Can AI Write Deep Strategic Business Reports? We Compare Claude and ChatGPT on "Digital Pathology"

 Can AI Write Deep Strategic Business Reports?

More and more, I am seeing complex reports automatically generated by Claude Opus 4.7 and by Chat GPT, especially in its "deep research" mode.  (I have the $20 subscriptions to each).

I gave both AI's a prompt to do a business analysis of two competitors in the digital pathology space (I chose Philips and Roche).  Claude Opus took about 15 minutes to generate a 13-page report.   Chat GPT in "deep research" mode took about a half hour for a ten-page report.

Here's the take-home lesson, per Chat GPT:

• Two LLMs were given the same business-research prompt: compare Philips and Roche/Ventana in the U.S. digital pathology market. 
• The results were both useful, but revealingly different. 
• Claude Opus produced the more vivid strategy memo, framing Philips as a “platform-of-platforms” and Roche as an integrated diagnostic “system of record.” 
• ChatGPT Deep Research produced the more cautious diligence brief, emphasizing FDA status, RUO caveats, U.S. commercialization limits, and reimbursement realities. 
• In short: Claude gave the sharper narrative; ChatGPT gave the safer client memo.

Find the full 27-page PDF, which opens with a two-page comparison, and then reproduces each report in full.

https://drive.google.com/file/d/1XOSImqnlKnoZgEnVSw4KZizv_axCmGr7/view?usp=sharing

How MACs Price Major Services That Lack Fixed RVU's (Case Study: PET CT)

Header: AMA CPT has created some new Category III codes for whole-slide imaging proprietary tests.  However, as of May 26, CMS has NOT added these clinical laboratory codes to the summer CLFS pricing process.   What happens to Category III codes in terms of MAC pricing?

_____

CMS assigns RVU-based national prices to nearly all AMA CPT codes in Category I (aside from clinical laboratory tests).   However, one counter-example is PET-CT, which does have fixed, national prices for physician interpretation (around $100) but there is no national Part B price for the technical component, or for the global service (meaing, scan + interpretation) in the non facility setting.

In the facility setting, Medicare sets a hospital outpatient rate (about $1400 for 78815, varies with geography).  This is via the ambulatory payment classification (APC) of code 78815 under OPPS policy.

In the non-facility setting, where >800,000 PET CT scans are performed, MACs set prices for 78815 for the technical component (or the global claim).   

In a nutshell, high-end prices, probably for global claims, ran to the $2300 range in California.  There was a plateau of pricing from about $1300 to $1700.   There was a sharp drop-off with claims paid in the  $200-$900 range, and then many claims in the $120 range (interpretation only on the RVU fee schedule).

For comparison, CMS pays about $1400 for the technical component of 78815 PET CT in the hospital outpatient setting (with regional adjustments.)

What's it mean?

First, it's tricky, since the Part B database I'm pulling from lumps together all forms of Part B 78815 payment (whether PC only, TC only, or global).

California TC Can Be Rationalized. MAC behavior when freed from RVU pricing is confusing.   Generally, the maximum Part B RVU delta in low to high priced areas is +50% of the base RVU rate (e.g. for 88361, computer assisted IHC, TC or global). 

This +50% geographic hike for priced RVUs across geographies would lift the $1400 hospital outpatient APC to the $2000 range, exactly what we see in California.   So it's reasonable to guess that Noridian used the APC price of $1400 and the regional multiplier (in SF, LA, etc) of +50% to get around $2000.

Payments $200-900 Seem Damn Odd.  It's unclear how you get the prices between $200 and $1000.   

• If these are TC alone, the prices seem "too low," at least, compared to the  APC benchmark of $1400.   
• On the other hand, if these are professional component alone, it should be impossible to pay over the $100-150 range due to fee schedules for 78815-26.   

Take-Home Lesson for WSI Codes

Based on AMA CPT Summary of Panel Actions documents, at the February Palm Springs meeting they created two WSI codes (X568T, X569T), final code text to be released 7/1/2026.   At the May Boston meeting, they created additional WSI codes for breast and prostate cancer respectively (X623T, X624T), I believe also to be released 7/1/2026.   

If CMS continues to leave these "off" the summer lab meeting schedule, then they will be included in summer policymaking for the Hospital Outpatient (OPPS) environment, and assigned to pathology category prices or new tech category prices for OPPS.   

MACs may use these Category III OPPS prices as reference prices for Category III codes in Part B, just like MACs used the APC price $1400 for 78815 as a reference price for Part B.

###

###

Sidebar:
Using Chat GPT for Data Analysis

I downloaded all Part B payments for 78815, PET CT, for CY2024.  

I asked Chat GPT to compare pricing >$1978 with pricing in the band $909-1197 (low, but believable for TC).

Over $1978

Click to enlarge

$909-1197

Click to enlarge

Chat GPT adds this text:

The contrast is striking. The high-payment band is not a general national distribution; it is dominated by California, with a bit of New York. The low-payment band is much more of a multi-state pattern, led by SC, OH, FL, MN, TN, and scattered others.

Both bands in the spreadsheet are Place of Service “O” lines, so this comparison appears to be within the office/nonfacility physician-supplier world, not hospital outpatient OPPS. That makes the spread more interesting: for 78815 technical/global office PET/CT, allowed amounts are not converging around a single national TC price. Instead, the data look like a patchwork of MAC/locality/carrier-priced technical-component behavior, with California standing out at the high end.

The policy takeaway is that although the physician interpretation of 78815-26 is nationally PFS-priced, the nonfacility technical economics of PET/CT remain quite variable. In this slice, the high band is roughly double the lower band: about $2,183 weighted average versus $1,073 weighted average. That is a large swing for what is nominally the same PET/CT skull-base-to-mid-thigh service.

United Healthcare Publishes Variations from "Date of Service" Rule

Medicare's date of service (DOS) rule for laboratory tests - often given the name, "14 day rule" - gets a new spin in a publication from United Healthcare affecting millions of beneficiaries.

Thanks to Ashley Zarling for highlighting the event at LinkedIn.

CMS Date of Service.  Since about 2001, CMS has set the "date of service" rule for lab tests - both clin lab & pathology - as teh date of specimen collection, and separate rules (not the DOS regulation itself) often lead to hospital inpatient and outpatient lab tests being bundled and not separately payable.  CMS bundles all inpatient lab tests, even those ordered 13 days after hospital discharge, to the hospital inpatient DRG, and CMs bundles most lab tests for outpatients to hospital outpatient events like office or ER visits, tagging them with a "status indicator" for bundling or, rarely, separate payment.  42 CFR 414.510   See also an explanatory webpage at CMS.

Medicare Advantage plans don't have to follow this rule, although I suspect they usually do, and other commercial insurance plans definitely don't have to follow this rule - but they may often elect to, by stating that as a first principle, Medicare billing rules will be followed.

