MedTechEurope: 169-page eBook on Clinical Testing Requirements in Europe

 MedTech Europe has released a free, 169-page eBoook on "Clinical Evidence Requirements under the EU IVDR Regulation."

Find it here:

https://www.medtecheurope.org/wp-content/uploads/2020/05/clinical-evidence-requirements-ebook_fourth-edition.pdf

Here's the TOC (click to enlarge):

click to enlarge TOC

AI Corner:

Here’s a big-picture overview of the MedTech Europe eBook – Clinical Evidence under IVDR (4th ed., 2025).

---

The MedTech Europe eBook provides a practical, industry-oriented guide to meeting the clinical evidence requirements of the EU In Vitro Diagnostic Regulation (IVDR 2017/746). It explains how manufacturers must show that their devices are not only technically reliable but also clinically relevant and safe in the context of modern medical practice. 

 At the heart of IVDR is the concept of performance evaluation as a continuous process built on three linked elements: scientific validity (showing that the biomarker or analyte is genuinely associated with a clinical condition or physiological state), analytical performance (the ability of the test to measure that analyte accurately and reproducibly), and clinical performance (evidence that the test’s results correspond to the clinical state or outcome it claims to inform). Together, these elements constitute the clinical evidence that must underpin a device’s CE marking and ongoing benefit-risk profile.

A recurring theme is the central role of the device’s “intended purpose”—what is measured, in which sample, for which patient population, and for what clinical function (diagnosis, screening, monitoring, therapy guidance). This declared purpose drives the design of studies, the choice of performance indicators, and post-market follow-up. Equally important is the need to show that the device remains aligned with the “state of the art in medicine”—not necessarily the newest technology, but the practice currently regarded as good and effective care in Europe. As medical practice and competing technologies evolve, manufacturers are expected to monitor whether their own devices continue to offer a favorable benefit-risk profile.

The document stresses a risk-based approach. Higher-risk IVDs (notably Class C and D devices such as companion diagnostics, infectious-disease assays, and complex genomic tests) typically require more robust clinical performance data—often from prospective or multi-site studies—while lower-risk devices may rely more on literature and routine-use data. Real-world evidence and published experience can supplement or, in some cases, substitute for formal clinical performance studies, provided the data are of suitable quality. The guidance also clarifies that clinical utility—the impact of the test on patient management—is generally not demanded by IVDR, except in special cases such as companion diagnostics.

Overall, the eBook highlights the cultural and regulatory shift from the older IVDD regime, which emphasized technical specifications, to the IVDR framework, which places clinical relevance and continuous evidence generation at the center of IVD regulation. It positions compliance not as a one-time hurdle but as an ongoing process that links the declared purpose of a device to the scientific understanding of the biomarker, to demonstrable technical reliability, and finally to its role in real clinical decision-making within current medical practice.

AMA Releases October 1, 2025 Edition of New PLA Codes

 On October 1, 2025, AMA released the latest quarter's worth of new PLA codes.  In its summary, AMA states there are 10 codes deletions and 14 new codes.  Several other codes have minor revisions.

Find the code list here:

https://www.ama-assn.org/system/files/cpt-pla-codes-long.pdf

JAMA's Call for Better Artificial Intelligence Papers

On September 30, 2025, JAMA issued a call for papers on AI, with a long wish list.

https://jamanetwork.com/journals/jama/fullarticle/2839356

It's open-access, so I've clipped the request below, which shows the span of issues for these editors.  Since it's fairly long, I've also included a short AI summary at top.

### AI Summary

AMA highlights that AI in medicine is shifting from proof-of-concept to rigorous clinical evaluation. 

With over 1000 FDA-cleared AI/ML devices already in use, the field now needs evidence of impact, safety, equity, and efficiency. The JAMA Network seeks manuscripts on clinical applications of AI, including head-to-head outcome comparisons, workflow and equity effects, ambient documentation tools, public trust, and generalizability across populations. 

Submissions should follow TRIPOD+AI, DECIDE-AI, and CLEAR reporting standards, and external validation is emphasized. Readers and authors are invited to explore ongoing insights via JAMA+ AI (jamaai.org).

##

JAMA Editors:

Few technologies have generated enthusiasm in medicine as rapidly and intensely as artificial intelligence (AI), often outstripping the research community’s capacity to study its impact. The explosion in interest in both research and implementation of AI in medicine is driven by a leap in the ability of these tools to analyze and synthesize data derived from written text, wearable technology, and images. For example, deep learning–based image analysis has enabled the prediction of pathologic gene sequence variations directly from digital histopathology slides in oncology¹ and demonstrated that insights exceeding expert human capability could be derived in specific use cases. To date, the US Food and Drug Administration (FDA) has authorized for marketing more than 1000 AI- and machine learning–enabled medical devices² designed for a wide range of applications, such as predictive analytics, clinical decision support, and deep phenotyping. The ability of large language models to recognize patterns in unstructured text and speech has already been used to dramatic effect, such as ambient scribing in clinical settings.

The journals of the JAMA Network believe AI applications will play an increasingly important role in the science and practice of medicine. We, as a network, are committed to ensuring that clinicians and policymakers have the evidence they need to make informed decisions about how and when they apply these new technologies. The application of AI to medicine is at a point of inflection, transitioning from proof-of-concept studies that demonstrate the potential of these technologies to later-stage investigations that apply more rigorous methodologies and aim to establish impact, safety, equity, efficacy, and efficiency.

Across the JAMA Network, we are seeking submission of manuscripts describing original research that examines the application of AI in clinical settings across all of these stages. We seek to support innovation across medicine and public health while upholding the highest standards for scientific rigor. Although adequately powered, prospective randomized clinical trials remain the gold standard in clinical medicine, we also welcome alternative study designs for the evaluation of AI tools, such as real-world evidence studies and proof-of-concept studies that involve validation by independent, external datasets. External, independent validation is particularly crucial for AI models used for risk stratification and treatment recommendation. Although the JAMA Network journals are generally interested in clinically impactful studies, the following list of topics will also be of particular interest:

• Head-to-head comparisons of clinical outcomes between various workflows or strategies using AI, compared with standard of care or other gold standards.

• Translational, computational to bedside AI studies that evaluate how AI tools are used in practice (including “off-label use”) and how the deployment of FDA-cleared clinical AI tools affects population-level metrics beyond efficacy, including workflow, referral patterns, access, equity, bias, and quality metrics.

• AI products or tools focused on increasing efficiency and reducing health care professional workloads, such as ambient scribes, clinical documentation summarization, and generative AI products for revenue cycle management.

• Public or patient perception and trust in specific AI tools or use of these tools in clinical workflows.

• Evaluation of existing AI tools—rather than repeatedly reinventing them from scratch—across diverse populations to ensure their generalizability beyond the specific cohorts originally studied.

We expect AI manuscripts submitted to the JAMA Network will adhere to the appropriate reporting standards and guidelines outlined by frameworks³ such as TRIPOD+AI, DECIDE-AI, and CLEAR. For additional AI-related studies and insights, we encourage readers and authors to explore JAMA+ AI at jamaai.org. This channel features in-depth AI research, conducts exclusive interviews with leading experts and authors, and highlights the diverse studies published across the entire JAMA Network, including, we hope, those this call for submissions will elicit.

Judge Throws Out Medicare Rule that Changed Too Much, Proposal - to - Final

One of the principles of notice and comment rulemaking - and it's applied less formally to LCDs and NCDs - is that a rule cannot change too much from proposal to final, even though edits to the proposal are the point of notice and comment rulemaking.

In recent years, Novitas put out a draft LCD for oncology biomarkers, radically changed the final, and CMS required it to put out the "new" final as a "proposal" for public comment.   I think something similar may have occurred with a MolDx LCD regarding transplant rejection genomics.

Here's a new example of the same principle playing out in federal rulemaking.

CMS published a Medicare Advantage rule about audits, and changed some key points from draft to final.  Judge Reed O'Connor of the Texas Northern District tossed out the rule, based on excessive change.

(Litigants may have chosen Texas to file in the hopes it would rule against Yankee regulators.)

