NEJM AI

Medical expertise has always been scarce. Can #ArtificialIntelligence help change that?   Dr. Karan Singhal, research lead at OpenAI, believes it can and explains how in the latest episode of NEJM AI Grand Rounds.   Drawing on his work at OpenAI and earlier efforts behind Med‑PaLM, he discusses how clinicians and patients are already using AI to answer questions, support decisions, and navigate care. He argues that the future of health AI is not only about improving model performance, but also about helping people advocate for themselves more effectively.     Through HealthBench and ChatGPT for Clinicians, his team is exploring how to make these systems safer, more useful, and more trustworthy. The result is a vision of health care where expertise becomes more accessible without losing sight of clinical responsibility.   Listen to the full episode with hosts Arjun (Raj) Manrai, PhD, and Andrew Beam, PhD: https://nejm.ai/ep43   #AIinMedicine

The model has the paper, but not the answer. Dr. Travis Zack of OpenEvidence explains on NEJM AI Grand Rounds how hallucinations still happen. Listen to the full episode hosted by Arjun Manrai, PhD, and Andrew Beam, PhD: https://nejm.ai/ep42 #ArtificialIntelligence #AIinMedicine

In heterogeneous clinical populations, the clinical and economic value of therapies depends on how they are allocated. Predictive #ArtificialIntelligence (AI) models can improve the precision of treatment allocation beyond what earlier clinical decision support tools could achieve by integrating data across clinical, biomarker, imaging, and administrative domains simultaneously. This capacity can increase the realized value of existing and emerging therapies relative to the status quo by improving two types of treatment decisions: risk-based targeting, which concentrates treatment among patients at higher baseline risk, and response-based targeting, which identifies patients with greater expected treatment benefit. Both mechanisms reduce effective numbers needed to treat (NNT) and improve return on investment (ROI), reducing low-value care and strengthening incentives to develop therapies for well-defined patient subgroups. Jonathan Ketcham, PhD, illustrates this argument using remote physiological monitoring and Alzheimer’s disease — two settings where current deployment heuristics produce high effective NNT and limited ROI, and where improved predictive targeting could materially improve the economic conclusion. Realizing this potential depends on data governance, incentives, and regulatory pathways that support the development, continuous refinement, and clinical integration of predictive models. Framing predictive AI as an ROI multiplier on the therapies it guides has broader consequences for how those therapies are developed, regulated, priced, and covered, and for how predictive AI models themselves should be valued and evaluated. Read the Policy Corner “Artificial Intelligence as a Return on Investment Multiplier in Health Care” by Jonathan Ketcham, PhD: https://nejm.ai/4uNRshk #AIinMedicine

#ArtificialIntelligence (AI) scribes are increasingly deployed in U.S. health systems to document patient–physician encounters, with anticipated gains in administrative efficiency counterbalanced by emerging privacy risks. These systems capture audio of clinical encounters and generate transcripts and structured clinical notes, yet vendor practices differ substantially in how long each data type is retained and whether it is used for AI training. Federal and state privacy frameworks govern the collection, storage, and disclosure of such data, shaping institutional obligations and constraints. Effective implementation therefore depends on clear governance of data flows, appropriately calibrated retention and deletion policies, and transparent consent processes that align technical design with legal requirements and ethical principles. Read the Policy Corner “Privacy Considerations of Artificial Intelligence Scribes” by Sara Gerke, Dipl.-Jur. Univ., MA, and David A. Simon, PhD, JD, LLM: https://nejm.ai/3Pkqwaa #AIinMedicine

Most decisions aren’t uncertain — they’re almost certain. In the latest episode of the NEJM AI Grand Rounds podcast, Dr. Travis Zack of OpenEvidence explains why AI is a double-check tool. Learn more in the full episode hosted by Arjun Manrai, PhD, and Andrew Beam, PhD: https://nejm.ai/ep42 #ArtificialIntelligence #AIinMedicine