New Rule at United Healthcare

The new rule at United Healthcare was published May 1, 2026, as new "Question 9" of document  2026R0111B, rules for submitting a 1500 claim form.  United states that, "This reimbursement policy applies to all health care services billed on CMS 1500 forms and, when specified, to those billed on UB04 forms."

The rules have a number of entries about United's use of MolDx and Z-codes.

Q: Which date should be submitted on the claim for reflex testing?

A: According to CMS, when reflex testing is performed, the date of service (DOS) submitted for each test must be the date the reflex test itself is performed, not the date the original specimen was collected.

I'm not sure exactly where United gets this rule from ("Accoding to CMS...").   United has a blanket reference for all its rule, referring to CMS, CMS Manuals, or Other CMS Publications.

Without knowing a written exception, I would have said the date of reflex testing would generally be the date of specimen collection - the master CMS rule for DOS, unless the test was ordered > 14 days after inpatient or outpatient discharge.   

###

https://www.uhcprovider.com/content/dam/provider/docs/public/policies/comm-reimbursement/COMM-Molecular-Pathology-Policy-Professional-Facility.pdf

###

See a "deep dive" AI written research article on DOS back to the 1990s, here.

Upending the MAC System: Bold Idea Proposed for a Radically Different MAC

Unhappy with your local MAC?  Waiting til you-know-where freezes over before they can issue a new LCD?   Anderson/Jackson/Miller have a new proposal to up-end the current A-B MAC system through ongoing, operational competition.

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6624938

I've read it; you should read it; it's open access; Claude Opus 4.7 read it, and tells us about it below.

The paper is 16 pages long, with 79 footnotes and a number of interesting tables and illustrations. The first author Anderson is from Tufts, the last author Miller from Stanford/Hopkins, and the middle author Jackson appears to be at the WPS MAC itself.

Note that Miller is serving his second term as a board member of MEDPAC, the official advisory body on Medicare policy, which gives this publication additional interest to policy-watchers.

I'm not asserting the proposal is "likely" but it is interesting and "shakes the box" in terms of new ideas.

###

AI Corner: Claude Opus 4.7

###

Medicare's Hidden Backbone:
Why MAC Reform Is the Overlooked Lever Everyone in Healthcare Innovation Should Care About

Three-sentence summary: A new paper by Anderson, Jackson, and Miller argues that the most underused opportunity to modernize traditional fee-for-service Medicare isn't benefit redesign — it's overhauling the Medicare Administrative Contractor (MAC) system, which currently runs as twelve regional monopolies focused on grinding through claims rather than serving beneficiaries. 

The authors propose moving to a multi-MAC competitive model (roughly three per region) in which beneficiaries choose their administrator annually, MACs compete on Part B premium rebates, provider access, and speed of covering new technologies, and program integrity functions get folded back into claims operations. 

• Done right, this is budget-neutral, shrinks systemic operational risk, and finally rewards MACs for doing things like saying "yes" to a new device in something less than geological time.

---

The article: Anderson et al.

If you work in biotech, devices, or diagnostics, you have almost certainly run into the MAC system — usually at the worst possible moment. Your CE-marked, FDA-cleared, peer-reviewed-evidence-supported technology lands on the desk of a regional MAC medical director, and then… nothing. For months. Sometimes years. The Local Coverage Determination (LCD) process is famously the place where reasonable timelines go to die. To borrow the Charles Barsotti New Yorker cartoon: "How about never? Does never work for you?" That, in essence, is the timetable many innovators experience. Anderson, Jackson, and Miller's new paper, "The Forgotten Opportunity: Improving Traditional Medicare Operations," argues that this isn't a quirk of the system — it is the system, and it's fixable.

The status quo, briefly. CMS contracts with twelve Part A/B MACs and four DME MACs, each holding a regional monopoly over claims processing, coverage policy, and provider-facing operations for traditional Medicare. Together they touch about 34 million beneficiaries, 1.2 million providers, and roughly $431.5 billion in annual FFS payments. Contracts run up to ten years on a cost-plus-award-fee basis, and — strikingly — no new entrant has won a MAC jurisdiction since the program began in 2003. CMS, terrified of "crashing" Medicare claims processing during a transition, has become structurally risk-averse, which entrenches incumbents and stifles innovation. Meanwhile, program integrity has been carved off into a separate, fragmented contractor ecosystem (UPICs, CERT, Strike Force teams), improper payment rates sit at 7.66% ($31.7 billion in FY2024), and LCDs vary so much across jurisdictions that the same device can be covered in Florida and denied in Oregon. Beneficiaries — the people the program ostensibly exists to serve — have essentially no voice in any of this. They don't pick a MAC. They don't even know who their MAC is.

The proposal. The authors want to flip the model. Instead of one monopoly administrator per region, CMS would contract with roughly three MACs per jurisdiction, and beneficiaries would choose among them during the annual Medicare open enrollment window (Oct 15–Dec 7). MACs would then compete on three dimensions: (1) price — returning fraud-and-waste savings to beneficiaries as Part B premium rebates; (2) access — using performance dollars to boost provider payments in underserved specialties like behavioral health (where only about a third of psychiatrists currently bill Medicare Part B) or rural geographies; and (3) innovation — covering new technologies and services faster than the competition. Program integrity work would be reintegrated into MAC operations, with CMS's Center for Program Integrity repositioned as a regulator-of-MACs rather than a direct operator — mirroring how CMS already oversees Medicare Advantage plans. The any-willing-provider standard stays. The benefit package stays. What changes is that FFS Medicare finally starts behaving like a real health plan, with regional administrators accountable to the people whose care they're paying for.

Why this matters specifically for the innovation economy. Today, a device or diagnostic company has to run an LCD gauntlet that is slow, opaque, and inconsistent. Under a multi-MAC model, a manufacturer with strong evidence could potentially get to "yes" with at least one MAC in a region quickly, and competitive pressure would push others to follow. The current system rewards saying "no" or "later" because there's no downside — no MAC loses customers when it stalls. Add genuine choice and suddenly there's a downside. The authors also note that variation in coverage, currently treated as a bug, becomes a feature: a beneficiary who wants access to, say, AI-interpreted imaging or an emerging remote diagnostic could pick the MAC that covers it. For diagnostics companies especially — which often live or die on LCD timing — this would be transformative.

Opportunities worth highlighting. Beyond faster coverage, the model unlocks several things the current system structurally cannot deliver. Embedding fraud analytics directly into claims workflows (rather than bolting them on through separate UPIC contractors) would enable real-time detection rather than the current pay-and-chase model. Competition would push MACs to invest in machine learning, modern provider portals, and the kind of consumer-facing tooling that MA plans have spent a decade building. A revamped Medicare Plan Finder could finally give beneficiaries an apples-to-apples comparison across MA plans, MAC + Medigap + Part D bundles, with provider networks and drug formularies side by side — something that does not meaningfully exist today. And critically, the reform is budget-neutral: the financial engine is recovered fraud and waste, not new appropriations. In a fiscal environment where Medicare trust fund pressure is intensifying and any proposal requiring new spending is politically dead on arrival, that matters.