Here's coverage:

• Healthcare Dive here.
• The 16 page decision here.
• Stat Plus (subscription) here.

Healthcare Dive summarizes,

Humana said that regulators changed their justification for reversing the fee-for-service adjuster between the rule’s proposed and final versions. Without the adjuster, MA plans would be unfairly penalized since doctors in traditional Medicare may not be incentivized to log all of their patients’ health conditions, the insurer argued.

On Thursday, O’Connor agreed.

“Because there was no meaningful notice of Defendants’ ultimate finding that actuarial equivalence does not apply to RADV audits, there was no meaningful dialogue regarding the costs and benefits of the surprise changes,” O’Connor wrote in his ruling, arguing that the harm to insurers was exacerbated by the rule’s retroactive application back to 2018.

###

AI CORNER

###

I ask Chat GPT 5 to assess the 16 page ruling.

###

Here is a paragraph-style summary for a readership of health-care policy experts, most of whom are attorneys, emphasizing the limits the court placed on CMS’s authority to alter the content and justification of a final rule. I include the full controlling citation for the “logical outgrowth” doctrine.

---

The decision by Judge Reed O’Connor in Humana Inc. v. Becerra, No. 4:23-cv-00909-O (N.D. Tex. Sept. 25 2025), vacated and remanded the 2023 CMS RADV Final Rule because the agency violated the Administrative Procedure Act’s notice-and-comment requirement. The APA requires that the final rule be a “logical outgrowth” of the proposal placed before the public during rulemaking. The leading Supreme Court case is Long Island Care at Home, Ltd. v. Coke, 551 U.S. 158, 174 (2007), which held that “[a] final rule need not match the proposed rule exactly, but it must be a logical outgrowth of the proposal so that the affected parties should have anticipated the final course in light of the initial notice.” The Fifth Circuit has repeatedly applied this standard, for example in Texas Association of Manufacturers v. U.S. Consumer Product Safety Commission, 989 F.3d 368, 381 (5th Cir. 2021), describing the test as whether the proposal “adequately frame[d] the subjects for discussion such that the affected party should have anticipated the agency’s final course in light of the initial notice.” The doctrine protects both regulated entities and the integrity of the rulemaking process by ensuring meaningful participation.

In this case the court concluded that CMS crossed the line from permissible modification into impermissible surprise. The 2018–2019 proposed rules explained the plan to drop the long-standing Fee-for-Service (FFS) Adjuster on two empirical or equitable grounds: CMS’s own study suggested that diagnosis-coding error in traditional Medicare did not materially distort MA payments, and CMS feared it would be inequitable to correct audited MA plans but not others. Those were the propositions on which the public was invited to comment. In the 2023 Final Rule, however, CMS offered two different core justifications: that the statutory requirement of “actuarial equivalence” in 42 U.S.C. § 1395w-23(a)(1)(C) simply does not apply to RADV audits as a matter of law, and that the separate Coding-Intensity Adjustment already accounted for any bias. That switch in legal theory—away from the empirical and equity-based rationale of the proposal and toward a new statutory interpretation—was held not to be a logical outgrowth of what had been proposed. The court stressed that nothing in the proposal gave regulated parties fair warning that CMS was reconsidering more than a decade of precedent recognizing actuarial-equivalence obligations in RADV audits.

The court also found it significant that CMS’s new interpretation rested in part on a D.C. Circuit decision (UnitedHealthcare v. Becerra, 16 F.4th 867 (D.C. Cir. 2021)) that was issued after the comment period closed. Stakeholders could not have anticipated the agency’s shift, and there was therefore no meaningful opportunity to submit comments on the costs, benefits, and reliance interests affected by abandoning the FFS Adjuster. The harm from the lack of notice was compounded by the rule’s retroactive application to 2018, which exposed MA organizations to substantial unanticipated liabilities for prior years in which they had operated under the old rule.

Under the governing doctrine, agencies retain wide latitude to adjust details, respond to evidence in the record, and select among approaches squarely placed before the public in the proposal. But they may not use notice-and-comment as a placeholder and then unveil in the final rule an altogether new controlling legal theory or an interpretive reversal of long-standing practice. When the final rule “repudiates or is irreconcilable with a prior legislative rule, the second rule … must itself be legislative,” requiring full notice-and-comment rulemaking. See Flight Training International, Inc. v. FAA, 58 F.4th 234, 241 (5th Cir. 2023).

The practical message for both government and industry is that Medicare can refine and even significantly modify a proposed rule so long as the shift is a reasonably foreseeable outgrowth of the proposal, but it cannot substitute an unforeseen statutory theory for the one on which comments were invited. Courts will vacate such rules even if the agency’s new interpretation might prove substantively sound. The opinion underscores that procedural fidelity to the APA remains as important as substantive correctness when CMS attempts to revise the payment and audit architecture of the Medicare Advantage program.

##

• How did CMS try to justify the shift?

##

The court’s opinion makes clear that CMS and HHS did try to defend the change, but the justifications they relied on were, in the court’s view, too different from what they had proposed back in 2018-2019 and too weak to satisfy the APA.

First, the agencies argued that their new legal rationale—that the actuarial-equivalence requirement in 42 U.S.C. § 1395w-23(a)(1)(C) simply does not apply to RADV audits—was really implicit in the proposal all along. They pointed to isolated sentences in the 2018 proposed rule suggesting that a RADV-specific adjustment might not be “appropriate” because the purpose of RADV audits is to identify diagnoses unsupported by the medical record. They also relied on a short request for public comment in the 2019 notice that asked whether the statute “mandates or prohibits” an FFS Adjuster. CMS argued that these references were enough to alert the industry that it was reconsidering the statutory foundation for the adjuster.

Second, the government argued that even if there was a procedural misstep, it was “harmless” because its new reading of the statute was, in its view, the “best reading,” particularly after the Supreme Court’s decision in Loper Bright Enterprises v. Raimondo, 603 U.S. 369 (2024), which curtailed Chevron deference. Essentially, CMS contended that because it was now correctly interpreting the statute, there was no need to redo notice-and-comment.

Third, in support of the change CMS cited the D.C. Circuit’s decision in UnitedHealthcare Insurance Co. v. Becerra, 16 F.4th 867 (D.C. Cir. 2021), which emphasized that the actuarial-equivalence mandate governs prospective payments and not necessarily retrospective audits. CMS referred to that case nineteen times in the preamble to the 2023 final rule, but the Fifth Circuit-bound Texas court found that relying on a case decided after the comment period closed only underscored that the public had lacked notice of the intended shift.

Judge O’Connor described this as a classic “surprise-switcheroo.” The government had invited comments on the equitable and empirical rationale for dropping the adjuster—whether it was fair or unfair to apply it only to audited plans—but then adopted a legal rationale that had never been squarely presented for comment. 

 The court rejected the argument that stray comments by a few stakeholders who anticipated the statutory argument were enough; under Texas Ass’n of Manufacturers v. CPSC, 989 F.3d 368, 381–83 (5th Cir. 2021), the agency must give notice sufficient for the general body of interested parties to foresee the agency’s final course.

Finally, the court dismissed the harmless-error argument as inconsistent with the APA’s requirement that agencies must not only reach the right substantive result but must do so through a procedure that gives the public a meaningful chance to comment on the costs, benefits, and reliance interests at stake. The procedural defect was especially serious here because CMS made the new policy retroactive to 2018, exposing MA organizations to large liabilities for past years in which they had acted under the old policy.

In short, CMS tried to justify its shift by saying that the seeds of the legal argument were already in the proposal and by invoking intervening case law and a post-Chevron view of the statute. The court found those efforts unpersuasive because the rulemaking record gave no meaningful notice that the agency might abandon its long-standing view that actuarial-equivalence applied to RADV audits, and because the lack of notice deprived stakeholders of the opportunity to comment on the costs and consequences of that change.

• But remember, too, the whole point of N&C rulemaking is to encourage changes in the proposal and incorporate public comment into the final.

