Anyone can see that, on the upcoming AMA CPT agenda, the AMA CPT "Appendix S" is up for revisions [2-5-26, Tab 68]. This is the Appendix that deals with software terminology issues in the context of the AMA CPT.
While the proposed revisions to Appendix S are confidential, I asked Chat GPT to provide us a plain-language reading of Appendix S (using the current public version and getting some context from public AMA writings here, here).
Here it is [Chat GPT 5.2]
Appendix S: Understanding It in Plain Language
Appendix S is the AMA’s attempt to answer a surprisingly simple question: when software helps make medical decisions, what kind of help is it actually providing?
Appendix S is not a payment rule, an FDA regulation, or an insurance coverage policy. Think of it as a labeling system — a way for the CPT coding world to describe the role that computers and algorithms now play in clinical care. As medicine increasingly includes machines that look at data, analyze patterns, and sometimes even suggest actions, CPT needed a shared language to describe what the “machine part” of care is doing.
Historically, medical care followed a straightforward pattern: a human clinician looked, thought, and decided. Today, that process often includes another participant. A machine may look at an image, calculate a risk score, or generate a probability before a clinician ever weighs in. Appendix S was created because the CPT system did not have clear, consistent terms to describe this growing range of digital involvement. Without such a framework, every new technology would require a fresh debate over what AI is and what it does. Appendix S provides a common structure so those conversations can start from the same definitions.
At its core, Appendix S sorts medical software into three broad categories, based on how much “thinking” the computer is doing. The first category, called assistive, describes software that acts like a spotlight. It highlights something important, flags an abnormality, or draws attention to a region of interest, but it does not interpret what that finding means. A clinician still must make the medical judgment. The software is essentially saying, “You may want to look here.”
The second category, augmentative, moves further along. Here, the software is more like a calculator or modeling engine. It analyzes data and produces new numbers, scores, or probabilities that a clinician could not easily generate by inspection alone. It might compute a risk score, quantify a measurement, or generate a predictive index. However, it still does not make the final clinical decision. It provides a derived parameter — a number or category — and the clinician determines how to act on it.
The third category, autonomous, represents the most significant shift. In this case, the software does not merely highlight or calculate; it interprets and reaches conclusions. It may recommend a diagnosis, suggest a management step, or in some systems even initiate an action unless a clinician intervenes. At this level, the computer begins to resemble a junior clinician, performing functions that previously belonged exclusively to human judgment.
The broader goal of Appendix S is to create a neutral vocabulary for describing these roles. As AI tools appear in imaging, monitoring, pathology, genomics, and many other areas, CPT committees need a way to talk about what the software is doing without reinventing definitions each time. Appendix S provides that shared language. It acknowledges that software is now a routine part of how care decisions are made and that its role can range from simple assistance to active interpretation.
Importantly, Appendix S does not itself determine payment or coverage. It does not say what insurers must reimburse or how much a service is worth. But the way a technology is described can influence how codes are later built, and that means Appendix S indirectly shapes the future structure of medical coding. It reflects a broader recognition within medicine: clinical decision-making is no longer purely human. Software has become another actor in the process, and Appendix S is CPT’s way of labeling how much of the clinical “thinking” is being done by the computer rather than the clinician.
[End Chat GPT 5.2]
I'm unsure how some of this fits to genomics. For example, a CMS NCD from 2018 (NCD 90.2) requires NGS tumor reports to call out driver mutations AND the recommended drug for that mutation. This generates a result which not obvious from the input (millions of tiny DNA SEQ fragments)--so it is Augmentive, but it also suggests a management step (an exact drug choice), so it is Autonomous.
Regardless of the decision, the service and code have been software-intensive and happily coded and priced since 2015 (81455, $2900) with no dilemmas caused by not having a physician work component or being expressed on the Clin Lab Fee Schedule, not the RUC RVUs.