The risks and drawbacks — and they're real. I'd be doing you a disservice not to flag them. First, beneficiary choice fatigue. Medicare beneficiaries already face a notoriously complex annual decision among MA plans, Part D plans, and Medigap policies; adding MAC selection to the pile could overwhelm seniors, especially those with cognitive decline or limited digital literacy. The paper acknowledges this and calls for investment in SHIP counselors and Plan Finder upgrades, but the implementation risk is non-trivial — and broker incentives in MA have shown how easily "choice" architecture can be manipulated. Second, fragmentation of coverage policy. The current LCD inconsistency the authors criticize could, under a multi-MAC model, get worse before it gets better: now a beneficiary in the same zip code could have different coverage than their neighbor based on which MAC they picked. For physicians treating patients across MACs in one region, this would be an administrative headache. Third, the cherry-picking problem. MACs bear no actuarial risk in this model, but they may still face soft incentives to attract healthier, lower-touch beneficiaries (fewer claims, fewer complaints, easier metrics). MA already struggles with risk selection despite risk adjustment; importing competition into FFS could replicate that pathology. Fourth, provider whiplash. Providers would now need to deal with three MACs per region instead of one, each potentially with different LCDs, different prior auth rules (if WISeR-style utilization review expands), different portals, and different appeals processes. The paper's vision of "non-price competition" on provider payment is appealing, but variable fee schedules within a single region would create real administrative burden and could distort referral patterns. Fifth — and this is mine, not the authors' — there's the question of what happens to small or rural providers if a MAC fails or exits. The paper argues multi-MAC reduces single-vendor risk, which is true at the regional level, but a failing MAC's beneficiaries still need to move somewhere, and the transition mechanics in a system with active beneficiary enrollment are more complex than the paper fully works through.

The bottom line. Whether you're building a Class III implant, a novel molecular diagnostic, a digital therapeutic, or you're just a Medicare beneficiary who'd like your claims questions answered by a human who knows your name, the MAC system is upstream of almost everything that frustrates you about traditional Medicare. Anderson, Jackson, and Miller have written the rare health policy paper that identifies a genuinely overlooked lever — one that doesn't require new mandatory spending, doesn't blow up the any-willing-provider promise that defines FFS Medicare, and aligns nicely with the agency's stated direction of being a regulator more than an operator. The implementation risks are real and the paper is more confident about the upside than I'd be. But the status quo — "How about never?" — is no longer defensible, and a serious conversation about what comes next is overdue.

---

Reference: Anderson LT, Jackson K, Miller BJ. "The Forgotten Opportunity: Improving Traditional Medicare Operations." SSRN, April 21, 2026.   

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6624938

See a humorous view auto-written by Chat GPT  "Whazzup."

Dunn Introduces H.R. 8890, Enhanced CLIA - Deja Vu Again

There have been a few attempts at "Enhanced CLIA" legislation, and there's a new kid in town: H.R. 8890.

It's introduced by Congr. Dunn of Florida, who recently introduced an "Improve LCDs" bill (here).

The legislative language of H.R. 8890 is NOT posted yet but will appear here.

(See page pdf 74 of Congressional Record here.)

See early coverage at Genomeweb here.   See an article by Alex Dickinson at LinkedIn here.

See Congr. Dunn press here.   And see a generally supportive comment by ACLA, here.

I'll reserve more detailed comment for when the actual legislative language is public.  In the meantime, here are the five bullets used in the Dunn press release.

____

The Enhancing CLIA Act would:

1. Direct CMS to establish a transparent and publicly accessible database with key information regarding laboratory developed testing services (LDTs), including summaries of performance specifications and validation information
2. Reaffirm that LDTs are professional services regulated by CMS under CLIA
3. Clarify that LDTs include innovative analyses of digital laboratory information, such as genomic data derived from previously-sequenced patient samples
4. Leverage existing review and oversight mechanisms to allow laboratories to voluntarily obtain confirmation from a recognized third party that an LDT is analytically and clinically valid
5. Establish centralized error reporting and bolster CMS oversight authority if credible information indicates an LDT lacks analytical or clinical validity

Senate Resolution would put Part B Prior Auth (WISeR) on Chopping Block

HEADER: A  new Senate resolution seeks to overturn CMS’s WISeR pilot.  This is the CMMI Medicare model that applies AI-supported prior authorization to selected services considered vulnerable to waste or inappropriate use.

• I wrote about some glitches in WISeR on May 11.

Senate Democrats, led by Ron Wyden, Maria Cantwell, Richard Blumenthal, and Kirsten Gillibrand, argue that the model risks importing the burdens of private-plan prior authorization into traditional Medicare, where such controls have historically been uncommon. 

CMS describes WISeR as a targeted fraud-and-waste initiative focused on services such as skin and tissue substitutes and epidural steroid injections (in its Version 01, no genomic tests.) Critics counter that prior authorization can delay care, increase administrative burden, and place algorithmic tools between physicians and beneficiaries. The political opening came after GAO concluded that WISeR is agency rulemaking subject to the Congressional Review Act, despite HHS’s earlier position that the notice was merely guidance. That decision gives Congress a procedural path to force a vote on repealing the model.

Falls under CRA Congressional Review Act

Augmented AI and AMA Thinking: Wishful Thinking?

There's a new paper going the rounds of commentary at Linked In:

Open access; find it here.

See comment at LinkedIn by Jan Beger here.  And a view from Robert Lauritzen here.  (And see another, related, from Lauritzen, here.)

Cuocolo & Huisman are from Italy and the Netherlands; the paper is in European Radiology; the thesis is that many clinicians and policymakers strike the position (or strike the pose) that AI will not replace physician work but "augment" it.  

This thinking, or rhetoric, isn't only across-the-pond.   The AMA has a position paper on AI ("Appendix S") that makes the biggest section of AI something called "augmentative" software services.  These by definition do not replace physician work but "augment" it.

To see this position taken apart, read the links above.

I have seen the same position taken a couple times, at least, in press releases or speeches from AMA in the U.S.    Below, I asked Chat GPT agentic/internet mode to find quotes where the AMA says this.   By the way, the Chat GPT replaced, rather than augmented, my work on that score.