You are exactly right — the APA does not freeze agencies to the wording of the proposed rule. The entire purpose of notice-and-comment rulemaking is to give agencies the benefit of public input and allow them to refine, improve, and even substantially alter their proposals before they are finalized. Courts have long recognized this. The Supreme Court in Long Island Care at Home, Ltd. v. Coke, 551 U.S. 158, 174 (2007), emphasized that a final rule “need not match the proposed rule exactly,” so long as it is a “logical outgrowth” of the proposal such that affected parties “should have anticipated the agency’s final course in light of the initial notice.”

What the Texas court underscored in Humana v. Becerra is that the kind of change matters. Changes in details — for example, the sampling size for audits, the choice of statistical formula, or even the weight given to certain empirical findings — are exactly what the notice-and-comment process is meant to elicit. If commenters raise concerns about data quality or fairness, the agency can respond by adjusting the methodology or even reversing its tentative approach, and those changes will usually be upheld as a legitimate outcome of public participation.

By contrast, the court viewed CMS’s 2023 RADV rule as an abrupt shift in the basic legal theory underlying the rule: the agency moved from saying that the FFS Adjuster was empirically unnecessary and potentially inequitable to saying that the statute never required actuarial equivalence in audits at all. That change was not a refinement of the same concept but, in the court’s words, a “surprise-switcheroo.” Because nothing in the proposal clearly signaled that the agency was reconsidering the statutory mandate itself — a 13-year-old premise of the RADV program — affected parties had no fair opportunity to focus their comments on that fundamental question. The court therefore concluded that the final rule failed the logical-outgrowth test.

So the case does not hold that agencies must freeze their proposals in place. It reiterates the traditional line: changes that stakeholders could reasonably foresee and comment on are permitted; changes that introduce a wholly new rationale or legal foundation without notice are not. That distinction preserves both the flexibility agencies need to learn from comments and the fairness and deliberation that the APA is designed to guarantee.

##

Prompt

For health care policy experts, most of whom are also attorneys, please summarize the ruling. Focus on when Medicare can, and cannot, adjust and edit proposals for the final version. Please write in paragraphs rather than bullets. Please give a full citation for the rule about 'natural outgrowth' of proposal.

##

Judge O'Connor ruled the Affordable Care Act unconstitutional in 2018.

##

Here's a footnote from the current ruling.

Rapid Advances in Alzheimer's Disease: A Quintuplet of Articles

There's an interesting quintuplet of articles on Alzheimer's disease in recent weeks.  The message is: this field is updating fast.

• At STAT-PLUS, Karlawish (U Penn) summarizes recent changes here.
• At Alz Dis Rel Disord, Palmqvist and colleagues provide a new 2025 guideline for blood-based Alzheimer diagnostics here. 
• At Lancet, see a three-part open-access series on Alzheimer's:
• Frisoni, new diagnostics here.
• Fox, new therapies here.
• Firsoni, new challenges here.

While the focus here is on Alzheimer's disease, advances in Parkinson's are significant too.   

• In JAMA Neurology,  Starr et al. report long-term favorable five year outcomes for PD patients treated with subthalamic deep brain stimulators. Multiple clinical axes of benefit.  Here.
• Also in JAMA Neurology, Bronte-Steward et al. report favorable outcomes for "adaptive" deep brain stimulation in PD.  Here.
• Multiple groups report high accuracy of alpha-synunclein seed assays in PD diagnosis, including differentiating related disorders.  E.g. Orru et al. Lancet Neurol 2025 here.  Ma et al (in multi-system atrophy) here.   Siderowf in Lancet Neurol 2023 here.

###

AI CORNER

I've taken the Alzheimer articles listed above, given them to Chat GPT 5, and asked for a summary.

###

The field of Alzheimer’s disease is undergoing its most significant transformation since the discovery of amyloid and tau pathology. For the first time, widely accessible blood-based biomarkers (BBMs) have achieved diagnostic performance sufficient to guide clinical decision-making, while disease-modifying anti-amyloid therapies—though still of modest effect and with notable safety and cost concerns—have entered routine practice in multiple countries. This simultaneous advance in diagnostics and therapeutics is reshaping clinical pathways, reimbursement policies, and the culture of dementia care. Understanding this inflection point requires integrating insights from the new Alzheimer’s Association guideline for BBM use, the Lancet three-part Series that frames the evolving landscape of diagnosis, treatment, and ongoing controversies, and the STAT keynote commentary by Karlawish, which highlights the profound societal and policy implications. Together, these sources chart how Alzheimer’s medicine is moving from late-stage recognition and largely supportive care toward an earlier, biology-driven, and potentially preventive model—a shift that demands attention from neurologists, health-system leaders, and policymakers alike.

###

###

Here is a comprehensive postgraduate-level synthesis of the five requested sources—the Alzheimer’s Association 2025 Clinical Practice Guideline on Blood-Based Biomarkers (Palmqvist et al.), the Lancet three-part Series on the new landscape of Alzheimer’s disease (diagnosis – Frisoni et al.; treatment – Fox et al.; controversies & future directions – Frisoni et al.), and the STAT “Neurotransmissions” keynote essay by Karlawish.

The summary is tailored for clinicians, neuroscientists, and policy leaders in dementia and Alzheimer’s care.

---

1. New Blood-Based Diagnostic Guideline (Palmqvist et al., Alzheimer’s & Dementia 2025)

Scope and rationale
The guideline responds to the clinical and policy need for widely accessible, less-invasive diagnostic tools as disease-modifying therapies now require biomarker confirmation of β-amyloid pathology for eligibility. Traditional methods—amyloid-PET and CSF assays—remain expensive or logistically limited.

Methodology
An Alzheimer’s Association-convened panel applied a GRADE-based systematic review of diagnostic-test-accuracy studies comparing plasma p-tau isoforms (p-tau217, p-tau181, p-tau231), Aβ42/40 ratios, or derived indices against PET, CSF, or neuropathology.

Key performance-linked recommendations

• Triaging role: BBMs with ≥90 % sensitivity and ≥75 % specificity can be used to identify patients with cognitive impairment in specialty memory clinics who should proceed to confirmatory CSF/PET testing.

• Substitution role: BBMs with ≥90 % sensitivity and specificity may replace CSF or PET in specialized-care diagnostic work-ups.

• Clinical integration: BBMs are adjuncts, not stand-alone tests; interpretation must consider pre-test probability, comorbidities, and local assay performance.

• Variability and evidence gaps: Many commercial assays do not yet meet thresholds across diverse populations; ongoing living-guideline updates will track new evidence and multiplex approaches (eg ratios, multi-threshold algorithms).

Policy significance
The guideline is brand-agnostic and performance-based, enabling payers, regulators, and health-system leaders to align coverage decisions with objective accuracy benchmarks and with equitable access goals.

---

2. Lancet Series, Part 1 – Diagnosis in the New Era (Frisoni et al., Lancet 2025)

Paradigm shift
Alzheimer’s diagnosis is moving from clinical-syndromic to clinical-biological: cognitive impairment plus biomarker evidence of β-amyloid and tau is now required both for diagnostic certainty and for eligibility for anti-amyloid monoclonals.

Impact of biomarkers

• Traditional clinical assessment alone had ~60–70 % accuracy; adding PET/CSF raises it to ~90–95 %, with earlier detection.

• The arrival of scalable blood biomarkers is anticipated to be a “diagnostic revolution,” enabling earlier, broader, and more equitable case-finding across health-care systems.

• Revised 2024 Alzheimer’s Association criteria propose an integrated bio-clinical staging framework—recognizing that pathology and symptoms evolve asynchronously, affecting clinical trial design and care planning.

Clinical pathway
A typical patient journey in memory clinics still begins with clinical and neuropsychological evaluation and structural imaging; biomarker confirmation (traditionally CSF/PET, increasingly plasma) now plays a decisive role for prognosis, counseling, and treatment access.

---

3. Lancet Series, Part 2 – Treatment Landscape (Fox et al., Lancet 2025)

Background therapies

• Cholinesterase inhibitors and memantine—long-standing symptomatic agents—remain standard but have modest and context-dependent benefits and variable reimbursement across countries.

