AI Exposure Varies Widely Across Medical Specialties, Study Finds

November 16, 2026
AI Exposure Varies Widely Across Medical Specialties, Study Finds
  • A study published in NEJM AI analyzed nearly US$45 billion in 2023 Medicare Part B payments to 589,648 physicians across 56 US specialties, assessing how much of their billed work might be exposed to software-based AI substitution.

  • The researchers emphasized that exposure is not evidence that today’s AI can perform the work independently or that physicians will lose their jobs; AI could replace or complement physicians depending on its design, deployment and reimbursement.

  • Pathology and diagnostic radiology had the highest estimated exposure, with 99% of pathology services and 96% of diagnostic radiology services classified as potentially substitutable because they center on interpreting tissue samples and medical images.

  • Evaluation and management services—including consultations, assessments, histories and treatment management—accounted for 55% of the payments analyzed and 87% of family physicians’ and 90% of internal medicine specialists’ payments; AI may help with documentation and routine follow-ups, but the work also involves complex decisions and physical examinations.

  • At the other end of the spectrum, only 7% of dermatology services and 10% of orthopaedic surgery services were categorized as potentially substitutable, reflecting the physical intervention required for procedures such as biopsies, excisions and surgery.

  • Physical services made up 68% of dermatology payments and 53% of orthopaedic surgery payments, while ophthalmology followed at 45%; potential substitutability was estimated at 62% for radiation oncology, 39% for cardiology and 32% for interventional cardiology.

  • The study linked longer postgraduate clinical training in more hands-on specialties with greater protection from software substitution, while cautioning that robotics and hardware-integrated systems could eventually reduce that barrier, even though hardware advances more slowly than software.

  • For primary care, AI could reduce administrative work and help physicians manage more patients, but replacing tasks or changing reimbursement could lower incomes and affect medical graduates’ specialty choices.

  • The analysis was based on US Medicare billing data, may have undercounted physical work bundled into consultation payments, and reflects current rather than future practices; the researchers urged policymakers, educators and healthcare institutions to account for differing exposure levels in training and workforce planning, particularly in primary care.

Summary based on 1 source


Get a daily email with more AI stories

More Stories