--- title: "The Physician AI Handbook" source: "https://physicianaihandbook.com/" author: "Bryan Tegomoh, MD, MPH" last_updated: "2026-10-21" doi: "10.5281/zenodo.18251404" license: CC BY 4.0 generated-by: handbook-markdown-export --- # The Physician AI Handbook Peer-Reviewed Evidence for Every Specialty What works. What doesn’t. What happens when AI is wrong. Author [Bryan Tegomoh, MD, MPH](https://bryantegomoh.com/) Updated November 2026 · [Changelog](./changelog.html) [Free PDF](./physician-ai-handbook.pdf) · [Print and Kindle editions](https://www.amazon.com/dp/B0HPXRFPFR) ## Welcome to The Physician AI Handbook AI performance in real-world clinical settings often falls short of published validation studies. Written by a physician for clinicians, health system leaders, and anyone building or deploying clinical AI. **Three questions drive this handbook:** What does the peer-reviewed evidence actually show? How do you evaluate claims against that evidence? What are the medico-legal implications when AI is wrong? **This resource is continuously updated as new research emerges.** Important Disclaimers **This handbook is for educational and informational purposes only.** It does not provide medical advice, diagnosis, or treatment. AI systems discussed here are decision-support tools, not substitutes for clinical judgment. **Clinical use remains context-specific.** Physicians should validate AI outputs before clinical use, follow applicable FDA and HIPAA requirements, and meet the standard of care in their jurisdiction. **Information may become outdated** as AI tools, evidence, and clinical guidelines change. Verify recommendations with current clinical guidance before application. **This handbook does not provide legal advice.** Consult qualified legal counsel for questions about malpractice, liability, and regulatory compliance. ------------------------------------------------------------------------ ## Recent highlights - 29 Oct [Bai et al. in Nature Medicine: EVLF-FM, an explainable vision-language model for imaging (opens the source in a new tab)](https://doi.org/10.1038/s41591-026-04655-5 "Bai et al., Nature Medicine, 2026") Showing where in the image an answer comes from helps radiologists check it; explanations are post hoc and the model reads only 2D images. In [Diagnostic Imaging, Radiology, and Nuclear Medicine](./specialties/radiology.html) New evidence [Bai et al., Nature Medicine, 2026](https://doi.org/10.1038/s41591-026-04655-5) - 28 Oct [Neprash and Barnett in JAMA: service use under Medicare’s WISeR prior authorization model (opens the source in a new tab)](https://doi.org/10.1001/jama.2026.21230 "Neprash and Barnett, JAMA, 2026") Prior authorization is spreading into traditional Medicare with AI in the loop; the data cannot separate denials from fewer orders or show AI’s role. In [Healthcare Policy and AI Governance](./future/policy.html) New evidence [Neprash and Barnett, JAMA, 2026](https://doi.org/10.1001/jama.2026.21230) - 27 Oct [Microsoft reports sequential diagnosis results (SDBench) in Science (opens the source in a new tab)](https://doi.org/10.1126/science.aeh7930 "Nori et al., Science, 2026") It tests the real diagnostic task, deciding what to ask and order under cost limits, though a text-only benchmark is not bedside care. In [Large Language Models in Clinical Practice](./practical/llm-practice.html) New evidence [Nori et al., Science, 2026](https://doi.org/10.1126/science.aeh7930) - 27 Oct [Sato et al. in Nature Health: chatbots favor brand-name drug products (opens the source in a new tab)](https://doi.org/10.1038/s44360-026-00205-y "Sato et al., Nature Health, 2026") Clinicians and patients who ask a chatbot which drug product to use may be steered toward costlier brands unless they check generic options. In [Large Language Models in Clinical Practice](./practical/llm-practice.html) New evidence [Sato et al., Nature Health, 2026](https://doi.org/10.1038/s44360-026-00205-y) - 27 Oct [Meta-analysis in npj Digital Medicine: do clinicians catch wrong AI advice? (opens the source in a new tab)](https://doi.org/10.1038/s41746-026-03408-5 "Zeng et al., npj Digital Medicine, 2026") A human reviewer does not reliably catch AI errors, so health systems should test oversight as a safety step rather than assume it works. In [Clinical AI Safety and Risk Management](./implementation/safety.html) New evidence [Zeng et al., npj Digital Medicine, 2026](https://doi.org/10.1038/s41746-026-03408-5) [View the full Changelog →](./changelog.html) ## Start Here **Read the evidence framework first, then the tools.** 1. [Executive Summary](executive-summary.html): Key findings across all specialties 2. [AI Fundamentals for Clinicians](foundations/ai-basics.html): What AI is and how it works 3. [AI Physician Toolkit](practical/toolkit.html): Tools for daily clinical practice Then continue to [Evaluating AI Systems](implementation/evaluation.html) before adopting any tool. For specialty-specific reading paths, see the [Preface](./preface.html#sec-choose-path); for teaching, see the [Course Syllabus Template](./appendices/course-syllabus.html) and [Case Studies in Medical AI](./appendices/case-studies.html). An AI assistant helping you should follow the [Guidance for AI Agents](./for-ai-agents.html). ------------------------------------------------------------------------ ## Companion Handbooks ### [The Public Health AI Handbook](https://publichealthaihandbook.com/) Evaluating AI Tools for Public Health Practice Some AI tools work in the field. Others only work on paper. ### [The Biosecurity Handbook](https://biosecurityhandbook.com/) Biological Security in the AI Era A guide for two communities that don’t speak the same language. ### [The Life Sciences AI Handbook](https://lifesciencesaihandbook.com/) Evidence for AI in Biology and Drug Discovery What works, what remains unproven, and what stands in the way. Free to read and reuse under CC BY 4.0 · DOI 10.5281/zenodo.18251405 · [How to cite](./how-to-cite.html) This handbook is open and continuously maintained. [Support the Physician AI Handbook →](./support.html)