• Non-pharmacologic strategies for behavioural & psychological symptoms of dementia (BPSD)—eg personalized activity, caregiver training, environmental adaptation—have demonstrated benefit but are under-implemented in real-world care due to workforce and resource barriers.

Disease-modifying agents

• Anti-β-amyloid monoclonal antibodies (lecanemab, donanemab) now have regulatory approval in the US, EU, UK, China, Japan, and others, representing the first class to slow cognitive-functional decline (by ~27–39 % over 18 months in phase-3 trials).

• Integration with routine care remains challenging because of eligibility constraints (confirmed β-amyloid positivity), infusion logistics, ARIA risk monitoring, costs, and health-system fragmentation.

Care priorities
Guidelines emphasize first addressing social, somatic, and behavioural issues; cognitive-symptom and disease-modifying therapies should be delivered in the context of comprehensive, individualized care plans.

---

4. Lancet Series, Part 3 – Controversies and Next Steps (Frisoni et al., Lancet 2025)

Debate over benefit–risk and value

• Early anti-amyloid experience with aducanumab bred skepticism; lecanemab/donanemab show replicable though modest slowing of decline, with treatment-related ARIA in 10–18 % and treatment-associated deaths in a small fraction, especially in patients receiving anticoagulants or thrombolytics.

• Policy and expert communities remain split between “breakthrough” and “insufficient benefit-for-risk-and-cost” camps, unlike in oncology or multiple-sclerosis where similar-magnitude biologic effects were accepted.

Converging innovations

• Blood-based biomarkers (p-tau217, Aβ42/40) now demonstrate PET-level accuracy and are already in clinical use in several countries; they lower barriers to biologic-therapy eligibility and to large-scale secondary-prevention trials.

• Digital biomarkers and remote cognitive assessments promise more sensitive, scalable screening.

• A long-term vision is precision, multi-target therapy (amyloid plus non-amyloid pathways) coupled with population-level risk-reduction and prevention programs.

Societal perspective
The paper highlights the interplay of science, economics, and culture—for decades Alzheimer’s care relied on low-tech tools; now the field faces expensive diagnostics and therapeutics, demanding new value frameworks and equitable resource allocation.

---

5. STAT “Neurotransmissions” Keynote – A Cultural & Policy Turning Point (Karlawish, STAT 2025)

Framing the challenge
Karlawish underscores that care follows diagnosis, yet historically ~40 % of US patients with documented dementia were not even told the diagnosis, and worldwide under-diagnosis approaches 60 %.
Cultural reluctance to name Alzheimer’s disease—linked to fear, stigma, and use of “dementia” as a political epithet—has impeded timely evaluation and planning.

Impending change

• With accessible biomarker tests and available disease-modifying therapies, the “walls are coming down”: the field is entering a period when getting a diagnosis becomes an enabling, not a dooming, act.

• Karlawish argues that society must now grapple with ethical, social, and policy challenges of living well with diagnosed cognitive disorders, including support for patients and caregivers and the equitable deployment of novel tools.

• He calls this moment the dawn of a revolution in how we talk about, detect, and manage dementia.

---

Discussion.
Integrated Perspective for Neurology & Policy Leadership

1. Convergence of Diagnostics and Therapeutics
    The simultaneous maturation of reliable plasma biomarkers and disease-modifying biologics re-defines the Alzheimer’s care pathway: from late-stage recognition to earlier biological detection and targeted intervention.

2. Implementation Science & Health-System Readiness
    Scaling BBMs and biologics requires validated high-throughput assays, clinician education, diagnostic-therapeutic algorithms, infusion infrastructure, ARIA-monitoring capacity, coverage policies, and cost-effectiveness frameworks.

3. Equity & Global Health Considerations
    Given the worldwide diagnosis gap and variation in access to symptomatic drugs and specialist care, the democratizing potential of BBMs is high but contingent on quality-assured assays, workforce training, and coverage policy—especially in LMICs.

4. Evolving Policy & Regulatory Landscape
    Living-guideline models (as exemplified by the Alzheimer’s Association CPG) and adaptive payer policies will be crucial to keep pace with rapid evidence evolution and to balance access with value and safety oversight.

5. Cultural & Ethical Dimension
    Normalization of biomarker-based diagnosis will challenge societies to destigmatize Alzheimer’s disease, address advance-care planning, and invest in long-term care and caregiver support infrastructures.

---

Bottom-Line

These five contributions collectively depict Alzheimer’s disease at an inflection point:

• Biological diagnosis is becoming routine, led by blood-based biomarkers that can triage or substitute for invasive and expensive tests.

• Disease-modifying therapy is clinically real but modest in effect and not without risk and cost, demanding nuanced risk–benefit assessment and system-level preparedness.

• Policy, ethics, and culture now matter as much as molecular science—successful translation will hinge on equitable access, health-system adaptation, transparent communication, and sustained investment in supportive care and prevention.

Together, these sources argue that the management of Alzheimer’s disease is moving from a late-recognition, largely supportive paradigm to an early-detection, biology-informed, progressively preventive model—a shift with profound implications for neurology practice, health-policy strategy, and societal attitudes toward cognitive aging.

#

#

Prompt. You are writing at the medical postgraduate level and for experts in neurologic disease and policy for Alzheimer's. Summarize these articles, which include a new guideline for blood-based diagnostics in Alzheimer's. There is also a timely new set of 3 articles from Lancet on diagnostics, on therapies, and on controversies and next steps. Finally, there is a keynote op ed in STAT, discussing that we are on the brink of major changes in management of Alzheimer disease. Write an informative article summarizing the five sources.

Ryan Fukushima Essays on Precision Medicine

 Ryan Fukushima (Linked In here) is COO of TEMPUS AI, which has a $13B market cap and $700M in 2024 revenue.  

I've seen his occasional Linked In essays in the past year on precision medicine & innovation & strategies.  I just discovered his longer set of essays at Substack - https://ryanfukushima.substack.com/about  

Here are a few titles followed by an AI summary.

• About Ryan Fukushima here.
• The open source movement in pathology AI here.
• AI and Lab in the Loop medicine here.
• Multi modal AI beats Single source for cancer predictions here.
• Deploying clinical multi modal AI here.
• Typhoons to Tumors - Precision medicine's future here. 

####

AI CORNER

####

##
GOOGLE NOTEBOOK LM

##

Based on the provided articles and biographical statement, Ryan Fukushima, COO of Tempus AI, presents a sophisticated and forward-looking perspective on the intersection of precision medicine and artificial intelligence. His thinking, targeted at a scientifically literate audience, revolves around several core principles: the necessity of multimodal data integration, the power of AI to bridge the translational gap between preclinical research and clinical reality, the shift from deterministic to probabilistic models in patient care, and a pragmatic approach to deploying AI in complex healthcare ecosystems.

The Centrality of Multimodal Data Integration

A cornerstone of Fukushima's philosophy is that transformative insights in oncology arise from the intelligent integration of multiple data modalities, not from any single source. He argues that focusing on one data type, whether genomics or digital pathology, provides an incomplete picture of a patient's disease.

• Complementary Biological Signals: Fukushima emphasizes that different data types capture distinct and complementary aspects of tumor biology. For instance, a study he highlights showed that combining transcriptomics (dendritic cell abundance), genomics (NTRK1 expression), radiomics (metabolic tumor volume), and clinical data (neutrophil counts) painted a complete prognostic picture that no single biomarker could provide. This complementarity is crucial, as different modalities can provide opposing information for the same patient, and integrating them leads to significantly more powerful predictive models. A model combining clinical, pathomic, and RNA data yielded a log-rank p-value of 3.51e-09 for predicting 1-year mortality, far outperforming single-modality approaches.
• Early vs. Late Fusion: He champions "early fusion" methods in AI model development, where different data streams are integrated at the beginning of the analytical process. This allows the model to learn the "fine-grained interactions" between modalities—for example, how specific molecular pathways directly influence tissue morphology. Citing the SURVPATH model, he notes that this approach yields significant performance improvements (a 7.3% gain over leading models) compared to "late fusion" methods that analyze data separately and combine conclusions at the end.
• The Data Problem: Fukushima sees the primary challenge in the field not as an algorithmic one, but as a "data problem". He stresses the importance of bridging real-world, multimodal data at scale, using "source data" that is as unprocessed as possible to avoid introducing artifacts that could doom an AI model from the start. Organizations like Tempus, he notes, are working to solve this by building expansive, de-identified multimodal datasets that can power the next generation of models.

