You may have seen it cited on Linked In - a new position paper on the future of pathology called, “Adapt or Become Irrelevant: The Pathologist as the New Diagnostic Architect.”
It is co-authored by Mariano De Socarraz, MD, Founder and CEO of CorePlus, and Jochen K. Lennerz, MD PhD, Medical Director for Pathology Innovation at Natera.
See a Linked In article here.
See the original article here - subscription access.
And see this blog as an 11-page white paper here.
Here's a Chat GPT 5.6 mediated discussion. Below, at 'Questions Arising," pretend I am sitting in the front row and the first to ask a hard-hitting question.
##
50-word summary
Pathology must evolve from issuing isolated test results to architecting integrated diagnostic systems, De Socarraz and Lennerz argue. Across seven cases—PD-L1, NGS reimbursement, AI triage, integrated diagnostics, DPYD testing, trial design, and AI assurance—they call for pathologists to synthesize multimodal evidence, govern quality, shape workflows, and secure accountability.
The paper in review
In “Adapt or Become Irrelevant: the Pathologist as the New Diagnostic Architect”, Mariano De Socarraz and Jochen Lennerz offer a forceful manifesto for the future of pathology. Their central argument is not simply that pathology must adopt artificial intelligence, digital pathology, genomics, or other new technologies. It is that the profession must redefine its intellectual and organizational role.
Modern diagnosis increasingly draws on morphology, molecular testing, imaging, clinical history, therapeutic evidence, informatics, and computational outputs. Yet these components are generally produced and reported through separate workflows. The authors see a widening mismatch: diagnostic information has become multimodal and interconnected, while pathology remains organized around individual specimens, procedures, reports, and fee-schedule codes.
This is what they call a “cognitive crisis.” Pathologists are still too often treated—and sometimes treat themselves—as downstream interpreters who issue technically accurate reports but do not take responsibility for how the total diagnostic system fits together. The proposed alternative is the pathologist as “diagnostic architect”: the person who integrates multiple information streams into a coherent, clinically actionable account and helps design the structures through which that information is generated, validated, communicated, reimbursed, and used.
What Kind of Paper Is This?
Although labeled a review, the paper is better understood as a conceptual review or professional position essay. The authors combine selected literature, policy analysis, institutional experience, claims observations, qualitative interviews, and illustrative cases. They do not undertake a systematic review, present a reproducible dataset, or quantitatively compare competing models.
The authors acknowledge this. Their seven examples are intentionally heterogeneous and reflect their professional experience and institutional exposure. They describe the subjectivity as deliberate and say that the purpose is to provoke reflection and debate rather than establish a definitive consensus.
Its important contribution is therefore not a new empirical finding. It is a vocabulary and framework for describing where pathology might position itself within increasingly computational and multimodal medicine.
The Seven Diagnostic Scenarios
| Scenario | Problem identified | Proposed pathology role |
|---|---|---|
| PD-L1 testing | Complex, gradated biomarker reduced to a binary procedural result | Integrate assay, specimen, scoring, biology, and therapeutic context |
| NGS reimbursement | High denials and inadequate payment for interpretive and coordination work | Participate in payer policy, reimbursement design, and institutional planning |
| AI-based triage | Automation creates efficiency but can separate throughput from oversight | Govern algorithms and redirect effort toward difficult cases and quality assurance |
| Integrated diagnostics | Pathology, radiology, genomics, pharmacy, and oncology operate in silos | Build coordinated, disease-specific diagnostic pathways |
| DPYD testing | Proven pharmacogenomic intervention remains poorly implemented | Move upstream from test performance to implementation leadership |
| Trial design | Trials fail to use biomarkers for treatment adaptation and de-escalation | Embed pathology and biomarker strategy at trial inception |
| AI assurance | Diagnostic quality standards may be developed outside laboratory medicine | Preserve pathology’s role in validation, monitoring, and accountability |
1. PD-L1: From a Stain to a Contextual Biomarker
PD-L1 testing is the authors’ clearest example of how a technically valid test can lose clinical precision when it is operationalized as a procedural checkbox. PD-L1 expression is not a simple biological yes-or-no property. Its meaning depends on tumor type, drug indication, specimen selection, antibody clone, platform, scoring system, cutoff, immune-cell contribution, tumor heterogeneity, and preanalytic conditions.
