The advent of artificial intelligence (AI) and its expansion into clinical analysis, decision support, and autonomous clinical actions raises questions about how physicians’ role in clinical medicine will evolve. These questions are not abstract—the way that the Centers for Medicare and Medicaid Services (CMS) pay for physician services means that tools that affect physicians’ work, as defined by CMS, can impact how much they get paid. While CMS has devoted substantial resources to understanding the precise nature of the work involved in performing clinical procedures to date, and in quantifying the specific ways that new technology affects that work, it has not yet developed a similar understanding of which evaluation and management (E&M) services may be most affected by AI, and how they will be affected.
CMS, and the physicians (disproportionately primary care providers) who deliver these services, will need to address this gap. As part of this process, physicians, payers, and policy makers will need to develop a common, evidence-based understanding of:
- Where physicians provide unique and irreplaceable value in evaluation and management, and how that value contributes to their patients’ overall well-being.
- Whether and how use of AI reduces physician effort required to perform certain tasks, improves the quality of their output, or both.
- How practice expenses will evolve as physicians rapidly accelerate their purchasing of AI technologies.
- How access to AI tools should affect scope of practice for evaluation and management tasks among nonphysician practitioners.
- How to consider medical malpractice liability in areas where AI is granted increasing autonomy to evaluate patients and propose or even implement treatment.
In this article (part 1), we look at the development of Medicare’s physician payment system and its limitations in accurately accounting for the mental effort involved in E&M services. In part 2, we describe the growing importance of AI tools in E&M services and the challenges this poses for the physician reimbursement. We note that the science of quality measurement will have to evolve as AI becomes a growing presence in providing E&M services, and we emphasize the importance of physicians proactively considering how they can best add value in an emerging new, hybrid E&M system.
How CMS’s Focus On Procedures In Physician Reimbursement Is Being Challenged
For a long time, CMS’s focus on understanding the nature of procedural work has made sense, as the agency sought to improve the accuracy of physician payments. In the 2026 Physician Fee Schedule (PFS) rule, CMS finalized a policy to reduce physician payment for non-timebound (that is, primarily procedural) services by reducing their estimate of the work involved. Specifically, CMS finalized use of an assessment of the Medicare Economic Index productivity adjustment percentage over the previous five years to generate an efficiency adjustment that would be applied to the service valuation. In future years, CMS may move to empiric studies of time use and away from subjective surveys with low response rates to “help address some of the distortions that have occurred in the PFS historically.”
CMS cited evidence suggesting that, as physicians learn and adapt to new procedural technology, they become faster and more efficient, and that relying on estimates of work early in a procedure’s adoption leads to overvaluation—particularly when a decade or more can pass before re-valuation. CMS’s change has been seen by some as vindication for representatives of the “cognitive specialties,” who have long claimed that the resource-based relative value scale (RBRVS) approach to fee setting and the zero-sum nature of the PFS have systematically undervalued their work.
The rise of AI, however, poses a question: What is the fundamental nature of cognitive work in medicine, and what is the specific value that human physicians bring? The answers to this question will have concrete effects on access to care, as policy makers determine who should be allowed to perform E&M services and how much they should be paid for doing so. The Center for Medicare and Medicaid Innovation’s new ACCESS model will pay for certain AI tools based on achievement of disease-specific outcome and will generate valuable new understanding of individual tools and their impact, but the speed of change in AI is so rapid that the scope of tools, and physicians’ approach to using them, may outpace lessons learned from the model.
Already, AI tools can identify relevant information from clinical interviews, maintain familiarity with an evidence base, develop a differential diagnosis and treatment plan, and communicate with patients in ways that patients rate more highly than they rate communications from human physicians. AI-augmented glasses and stethoscopes may soon revolutionize the physical exam. New AI tools are coming into practice that act increasingly autonomously—such as Doctronic’s medication prescribing tool which was recently authorized in Utah. Indeed, AI developers seem eager to tackle any E&M-related challenge (other than accepting medicolegal accountability). Amid a looming physician shortage, insurers and policy makers may grow willing to let them.