Bridging the Translational Gap with AI

Fukushima identifies the "translational gap"—the failure of promising preclinical findings to translate into effective human therapies—as one of medicine's most persistent and costly challenges. He sees AI as an essential tool for bridging this gap.

• Improving Preclinical Models: He highlights the fundamental transcriptional differences between mouse models and human tumors, which often render preclinical efficacy studies unpredictive. To address this, he points to advanced deep learning methods like Novartis's MOBER (Multi-Origin Batch Effect Remover). MOBER can simultaneously analyze data from cell lines, patient-derived xenografts (PDXs), and clinical tumors to identify which lab models have the greatest "transcriptional fidelity" to real cancers. This allows researchers to select better models, potentially saving years of research and millions of dollars.
• Transforming and Harmonizing Data: Beyond selection, AI tools like MOBER can computationally "transform" the expression profiles of preclinical models to better resemble clinical tumors. Fukushima finds this capability particularly compelling, citing an example where biomarkers identified in MOBER-transformed cell line data showed dramatically improved clinical translation and survival stratification in patient cohorts (P = 6.2× 10^-29). This AI-driven harmonization allows for more relevant preclinical experiments that can better de-risk future clinical trials.

Shifting from Deterministic to Probabilistic Frameworks

Fukushima argues for a paradigm shift in clinical decision-making, moving away from the "one-size-fits-most" approach of traditional evidence-based medicine toward a more personalized, probabilistic framework.

• The "Average Patient" Fallacy: He notes that evidence-based medicine is inherently deterministic, providing guidance that works "on average" but often fails to account for the unique biological complexity of individual patients. Most patients exist in a "clinical gray zone," where their individual characteristics may question the applicability of population-level findings.
• Learning from Other Complex Systems: Drawing powerful analogies from weather forecasting and traffic simulation, Fukushima illustrates the superiority of probabilistic models in complex, dynamic systems. He points to DeepMind's GenCast, a weather model that generates an ensemble of possible scenarios rather than a single deterministic forecast. The ensemble mean often "blurs out" critical details (like the precise path of a typhoon), much like population averages in medicine obscure the reality of an individual patient's disease trajectory.
• Quantifying Uncertainty: By embracing uncertainty, probabilistic AI models can quantify confidence in treatment recommendations, identify patients needing closer monitoring, and design adaptive treatment protocols. This approach is better suited for the inherent stochasticity of disease progression and treatment response.

Pragmatic Deployment of Clinical AI

While optimistic about AI's potential, Fukushima maintains a pragmatic and clear-eyed view of the challenges involved in deploying these tools in real-world clinical settings.

• AI as an "Intelligence Amplifier": He refutes the idea that AI is meant to replace clinical judgment, instead framing it as an "intelligence amplifier" that helps clinicians see hidden patterns in existing data. He uses the analogy of Google Maps: it doesn't drive your car but provides real-time information to help you make better decisions. The goal is to empower doctors with smarter systems that can make truly personalized decisions.
• The Open-Source Ecosystem: Fukushima is a strong advocate for the open-source movement in AI, particularly in digital pathology. He highlights foundation models like Paige's Virchow, Microsoft's Prov-GigaPath, and Bioptimus's H-Optimus-0, which are democratizing access to state-of-the-art tools and creating a global ecosystem where innovation can accelerate exponentially. This allows research teams to build on established foundations rather than reinventing the wheel.
• Beyond Model Building: He stresses that building a sophisticated model is meaningless if it cannot be deployed to impact patients. The true challenge lies in navigating technical barriers, regulatory hurdles (like the FDA clearance process), and workflow integration. His strategic framework for AI leaders emphasizes building on open foundations, focusing on clinical integration and validation, prioritizing multimodal data, and scaling through partnerships.

In summary, Ryan Fukushima's thought leadership is defined by a vision where the integration of comprehensive, multimodal data through sophisticated yet interpretable AI is not merely an academic exercise, but the fundamental driver of a more precise, personalized, and effective standard of care in oncology.

CHAT GPT 5

Here’s a synthesis of Ryan Fukushima’s recent essays and bio—the through-line is data realism + pragmatic deployment: use multimodal evidence, design for translation into clinic, and bias toward systems that adapt over time rather than static “point” models.

1) Why multimodal beats unimodal (and how to do it right). Across posts, Fukushima argues that single-source models (e.g., WSI-only) rarely capture the causal texture of disease. He highlights work like SURVPATH to show early-fusion architectures (a joint feature space where histology tokens “converse” with pathway tokens) outperform late fusion and deliver interpretable links between morphology and biology (e.g., EMT pathways ↔ image features). The practical takeaway for biotech teams: stop expecting any one modality to “rule them all”; build data pipelines that integrate transcriptomics + imaging + clinical from day one and keep interpretability first-class for translational adoption.

2) Deployment mindset: AI as an “intelligence amplifier,” not an intervention. He pushes back on calls for RCT-style evidence of the tool itself, reframing clinical AI as decision-support (like Google Maps): it surfaces patterns clinicians would otherwise miss, updates with new information, and helps tailor choices to patients whose profiles only partially overlap trial cohorts. He illustrates how feature importance is context-dependent (e.g., CRP risk flips with platelet strata) and how the salience of biomarkers shifts over the cancer journey—hence the need for adaptive, longitudinal models and robust guardrails (validate against outcomes, ensure explainability, bias monitoring, UI that turns signals into actions).

3) Lab-in-the-loop translation: fix the model-to-patient gap. Fukushima centers the preclinical translation problem and spotlights MOBER as a deep-learning bridge across cell lines, PDXs, and tumors (n≈932/434/11,159). MOBER aligns datasets by biological rather than batch origin, identifies which lab models actually resemble human tumors, and can transform preclinical expression profiles to better match clinical biology—improving survival stratification in TCGA (P≈6.2×10⁻²⁹). The point isn’t a single algorithm; it’s a playbook: pick the right models, transform when needed, and iterate against real clinical outcomes.

4) Systems thinking: from “point measurements” to path-level, probabilistic digital twins. Borrowing from traffic calibration and DeepMind’s GenCast weather work, he argues precision medicine should model journeys (trajectories across scales and time) rather than isolated markers. That implies path-level data integration, multi-scale calibration, and probabilistic forecasts that quantify uncertainty—exactly what clinicians need for rare but high-impact “tail events” (e.g., atypical responses, adverse events). Biological complexity and fragmented data remain hard problems; still, hybrid mechanistic-statistical modeling and improving compute suggest a tractable roadmap.

5) Digital pathology and open source: build on foundations, ship to clinic. He sees an inflection point as open models (e.g., Virchow/PRISM, PROV-GigaPath, UNI/CONCH, H-Optimus-0) push benchmark→ecosystem. But he’s blunt: great models mean nothing if they never reach patients. His “leader’s framework”: (i) build on open foundations to move fast; (ii) focus on clinical integration (validation, workflows); (iii) prioritize multimodal (pathology+genomics+clinical); (iv) scale through partnerships (regulatory-proven companies, hospital integrations). He underscores Paige’s FDA experience and Tempus’ data + deployment infrastructure as the route from proof-of-concept to clinical-grade.

6) “What elite teams know”: let the data choose the biomarker strategy. In oncology (e.g., NSCLC, variable immunotherapy response), he urges moving beyond simplistic single-gene heuristics toward complex multimodal biomarkers tuned for clinical applicability (sample availability, assayability) and predictive lift. With large, well-annotated de-identified datasets and tooling (e.g., curation, automated pipelines), organizations can industrialize biomarker discovery→validation and avoid chasing narratives that don’t survive contact with longitudinal outcomes.

