NEJM AI

@ai.nejm.org

1424 Followers 91 Following

NEJM AI, a new monthly journal from the publisher of @nejm.org, explores the cutting-edge applications of artificial intelligence and machine learning in clinical medicine. Online at ai.nejm.org.

This infographic illustrates a study on predicting patient outcomes from surgical video data using Swarm learning. It includes sections on data collection, preprocessing, and model development with flowcharts. Diagrams detail the process for non-perforated and perforated appendicitis alongside intraoperative images showing different grades. The figure highlights traditional and Swarm learning approaches across multiple centers.
Flowchart illustrating the CBC Utilization SmartAlert Development and Governance Process. It begins with lab utilization preparation in November 2022 and continues through stages including stakeholder meetings, committee reviews, and pilot implementations. Key milestones include a planned SEP SmartAlert enterprise go-live in September 2025. The chart features color-coded diamonds and rectangles representing different phases and actions, connected by arrows to indicate process flow.
This image contains a flowchart and line graphs illustrating use of the machine-learning system for the detection of Hirschsprung disease in the validation and external testing datasets. The flowchart in Panel A shows training and testing of machine-learning system for detection of Hirschsprung disease risk. Panel B presents the area under the ROC curves for Hirschsprung disease risk detection models.
Table outlining potential applications of artificial intelligence for clinical trial endpoints. It includes focus areas such as histopathology, anatomical imaging, and physiological imaging. Descriptions and current examples are provided alongside AI-enabled future possibilities for each category.
A digital illustration showing a medical diagram of human organs overlaid with surgical tools. At the bottom, there's a text box containing the article title: "Privacy-Preserving Surgical Video Analysis with Swarm Learning — Results from a Multinational Appendectomy Cohort" by O.L. Saldanha and Others.
Promotional image for episode 44 of the NEJM AI Grand Rounds podcast featuring Weave’s Brandon Rice on rebuilding drug regulation with AI. Includes a photo of the guest.
Table depicting demographics and outcomes of CBC Utilization SmartAlert Pilot. It displays data for treatment and control groups, including number of encounters, alerts, ages, and race percentages. Rates for ICU admission and mortality are included.
This image is a diagram titled "Potential Downstream Applications of ReXGroundingCT." It is divided into two main sections: Training and Inference Tasks. In the Training section, a stack of cylindrical objects representing the ReXGroundingCT is connected to a neural network diagram. In the Inference Tasks section, two processes are shown: Report Grounding and Grounded Report Generation, each with a medical image or document icon. Below, Use Cases are shown with icons depicting Education, Patient Explanation, and Report Visualization involving medical professionals and patients.
Image of a document titled "Family 0955 – Case Review for PATIENT_0955-01 (Hypophosphatemic Rickets)." The document outlines clinical background, an interpretation, and a conclusion. The following figure caption sits at the bottom: "Figure 1. Example Output from the OpenAI Tool from a Known Diagnosed Control Case."
A quote from an editorial by Samuel G. Finlayson, M.D., Ph.D., and Heidi L. Rehm, Ph.D., published in NEJM AI. The quote reads as follows: "The appropriate posture is . . . neither to encourage patients to assume a duty they are not trained for, nor to obstruct an exercise of autonomy that is both legally grounded and technically feasible. Instead, the field must build the infrastructure that makes patient-generated findings safe, useful, and equitably available." The background has horizontal colored stripes and abstract patterns.
Figure 1 illustrates the ReXGroundingCT Dataset. Panel A displays lung CT scan findings. Panel B highlights a bar chart of findings per category across the dataset, grouped by typical pattern (nonfocal vs. focal). Panel C details an example of anatomical chain-of-thought reasoning.
Diagram showing protein structure predictions related to the pathological mechanism of the Sphingosine-1-Phosphate Receptor 1 (S1PR1) c.850_882del variant in a vitiligo patient. Image includes a protein sequence diagram, ribbon model structures, and molecular surface models.
The image is a quote from an editorial published in NEJM AI asking, "Is clinical genetics prepared for a world in which patients and families can generate analyses of this kind on their own?" Authors: Samuel G. Finlayson, M.D., Ph.D., and Heidi L. Rehm, Ph.D. The background is a gradient orange.
This image shows page 1 of a Perspective from NEJM AI titled "The Medical AI Assistant as Publication, Not Device — Why Peer-Reviewed, Open-Source AI Belongs in the Standard of Care" by Adrian Gropper, M.D. It includes publication dates and part of the abstract. A blue banner at the bottom invites readers to read the full article at ai.nejm.org.
