AI on CAP on DIGITAL PATHOLOGY and REMOTE BIOINFORMATICS
https://chatgpt.com/share/6ab35108-06e8-83e8-b0f2-4c09492ebb45
Yes. The CAP record is unusually clear—and it directly
conflicts with CMS’s categorical premise that these computational-pathology
analyses “do not require” a CLIA laboratory and may be performed by “any
non-regulated entity.”
Strongest CAP authority
The most useful document is CAP’s February
23, 2026 response to the HHS Health Sector AI RFI. It is recent, formal,
addressed to HHS, and explicit:
- CAP
identifies itself as a CLIA-deemed accrediting organization that accredits
more than 8,300 laboratories.
- CAP
states that its accreditation checklists require laboratory-director
validation or verification before patient testing, “including AI and
machine learning algorithms.”
- CAP
describes FDA regulation and CLIA oversight as “complementary and
necessary” for clinical AI.
- CAP
says laboratories must validate or verify local AI performance under CLIA
before implementation.
- Later
in the letter, CAP states directly that CLIA requires pathologists, acting
as laboratory directors, to assess AI systems before local implementation.
- CAP
recommends using its accreditation and proficiency-testing infrastructure
to evaluate, monitor, and safely deploy AI tools.
This is not an inference from general laboratory principles.
CAP expressly places clinical AI under CLIA laboratory-director oversight and
CAP accreditation requirements.
CAP’s actual inspection checklist
CAP’s 2025 accreditation checklist edition is described in
detail in the February 2026 CAP TODAY article, “Digital
path practices reflected in latest checklist changes”.
The Laboratory General Checklist now contains a section
expressly titled:
Digital Pathology Including Remote Data Assessment
CAP explains that it revised this section specifically to
encompass developments in digital algorithms and artificial intelligence. Its
definition expressly includes pathologists’ review and diagnostic
interpretation of tissue using digital pathology and whole-slide images. It
also ties the activity to the laboratory’s CLIA address and explains when a
remote interpreting location becomes a separate referral laboratory.
Page
two of the CAP article identifies particularly useful checklist
requirements:
- Inspectors
sample digital-pathology policies and procedures.
- Inspectors
sample reports generated through digital pathology.
- Inspectors
review digital-pathology validation and verification records.
- GEN.50630
requires the laboratory to validate or verify digital-pathology systems
used for clinical diagnostic purposes, with approval by the laboratory
director or a qualified designee.
- GEN.52860
requires digital-pathology services to be included in the laboratory’s
quality-management system.
- CAP
inspectors may examine scan-failure rates, image-quality criteria,
risk-mitigation procedures, discordance reconciliation, and other ongoing
performance measures.
That is probably the best operational evidence for your
letter. CAP is not merely saying that a pathologist should use good judgment
when purchasing software. CAP has placed clinical digital pathology inside the
inspection checklist used to determine continuing laboratory accreditation.
CAP’s specific AI-validation guidance
CAP’s educational guidance, “How
to Validate AI Algorithms in Anatomic Pathology”, is also remarkably
direct:
- Both
CLIA and CAP require validation of a new test, device, or diagnostic aid
before reporting patient results.
- CAP
applies its whole-slide-imaging validation principles to image-analysis
algorithms.
- CAP
states that any image-analysis or image-recognition system—FDA-authorized
or not—must be validated before clinical use.
- Validation
occurs in the laboratory, using cases representative of the laboratory’s
intended clinical use and patient population.
- The
medical director establishes acceptance criteria and approves
implementation.
- CAP
even supplies suggested report language for a non-FDA-authorized AI system
treated as a laboratory-developed test.
Although this is educational guidance rather than the
checklist itself, it demonstrates how CAP interprets and operationalizes CLIA
and CAP requirements for H&E-based AI.
Second formal CAP policy letter
CAP made essentially the same point in its December
1, 2025 comments to FDA on measuring and evaluating AI performance.
CAP states that:
- CAP is
a federally deemed CLIA laboratory accrediting organization.
- CAP
checklist validation and verification requirements expressly include AI
and machine-learning algorithms.
- AI
represents a new component of pathologists’ responsibilities as CLIA
laboratory directors and section directors.
- CLIA
and CAP impose quality requirements beyond the manufacturer’s
FDA-authorized operating instructions.