7) “Multisource > single-source” is also a culture and process claim. He describes real rooms where pathology, genomics, and radiology “talk past each other” and argues the win condition is operational: converge teams and tools around shared multimodal abstractions, enforce interpretability, and stage roll-outs where benefits clearly exceed integration costs. That’s how you cross the research→clinic chasm.

8) Biography → strategic lens. Fukushima’s bio frames this as a builder’s agenda: Tempus (AI-enabled diagnostics + massive clinical/molecular library for real-time decisions and biopharma R&D), Pathos (AI to understand why patients respond/don’t, lowering trial failure), and >$1.4B raised across companies focused on “right treatment, right time.” His newsletter promises implementation frameworks, tech evaluations, and market/regulatory analysis—i.e., an operator’s view of where AI + precision medicine actually change care.

---

What this adds up to (for expert readers)

• Epistemology: Prefer path-level, probabilistic modeling over point estimates and averages; treat uncertainty as a signal to manage, not a nuisance to ignore.

• Architecture: Invest in early fusion multimodal models with transparent tokens/factors that map onto biology and clinic; make interpretability non-negotiable to speed validation and payer/regulator uptake.

• Translation: Operationalize lab-in-the-loop (model selection + transformation + outcome feedback) to shrink the preclinical↔clinical gap; measure gains in calibration and tail-risk detection.

• Strategy: Stand on open foundations, but win on deployment: workflow fit, regulatory path, and partnerships that get models into care settings—because impact = model × (integration & scale).

If your team already lives in WSI/NGS/registry land, Fukushima’s message is not “add one more model,” it’s re-platform around multimodal, adaptive, clinically-tethered systems—and judge success by what reaches patients, not by leaderboards.

##

##

SIDEBAR

I asked Google LM and Chat GPT 5 to answer my question, does Fukushima risk underestimating the problem of Overfitting, since he aims to apply ultra high dimension data to simple questions like "survival" or "best drug."  

I place the Google and CHAT answers in a sidebar.   

Then, I asked both Google and CHAT to compare the overfitting answers they generated, with each other.  

https://bqwebpage.blogspot.com/2025/09/does-high-multiplicity-ai-data-plus.html

##

And here, in another Sidebar, comedian Lewis Black summarizes it all.

https://bqwebpage.blogspot.com/2025/09/lewis-black-on-overfitting-high.html

Ferber et al: Autonomous Agents Reach Clinical Decision Making (Nature Cancer)

Thanks to Joseph Steward at Linked In for pointing out this paper.

Before I link it, you would also enjoy this important essay by Ethan Mollick on active agentic AI- here.

##

See:   https://www.nature.com/articles/s43018-025-00991-6   Ferber et al.

Here's the authors' abstract.  A special agentic tool made a big improvement in the success rate for GPT-4.

##

Clinical decision-making in oncology is complex, requiring the integration of multimodal data and multidomain expertise. We developed and evaluated an autonomous clinical artificial intelligence (AI) agent leveraging GPT-4 with multimodal precision oncology tools to support personalized clinical decision-making. 

The system incorporates vision transformers for detecting microsatellite instability and KRAS and BRAF mutations from histopathology slides, MedSAM for radiological image segmentation and web-based search tools such as OncoKB, PubMed and Google. Evaluated on 20 realistic multimodal patient cases, the AI agent autonomously used appropriate tools with 87.5% accuracy, reached correct clinical conclusions in 91.0% of cases and accurately cited relevant oncology guidelines 75.5% of the time.

Compared to GPT-4 alone, the integrated AI agent drastically improved decision-making accuracy from 30.3% to 87.2%. These findings demonstrate that integrating language models with precision oncology and search tools substantially enhances clinical accuracy, establishing a robust foundation for deploying AI-driven personalized oncology support systems.

###

They describe the operations this way:

We build and evaluate an AI agent tailored to interact with and draw conclusions from multimodal patient data through tools in oncology. Contrarily to the philosophy of an all-encompassing multimodal generalist foundation model, we see the achievements that specialist unimodal deep learning models have brought to precision medicine[21] as a viable template even in the future by equipping an LLM, specifically GPT-4, with additional functions and resources. These could be precision oncology deep learning models or the abil-ity to perform web search, all referred to herein as tools. Specifically, this study includes the vision model application programming interface (API) dedicated to generating radiology reports from magnetic resonance imaging (MRI) and computed tomography (CT) scans, MedSAM[22] for medical image segmentation and in-house developed vision transformer models trained to detect the presence of genetic alterations directly from routine histopathology slides[23], in particular, to distinguish between tumors with microsatellite instability (MSI) and microsatellite stability (MSS)[24] and to detect the presence or absence of mutations in KRAS and BRAF. Additionally, the system encompasses a basic calculator, capabilities for conducting web searches through Google and PubMed, as well as access to the precision oncology data-base OncoKB[25]. To ground the model’s reasoning on medical evidence, we compile a repository of roughly 6,800 medical documents and clinical scores from a collection of six different official sources, specifically tailored to oncology.

click to enlarge

###

AI CORNER  Chat GPT 5

###

Here’s a ~400-word expert-oriented summary of the Nature Cancer paper for your genomics-industry readers. It highlights what’s genuinely new and interesting beyond the abstract.

---

A Blueprint for Agentic AI in Oncology

Ferber et al. demonstrate that a modular AI agent built around GPT-4 and a suite of precision-oncology “tools” can outperform a stand-alone LLM for tumor-board–style decision-making. This is less about yet another LLM benchmark and more about how to orchestrate existing components—LLM, retrieval, domain-specific models, search, calculators—into an autonomous clinical-reasoning pipeline.

What the Agent Does

Rather than relying on a monolithic “generalist” multimodal model, the authors equip GPT-4 with specialized functions:

• Vision transformers trained on TCGA slides to infer MSI vs MSS and KRAS/BRAF mutation status.

• MedSAM for segmentation of CT/MRI lesions and a calculator to apply RECIST-style %-change.

• OncoKB API, PubMed, Google search, and a 6,800-document oncology-guideline library accessed through a retrieval-augmented generation (RAG) pipeline.

• Basic reasoning and planning—e.g., recognize a lesion in two scans, segment both, compute growth, then look up targeted-therapy options.

The agent autonomously chooses and sequences these tools (up to 10 per case), integrates results, and cites supporting evidence.

Performance Highlights

• On 20 realistic gastrointestinal-oncology cases with multimodal inputs, the agent achieved 87.2% completeness on expert-defined decision criteria versus 30.3% for GPT-4 alone.

• Tool use was successful in 87.5% of required calls; 91% of individual statements were judged correct, with only 2.4% potentially harmful.

• Citations were accurate in 75.5% of instances—an important step toward auditability and trust.

• Sequential tool chaining (e.g., MedSAM → calculator → RAG) proved critical to solving multi-step tasks.

• Open-weights LLMs (Llama-3 70B, Mixtral 8×7B) failed badly at function calling—success in only 39% and 8% of required calls—reinforcing that reasoning + tool orchestration remains a differentiator for proprietary frontier models.

Why It Matters

• Modular, update-friendly architecture: tools and guideline corpus can be swapped or refreshed without retraining the core LLM—key for keeping up with rapidly changing oncology standards.

• Explainability & regulatory alignment: each tool can be validated separately and its output inspected—more transparent than a single black-box model.

• Blueprint for workflow integration: the proof-of-concept suggests a path to embedding such agents into tumor-board software or EHRs, contingent on data-privacy, interoperability, and device-approval hurdles.

Limitations and Next Steps

Small-scale evaluation (20 cases) and reliance on single-slice images limit clinical readiness. Future work will need:

• Better-validated tools (e.g., clinical-grade MSI detectors, 3-D radiology models such as Merlin)

• Local, secure open-weights models for HIPAA/GDPR compliance

• Human-in-the-loop, multiturn interaction to capture real tumor-board dynamics

---

Bottom line: The paper moves the conversation from “Can a big LLM answer board-style questions?” to “Can we orchestrate specialist tools around an LLM to make safe, auditable, up-to-date recommendations?”—a shift with immediate relevance for developers of clinical-grade decision-support in genomics and oncology.