1. “Augmented intelligence … [is] designed to enhance—not replace—physicians.”
   —AMA CEO John Whyte, MD, MPH, 2026 press release.
   https://www.ama-assn.org/press-center/ama-press-releases/ama-ai-usage-among-doctors-doubles-confidence-technology-grows

2. “Health care AI should be understood as a tool to augment professional clinical judgment, not a technology to replace or override it.”
   —AMA Journal of Ethics discussing AMA policy.
   https://journalofethics.ama-assn.org/article/making-policy-augmented-intelligence-health-care/2019-02

3. AMA uses “augmented intelligence” rather than “artificial intelligence” because AI tools “support rather than explicitly replace human decision-making.”
   —AMA report: Future of Health: The Emerging Landscape of Augmented Intelligence in Health Care.
   https://www.ama-assn.org/system/files/future-health-augmented-intelligence-health-care.pdf

4. AI is “intended to co-exist with human decision-making” and “should not be used to replace physician reasoning and knowledge.”
   —AMA House of Delegates augmented intelligence principles.
   https://www.ama-assn.org/system/files/a24-246.pdf

5. “Augmented intelligence must not replace or diminish the patient-physician relationship.”
   —AMA House of Delegates principles.
   https://www.ama-assn.org/system/files/a24-246.pdf

6. “The real promise of health care AI is to enhance the work of physicians … not to replace them.”
   —AMA article on AI workflow.
   https://www.ama-assn.org/practice-management/digital-health/3-ways-medical-ai-can-improve-workflow-physicians

7. AMA says AI should “enhance physician intelligence, not replace it.”
   —AMA article on healthcare AI.
   https://www.ama-assn.org/practice-management/digital-health/4-ways-health-care-ai-could-help-physicians-work

8. “AI is not going to replace doctors—but doctors who use AI will replace those who don’t.”
   —AMA President Jesse Ehrenfeld, MD, MPH.
   https://www.ama-assn.org/practice-management/digital-health/ai-already-reshaping-care-heres-what-it-means-doctors

9. “AI’s role is to help health care professionals, not replace them.”
   —AMA implementation guidance.
   https://www.ama-assn.org/practice-management/digital-health/implement-health-ai-first-decide-who-s-accountable

[Chat GPT added] A subtle but important point in AMA rhetoric is that they deliberately prefer the term “augmented intelligence” over “artificial intelligence.” That terminology itself is almost a policy statement: AI is framed as an assistant to physician cognition and workflow rather than as an autonomous substitute for physicians. 

###

###

A 150-word AI version of this blog.

A new paper in European Radiology by Cuocolo and Huisman is stirring unusually sharp debate on LinkedIn and elsewhere. The paper questions a now-familiar refrain in medicine: that artificial intelligence will “augment” physician work, rather than replace it. Critics argue that this language may be less a prediction than a rhetorical strategy designed to make AI adoption more politically and professionally acceptable.

The issue is hardly confined to Europe. In the United States, the AMA has repeatedly framed AI as “augmented intelligence,” emphasizing that these systems are intended to support—not replace—physician judgment. In fact, the AMA’s evolving AI framework (“Appendix S”) places major emphasis on “augmentative” software services.

Yet the economic logic of AI in other industries has often centered on labor substitution, not coexistence. As noted in Quinn's blog, even gathering AMA quotations on “augmentation” was efficiently handled by ChatGPT itself—ironically replacing, rather than augmenting, his own research work.

Dr Bien-Willner Is Departing MolDx - An Update

 

Gabriel Bien-Willner to
Transition Out of
MolDX Medical Director Role

May 15, 2025  (Columbia, SC)
LinkedIn by GBW: here.

Gabriel A. Bien-Willner, MD, PhD, the medical director most closely associated with the modern expansion of Palmetto GBA’s MolDX program, has announced that he will transition out of the MolDX Medical Director role as Palmetto seeks a successor.

The move is significant because MolDX is no longer a niche Medicare contractor initiative. Under Bien-Willner’s tenure, it became one of the most influential operating systems for molecular diagnostics reimbursement in the United States — combining test registration, Z-Code identification, evidence review, coverage policy, pricing, and claims edits into a single payer-facing architecture.

Bien-Willner said he will remain at Palmetto during the transition and emphasized that MolDX operations will continue. He also noted that his recent role has been largely one of oversight, suggesting that day-to-day program functions are now embedded in a broader MolDX team rather than dependent on a single physician leader.

An "Unusually Relevant" Background 

Bien-Willner brought an unusually relevant background to the role. He is a physician-scientist trained in pathology and molecular genetics, with an MD/PhD from Baylor College of Medicine and clinical training at Washington University in St. Louis. Before joining Palmetto, he worked in molecular pathology, genomics, and diagnostic laboratory leadership, giving him fluency in both the science of advanced testing and the operational realities of laboratory medicine.

Digital Pathology Dollars. Re-live the Film-Digital Switchover at the RVU Ranch.

What’s Past Is Prologue: 

Medicare, Radiology, and the Strange Migration from Film to Digital

##

The CMS-RVU migration from film to digital radiology was not simply a technology upgrade; it became a revealing Medicare payment episode. 

Around 2013–2017, CMS and the AMA/RUC translated the disappearance of film, processors, view boxes, and related supplies into the mechanics of practice expense RVUs. Film-era inputs were removed from hundreds of imaging codes, PACS workstations were added, and CMS estimated about $240 million in annual budget-neutral redistribution—small relative to the whole Physician Fee Schedule, but roughly 5% of radiology’s Medicare allowed charges.

The episode had a distinctive mood: inevitability, suspicion, and invoice-driven trench warfare. Radiology accepted that film was obsolete but argued that CMS was underpricing the digital replacement by treating PACS like a desktop computer. 

The history is not a perfect parallel to digital pathology or AI, but it is a useful prologue. It shows how Medicare recognizes technology transitions: slowly, mechanically, and only when costs can be named, priced, and mapped to codes.

Updated: Amal Thommil's MRD Roadmap

 Amal Thommil of DeciBio updates his map of MRD coverage in oncology.

Find it here:

https://www.linkedin.com/posts/amalthommil_reimbursement-of-mrd-in-solid-tumors-may-ugcPost-7460780354581303298-xIVG? 

see original at linked in

AI Guest Column: Chat GPT on TROP2 CDx from Roche Diagnostics Day

 A previous blog summarized the 3-hour, annual Roche Diagnostics Day.

I asked Chat GPT to go back to the transcript and discuss the news as far as TROP2 CDx coming through FDA using the AZ QCS technology.

As always, take this as an example of the current state of "AI thinking and writing" - not as gospel truth.

###

Roche’s TROP2 / AstraZeneca CDx Strategy

Why Roche is treating this as more than “another IHC companion diagnostic”

1. The short version

Roche is positioning VENTANA TROP2 RxDx, developed with AstraZeneca and incorporating AstraZeneca’s QCS — quantitative continuous scoring — as the first major proof point for a new generation of computational pathology companion diagnostics. 

In Roche’s framing, this is not simply a TROP2 IHC assay. It is a regulatory, commercial, and technical template for complex IHC-based CDx in which the final clinical result depends on a tightly integrated system: VENTANA staining, digital slide acquisition, image management, algorithmic scoring, pharma trial linkage, and FDA-reviewed RxDx labeling. 

Roche says the TROP2 product is currently launched as RUO, with a CDx launch planned for Q1 2027, and describes it as the first computational pathology RxDx device.

