01106T - exclusive or inclusive algorithm text

 

1106T  Oncology (breast), augmentative algorithmic analysis of histomorphologic features in digitized hematoxylin and eosin stained slides from formalin-fixed paraffin-embedded (FFPE) breast cancer tissue, algorithm-derived parameters reported as 5-year and 10-year risks of distant metastasis, including digitization of glass microscope slides and hematoxylin and eosin staining of tissue sections, when performed.

This CPT code is silent as to whether the algorithm EXCLUDES a test which uses both histomorphology and cliniccal features like "tumor size".   Does it imply only "histomorphologic" algorithms be coded here, or can algorithms with also clinical features be aloud on the groounds they are not excluded and merely silent.   PS _ i read the question as difficult and perhaps splitting experts.

###

CHAT GPT

I agree this is a difficult coding interpretation. It is exactly the sort of question that reasonable CPT experts could disagree about because the descriptor does not expressly answer it.

Here's how I would analyze it.

The descriptor

"Augmentative algorithmic analysis of histomorphologic features in digitized H&E slides..."

The object of the algorithm is stated to be histomorphologic features. It does not say:

• histomorphologic and clinical features,

• multivariable prediction model,

• clinicopathologic model,

• or "using all available patient information."

That tends to favor a narrow reading.

But it does not say "only"

On the other hand, CPT descriptors often describe the principal work without attempting to enumerate every internal computational step.

Suppose an algorithm:

• extracts hundreds of image-derived features from H&E,

• then combines them with

  • age,

  • tumor size,

  • nodal status,

  • ER status,

before generating the final risk score.

Nothing in the descriptor explicitly says this is prohibited.

Arguments for inclusion

Someone favoring use of 1106T would probably argue:

• The service remains an algorithmic analysis of histomorphologic features.

• Clinical variables merely calibrate or refine the prediction.

• The dominant source of information remains the digital pathology algorithm.

• CPT descriptors generally do not police internal mathematical architecture unless they specifically do so.

This is actually common in laboratory medicine. Many genomic algorithms incorporate age or sex without changing the CPT code.

Arguments against inclusion

The opposing argument is also strong.

One could say:

The descriptor intentionally identifies the analyte as histomorphologic features.

If the reported risk is materially generated from a clinicopathologic model rather than image analysis alone, then the reported result is no longer simply an algorithmic analysis of histomorphology.

The service has become a different category of test.

That argument gains force because the CPT Editorial Panel has recently become much more concerned about distinguishing:

• laboratory measurements,

• algorithmic analyses,

• software-only analyses,

• multicomponent predictive models.

Analogies

This resembles other CPT interpretation problems.

For example:

• A genomic assay does not stop being a genomic assay merely because age is incorporated into the report.

• Conversely, if an algorithm combines genomic data with imaging, laboratory values, and EHR variables into a single integrated prediction, many CPT experts would hesitate to call it solely a genomic test.

Where that line falls is often judgment rather than explicit CPT instruction.

My reading

If I were asked to interpret the descriptor without additional AMA guidance, I would probably adopt this principle:

The algorithm may incorporate limited ancillary clinical variables, provided the essential analytical work remains the extraction and interpretation of histomorphologic features from the digital slides.

However,

if the clinical variables materially contribute to the prediction such that the result is better characterized as a clinicopathologic prediction model than as histomorphologic analysis, then the fit with 1106T becomes questionable.

Confidence

I would characterize this as roughly:

• 40%: descriptor should be read narrowly—image-derived features only.

• 60%: limited clinical covariates are permissible if the algorithm remains fundamentally a histomorphology algorithm.

The real answer would likely depend less on the wording of the descriptor itself than on the CPT Editorial Panel's intent, which might be reflected in:

• the original Category III application,

• CPT Assistant commentary,

• Editorial Panel discussion minutes,

• or future coding guidance.

So I think your instinct is right: this is not a question with an obvious textual answer, and it is one that could legitimately split experienced CPT coding experts.

##

##

CLAUDE

#

#

Short answer: experts are likely to differ — but I'd put it 70/30 against, and the strongest evidence is sitting in the same tab.

The decisive point isn't abstract canon-of-construction reasoning; it's internal to the code family. The bladder codes released alongside 1106T read "augmentative algorithmic analysis of histomorphologic features in digitized slides from FFPE bladder cancer tissue (high-grade, non-muscle invasive), with clinicopathologic variables entered by the qualified health care professional." The Panel demonstrably had the vocabulary, understood the distinction, and deployed it where it applied. Its absence from 1106T is therefore not innocent silence — it is an omission made against a live, adjacent alternative that the drafters were staring at the same afternoon. That's about as clean an expressio unius setup as CPT ever gives you.

The second-order support is the placement of "when performed." The descriptor flags optionality exactly once, and it attaches to digitization and H&E staining — the workflow envelope — not to the analytic inputs. The drafters showed you where they meant to leave slack, and it isn't in the algorithm.