CLIA Issues Extensives Updates to Interpretive Manuals

 The statute and the regulations for CLIA are very dry, and often vague ("Lab director shall ensure appropriate controls and validations...").

However, CMS also issues a richly textured 379-page manual for actually running CLIA labs and CLIA inspections.  This is called, "Pub. 100-07 State Operations Provider Certification.'  It just got a 379-page update with numerous small changes, issued September 12, 2025.   

CMS writes,

• Revisions have been made to reflect comments and recommendations from regional office and State Agency surveyors, professional and accrediting organizations, other Health and Human Services components, and a GAO audit. 
• Changes have also been made to include partial deletions of certain subsections affected and to reflect the recent publication of regulations and updates in technologies within the field of laboratory medicine. 

For example, there are lots of little redline edits around failure of proficiency testing (PT).

page 80, PT rules, 493.803 edits

page 84, 493.821 microbiology

Guardant Health - Analysts Day - 163 Page Deck Online

This week, Guardant Health held its InvestorDay, with a multi-hour presentation of the company's latest plans.

Response seems favorable, with JP Morgan reiterating its favorable "overweight" rating.  Some of the key resources are online:

• News coverage here.
• Web page at Guardant here.
• There's a link for the 163-page PowerPoint, as well as for the streaming webcast.
• Direct to PPT here.
• It looks like Seeking Alpha has a full transcript free with registration, here.
• Shares have traded $58-62, for a market cap of about $7.5B.

I highlight two slides, an agenda slide and a strategy-overview slide, below, click to enlarge.

click to enlarge

enlarge

CAP TODAY: Where Does the FDA LDT Status Leave Labs?

 

The cover story in the September CAP TODAY features discussion with experts over the status of LDTs in the FDA and in other policies.   Enforcement of RUO kits, marketing claims, direct-to-consumer special rules, still make a complex environment.

Find it here:

https://www.captodaymag.com/captoday/library/item/september_2025/4294593/

Digital Pathology Too

Also of interest, there's an article on the importance of huge pathology slide archives in the new AI world.  And, an expert panel discusses momentum and issues in digital pathology.

All CAP TODAY articles are open-access.

###

AI CORNER

###

FDA and LDTs (CAP TODAY, Sept 2025, Karen Titus)

A federal court struck down the FDA’s sweeping plan to regulate laboratory-developed tests. The ruling focused only on the 1976 definition of "a medical device." Experts warn FDA authority remains and labs still face risks in areas like selling or using RUO products, Direct to consumer (DTC) testing, and intellectual property licensing.

#

Digital Pathology Momentum (CAP TODAY, Sept 2025; Bob McGonnagle interviews)

Leaders from the Digital Pathology Association report rapid adoption of digital platforms, with institutions like Moffitt and Mayo moving to fully digital workflows. Panelists noted strong drivers—AI integration, workforce needs, and consolidation—though reimbursement gaps and uneven adoption across regions remain challenges.

#

Pathology Archives and AI Future (CAP TODAY, Sept 2025, Ann Paxton)

The [US Military] Joint Pathology Center’s vast repository of 55–60 million slides and tissue blocks is being digitized to train AI tools for cancer, infectious disease, and rare disorders. Leaders expect these algorithms to improve diagnostic accuracy and speed, while also raising questions about data governance, ethics, and equitable access.  In Silver Spring MD; see  https://www.jpc.capmed.mil/

Category III Codes in CY2024: Are They Working?

See my September 4, 2025 blog, where CMS releases Part B utilization for CY2024 by CPT code.  I discussed utilization of lab codes there.

Category III Codes

Let's revisit the 2024 data and ask about Category III codes.  Category III codes are now breaking 900 in terms of total codes issued.  While many have been "sunsetted" some Category III codes as early as 0042U are still in use.

For 2024, CMS lists about 370 Category III codes in the utilization data, but about 100 of those have utilization of $0 and another 100 or so are below $1000.  So, only 170 of 370 codes had utilization over the low bar of $1000.

Total payments in 2024 for Cat III were about $225M.  

Generally, prices are set locally, so they may differ MAC to MAC.  Of the top 7 codes, 1 paid over $5000, while 3 paid in the $2100-2700 range.

Utilization of Top Category III Codes

Let's look at utilization from the top down.

Category III code utilization is extremely skewed, with two-thirds of the money (66%) going to just the  top 5 of 370 codes. The top 7 codes get 75% of the money ($166M).   The top 25 codes are 95% of the money, leaving just 5% of the money for the other ~350 codes.

click to enlarge

Most of the top codes are surgical, either eye surgery or back surgery.   

• Code 0275T for lumbar laminectomy had a mean fee of $2723, 26,108 services, paying $71,087,000.
• Code 0627T for cellular product injection, disk, had a mean fee of $5,263, 8,469 services, and paying $44,574,000.
• 0394T for skin brachytherapy (radiation) had a mean fee of $186, 69,000 services, and paying $12,864,000.

##

I understand AMA CPT has been putting (non pathology) codes using software, AI, machine learning, here in Category III.   I will try to come back to the spreadsheet, flag those codes, and run a separate spending analysis on them.

##

AMA and the Shrinking RVU: Medicare Payments to Physicians Badly Lag Inflation

A major theme at the AMA is the severity of the gap between inflation and physician payments based in RVUs.   This affects a lot of things that are paid outside of hospital budgets - for example, community imaging services and community pathology services.    

I would argue this also impacts how AMA views innovation, such as artificial intelligence-based diagnostics in radiology and cardiology.  One issue is whether automation will speed up physician work (and therefore shrink time-based reimbursement).  Another issue is whether AI-based services will get expanded payments in a way that literally "shrinks" the available RVUs.

Here are some 2025 entry points.

• Medicare experts back tying physician payments to inflation.  Here.
• AMA web resources on shrinking RVUs relative to inflation.  Here.
• Medicare volume groowth; payments shift to non-physician practitioners.  Christensen et al. here.
• Doctors slam 2026 pay cuts.  Here.
• AMA slams two CMS proposals; August 2026, here.
• Kaiser Briefing: How CMS pays physicians.  Here.
• Medpac 2025 re: Physician payments.  Here.
• Why Medicare reimbursement keeps declining.  Here.
• How Inflation impacts the Medicare physician fee schedule.  (Amer Coll Radiol), here.
• Report whitepaper by "Omniscient" entity.  Here.  Covered in Becker's.  Here.
• Cigna downcoding here.   
• Reimburesment in the age of AI radiology, Dogra et al.  Here.
• And here's an AI blog by me on the complex stages of "SGR" reform.  Here.

###

 Google Notebook LM and  Chat GPT 5 discusses implications for AI and RVUs.

###

AI CORNER

###

GOOGLE NOTEBOOK LM

Based on the sources you provided regarding Medicare physician pay, there are significant implications for specialties like radiology and pathology, especially in the context of artificial intelligence (AI) development and integration. The current Medicare Physician Fee Schedule (MPFS) operates under a budget neutrality requirement, meaning any increase in payments for one area must be offset by decreases elsewhere, often through a reduction in the monetary conversion factor that translates Relative Value Units (RVUs) into payments.

This system has led to a substantial decline in real, inflation-adjusted physician payments over the past two decades. From 2001 to 2025, Medicare physician pay has effectively dropped by 33% when adjusted for practice cost inflation. The core of your concern lies in how AI could exacerbate this deflationary pressure on specialties like radiology and pathology through two primary mechanisms.

Concern 1: AI-Driven Efficiency Leading to RVU Reductions

Your first concern is that AI could make physician work more efficient, and this increased efficiency could lead to future reductions in the work RVUs assigned to various procedures. The sources strongly support the basis for this concern.