The Deeper Message

The deeper message is that Roche wants to use TROP2 as the wedge product for a much larger AI-enabled CDx franchise. The transcript is explicit: Laura Apitz calls attention to the “TROP2 franchise,” says it will be the “very first computational pathology companion diagnostic” launching in lung cancer next year, and says the RUO TROP2 footprint is being grown globally to set up later companion diagnostic algorithms.

---

2. What Roche is actually claiming

Roche’s claims fall into four layers.

First, TROP2 is a specific AstraZeneca-linked CDx project. The deck says VENTANA TROP2 RxDx is developed in collaboration with AstraZeneca and incorporates AstraZeneca’s proprietary QCS, or quantitative continuous scoring. The footnote also cautions that PHCS launches depend on the success of the related pharmaceutical trials and drug launches, which is important: this is not a generic assay launch independent of a drug program. It is a drug-diagnostic co-development story.

Second, TROP2 is the first computational pathology RxDx device in Roche’s pipeline. Roche highlights TROP2 as the first computational pathology RxDx device, currently launched as RUO, with CDx launch planned for Q1 2027. It also says the QCS platform can support multiple indications and biomarkers and can fuel future pharma collaborations.

Third, TROP2 is part of a larger digital pathology buildout. Roche ties TROP2 to AISight, PathAI, scanner updates, and digital pathology adoption. AISight Dx is described as a cloud-based image-management solution with a pathologist-centric collaborative interface, designed to be interoperable with scanners, LIS, and PACS systems.

Fourth, TROP2 is a commercial proof point for Roche’s PHCS/CDx flywheel. Roche says it has more than 25 years of experience in personalized healthcare solutions, more than 85 pharma partners, more than 150 ongoing IVD programs, over 25 digital pathology IVD development projects, seven biomarkers, and 13 indications. That scale is central to the argument: TROP2 is not a one-off; it is the first visible example of a pipeline.

---

3. Why TROP2 is more complex than ordinary IHC

Traditional IHC CDx often reduces to a standardized stain, a pathologist-read score, and a cutoff embedded in a drug label. That model is already complicated — specimen handling, fixation, clone, platform, staining conditions, reader training, and cutoffs all matter. But Roche is implying that TROP2 moves beyond that classic model.

The phrase quantitative continuous scoring is the giveaway. Instead of a relatively simple binary or ordinal manual interpretation, QCS suggests a more granular quantitative or semi-quantitative readout, presumably linking TROP2 expression patterns to drug eligibility or treatment-response enrichment. Roche does not fully disclose the scoring architecture in the investor deck, but it makes clear that AstraZeneca’s QCS is proprietary and incorporated into the VENTANA TROP2 RxDx device.

That makes this a complex IHC CDx in several senses:

Complexity layer	Why it matters
Biology	TROP2 expression may not behave like a simple mutation-positive / mutation-negative marker.
Analytical pathology	IHC intensity, distribution, tumor heterogeneity, and compartment recognition may all matter.
Scoring	Quantitative continuous scoring implies algorithmic or computational assistance rather than a simple manual category.
Device integration	The final result may depend on stain, scanner, image management, and software.
Regulatory review	FDA must evaluate a drug-linked diagnostic system, not merely an antibody reagent.
Commercial deployment	Labs need validated digital workflows and possibly scanner/software compatibility.

This is why Roche repeatedly links TROP2 to digital pathology, not merely to advanced staining. In the transcript, Apitz discusses new multiplexing and computational pathology algorithms, then immediately pivots to the TROP2 franchise as the first computational pathology CDx in lung cancer.

---

4. Why AstraZeneca matters

AstraZeneca’s role is not incidental. Roche is not saying, “Here is a Roche TROP2 stain that might be useful for many drugs.” It is saying the VENTANA TROP2 RxDx device was developed with AstraZeneca and incorporates AstraZeneca’s proprietary QCS.

That matters because pharma companies increasingly need biomarkers that do more than identify a target’s presence. For antibody-drug conjugates, immuno-oncology combinations, and complex targeted therapies, a simple “marker present” assay may be inadequate. Pharma may want richer tissue information: expression intensity, distribution, tumor-cell localization, microenvironment context, heterogeneity, or a continuous score tied to response.

So the TROP2 / AstraZeneca project is a signal to other pharma companies: Roche can take your complex tissue biomarker and turn it into a regulated, globally deployable RxDx. Roche is not merely offering a stain. It is offering a development partner that can combine VENTANA IHC, digital pathology, algorithmic scoring, global lab reach, and FDA-facing CDx experience.

This is the “partner pharma” message. Roche is marketing itself as the place where pharma can bring complex tissue biomarkers that are too hard for simple manual IHC and need computational pathology to become clinically usable.

---

5. Why Roche calls this a franchise

The phrase “TROP2 franchise” is significant. It implies that Roche sees TROP2 not just as one assay for one drug, but as a platformable class of work.

In the transcript, Apitz says Roche is growing the RUO TROP2 footprint globally now, which “sets us up really nicely” for the companion diagnostic algorithms. This will allow experience and help prepare the laboratories before the regulated CDx launch. RUO deployment can support familiarity, pharma development, workflow testing, institutional readiness, and potentially evidence generation, while the formal CDx path proceeds.

The deck reinforces this by saying QCS is a platform for multiple indications and biomarkers and that TROP2 can fuel a pipeline of future pharma collaborations.

That is the franchise logic:

TROP2 RUO footprint → pharma trial use → computational scoring validation → FDA RxDx launch → broader QCS / algorithmic CDx platform → more pharma collaborations.

---

6. Why FDA significance is high

The regulatory significance is that Roche is describing this as the first computational pathology RxDx device and elsewhere as an FDA Breakthrough Device Designation for the first AI-based CDx in oncology. The deck’s AI timeline specifically notes “VENTANA TROP2: FDA BDD for 1st AI-based CDx in oncology.”

That positions TROP2 as a test case for FDA review of a new CDx category: not just IHC, not just image analysis, not just software, and not just a digital pathology workflow, but an integrated AI-enabled or computational pathology companion diagnostic linked to a therapy.

For FDA, the hard questions likely include:

FDA issue	Why TROP2 may matter
Algorithm performance	How accurate and reproducible is the QCS output?
Clinical validity	Does the score identify patients who benefit from the AstraZeneca therapy?
Analytical reproducibility	Does the result hold across sites, tissue handling, instruments, scanners, and operators?
Human factors	How does the pathologist interact with the algorithmic output?
Device boundaries	What exactly is the regulated device: antibody, stain, scanner, software, algorithm, IMS, or all of the above?
Change management	How are future software, scanner, algorithm, and scoring updates handled?
Labeling	How will the drug label and diagnostic label specify use, cutoff, and eligible population?

This is why TROP2 is a bellwether. If Roche can get FDA comfortable with this integrated model, it creates a precedent for additional computational IHC CDx devices. 

If it struggles, the entire AI-enabled CDx category may move more slowly.