Third, note that "tumor size" almost certainly isn't a histomorphologic feature under the Panel's own working vocabulary. Grade, mitotic index, tubule formation, stromal TIL density — those are histomorphologic. Size and nodal status are precisely what the bladder descriptor calls clinicopathologic. The taxonomy is right there in the family.

Now the case for the permissive reading, which is not frivolous. The descriptor lacks any restrictive limiter — no "solely," no "alone," no "without." CPT descriptors are definitions, not exclusion lists, and elsewhere in the code set additional work not named in the descriptor doesn't disqualify the code; it's just uncompensated. A multimodal breast test genuinely does everything 1106T recites: digitizes the slide, runs an augmentative algorithm over histomorphologic features, and reports 5- and 10-year distant-metastasis risk. On that view the clinical variables are a refinement, not a different service. And the alternative — 88399 or 81599 unlisted — is commercially close to a death sentence, which supplies considerable motivated-reasoning pressure toward "not excluded, therefore allowed."

The rebuttal is that for algorithmic codes the algorithm is the service. In surgery, an extra maneuver is incidental to a described procedure; here, changing the input set changes the intellectual content of the reported output. That's the organizing logic of MAAA coding and Appendix O generally: test identity is defined by what goes in, not merely what comes out. Two tests reporting the same endpoint from different input sets are two services.

Where this actually gets resolved, though, is not textual. Under MolDX the Z-code/DEX registration does the discriminating work, and a contractor can perfectly well pay 1106T-plus-Z-code for a multimodal test without ever adjudicating the descriptor question — the code becomes a container and the registry becomes the definition. So "correct coding" and "what gets paid" can diverge here for years without anyone forcing the issue. Meanwhile the AMA's own answer, if you put it to CPT Network, would almost certainly be: the descriptor describes analysis of histomorphologic features; if your test requires entry of clinical variables, file a code change application — and here is the bladder code proving we'll write one for you.

Which is the real tell. The permissive reading isn't wrong because it's illogical; it's weak because the Panel has already shown, in the same release, exactly how it says the thing 1106T doesn't say.

WAIT

I agree with you, but I think the question is subtlely different. it is not whether the drafters had a habit of including those variables, when they were aware of them. The question is whether this code can be used with a est that includes tumor size as an (unstated0 input. Perhaps we could think of a lot of unstated inputs necessary to run a test or do a procedure.

##

You're right, and the reframing is the better question. Drafting habit tells you what the Panel would have written; it doesn't tell you whether an unnamed input is disqualifying. And you're correct that every code is silent about a great many things: 81519 doesn't mention reference genes, a colonoscopy code doesn't mention insufflation, 88360 doesn't mention that someone had to pick the field. Silence about inputs is the normal condition of CPT, not an anomaly demanding explanation.

So the real test isn't stated-vs-unstated. It's enabling versus constitutive — and I'd put three axes under that.

First, does varying it change the reported number? Hold the slide fixed and change the input. Scanner model, stain lot, nuclei count, tumor area annotation, QC thresholds — the answer doesn't move, or moves as noise. Tumor size in a Cox-type model moves the 10-year risk materially and monotonically. An input that is patient-specific, independently prognostic, not derivable from the slide, and additive in the model is a determinant of the answer, not a means of obtaining it. That's the category the descriptor is defining when it names "histomorphologic features" as the object of analysis.

Second, gating versus weighting. This is where a lot of the intuitive force of your point lives. A lab needs ER and HER2 status to know the patient is even eligible for the test; it needs specimen adequacy, fixation time, block selection. Those are unstated inputs genuinely necessary to run the test, and nobody thinks they belong in a descriptor. But they gate — they don't weight. They select who gets the analysis; they don't enter the computation. Tumor size as a model term weights.

Third — and this is the strongest argument for your permissive reading, better than anything I gave you last round — upstream input versus downstream calibration. The descriptor demands output "reported as 5-year and 10-year risks of distant metastasis." Not a score. Not a stratum. Absolute, time-indexed risk. You cannot get absolute risk out of morphology alone; you need a baseline hazard, and a baseline hazard implies a population anchoring that in practice usually means age, at minimum, and often nodal and endocrine-therapy context. The descriptor's own output specification arguably presupposes clinical inputs. If a test's algorithmic analysis is purely histomorphologic and clinical variables are applied only afterward to convert a morphologic score into an absolute risk, an expert could say — quite defensibly — that the algorithmic analysis of histomorphologic features is exactly what the code describes, and the calibration step is the unremarkable arithmetic of turning a hazard ratio into a percentage.

That's a real carve-out, and it's narrower than "tumor size is fine because nobody excluded it." A test that runs a multivariable model with size as a coefficient sits outside; a test whose morphologic classifier is calibrated to absolute risk using age plausibly sits inside.

So I'd revise my verdict this way: experts will still differ, but the disagreement should be about whether the clinical variable is constitutive of the risk estimate or ancillary to producing it — not about whether silence implies permission. Which also happens to be the fight that CPT Assistant or a code change application could actually settle, and the one a lab's coding memo should be arguing on the merits rather than on the absence of the word "solely."