CMS’s Approach To Price Setting Does Not Clearly Define The High-Value Elements Of The Mental Effort Key To E&M Services
In the Medicare fee-for-service system, which has widespread influence on reimbursements in other settings, physician payments are determined using a process based on research begun in the 1980s. As described in a historical summary from the American Medical Association (AMA), CMS funded the Harvard University School of Public Health to develop an RBRVS for physician services across 18 specialties by quantifying the work needed to provide each service compared with another reference service. To develop the scale, researchers worked with the AMA and specialty societies to convene technical consulting groups for each specialty, which informed the development of vignettes describing the services physicians performed and a survey asking physicians to rate the work required for the service in each vignette.
The study considered the following elements as contributing to the amount of physician work a service requires:
- Time
- Technical skill and physical effort
- Mental effort
- Psychological stress related to risk of causing harm
The study considered work done face to face with the patient (described as intraservice), work done in preparation (preservice), and follow-up tasks such as documentation (post-service). The survey required physicians to calculate the work of each service compared to a reference service. These relative estimates of work were converted into relative value units, which, when combined with an estimate of the costs of maintaining a practice and paying for malpractice insurance, comprised the elements of a physician payment.
Under the Omnibus Budget Reconciliation Act of 1989, updates to the PFS must be budget neutral—that is, if additions of new services and changes in valuation of existing services would increase overall Medicare spending by more than $20 million, CMS must reduce payment on all other services correspondingly. Because services are intended to be compared both within and between specialties, and because relative valuations are applied to a fixed amount of funds, this approach gave rise to persistent historical concerns that the relative value of the work required for surgeries and other procedures have been inflated, leading to a corresponding devaluation of the non-procedural E&M services often performed in primary care.
The main effort by physicians for these E&M services involves the mental labor of talking with and examining the patient, reviewing previous records, thinking through presentations, and developing a differential diagnosis and treatment plan. The physician payment system originally classified these E&M services based on the expected length of time it would take to assess new and established patients at varying levels of complexity, and physicians’ coding was expected to correspond to the actual duration of the visit; more recently, physicians can elect to bill for a service based on the complexity of their medical decision making by documenting specified elements in the clinical record, regardless of how much time they spend in the clinical encounter.
Of note, the documentation for the RBRVS does not provide a detailed analysis or justification of “mental effort,” which in some related publications is mentioned in parallel with the concepts of “knowledge,” “judgment,” and “diagnostic acumen.” No further details are provided about the nature of mental effort, and no reference is made to whether this work must be done by the physician alone or if it may be augmented by other tools.
This lack of detail may reflect the milieu in which the foundational studies for the RBRVS were conducted when computer use was at a much earlier stage. A 1987 paper on the promise and limitations of computerized decision support, published shortly before William Hsiao and his team submitted their final Phase 1 report to the Health Resources and Services Administration in 1988, lists several clinical decision support modules (including an early AI model) then available to physicians in research and academic settings. However, the paper describes “ubiquitous computer-based aids that routinely assist most aspects of their practice” as “still the stuff of science fiction;” it identifies as a key barrier to adoption physicians’ persistent distaste for typing. Over the subsequent decades, as reimbursement debates between “proceduralists” and “cognitive specialists” have smoldered and flared and physicians have become expert typists, CMS has provided no further regulatory guidance about what specific aspects of “mental effort” they consider valuable.
In Part 2 of this article, we will discuss how current payment approaches for cognitive services will be challenged in a world of more ubiquitous AI. “Input-based” approaches to valuation (e.g., physician time, technology costs, malpractice costs) are at best today only a rough approximation of value and risk becoming increasingly disconnected from value when AI is doing more and more of the “work.”
——————————————————
Originally Published On: Health Affairs
Photo courtesy of: Health Affairs
Follow Medical Coding Pro on Twitter: www.Twitter.com/CodingPro1
Like Us On Facebook: www.Facebook.com/MedicalCodingPro
