Today, July 14, CMS held its annual experts advisory committee for the 100-odd codes it is pricing for CY2027.
In a previous blog, I showed how AI could quickly make a structured complete transcript of the day, including features like writing the code descriptor just before each discussion. Here.
In this blog I took the whole transcript of the day - 82 pages and 32,000 words - and I asked Chat GPT to analyze it via several questions.
- (AI can make mistakes so take these as directional or first-pass only. Also, some of the questions, two human raters might tally different numbers againts the 82 page transcript.)
Answer: I treated each FACA item as one code and used the later correction, resumed discussion, or revote as the controlling account when a code appeared twice. These are transcript-based judgments rather than an official CMS tally.
How many codes received an explicit initial Panel or Subcommittee recommendation for gap-fill?
27 codes. I counted only statements in which the Panel presenter or Subcommittee expressly recommended gap-fill. I did not count gap-fill merely because it appeared as a standard ballot option, or because the sponsor requested gap-fill while the Panel recommended a crosswalk.
The 27 were: 0609U, 0605U, 0634U, X251U, X283U, 0604U, 0614U, 0654U, 0655U, 0645U, 0646U, 0644U, X261U, X263U, X265U, X273U, X286U, X293U, 0607U, 0608U, 0601U, 0610U, 0629U, X269U, 0602U, 0649U, and 0631U.
This distinction matters in such examples as the recombinant-antigen assays, where the sponsors requested gap-fill but the Subcommittee recommended multiplied crosswalks, and X271U, where the laboratory favored gap-fill but the Panel’s starting position was a crosswalk.
How many codes had significant debate rather than a short presentation followed immediately by voting?
My best single number is 8 codes.
I counted a code as significantly debated when there was substantive back-and-forth, a new or corrected ballot alternative, direct stakeholder clarification, or a procedural question that had to be resolved. Those eight were:
8XX19, X271U, X302U, X291U, X254U, X280U, 0643U, and 0631U.
The largest discussions were clearly X271U, X254U, and the related X280U. The discussion of 0631U also became substantial after omitted sponsor alternatives were restored and the Panel examined whether a nominally “solid-tumor” test was actually a clonal-hematopoiesis assay. 8XX19 generated a shorter but real procedural debate over whether Tier 2 molecular-pathology codes could or should appear on the ballot.
A somewhat looser definition would produce 11, by also counting the correction-and-revote discussions for 0657U, 0658U, and 0659U. I excluded those because they were primarily corrections and explanations rather than genuine disagreement among the Panelists.
How often did the Panel receive direct verbal input from the laboratory owner or another stakeholder?
On a code-by-code basis, 6 codes received direct verbal stakeholder or laboratory-owner input:
X271U, X302U, X291U, X254U, X280U, and 0643U.
Those six code discussions occurred in roughly four conversational episodes, because several related Variantyx codes were discussed together:
The short-read/long-read whole-genome family: X271U, X302U, and X291U.
The initial correction and later resumed discussion of X254U and X280U.
The separate sponsor intervention on 0643U requesting a 1.3 multiplier for urine extraction and sequencing.
This is a strikingly small fraction of the 110 codes: direct live laboratory participation was exceptional, not routine. Most codes were decided entirely from the written submissions, prior Annual Laboratory Meeting presentations, and Subcommittee review.
What were the main categories of confusion or competing alternatives that caused discussion instead of immediate voting?
The live discussions fell into four broad categories.
First, errors or omissions in the meeting materials. Several ballot choices had been copied into the wrong rows, public recommendations were missing, or a descriptor and its proposed crosswalk did not match. This was especially prominent in the whole-genome group, where the Panel repeatedly referred to spreadsheet or “scroll-down copy-and-paste” errors.
Second, uncertainty about exactly what service the code represented. The Panel had to distinguish proband from comparator services, singleton from duo or trio analysis, prenatal from nonprenatal testing, rapid from ultrarapid testing, and short-read sequencing from combined short- and long-read sequencing. Until those distinctions were settled, the proposed crosswalks could not be evaluated reliably.
Third, disagreement over a single crosswalk versus a stacked crosswalk or multiplier. The recurring question was whether one existing panel code reasonably represented the service, or whether additional codes should be added for whole-genome methodology, skin-fibroblast culture, long-read confirmation, difficult genomic regions, or other components. The counterargument was that stacking these services might double-count sequencing, copy-number analysis, or work already inherent in the principal panel code. X254U and X280U were the clearest examples.