- Laboratories
should establish controls, performance metrics, corrective actions, and
procedures addressing AI performance drift.
Again, CAP treats AI applied in clinical pathology as part
of the laboratory test system—not as an unregulated computer transaction
detached from the laboratory.
CMS’s own CLIA guidance
There is also an apparent inconsistency within CMS itself.
In its May
11, 2023 CLIA Post-PHE Guidance, CMS treated the remote review of digital
laboratory data, results, and images as testing performed under a primary
laboratory’s CLIA certificate:
- The
remote work must fall within the specialties or subspecialties on the
primary laboratory’s certificate.
- The
laboratory director remains responsible.
- Records
are retained by the primary laboratory.
- Survey
findings and enforcement actions attach to that laboratory’s CLIA
certificate.
- Digital
proficiency-testing review is treated as occurring under the primary
site’s CLIA number.
This does not decide every possible algorithm-only service,
but it refutes any general theory that converting tissue information into
digital images automatically removes subsequent clinical interpretation from
the CLIA framework.
The precise conflict with the OPPS proposal
CMS states in the CY
2027 OPPS proposed rule that stand-alone algorithmic analyses are separate
from a CLIA laboratory’s examination of human material. CMS then concludes that
they:
- do not
require CLIA-regulated laboratory services or entities;
- can be
performed by any “non-regulated entity” possessing the software;
- are
“other diagnostic tests,” rather than clinical diagnostic laboratory
tests; and
- should
therefore be removed from the CLFS and paid as SaMS under OPPS.
For a purely secondary calculation performed on a finalized
laboratory result, CMS’s distinction may sometimes be defensible. It is much
less persuasive for computational pathology applied to an H&E whole-slide
image. The WSI is a direct digital representation of the patient’s tissue
specimen, and the algorithm produces a patient-specific diagnostic, prognostic,
or predictive result. CAP expressly subjects that clinical process to
laboratory validation, laboratory-director oversight, quality management,
reporting requirements, and inspection.
Payment classification and regulatory classification need
not always be identical. But CMS’s proposal does not merely select a different
payment system. It makes the broader factual and regulatory assertion that no
CLIA entity is needed. That is where the collision becomes unavoidable.
Suggested comment-letter language
CMS’s proposed treatment of computational pathology as
Software as a Medical Service creates a direct conflict with the existing
federal laboratory-quality framework. CMS states that these algorithmic
analyses do not require a CLIA-certified laboratory and may be performed by an
otherwise unregulated entity. In contrast, the College of American
Pathologists—the nation’s largest federally deemed CLIA accrediting
organization—expressly applies its laboratory accreditation requirements to
artificial intelligence and machine-learning algorithms used for patient
testing. CAP requires laboratory-director validation or verification before
implementation, and its 2025 Laboratory General Checklist includes a dedicated
section entitled “Digital Pathology Including Remote Data Assessment.” CAP
inspectors review digital-pathology policies, patient reports, validation
records, and quality-management activities; checklist requirements GEN.50630
and GEN.52860 specifically address laboratory validation, laboratory-director
approval, and continuing quality management of digital-pathology systems.
This conflict is especially pronounced for
computational-pathology tests that apply an algorithm to an H&E whole-slide
image to produce a patient-specific diagnostic, prognostic, or predictive
result. Such a service is not simply an arithmetic calculation performed on a
previously reported laboratory value. It is the clinical examination and
interpretation of a digital representation of human tissue. CAP’s current
accreditation framework treats that activity as patient testing conducted under
the responsibility of the CLIA laboratory director.
CMS should therefore withdraw or qualify its categorical
conclusion that SaMS analyses performed on laboratory-test data require no
CLIA-regulated entity. At minimum, CMS should clarify that classification and
payment of a service under OPPS does not displace otherwise applicable CLIA
requirements; distinguish purely secondary analysis of finalized data from
computational examination of whole-slide tissue images; and consult with CMS’s
CLIA program and federally deemed accrediting organizations before finalizing
the regulatory status of computational-pathology services.
One terminology refinement: CAP is not universally “CLIA’s
inspection body.” The technically exact phrase is “a CMS-approved, federally
deemed CLIA accrediting organization.” CAP states that, for CAP-accredited
laboratories, a CAP
inspection occurs in lieu of a CMS inspection. That wording will be harder
for CMS to evade.