Medicine is fundamentally an information enterprise in which clinicians integrate patient data with prior knowledge and external sources to guide decisions. Over time, the delivery of medical knowledge has evolved from apprenticeship and print to digital search, improving access but not necessarily efficiency. The emergence of large language models (LLMs) marks a qualitative shift — from retrieval to synthesis — enabling rapid, contextually tailored responses to clinical questions. This transformation offers substantial gains in efficiency, but also introduces new epistemic risks. LLMs generate fluent, authoritative-seeming outputs based on statistical patterns rather than true understanding, and they have limitations in reasoning, calibration, and transparency. As a result, distinguishing evidence-based conclusions from plausible inferences becomes challenging. This shift redefines the role of clinicians and medical journals, which now function both as curators of validated knowledge and as upstream inputs to AI systems. Ensuring safe integration will require preserving critical appraisal, accountability, and standards of evidence in LLM-mediated clinical decision support. Read the Editorial “Medicine as an Information Industry in the Age of Language Models” by Jeffrey M. Drazen, MD, and Charlotte J. Haug, MD, PhD, MSc: https://nejm.ai/4ddmTMh #ArtificialIntelligence #AIinMedicine

“The answer is not the important part.” In the latest episode of the NEJM AI Grand Rounds podcast, Dr. Travis Zack of OpenEvidence challenges how we measure both clinicians and AI. Hear more from Dr. Zack in the full episode hosted by Arjun Manrai, PhD, and Andrew Beam, PhD: https://nejm.ai/ep42 #ArtificialIntelligence #AIinMedicine

The rapid proliferation of ambient #ArtificialIntelligence (AI) platforms has created a precarious legal landscape, as highlighted by the 2025 class action lawsuit Saucedo versus Sharp Healthcare. In the absence of legal precedent, the application of existing legal restrictions to the capture and transmission of audio recordings in the clinical context remains unclear. State “eavesdropping” statutes create a patchwork of consent requirements — 34 states follow one-party consent policies, while the remaining 16 states require all-party consent or apply mixed standards. In the latter jurisdictions, institutions and clinicians face potential civil and criminal penalties for failing to obtain consent from all recorded parties, a risk compounded by “bycatching” incidental conversations in shared clinical spaces. Defining and mitigating these risks requires the collaborative effort of institutions, clinicians, vendors, and physician advocacy groups. Ultimately, responsible deployment of ambient AI necessitates clear institutional safeguards and advocacy for legislative clarification to protect both patients and clinicians. Read the Perspective “Ambient AI in Clinical Practice — The Legal Landscape of Recording Consent Requirements” by T.N. Anderson et al.: https://nejm.ai/4eEo0G3 #AIinMedicine

AI in health care presents persistent challenges for patients, clinicians, and policy makers owing to its rapid evolution and conceptual complexity. The field has progressed swiftly from early generative chatbots to more advanced autonomous agents and, increasingly, to integrated agentic AI systems capable of coordinating complex tasks autonomously. These technologies are often introduced without clear definitions or practical guidance, leaving health care organizations to interpret their capabilities and implications independently. A new Policy Corner article aims to clarify key distinctions among emerging AI paradigms and provide a lens for understanding their functional differences, risks, and potential applications in clinical and administrative contexts. Read “A Typology of Generative Health Care Artificial Intelligence — Definitions and Policy Implications” by David Blumenthal, MD, MPP, and Vivian S. Lee, MD, PhD, MBA: https://nejm.ai/3OTTFZE #ArtificialIntelligence #AIinMedicine

Informed consent forms (ICFs) are foundational to ethical clinical research consent, yet they remain lengthy, cognitively demanding, and written at reading levels that are not comprehensible to most participants. Recent advances in #ArtificialIntelligence, particularly large language models (LLMs), offer new opportunities to improve clarity, comprehension, accessibility, and participant engagement throughout the consent process. In a new narrative review, the authors map the emerging landscape of LLM applications for ICFs across seven domains: glossary generation, plain language revisions, visual aid generation, accessibility and document formatting, translation into other languages, teach-back and comprehension validation, and interactive chat-based consent. The authors conclude by outlining considerations for their proper usage and integration with research and ethical review frameworks. Read the Review Article “Large Language Models in Informed Consent — Opportunities, Evidence, and Challenges” by R. Goel (Rishabh Goel) et al.: https://nejm.ai/3QW1Oxm #AIinMedicine