The authors report substantial heterogeneity in test execution and scoring, even among institutions using the same antibody clone. Inadequate tissue, poor specimen selection, lack of clinical context, and unfamiliarity with tumor-specific scoring methods can weaken the relationship between the reported result and the likelihood of therapeutic benefit.
Companion-diagnostic approval formally ties a particular assay to a drug, but analytical validity does not ensure optimal clinical use. The pathologist’s role should therefore extend beyond correctly performing and scoring the assay. Pathologists should act as biomarker stewards, connecting the technical result to the biological and therapeutic context and preventing a complex biomarker from being flattened into a misleading binary output.
2. NGS Reimbursement: Diagnostic Complexity Without an Economic Model
Comprehensive genomic profiling illustrates the mismatch between the complexity of precision medicine and the procedure-based economics supporting it. The authors cite a Medicare NGS denial rate of approximately 23%, with denials reaching 37% in their own experience.
The problem is broader than the payment assigned to the sequencing procedure. Effective NGS use may require selecting appropriate tissue, assessing tumor content, integrating morphology and molecular findings, determining whether a variant is actionable, participating in tumor boards, supporting prior authorization, documenting medical necessity, and communicating treatment implications. Much of this cognitive and coordination work falls outside traditional fee-for-service payment.
The authors compare the physician work attached to molecular interpretation code G0452 with established evaluation-and-management and surgical-pathology codes. Their point is that conventional coding treats interpretation as a discrete, time-limited procedure, whereas modern molecular oncology requires ongoing synthesis and coordination.
Consequently, pathologists should participate in payer discussions, reimbursement design, coverage policy, and institutional planning. Precision oncology, they argue, cannot scale on an unstable financial infrastructure.
This is a notable expansion of the professional claim. The pathologist is no longer merely responsible for an accurate genomic report but also for helping make the service economically and operationally sustainable.
3. AI-Based Triage: Automation as Cognitive Redistribution
The authors use AI triage of prostate biopsies to argue that automation need not replace pathologists. In reported implementations, deep-learning systems identifying probably benign cores achieved negative predictive values exceeding 99%. This allowed pathologists to concentrate on suspicious, difficult, or clinically important cases while retaining final interpretive responsibility.
The envisioned benefit is not simply faster slide reading. It is a redistribution of professional effort:
Less time on repetitive, low-yield review.
More time on difficult interpretations.
More attention to quality assurance and discordant cases.
Responsibility for local validation and performance monitoring.
Oversight of model drift and lifecycle changes.
Participation in institutional AI governance.
Within the AMA taxonomy, this is augmentative rather than autonomous AI: the algorithm analyzes and prioritizes material, but the physician remains responsible for interpretation and reporting.
The authors’ larger point is that AI increases the need for pathology governance. An automated system deployed across thousands of cases may create more aggregate diagnostic risk than an individual human error. Someone must define acceptable performance, detect drift, investigate failures, and connect algorithmic output to clinical reality.
- NOTE: SEE QUINN AI WHITE PAPER ON machine-generated initial path reports.
- https://www.discoveriesinhealthpolicy.com/2026/08/seven-days-ago-we-published-white-paper.html
4. Integrated Diagnostics: From Parallel Reports to a Care Pathway
The authors broaden “integrated diagnostics” beyond combining pathology and radiology findings. They define it as aligning diagnostic information, professional authority, workflows, informatics, administration, and payer requirements around a patient-centered care pathway.