---

7. How PathAI changes the story

PathAI is strategically important because it gives Roche more of the missing software and AI layer. Roche already had VENTANA staining, scanners, and pharma CDx relationships. PathAI adds image management, AI-driven analysis, workflow capabilities, clinical-trial services, translational research, and biomarker discovery. Roche describes PathAI as complementary to Diagnostics and synergistic with Pharma, including workflow enhancement, regulated AI-enabled CDx, translational research, biomarker discovery, and drug-target discovery.

In the transcript, Andy Beck goes further. He says AI-powered CDx is expected to accompany most new approvals by 2035, that AI may be run on every slide, and that massive real-world datasets from millions of slides could drive R&D, regulatory, and policy decisions. He also describes AISight as a best-in-class digital pathology image-management system that received CE/IVDR and FDA clearance and a predetermined change control plan to help expand the label and add new scanners and monitors.

That matters for TROP2 because the product is not just an antibody. Roche is trying to create the end-to-end environment in which algorithmic CDx can be developed, deployed, monitored, and scaled.

---

8. The PHCS/CDx flywheel

Roche’s pathology strategy is built around a flywheel. The deck says Roche has a pathology menu with more than 250 ready-to-use IHC/ISH assays and personalized healthcare services involving more than 85 pharma partnerships and more than 150 ongoing IVD programs. It then describes pharma CDx partnerships as a flywheel: partnerships accelerate menu growth, menu growth drives instrument placements, placements increase access to Roche’s diagnostic menu, and increased access makes Roche more attractive for CDx development.

TROP2 fits perfectly into that flywheel. A high-profile AstraZeneca-linked computational CDx can do several things at once:

1. Strengthen Roche’s pharma-partner credibility.
   Other pharma companies can see Roche as a partner for complex tissue biomarkers.

2. Increase demand for VENTANA digital pathology infrastructure.
   If the CDx requires digital scoring, labs need compatible scanners, image management, software, and validated workflows.

3. Drive IHC/ISH menu differentiation.
   Roche can argue its pathology menu is moving beyond conventional stains toward AI-enabled therapeutic decision tools.

4. Create pull-through for PathAI / AISight.
   TROP2 gives a concrete reason for labs and pharma partners to adopt the computational layer.

5. Create a repeatable FDA pathway.
   If TROP2 succeeds, it becomes the model for future AI-enabled RxDx products.

---

9. Why this could be the first of many “complex IHC” CDx products

Roche also discusses translucent chromogens and multiplexing, including P40/TTF1 in lung cancer and SOX10/Ki-67 in melanoma / lymph node assessment. Apitz describes these as the first of many multiplexes Roche plans to bring out in companion diagnostic, companion digital pathology, and non-gynecologic cytology spaces.

This is important. TROP2 is not isolated from the multiplexing roadmap. The general direction is toward IHC that becomes more information-rich:

single marker → multiplex marker → quantitative score → digital pathology algorithm → companion diagnostic → pharma-linked treatment selection.

That path is a major shift for IHC. The historical strength of IHC is that it is cheap, familiar, visual, and installed everywhere. The historical weakness is that interpretation can be subjective, semi-quantitative, and hard to standardize across sites. Roche is trying to preserve the installed-base advantage of IHC while adding algorithmic reproducibility and pharma-grade scoring.

---

10. An interpretation

Roche is using TROP2 as a category-creating product. The immediate commercial story is an AstraZeneca-linked lung cancer companion diagnostic. The larger strategic story is that Roche wants to own the regulated infrastructure for computational IHC companion diagnostics.

That infrastructure includes:

VENTANA stain + BenchMark instruments + DP scanners + AISight image management + PathAI algorithms + pharma CDx development + FDA regulatory pathway + global pathology installed base.

If successful, TROP2 becomes the proof that Roche can take a complex tissue biomarker, convert it into a quantitative computational pathology result, support a drug approval, and deploy the diagnostic globally. That would be a major competitive advantage versus companies that own only the antibody, only the scanner, only the image-management software, or only the algorithm.

The risk is that this is hard. FDA will need to be comfortable with the device boundaries, scoring method, clinical cutoff, software updates, scanner dependencies, and pathologist role. Labs will need digital pathology infrastructure. Pharma trials must succeed. The clinical label must be meaningful. And the algorithm must be robust across real-world tissue variability.

But Roche’s message is clear: TROP2 is the opening act for AI-enabled CDx IHC. If TROP2 lands, it will not just be a successful TROP2 assay. It will be the first credible example of Roche’s next-generation pathology strategy: complex IHC made quantitative, digital, algorithmic, pharma-linked, and FDA-reviewable.

AI Guest Column: ChatGPT & Opus: Roche Diagnostics Day

On May 12, 2026, Roche held its annual Diagnostics Day with a 3-hour webinar and a 150-slide deck.

Find them here:  https://www.roche.com/investors/events/diagnostics-day-2026

I gave the materials separately to Chat GPT and Opus 4.7, which provided 13p and 24p reports.  I asked Chat GPT to produce a single combined report with a cover page (15pp).

Remember, this isn't an example of gospel-truth, it's an example of the current state of AI reading and writing.

Find the PDF at Google Cloud here:

https://drive.google.com/file/d/1GcbrjhXCujRxJ7Wg8v9UbjJ32nJH457f/view?usp=sharing

See also a deep dive on Trop2 CDx.

See an essay on Roche Dx Day by Swapnil Amin.

To Fix Software and AI Reimbursement - Do We Need to Understand the RUC?

To Fix Software and AI Reimbursement - Do We Need to Understand the RUC?

Maybe.

For at least five years, CMS has had difficulty pricing software-intensive services like AI.  In 2025, CMS included an RFI for "public comment" on these problems in both the July 2025 Physician (Part B) rulemaking and the Hospital Outpatient rulemaking.  

One Damn Thing After Another

One case study was the great difficulty CMS had when in pricing "Heartflow" software in both the OPPS and Part B settings.

Another case study was the difficulty in pricing a photographic device that gives retinal diagnostics. Processed through the RUC,  the AMA valuation committee gave 92229 a few dollars for technician time ($9), seventy cents for capital equipment ($14,000 device x 13 minutes), and a software fee.  

CMS  declined the software fee as a payable item, leaving the $40 service valued at about $10.   CMS arbitrarily assigned an RVU value paying $46.   This dates back to 2019/2020.

On the Clin Lab Fee Schedule side, things used to work, then broke.  AMA issued and CMS accepted around 10 codes for complex whole slide imaging (WSI)  tests, and generally priced at $700.  Then, AMA stopped issued WSI PLA codes, so nothing happened for a couple years.

Amidst many AMA initiatives around AI policy, AMA is now revising its Appendix S, and has begun putting new WSI tests into "Category III."  But we don't know if CMS will include these Category III lab test codes in its summer lab crosswalk/gapfill process.  I think they can and should, since CMS has classed WSI tests as CLFS tests, even when they are closely reviewed as ADLTs.  And FDA and CLIA and, say, New York State, classifies computatioinal pathology tests as CLIA tests.   So they should go into the summer CLFS crosswalk/gapfill process.   File that under "track this."