Fourth, procedural and transparency questions. The Panel debated whether every stakeholder suggestion must be displayed, whether Tier 2 codes could appear as alternatives, whether a previously priced code could still be reconsidered, and whether newly introduced recommendations required postponing the vote. The governing instinct was that even weak alternatives should be visible on the ballot when they had actually been submitted, while codes merely mentioned and rejected by the sponsor did not necessarily belong there.
Were there any humorous moments?
Yes, although the humor was mostly mild, spontaneous meeting humor rather than intentional jokes.
The voting software supplied several moments. When Dr. Hooker joined, staff observed that “all of a sudden we had an extra vote.” Later, after repeatedly waiting for twelve ballots, a staff member said, “Can’t count to 12 anymore.”
The genome-code corrections also became gently self-mocking. After the Chair accurately reconstructed complicated proband/comparator relationships, Dr. Hooker twice responded that she had “nailed it.” The Chair referred to the “unfortunate scroll-down copy-and-paste,” described an ultrarapid genome test as having “all the bells and whistles,” and the group repeatedly acknowledged that there were simply a great many nearly identical codes.
The overall tone was good-humored rather than comic: people were visibly tired and occasionally exasperated, but the jokes generally acknowledged polling glitches, spreadsheet errors, and the nearly impossible task of keeping dozens of similar genomic codes straight.
##
##
Overall assessment
Yes. I see one fairly strong internal inconsistency and several softer inconsistencies in how the Panel weighted different characteristics. I would not say the recommendations were simply irrational; rather, the Panel did not apply a visibly stable hierarchy among methodology, clinical purpose, specimen, analyte, output, and resource use.
Your friend’s recollection is directionally correct but not literally exact. I did not find the Panel saying that no digital-PCR or ddPCR codes existed anywhere in CPT. The wording was generally narrower: no existing code shared the same ddPCR methodology and resource profile, or the same HPV ddPCR methodology. Yet in another case the Panel readily crossed a digital-PCR assay to an NGS assay when the clinical service was otherwise nearly identical.
1. The clearest inconsistency: when does a digital-PCR methodology mismatch prevent crosswalking?
For 0634U, the 11-variant ESR1 breast-cancer assay, the Panel said there was no existing code sharing its ddPCR methodology and resource profile and therefore recommended gap-fill. It considered 0332U, but emphasized that it used qPCR and differed in specimen and clinical intent. The same reasoning was used for X251U and X283U, the blood-based HPV ddPCR assays: no existing code shared the HPV ddPCR methodology, and an NGS-based HPV comparator did not represent ddPCR resources.
The four methylation-sensitive digital-PCR codes—X261U, X263U, X265U, and X273U—were also gap-filled after the Panel rejected 0566U partly because quantitative PCR was considered fundamentally different from digital PCR.
But later, for X253U, the Panel recommended a crosswalk to 0493U. X253U quantifies donor-derived cfDNA using digital PCR, whereas 0493U is explicitly an NGS donor-derived cfDNA assay. The sponsor’s rationale was that the two tests had essentially the same patient population, plasma specimen, analyte, result, and clinical purpose, and that an earlier version of the test had already received the same crosswalk.
That is a real inconsistency in the decision rule:
In the HPV and ESR1 cases, a ddPCR-versus-NGS or ddPCR-versus-qPCR difference was treated as disqualifying.
In the transplant case, a digital-PCR-versus-NGS difference was overridden by clinical and descriptive similarity.
The outcomes might both be defensible, but the Panel never articulated why methodology should dominate in one family and yield to clinical equivalence in another.
2. Methodology was expressly tolerated for 0613U, though it defeated other crosswalks
For 0613U, a urine urothelial-cancer assay combining methylation-specific PCR and targeted NGS, the Panel crosswalked to 0420U because the assays served the same clinical population and shared biomarkers. The presenter specifically acknowledged that the comparator platforms used digital PCR or real-time PCR rather than the targeted NGS used by 0613U, calling 0613U a “methodological outlier,” but retained the crosswalk.
