests were ever paid under 81445, 81450, or 81455. For several years, claims were paid under the special code for Foundation Medicine's F1 test, then add additional claims for Guardant's and FMI's liquid biopsy FDA CDx tests. Gradually there have been others - like Tempus and Caris lab-based NGS CDx tests (and not its from TFS Oncomine and ILMN Trusight), but those are outside the time range.
The NCD allowed local coverage for LDT tests as long as they were used once, in patients with advanced cancer. However, in 2018 era, I don't believe NGS MAC was covering code 81455 yet, and MolDx never used code 81455 (or almost never). So there are a lot of strangenesses that emerge from tracking only 81445, 81450, 81455, and calling this a good observation of what NCD 90.2 accomplished.
Here is an AI discussion of the Kang article.
###
Did Kang et al. Actually Study the 2018 NGS NCD?
Kang et al. frame their paper as an evaluation of how Medicare’s 2018 National Coverage Determination (NCD 90.2) affected claim denials for next-generation sequencing. That framing is plausible on first read. On closer inspection, however, the study design does not meaningfully observe the NCD’s core coverage pathway at all. Instead, it examines a different—and largely orthogonal—slice of Medicare molecular testing.
This is not a small technical quibble. It goes to the heart of what the paper can and cannot tell us.
What the 2018 NCD actually did
The main operative clause of the 2018 NCD granted national coverage for:
FDA-approved or FDA-cleared NGS companion diagnostic tests
Used on-label
Once per patient
In patients with advanced cancer (stage III/IV, metastatic, relapsed, refractory)
In practice, during the study window, this pathway overwhelmingly meant Foundation Medicine’s FoundationOne CDx, billed under PLA code 0037U, later joined by a small number of additional FDA-approved CDx assays (including liquid biopsy CDx from Foundation Medicine and Guardant, 0239U and 0242U).
Those claims are essentially absent from Kang et al.’s analytic universe.
The codes Kang et al. actually study
Instead, the authors restrict their analysis to three generic genomic sequencing procedure codes:
81445 (solid tumors, 5–50 genes)
81450 (hematologic malignancies, 5–50 genes)
81455 (>50 genes)
These codes were not the primary billing pathway for NCD-covered FDA CDx tests. With rare exceptions, FDA-approved CDx claims during this period were not paid under 81445/81450/81455, but under PLA codes or unique HCPCS codes tied to the approved assay.
As a result, Kang et al. are not observing payment or denial behavior for the NCD’s nationally covered tests. They are observing something else.
What they are actually observing
The NCD had a second, subordinate provision: it explicitly allowed Medicare Administrative Contractors (MACs) to determine coverage for non-FDA-approved NGS laboratory-developed tests, provided those tests were used:
once per patient, and
in advanced cancer.
Crucially, this provision did not itself grant coverage. It merely preserved local discretion.
Claims billed under 81445/81450/81455 therefore fall into one of three buckets:
LDTs paid (or denied) under local LCDs
LDTs paid under alternative codes (e.g., 81479 or PLA codes), which Kang et al. do not track
Claims that never realistically had a pathway to payment under the NCD framework
In other words, the study largely measures MAC-level LDT adjudication behavior, not the effect of national coverage for FDA CDx assays.
The MolDX problem the paper never confronts
This misalignment is especially stark in MolDX jurisdictions.
During much of the study period:
MolDX did not routinely cover 81455 for large tumor panels.
When MolDX did cover large panels, it often required 81479 or PLA codes, not 81455.
As a result, many of the largest, most clinically prominent panels never appear under 81455 claims at all.
Thus, when Kang et al. report that >50-gene panels have especially high denial rates, they may be observing a coding artifact as much as a coverage signal: the panels that successfully navigated MolDX policy often left the 81455 universe entirely.
What remains in the data are the stragglers—the tests and labs that either lacked alternative coding pathways or were operating outside favorable LCD structures.
Why the time-trend conclusions are especially fragile
The paper emphasizes rising denial rates after 2018 and again after 2020. But once we recognize that:
the NCD-covered CDx claims are mostly invisible, and
the composition of tests billed under 81445/81450/81455 was changing rapidly,
the time trends become ambiguous.
What looks like an “effect of the NCD” may instead reflect:
migration of successful assays to PLA codes,
increased scrutiny of residual LDT billing,
or shifting MAC enforcement patterns unrelated to the national policy.
The authors acknowledge the absence of clinical data such as cancer stage, but the absence of code-pathway fidelity is the deeper limitation—and it is not clearly acknowledged.
What the paper still does well
None of this means the paper is useless. It provides a solid descriptive account of:
denial behavior for generic NGS CPT codes,
strong site-of-service effects,
and something about the financial exposure associated with denied molecular claims - but they don't make clear enough this is only IF the patient has signed an "ABN" and the "physician who ordered" is never responsible for paying it.
Plus, any Medicare patient HAD first-dollar coverage for FMI Foundation One - there was no reason for him/her to get stuck with a $5000 bill for a denied local lDT.
The paper does not tell us whether the NCD “worked,” failed, or increased uncertainty for its primary target: FDA-approved CDx NGS testing.
The core interpretive error
The paper’s central analytic move—treating 81445/81450/81455 as a proxy for “NGS under the NCD”—is a category mistake.
It conflates:
national coverage of FDA CDx tests
withlocal adjudication of LDT-style NGS panels.
Those are different policy regimes, operating in parallel, with different coding pathways, evidentiary standards, and contractor behaviors.
Bottom line for expert readers
Kang et al. do not show that the 2018 NCD increased denial rates for NCD-covered NGS tests. They show that denials are common—and rising—within a residual, locally adjudicated CPT universe that the NCD largely sidestepped.
Read properly, the paper is not an evaluation of NCD 90.2’s success or failure. It is an inadvertent study of what happens to molecular testing that never truly entered the NCD’s protected lane.
That is still interesting—but it is a very different claim than the one implied by the paper’s framing.
Citations to Kang
Melly et al, 2025 Epilepsia Open
Elbhiry et al., 2025 Diagnostics
Kadakia et al., 2026 NEJM