All About the RUC

In 2016, Miriam Laugesen of Columbia published a wonderful book about the AMA RUC process, "Fixing Medical Prices," (Harvard University Press.)

It turns out there are a lot of RUC documents, some intricately detailed, on the AMA website.

For example, there is an encyclopedic 27 page article about practice expense components - which will be relevant, if only for contrast, to future software valuation.   Find it here:

https://www.ama-assn.org/system/files/practice-expense-component.pdf

Find it all on a sidebar blog - I've listed about 20 AMA RUC resources there.  

https://bqwebpage.blogspot.com/2026/05/ruc-resources-are-numerous-at-ama.html

At the bottom of the long sidebar, you can get files for RUC decisions and meeting minutes going back to 1993 (!!!).   The AMA RBRVS subscription database (about $300) lists the last RUC review by date for every CPT code, and with that date, you can get the original files at AMA.   And you can also track the most recent debates for how the RUC is handling new Category I codes with software.  Currently as recent as...May 2026!

Understanding FDA Regulatory Documents for Digital Pathology: PAIGE and ARTERA-AI

We often hear there are 1000 FDA software/AI devices or SaMD for radiology, and only a few for pathology.

Let's look at pathology, and see what the FDA regulatory publications are.  We also look at the sometimes-confusing delays and timelines in the appearance of documents.

Classification Letter and "Safety and Effectiveness" Review

FDA issues a clearance or classification letter, reflecting the date of clearance and documenting the date of submission.   FDA lists the regulatory category (e.g. 864.3755).   

The above letter enables the company to issue a press release about its new FDA product.

At a delay of 2-12 months, the FDA publishes a "Safety and Effectiveness" review, a technology assessment, often circa 20 pages long.

Regulatory Category and Product Code Quick; Federal Register "Someday"

For a new de novo category (which will be a future producdt 510(k) category), FDA assigns a name of the product category and a CFR category citation (such as 21 CFR 864.3750 and 864.3755).   

However, that CFR is a placeholder.  It make take 1-3 years for FDA to publish a several-page announcement in the Federal Register that the regulatory category is officially created.  

At that point, the regulatory category will appear in the eCFR.   Some  regulatory categories in the eCFR are relatively short and some run several detailed paragraphs long.

At the time of the classification letter, FDA will also have assigned a Product Category (a 3-letter code, such as QPN).   

Remember; The regulatory code (864.3755) and the product code (QPN) will be issued right away.  It's the Federal Register publication, which actually puts that number into the eCFR, that can lag a couple years.

FourExamples:

1) Paige Prostate (9/2021)

 Letter YES, S&E Review YES, Category 3750 YES, Product Code QPN YES, Fed Reg Yes (2/23)

2) Artera Prostate Prognostic (7/2025)

 Letter YES, S&E Review NO, Category 3755 YES, Product Code SFH YES, Fed Reg NO

3) Artera Breast Prognostic (5/2026)

Clearance Press release YES
Letter NO, S&E Review NO, Category 3755 YES, Product Code SHW YES, Fed Reg NO

4) Paige Pan-Cancer (in review as "breakthrough," since 4/2025)

(Only: Press release states under breakthrough review)

###

Paige Prostate

Home Page

https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN200080

DEN 200080 Classification Order (Letter)

https://www.accessdata.fda.gov/cdrh_docs/pdf20/DEN200080.pdf

DEN 200080 25pp Safety and Effectiveness

https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN200080.pdf

Fed Reg 88 FR 7007, 2/2/2023

https://www.govinfo.gov/content/pkg/FR-2023-02-02/pdf/2023-02141.pdf

eCFR

https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-864/subpart-D/section-864.3750

Product Code

https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpcd/classification.cfm?ID=QPN

Artera Prostate Prognostic

Home page

https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN240068

Classification Order (Letter)

https://www.accessdata.fda.gov/cdrh_docs/pdf24/DEN240068.pdf

DEN 240068  24pp Safety and Effectiveness

https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN240068.pdf 

eCFR: NOT YET, but reserved for 864.3755

FED REG: NOT YET

Product Code

https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPCD/classification.cfm?id=SFH

Artera Breast Prognostic

Product Code (under 864.3755)

https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpcd/classification.cfm?id=SHW

#####

Product Code definitions

QPN - Paige Prostate

• Software algorithm device to assist users in digital pathology
• A software algorithm device to assist users in digital pathology is an in vitro diagnostic device intended to evaluate acquired scanned pathology whole slide images. The device uses software algorithms to provide information to the user about presence, location, and characteristics of areas of the image with clinical implications. Information from this device is intended to assist the user in determining a pathology diagnosis.

SFH - Artera Prostate

• Pathology software algorithm device analyzing digital images for cancer prognosis
• A pathology software algorithm device analyzing digital images for cancer prognosis is a software intended to analyze scanned whole slide images (WSIs) from prostate biopsies prepared from formalin fixed paraffin-embedded (FFPE) tissue and stained using Hematoxylin & Eosin (H&E) stains. The device provides prognostic risk estimates which are intended to assist physicians with prognostic risk-based decisions along with other clinicopathological factors. The device is not intended to determine a clinical diagnosis.

SHW - Artera Breast

• Pathology software algorithm device analyzing digital images for breast cancer prognosis
• A pathology software algorithm device analyzing digital images for cancer prognosis is a software intended to analyze scanned whole slide images (WSIs) from breast cancer specimens prepared from formalin fixed paraffin-embedded (FFPE) tissue and stained using Hematoxylin & Eosin (H&E) stains. The device provides prognostic risk estimates which are intended to assist physicians with prognostic risk-based decisions along with other clinicopathological factors. The device is not intended to determine a clinical diagnosis.

21 CFR 864.3750

Paige Prostate falls under 864.3750.   Artera Prognostic Prostate and Prognostic Breast fall under 3755, which is a reserved number with no text.

864.3750 is lengthy.

21 CFR 864.3750 
Software algorithm device to assist users in digital pathology.

(a) Identification. A software algorithm device to assist users in digital pathology is an in vitro diagnostic device intended to evaluate acquired scanned pathology whole slide images. The device uses software algorithms to provide information to the user about presence, location, and characteristics of areas of the image with clinical implications. Information from this device is intended to assist the user in determining a pathology diagnosis.

(b) Classification. Class II (special controls). The special controls for this device are:

(1) The intended use on the device's label and labeling required under § 809.10 of this chapter must include:

(i) Specimen type;

(ii) Information on the device input(s) (e.g., scanned whole slide images (WSI), etc.);

(iii) Information on the device output(s) (e.g., format of the information provided by the device to the user that can be used to evaluate the WSI, etc.);

(iv) Intended users;

(v) Necessary input/output devices (e.g., WSI scanners, viewing software, etc.);

(vi) A limiting statement that addresses use of the device as an adjunct; and

(vii) A limiting statement that users should use the device in conjunction with complete standard of care evaluation of the WSI.