• Foundation of RVUs: The MPFS and its underlying Resource-Based Relative Value Scale (RBRVS) are designed to measure the resources necessary for a service, including physician time, technical skill, mental effort, and stress. The "work" component of RVUs is meant to reflect the relative levels of these inputs.
• CMS's Efficiency Adjustment: The Centers for Medicare & Medicaid Services (CMS) already operates on the assumption that physicians become more efficient over time with experience and improved technology. The proposed 2026 MPFS includes an "efficiency adjustment" that would cut work RVUs by 2.5% for over 7,000 services, based on the unproven assumption of greater efficiency and less time involved in each service. This proposal highlights CMS's willingness to reduce RVUs based on perceived efficiency gains, even without new data to support it.
• AI's Impact on Work: AI tools, particularly assistive or augmentative AI, could streamline workflows in radiology and pathology. If these tools reduce the time, mental effort, or technical skill required from a physician to interpret an image or a digital pathology slide, future surveys conducted by the American Medical Association/Specialty Society RVS Update Committee (RUC) could reflect this reduced physician "work". As a result, the work RVUs for these services would likely be revised downwards in subsequent fee schedules. This is a significant threat, as even a successful AI tool that improves care could inadvertently lead to lower reimbursement for the physicians using it.

Concern 2: Successful AI Reimbursement Squeezing Existing Payments

Your second concern is that if AI algorithms successfully obtain their own CPT codes and reimbursement, the new spending on these AI tools could trigger budget neutrality adjustments, leading to cuts in existing physician payments within those specialties.

• The Zero-Sum Game of Budget Neutrality: The MPFS is a zero-sum system. A provision in the Omnibus Budget Reconciliation Act of 1989 requires that any changes to the fee schedule that increase spending by more than $20 million must be offset by cuts elsewhere. These cuts are typically achieved by reducing the conversion factor, which lowers payments for all services on the fee schedule.
• AI Coding and Reimbursement Pathways: The sources detail the pathways for AI reimbursement, including new Category I and Category III CPT codes for "Software as a Medical Device" (SaMD). As AI tools get their own codes and valuations, the payments for these new services add to the total Medicare spending under the MPFS.
• Redistribution of Payments: If new AI services in radiology and pathology become widely adopted and billed, the resulting increase in overall Medicare spending will trigger these budget neutrality adjustments. This would lead to a reduction in the conversion factor, effectively taking money from existing physician services to pay for the new AI services.
• Historical Precedent: This is not a theoretical concern. In 2021, when CMS increased the RVUs for evaluation and management (E&M) services to better value primary care, it caused an offsetting budget neutrality adjustment to the conversion factor of -6.8% to -10.20%, which cut payments for all other services. Similarly, successful AI reimbursement could lead to a redistribution of payments away from physicians and toward the AI tools themselves, especially since software costs are difficult to characterize within the current practice expense methodology.

Implications for Future Pathology Coding Systems

These two concerns are highly relevant to discussions about new coding systems for pathology, particularly in the era of digital pathology and AI.

• Valuing Physician Work vs. Machine Work: As pathology digitizes, new coding systems will need to grapple with how to value the distinct contributions of the pathologist versus the AI algorithm. If a new coding system bundles the AI's contribution with the pathologist's work, there's a risk that the pathologist's work component will be devalued over time due to efficiency gains, as described in the first concern.
• Avoiding Cannibalization: If the system creates separate codes for AI analysis, it raises the second concern: these new codes could siphon funds from traditional pathology services due to budget neutrality. Stakeholders in pathology will need to carefully consider coding frameworks that recognize the value of AI without inadvertently triggering across-the-board cuts for pathologists. This may involve exploring alternative payment models that are not strictly tied to the fee-for-service, budget-neutral MPFS, such as value-based models that are better suited for situations where AI functions cannot be divided into discrete services.

In summary, the structure of Medicare reimbursement, particularly its budget neutrality requirement and its focus on physician work as a component of value, creates a precarious situation for specialties like radiology and pathology that are on the cusp of significant AI-driven transformation. The very success of AI, whether through improving physician efficiency or securing its own reimbursement, paradoxically threatens to accelerate the decline in real physician payment within the current system. These dynamics must be a central consideration in the development of future coding and reimbursement policies for these fields.

##

Chat GPT 5 takes a swing at the same topic.

Here’s a detailed, integrated analysis of physician pay cuts and their intersection with AI, particularly for radiology and pathology, based on the uploaded articles and current policy dynamics:

---

1. Context: Structural Decline in Physician Payment

Medicare physician reimbursement has lost over 20% of its value (inflation-adjusted) since 2001, with another 4% drop between 2021 and 2023. This is largely due to budget neutrality rules in the Medicare Physician Fee Schedule (MPFS), which require offsetting any $20M+ increase in spending (e.g., new codes, E/M increases) with across-the-board cuts through the conversion factor. Unlike hospitals, skilled nursing facilities, and other Part A providers (which get inflationary updates), physician payment updates are frozen or capped at 0.25–0.75% starting in 2026 — well below medical inflation.

This is why radiologists and pathologists have seen a 24.9% decline in per-beneficiary reimbursement since 2005 (inflation-adjusted) even as productivity rose — a silent “deflationary” force that pressures practice sustainability.

---

2. The Two AI-Driven Risks for Radiology and Pathology

(A) RVU Compression Risk

• CMS’s 2026 proposal includes a 2.5% “efficiency adjustment” cut to work RVUs for >7,000 services, justified by an assumption that clinicians are faster and more efficient — even though there is no new time-study data and some procedures are becoming more complex.

• If AI tools demonstrably speed up image reads or slide reviews, future RUC time surveys could ratchet RVUs downward, embedding efficiency gains as permanent payment reductions — the same way E/M updates permanently lowered the conversion factor for all physicians in 2021–2022.

(B) Code Volume & Budget Neutrality Risk

• Paradoxically, if AI generates new reimbursable codes (e.g., CPT codes for digital pathology AI triage, AI-detected incidental findings), this adds “new dollars” to the MPFS — which triggers budget neutrality and lowers the conversion factor for all physicians.

• This is exactly what happened with the 2021 E/M code revaluation: primary care gained, but specialties like radiology and pathology saw net cuts to maintain budget neutrality.

In other words, radiology and pathology face a “heads I lose, tails I lose” scenario:

• Efficiency from AI → fewer RVUs per case.

• Growth of AI-billed services → lower conversion factor for every other service.

---

3. Strategic Implications for New Coding Systems in Pathology

Because of these dynamics, the profession has an incentive to be cautious about how it codes AI:

• Bundling vs. Unbundling: New AI codes could be bundled into existing pathology services (no new dollars, thus no conversion-factor pressure) rather than stand-alone payable codes that expand spending.

• Appropriate Valuation: If pathology societies seek new RVUs for AI-assisted services, they may need to simultaneously argue that these should not trigger downward adjustment to existing slide review RVUs (e.g., via an add-on code rather than a revaluation).

• Policy Advocacy: Specialty societies (CAP, ACR) may push for:

  • MEI-linked annual updates (MedPAC proposal: MEI – 1%) to break the structural cycle of deflation.

  • Exemptions from budget neutrality for digital/AI transformation codes, analogous to how telehealth waivers were handled during COVID.

  • Guardrails against automatic “efficiency adjustments” unless backed by robust time data — to avoid preemptive payment erosion based on theoretical AI gains.

---

4. Broader Policy Considerations

This debate is already visible in AMA and MedPAC commentary:

• The AMA warns that automatic efficiency cuts “are not grounded in new data” and could “encourage consolidation and reduce access”.

• MedPAC supports re-basing RVUs with timely cost data, arguing current allocations use 2006 practice-expense data that may misvalue facility-based specialties.

• Stakeholders fear that chronic underpayment plus AI-driven deflation could accelerate radiology/pathology mergers into large health systems — reducing independent practice and possibly raising system-wide costs.

---

5. Key Takeaways

• AI magnifies the deflationary dynamics already baked into the MPFS. Without policy reform, any efficiency gains or new AI services risk translating into net pay cuts for radiologists and pathologists.

• Strategic coding and advocacy are critical. Specialty societies should shape coding pathways that recognize AI’s value without automatically triggering conversion factor cuts.

• Long-term fix required. Linking updates to MEI (even at MEI–1%) could stabilize physician payment and prevent AI from becoming a perverse driver of reimbursement decline.

---