Cat III Not on CLFS
Over the past two months, CMS omitted new Category III codes for Digital Pathology [Computational Pathology] from CLFS meetings June 10. (Blog).
OPPS Rule; Push Dig Path onto APC Policy System
Then, July 2, CMS published the OPPS rule, taking the position that computational pathology per-se was not a clinical lab service at all (not under CLIA) and any such codes should be pulled from the CLFS. (Blog).
In the OPPS rule, CMS proposes to take such codes off CLFS and for the hospital outpatient setting, putting them under regular APC (ambulatory payment category rules). If CMS take that category of test off the CLFS (off the list of CDLTs) they won't be eligible for ADLT, either. (That's IF, not when. And regular genomics, e.g. CGP, MRD, isn't involved).
PFS Rule: Make Dig Path Tests "Contractor Priced"
In the recent PFS rule, July 14, CMS proposed to take the same 10 tests (a list with probably errors, even on its own terms) off the CLFS and kick them into "Contractor Priced" codes. (Blog.)
CLIA RFI
In the CLIA RFI released on July 16, the same topic is discussed at length (for 2 columns). It's topic 6 & 7 on page 43588. (Blog). This blog, a couple hours later, focused more closely and specifically on digital/AI topics, topics 6,7.
##
AI CORNER for 6 & 7
##
CMS and CDC are asking unusually direct questions about how CLIA should apply to artificial intelligence, digital pathology, and laboratories that perform interpretation without handling a physical specimen. Importantly, these are requests for information, not yet proposed regulatory changes.
#
CLIA Turns to AI, Digital Pathology, and “Data-Only” Laboratories
Point 6 addresses postanalytic interpretation and the use of artificial intelligence. CMS notes that modern laboratory testing increasingly incorporates sophisticated software, automation, and AI, and asks where these technologies fit within the CLIA-regulated testing process. The questions cover the use of algorithms in postanalytic workflows; AI-assisted interpretation of results from fields such as next-generation sequencing, histocompatibility, and pharmacogenomics; and, most directly for pathology, the interpretation of histopathology slides and results.
CMS also asks how laboratories verify the performance of these systems. Its examples extend beyond validating an algorithm’s output to include image resolution, image quality, computers, monitors, and other components of the digital testing environment. The agencies additionally request comments on cloud analytics, automation, and AI applications used with high-complexity testing.
For digital and computational pathology, this signals that CLIA may eventually treat the software, viewing environment, and computational infrastructure as integral parts of the test system—not merely as administrative or communications tools. Laboratories and technology companies may therefore want to describe existing practices for algorithm validation, change control, quality assurance, human review, cybersecurity, cloud deployment, and monitoring of performance after implementation.
Point 7 focuses on “data-only facilities.” These are organizations that do not necessarily receive or examine a physical specimen but process analytical data, interpret genetic information or digital images, or calculate risk scores that contribute to a final laboratory result. Some may also be manufacturers of medical-device software.
CMS explains that CLIA already recognizes “distributive testing,” in which multiple laboratories share work on the same specimen or its aliquots to produce the final reportable result. The harder question is how that concept should apply when the second organization receives only data or digital images. CMS has apparently received inquiries about whether such facilities require their own CLIA certificates and now asks what activities they perform in generating or contributing to test results and interpretations.
This question goes directly to the business models of remote pathology services, computational pathology companies, genomic interpretation vendors, cloud-based analytics platforms, and developers of algorithmic risk scores. A future policy could distinguish between a vendor supplying software to a CLIA laboratory and a separate facility actively performing part of the analysis or interpretation. Depending on where CMS draws that line, some data-only organizations could potentially be viewed as participants in laboratory testing rather than merely as software suppliers or consultants.
Together, Points 6 and 7 show CMS and CDC examining both sides of the same emerging problem: when software helps interpret a test, what must the laboratory validate—and when the interpretation occurs elsewhere, which organization is legally operating the laboratory service? The answers could materially affect digital pathology networks, centralized AI services, cloud computing arrangements, and the division of responsibility among laboratories, pathologists, and software companies.
#
The essential theme is that CMS is beginning to examine not only whether an AI tool works, but also where the computational work occurs, who controls it, and who is responsible for the final result.