(2) The labeling required under § 809.10(b) of this chapter must include:

(i) A detailed description of the device, including the following:

(A) Detailed descriptions of the software device, including the detection/analysis algorithm, software design architecture, interaction with input/output devices, and necessary third-party software;

(B) Detailed descriptions of the intended user(s) and recommended training for safe use of the device; and

(C) Clear instructions about how to resolve device-related issues (e.g., cybersecurity or device malfunction issues).

(ii) A detailed summary of the performance testing, including test methods, dataset characteristics, results, and a summary of sub-analyses on case distributions stratified by relevant confounders, such as anatomical characteristics, patient demographics, medical history, user experience, and scanning equipment, as applicable.

(iii) Limiting statements that indicate:

(A) A description of situations in which the device may fail or may not operate at its expected performance level (e.g., poor image quality or for certain subpopulations), including any limitations in the dataset used to train, test, and tune the algorithm during device development;

(B) The data acquired using the device should only be interpreted by the types of users indicated in the intended use statement; and

(C) Qualified users should employ appropriate procedures and safeguards (e.g., quality control measures, etc.) to assure the validity of the interpretation of images obtained using this device.

(3) Design verification and validation must include:

(i) A detailed description of the device software, including its algorithm and its development, that includes a description of any datasets used to train, tune, or test the software algorithm. This detailed description of the device software must include:

(A) A detailed description of the technical performance assessment study protocols (e.g., regions of interest (ROI) localization study) and results used to assess the device output(s) (e.g., image overlays, image heatmaps, etc.);

(B) The training dataset must include cases representing different pre-analytical variables representative of the conditions likely to be encountered when used as intended (e.g., fixation type and time, histology slide processing techniques, challenging diagnostic cases, multiple sites, patient demographics, etc.);

(C) The number of WSI in an independent validation dataset must be appropriate to demonstrate device accuracy in detecting and localizing ROIs on scanned WSI, and must include subsets clinically relevant to the intended use of the device;

(D) Emergency recovery/backup functions, which must be included in the device design;

(E) System level architecture diagram with a matrix to depict the communication endpoints, communication protocols, and security protections for the device and its supportive systems, including any products or services that are included in the communication pathway; and

(F) A risk management plan, including a justification of how the cybersecurity vulnerabilities of third-party software and services are reduced by the device's risk management mitigations in order to address cybersecurity risks associated with key device functionality (such as loss of image, altered metadata, corrupted image data, degraded image quality, etc.). The risk management plan must also include how the device will be maintained on its intended platform (e.g. a general purpose computing platform, virtual machine, middleware, cloud-based computing services, medical device hardware, etc.), which includes how the software integrity will be maintained, how the software will be authenticated on the platform, how any reliance on the platform will be managed in order to facilitate implementation of cybersecurity controls (such as user authentication, communication encryption and authentication, etc.), and how the device will be protected when the underlying platform is not updated, such that the specific risks of the device are addressed (such as loss of image, altered metadata, corrupted image data, degraded image quality, etc.).

(ii) Data demonstrating acceptable, as determined by FDA, analytical device performance, by conducting analytical studies. For each analytical study, relevant details must be documented (e.g., the origin of the study slides and images, reader/annotator qualifications, method of annotation, location of the study site(s), challenging diagnoses, etc.). The analytical studies must include:

(A) Bench testing or technical testing to assess device output, such as localization of ROIs within a pre-specified threshold. Samples must be representative of the entire spectrum of challenging cases likely to be encountered when the device is used as intended; and

(B) Data from a precision study that demonstrates device performance when used with multiple input devices (e.g., WSI scanners) to assess total variability across operators, within-scanner, between-scanner and between-site, using clinical specimens with defined, clinically relevant, and challenging characteristics likely to be encountered when the device is used as intended. Samples must be representative of the entire spectrum of challenging cases likely to be encountered when the device is used as intended. Precision, including performance of the device and reproducibility, must be assessed by agreement between replicates.

(iii) Data demonstrating acceptable, as determined by FDA, clinical validation must be demonstrated by conducting studies with clinical specimens. For each clinical study, relevant details must be documented (e.g., the origin of the study slides and images, reader/annotator qualifications, method of annotation, location of the study site(s) (on-site/remote), challenging diagnoses, etc.). The studies must include:

(A) A study demonstrating the performance by the intended users with and without the software device (e.g., unassisted and device-assisted reading of scanned WSI of pathology slides). The study dataset must contain sufficient numbers of cases from relevant cohorts that are representative of the scope of patients likely to be encountered given the intended use of the device (e.g., subsets defined by clinically relevant confounders, challenging diagnoses, subsets with potential biopsy appearance modifiers, concomitant diseases, and subsets defined by image scanning characteristics, etc.) such that the performance estimates and confidence intervals for these individual subsets can be characterized. The performance assessment must be based on appropriate diagnostic accuracy measures (e.g., sensitivity, specificity, predictive value, diagnostic likelihood ratio, etc.).

###

864.3755:  For Cancer Prognostics; Pending since Paige Prostate in 2025...

WISeR Prior Authorization - It's Worse Than We Knew. Case study: DBS in PD.

Announced last summer, CMS has now instituted the WISeR program to force prior authorization into Medicare Part B using outside venders and their software.

There's a new piece about it in JAMA Internal Medicine currently.  Kannarkat et al.  They note, 

• CMS puts forth a thoughtful effort through WISeR to reduce health care waste....With WISeR, AI tools prescreen PA documentation for completeness and match against Medicare coverage rules before human review,

Well, wait a minute.  Does anyone really read what CMS has done?   The procedures include some with no CMS rules (for erection therapy, an old NCD says, basically, "Impotence therapy is covered when it is medically necessary.")   That AI Prior Auth software may well implement ten or twenty rules and rejection triggers, but they're not CMS rules, 'cause there aren't any.

Even more alarming is the inclusion of deep brain stimulation in Parkinson's.  CMS's NCD for DBS is approaching 25 years old - 2003 - and is grossly outdated.   See a Chat GPT assessment of this policy mess, below.  (See also an Opus 4.7 assessment).

Nobody seems to have even noticed these obvious problems in the months of large teams working on WISeR. 

(Admittedly, CMS has delayed implementation of DBS in WISeR, but it took nine months to do so, and the reason doesn't seem to have anything to do with the outdated NCD.)

Chat GPT on DBS Outdated Rules

It's a mess.  The CMS NCD is strikingly frozen in 2003 language. It still covers the basic durable points — idiopathic PD, levodopa responsiveness, STN/GPi targets, movement-disorders involvement, and exclusion of atypical parkinsonism — but a modern DBS center would think about DBS much differently than the NCD reads. CMS itself still lists NCD 160.24 as Version 1, effective April 1, 2003, with no ending effective date. (Centers for Medicare & Medicaid Services)