Current systems often produce fragmentation. Oncologists may order commercial molecular tests outside local pathology workflows. A pathologist may be unable to initiate reflex NGS even when the indication is evident. Specimens may be routed inefficiently, testing may be duplicated, and the clinical record may contain several technically correct but disconnected reports.
By contrast, disease-specific programs in which pathology, radiology, molecular diagnostics, oncology, pharmacy, and administration share governance can reduce redundancy and shorten time to treatment. The authors regard the pathologist as particularly well placed to connect these domains because pathology sits near the intersection of tissue, laboratory measurement, disease classification, and therapeutic biomarkers.
The intended endpoint is not merely a combined report. It is an integrated diagnostic strategy in which specimen use, test sequencing, information flow, clinical decisions, and payment requirements are designed together.
5. DPYD Testing: The Implementation Gap
DPYD genotyping identifies patients at elevated risk of severe or fatal toxicity from fluoropyrimidines such as 5-fluorouracil and capecitabine. The authors estimate that clinically important variants place approximately 3%–8% of patients at increased risk. Despite evidence of clinical utility and cost-effectiveness—and despite preventable deaths—routine testing has remained inconsistent.
The failure is not primarily analytical. It reflects fragmented ordering processes, inadequate reflex pathways, inconsistent reimbursement, limited clinician education, and the absence of infrastructure for preemptive pharmacogenomics.
This example supports an important distinction in the article: producing a valid test is not equivalent to producing clinical value. Pathologists should therefore help design the ordering, reporting, education, decision-support, and follow-up systems required to make a validated intervention routine.
6. Trial Design: Pathology Upstream, Not After the Fact
The authors use KEYNOTE-522 in early-stage triple-negative breast cancer to illustrate how successful trials may still impede individualized treatment. Although information such as tumor-infiltrating lymphocytes, PD-L1 expression, and early response was available, the trial did not use these markers to stratify therapy or test de-escalation. Patients received a prolonged regimen without a second randomization asking whether early responders needed continued adjuvant immunotherapy.
The authors see a structural commercial problem: biomarkers that expand a drug’s market are attractive to sponsors, while biomarkers that identify patients who can safely receive less treatment may not be. Consequently, potentially useful de-escalation strategies remain underdeveloped.
Pathologists generally do not control trial design, but they develop, validate, and interpret the biomarkers on which adaptive trials depend. The paper argues that they should be involved much earlier—in endpoint development, assay strategy, biomarker selection, and trial architecture—rather than receiving a nearly completed protocol and being asked to operationalize its laboratory component.
7. AI Assurance and “Assurance Displacement”
National AI-assurance initiatives create the paper’s most explicitly political concern. Organizations such as the Coalition for Health AI and proposed health AI assurance laboratories are developing methods for assessing safety, bias, transparency, and performance. These initiatives may be valuable, but much of their leadership and conceptual structure comes from outside pathology and laboratory medicine.
The authors introduce “assurance displacement risk”: the possibility that authority over diagnostic performance, bias management, model integration, and quality standards will migrate from pathology-led systems into external, algorithm-centered organizations.
Pathology has decades of experience with analytical validation, controls, proficiency testing, traceability, revalidation, postimplementation monitoring, and continuous quality improvement. The authors do not argue that pathologists should reject external AI governance. They argue that new frameworks should build upon this laboratory tradition and retain pathologists in standard-setting, clinical correlation, and lifecycle oversight.
Otherwise, pathology could become a consumer of standards created elsewhere rather than a steward of diagnostic quality.
The Proposed Professional Framework
The article distinguishes roles, competencies, and concepts.
The two principal roles are:
Diagnostic architect: Integrates morphology, molecular findings, imaging, and clinical information into a coherent, actionable assessment or care pathway.
Diagnostic steward: Assumes continuing responsibility for whether biomarkers, algorithms, and diagnostic workflows remain accurate and clinically appropriate over time.
The required competencies include:
Diagnostic synthesis.
Test-selection and utilization stewardship.
Institutional and policy leadership.
Systems thinking.