That sits awkwardly beside:
0634U, where ddPCR versus qPCR was an important reason not to crosswalk;
the four methylation-sensitive digital-PCR assays, where the digital-PCR versus quantitative-PCR distinction was also treated as decisive;
X251U and X283U, where an NGS HPV comparator was rejected as not reflecting ddPCR resources.
Again, this is not a perfect apples-to-apples contradiction. In 0613U, the clinical setting, specimen, overlapping biomarkers, and algorithmic output were unusually close. But it demonstrates that the Panel did not consistently answer the preliminary question: Is crosswalking primarily about the analytical platform, or primarily about the overall clinical laboratory service?
3. Disease and specimen mismatch mattered greatly for one assay and much less for another
For 0649U, the Alzheimer-disease prognostic assay, the Panel considered a Barrett’s-esophagus risk-score code because the apparent effort and number of genes were similar. It nevertheless recommended gap-fill because the diseases differed and one test used blood while the other used tissue.
But for 0628U, a saliva-based 449-gene renal-disease panel, the Panel recommended crosswalking to 81432, a hereditary breast-cancer panel, as the best match for methodology and resources.
Those decisions can be reconciled: 0628U is a very large germline panel, and 81432 may have been used as a broad laboratory-resource proxy, whereas 0649U is a small proprietary algorithmic assay whose value may not correlate well with a cancer-risk assay. Still, the rhetoric is inconsistent:
For 0649U, different disease and specimen defeated the resource analogy.
For 0628U, a profoundly different disease indication did not prevent use of a cancer panel as a resource proxy.
I would call this a soft but meaningful inconsistency.
4. Stacked crosswalks were accepted when components were viewed as additive, but resisted when they might overlap
For X282U, the pancreatic-cancer test combining a 5hmC-enriched genomic workflow with a CA 19-9 immunoassay, the Panel accepted a stacked crosswalk of 0410U plus 86301. Its reasoning was that the immunoassay was a separate additional laboratory component.
For X254U and X280U, however, the sponsor proposed combinations of a full-gene sequencing code, a hereditary-cancer panel code, and in one case a fibroblast-culture code. The Panel repeatedly worried that the sequencing and panel codes would count the same sequence-variant and copy-number work twice. It favored a single panel code or gap-fill instead.
This is probably not truly contradictory. The implied principle is sensible:
Stack codes for genuinely separate parallel procedures.
Do not stack multiple codes that are merely different proxies for the same underlying sequencing work.
The problem is that this principle was never stated clearly or applied through a standardized worksheet. Consequently, two stacked proposals can look arbitrarily different from the public’s perspective.
5. The meeting itself demonstrated that the code universe exceeded a manageable mental model
The whole-genome family produced actual mix-ups among fetal versus nonprenatal services, duo versus trio configurations, and the applicable proband and comparator codes. X291U had to be deferred, corrected, and revoted after the prenatal and nonprenatal ballot choices had apparently been mixed.
The resumed X254U/X280U discussion also documented missing sponsor recommendations, options apparently copied into the wrong rows, and the need to reconstruct the correct ballot late in the meeting.
That supports your broader point. The Panelists were not merely choosing among several familiar Category I codes. They were trying to remember hundreds of highly specific PLA descriptors, many of which differ by:
specimen;
one or two targets;
PCR versus digital PCR versus NGS;
baseline versus monitoring;
singleton, duo, or trio;
proband versus comparator;
rapid versus ultrarapid;
tissue-informed versus tumor-naïve;
one added algorithm or preanalytic step.
My conclusion
I would identify the ddPCR treatment as the strongest apparent inconsistency. The Panel effectively used two different standards:
Methodology mismatch means there is no adequate comparator.
and
Methodology mismatch is acceptable when the clinical service, analyte, specimen, and output are sufficiently close.
Both are legitimate approaches, but switching between them without an explicit weighting framework makes the process vulnerable to whoever happened to research or present a particular code.
The remedy would not merely be a better alphabetical list of PLA codes. CMS and the Panel need a searchable comparison matrix that gives every existing code structured fields for methodology, analyte, specimen, target count, clinical purpose, output, algorithm, preanalytic complexity, and major resource drivers. Each proposed crosswalk could then show which dimensions match, which do not, and why a mismatch that was fatal in one case is tolerable in another. That would make genuine distinctions visible—and expose actual contradictions before the vote.
