Header: AMA continues to work on revisions to "Appendix S" of the AMA CPT, which covers predominantly digital services.
The AMA held a discussion in September during the CPT meeting, then a fly-in (+virtual) meeting December 8. It released versions A and B in December. Now on February 3 (just prior to a February 5 meeting) AMA releases quite a few additional revisions.
At which point I slow down, as the exact content of the revisions are under confidentiality.
Here are some cloudy generalities, as seen by an AI service, Chat GPT 5.2 [start]:
____________________
At a very high level, Appendix S is CPT’s attempt to create a shared vocabulary for describing what “digital services” do in relation to clinical decision-making — whether software is merely highlighting information, deriving new parameters, or moving into interpretive and action-oriented roles.
In December, the early “A” revision largely focused on sharpening definitions and boundaries: clarifying what counts as meaningful analytic output, how categories differ conceptually, and where simple data handling ends and clinically relevant digital processing begins.
The subsequent “B” revision took a more structural turn, paying closer attention to how these digital functions might relate to existing CPT code architecture, including issues around whether a digital function is integral to another service or potentially distinct, and acknowledging that software increasingly changes how clinicians interact with information rather than simply adding traditional interpretive work. "Examples" were dropped, however.
What's New - with 48 Hours to the Meeting
The February “C” revision moves further in that direction, giving more explicit attention to the relationship between digital services and physician work — including performance, interpretation, and responsibility — while also trying to stabilize the framework so it can be applied consistently across different technologies and risk levels. In effect, it pushes Appendix S from a conceptual taxonomy toward a more operational set of principles about how digital outputs relate to clinical use and accountability.
Never Fire-Tested Against Genomics
What remains notable, however, is that these principles have not been stress-tested in domains like genomic testing, where CMS does not treat the activity as physician work at all.
Genomics operates under laboratory fee schedule conventions, where software (including AI) is often part of the assay itself rather than an overlay on a RVU-based, billable physician service. As a result, Appendix S is evolving within a physician-service mental model that fits some fields naturally, but no one has started to reconcile it with areas like genomics that sit completely outside the AMA's physician services framework.
Parts of Digital Pathology Migrate from Physician Service to Machine-Learning Event
Importantly, future forms of digital pathology may begin to resemble this AI-genomics model more than traditional anatomic pathology. Image data will be transformed directly into machine-generated clinical outputs — for example, recurrence risk scores on a 1–100 scale — rather than supporting a pathologist’s experience (visual impressions and narrative interpretation.)
As that shift occurs, the gap between Appendix S’s physician-work assumptions and the operational realities of data-driven laboratory diagnostics may become even more pronounced.
[end Chat GPT 5.2]