Digital and algorithmic fluency.
Strategic participation in reimbursement, regulation, and governance.
The broader concepts include integrated diagnostics, diagnostic governance, diagnostic intelligence, and assurance displacement risk. “Diagnostic intelligence” describes value that emerges only when multiple information sources are combined; it cannot be located in any single stain, sequence, image, or report.
The most consequential change is from episodic responsibility to lifecycle responsibility. The traditional pathologist signs out a case. The diagnostic steward is also concerned with how the test was selected, whether the algorithm remains calibrated, whether reporting promotes the right clinical action, whether unnecessary testing is occurring, and whether the system continues to perform after implementation.
The Kuhnian Claim
The authors invoke Thomas Kuhn to characterize the present moment as a paradigm crisis. Traditional pathology represents “normal science”: morphology as the principal ground truth, stable professional boundaries, location-based laboratories, siloed workflows, and incremental improvements in testing.
The seven cases are presented as anomalies that the traditional model can no longer comfortably absorb. Multimodal diagnosis, distributed testing, AI-based interpretation, external quality-assurance organizations, and value-based care require different forms of authority and collaboration. The new paradigm is therefore not simply digital pathology added to conventional pathology. It changes what counts as the professional product—from an accurate individual report to a coherent and accountable diagnostic system.
This is the article’s boldest claim. It is also the least empirically demonstrated. The authors establish that important changes are occurring, but whether these changes constitute a Kuhnian revolution rather than an enlargement of established consultation, laboratory-director, and quality-assurance functions remains debatable.
Strengths of the Article
The article succeeds in identifying a genuine structural problem: medicine produces increasingly sophisticated diagnostic components without reliably assigning responsibility for integrating them.
It is also persuasive in arguing that automation does not eliminate professional responsibility. AI may reduce routine review, but it creates new obligations involving validation, exception handling, performance monitoring, clinical correlation, and accountability.
Its vocabulary is useful. “Diagnostic architect,” “diagnostic steward,” and “assurance displacement” give names to functions that are real but frequently scattered across tumor boards, laboratory leadership, informatics committees, utilization programs, and informal consultations.
Finally, the paper connects clinical interpretation with economics and governance. A diagnostic technology can be scientifically excellent yet fail because ordering rules, reimbursement, specimen routing, reporting, or responsibility are incoherent.
Limitations and Unresolved Issues
The paper repeatedly says that pathologists are “uniquely positioned” to lead, but it does not rigorously compare pathology with oncology, radiology, clinical genetics, pharmacy, informatics, or multidisciplinary disease-management teams. Pathologists possess important expertise, but unique positioning does not automatically produce authority, time, training, or institutional resources.
Nor does the paper fully explain how the expanded work will be financed. It criticizes procedure-based reimbursement and shows that integrative work is underrecognized, but the proposed pathologist could simultaneously be a diagnostician, informatician, utilization manager, payer strategist, trial designer, quality officer, and AI governor. That is an attractive professional identity but not yet an operating or payment model.
The seven examples also represent different kinds of problems. PD-L1 concerns case-level interpretation; NGS reimbursement concerns economics; DPYD concerns implementation; KEYNOTE-522 concerns trial incentives; and AI assurance concerns institutional authority. Describing all seven as manifestations of one cognitive crisis is illuminating, but it may also impose unity on problems with quite different causes and solutions.
Questions Arising
BQ writes:
The central question I would put to the authors is this:
For complex biomarkers such as PD-L1, is each pathologist really making a separately obtained, de novo judgment—or primarily applying and transmitting the best available state-of-the-field knowledge, of the kind represented by UpToDate, CAP/IASLC guidance, etc?
If much of the work is disciplined application of recently curated knowledge, is this truly a transformation of the pathologist’s role or a new, unexpected kind of role?
Does it boil down to, the pathologist and oncologist need to check UpToDate every quarter?
The Case for This Skeptical Position
Your challenge identifies a real inflation in the paper’s rhetoric. High-quality biomarker interpretation should not depend on every pathologist independently reinventing the meaning of PD-L1. The objective is precisely to reduce idiosyncratic judgment through validated assays, standardized scoring systems, indication-specific cutoffs, guidelines, proficiency testing, and decision support.
Much of what the authors describe as “diagnostic intelligence” may consist of three relatively conventional activities:
Keeping current with published evidence and regulatory changes.
Applying standardized rules correctly to an individual specimen.
Communicating the result in a form that an oncologist can use.
If those activities can be encoded in protocols, report templates, reflex algorithms, and regularly updated knowledge systems, then the profession may need better information management rather than a new professional identity.
The article also sometimes conflates possessing knowledge with leading an entire system. Knowing that PD-L1 scoring varies by tumor and indication does not necessarily make the pathologist the natural leader of reimbursement policy, trial design, pharmacy integration, or the patient’s total treatment strategy. In some settings, a specialized oncologist, molecular tumor board, clinical pharmacologist, or informatics team may be better positioned.
“Check UpToDate every quarter” is therefore a pointed way of asking whether the paper has converted the ordinary professional duty to remain current into a supposed Kuhnian revolution.
The Case for the Authors
The strongest response is that curated knowledge cannot examine the actual tissue. UpToDate can describe which scoring system applies, but it cannot determine whether the available block is representative, recognize heterogeneous staining, distinguish tumor cells from immune cells, assess fixation artifacts, reconcile a discordant molecular result, or determine whether the assay has been locally validated for that specimen type.
Nor is the relevant knowledge contained in one source. PD-L1 interpretation may require the FDA label, tumor-specific guidelines, assay instructions, pathology standards, clinical-trial criteria, local validation data, and knowledge of the specimen’s history. A quarterly literature check may establish the general rule but does not apply it to the peculiarities of an individual patient and specimen.
The authors’ strongest argument is also not about possessing facts. It is about responsibility for the interfaces between facts. Someone must notice when:
The wrong specimen was selected.
The tissue is inadequate or unrepresentative.
Different PD-L1 assays or scoring systems are being treated as interchangeable.
PD-L1, tumor mutational burden, microsatellite instability, morphology, and clinical history point in different directions.
A commercial test bypasses local specimen-management safeguards.
A validated biomarker is never ordered because the workflow has no trigger.
An AI model’s performance deteriorates after deployment.
A technically correct report predictably produces the wrong clinical interpretation.
No quarterly update can by itself establish ownership of those failures. The authors are arguing that pathology should claim this integrative and governance responsibility before it is assigned elsewhere.
What's the AI Assessment
The paper is most persuasive if three different activities are kept separate:
Knowledge maintenance: Keeping guidelines, protocols, and decision support current. This is essential but not transformational.
Patient-specific interpretation: Applying standardized knowledge to a particular specimen and clinical setting. This involves genuine judgment, but it is an extension of traditional pathology.
Systems architecture and lifecycle governance: Designing reflex pathways, integrating modalities, governing algorithms, resolving responsibility, and monitoring performance over time. This is the truly expanded role.
Thus, your criticism lands against the paper’s occasional tendency to describe all sophisticated application of knowledge as a cognitive revolution. Every PD-L1 case should not become an act of solitary intellectual creation. Standardization and curated knowledge are virtues, not threats to professional status.
But “check UpToDate every quarter” is insufficient because the hard part often lies in local implementation, specimen-specific exceptions, cross-modal discordance, and accountability. UpToDate can supply the map; it cannot determine whether the patient, specimen, assay, local workflow, and reported conclusion are actually traveling along the right road.
The fairest conclusion is that the transformational element is not that the pathologist knows something the oncologist could never look up. It is that the pathologist accepts responsibility for making standardized knowledge work reliably in a particular patient and across an entire diagnostic system. The article is strongest when it makes that systems-governance claim—and weakest when it presents ordinary expert interpretation itself as a new paradigm.