Editorial image from NEJM AI featuring a side profile of a silhouetted face with digital elements and colorful lines. Includes a quote from Isaac S. Kohane, M.D., Ph.D., on the role of AI in medical guidelines, emphasizing the importance of AI in avoiding preventable harm while maintaining the art of good medicine. The NEJM AI logo is displayed.
The image contains a quote about medical artificial intelligence, emphasizing its value in methodology rather than being a device. The quote mentions embedding strategy, retrieval parameters, clinical workflow, and validation. It is attributed to Adrian Gropper, M.D., from the Perspective titled “The Medical AI Assistant as Publication, Not Device — Why Peer-Reviewed, Open-Source AI Belongs in the Standard of Care.” The NEJM AI logo is displayed in the lower-right corner. The background is a blue gradient.
Text from an editorial published in NEJM AI reads: “The danger is not that AI will fail to produce guideline-like prose. The danger is that it will produce too much of it or plausible ‘guidance’ for situations in which no trustworthy guidance exists. . . . It will be easy for health systems, payers, regulators, vendors, and professional societies to mistake linguistic competence for clinical legitimacy.” Editorial, “Guideline Machines” by Isaac S. Kohane, M.D., Ph.D.
Promotional image for episode 43 of the NEJM AI Grand Rounds podcast featuring OpenAI's Karan Singal discussing HealthBench and the future of medical AI. Includes a photo of Dr. Singal.
Figure 2 shows a world map highlighting 21 institutions using MEDS, marked with black pins. Panel B lists 27 published articles, 17 datasets, and 12 models that have used MEDS. Panel C displays icons representing the more than 14 tools publicly available in the MEDS ecosystem that enable workflows to be written with (in cases) up to 70% fewer lines of functional code and run orders of magnitude faster than prior research code.
Quote on a blue background from a Policy Corner article discussing the economic implications of treatment based on patient risk and responsiveness. It is cited from "Artificial Intelligence as a Return on Investment Multiplier in Health Care" by Jonathan D. Ketcham, Ph.D. The NEJM AI logo is visible.
A quote discussing the need for data governance policies for AI scribes relative to federal and state laws. Followed by "Policy Corner" and the article title, “Privacy Considerations of Artificial Intelligence Scribes,” by Sara Gerke, Dipl.-Jur. Univ., MA, and David A. Simon, PhD, JD, LLM. The NEJM AI logo is present.
The first page of an editorial titled "Medicine as an Information Industry in the Age of Language Models" published in NEJM AI, authored by Jeffrey M. Drazen, MD, and Charlotte J. Haug, MD, PhD. A banner at the bottom invites readers to view the full editorial at ai.nejm.org.
A three-part diagram illustrating the design principles and schema of MEDS.
A colorful graphic accompanies text from a Policy Corner article published in NEJM AI. The text highlights the impact of concentrated therapies among patients, indicating changes in treatment numbers, outcomes, and investment returns. The article is titled "Artificial Intelligence as a Return on Investment Multiplier in Health Care" by Jonathan D. Ketcham, Ph.D.
Table displaying data retention policies for AI scribes from various vendors. It includes rows for Freed, Heidi Health, Microsoft Dragon Copilot, Nabla, Suki, Sunoh.ai, Tali AI, and Upheal. Columns include the following: Audio Recording, AI-Generated Transcript, AI-Generated Clinical Note, and AI Training.
This image features a quote discussing the role of language models (LLMs) in health care, emphasizing their limitations in understanding patient needs. The quote is from an editorial titled "Medicine as an Information Industry in the Age of Language Models" by Jeffrey M. Drazen, M.D., and Charlotte J. Haug, M.D., Ph.D., published in NEJM AI, with a blue gradient background.
The image is a table titled "Table 1. States with All-Party or Mixed Consent Requirements for Audio Recording." It has three columns: "Category," "States," and "Implication for Clinical Ambient AI Deployment." The table lists different categories of consent, corresponding states, and implications for AI deployment. Notable entries include all-party consent required in states like California and Florida, and in-person consent in states like Oregon and Illinois.
A quote about large language models (LLMs) is displayed, highlighting their potential and associated risks. It refers to informed consent. The article is titled “Large Language Models in Informed Consent — Opportunities, Evidence, and Challenges” by R. Goel et al. The image features the NEJM AI logo.
This image is a screenshot of an NEJM AI article titled "A Typology of Generative Health Care Artificial Intelligence — Definitions and Policy Implications" by David Blumenthal , M.D., M.P.P., and Vivian S. Lee , M.D., Ph.D., M.B.A. A blue banner at the bottom invites readers to view the full article at ai.nejm.org.
The image features a stylized human profile with multicolored stripes, alongside a quote discussing litigation involving clinical ambient AI. It highlights the need for updated legal and regulatory guidance. The quote is from a Perspective by T.N. Anderson et al. published in NEJM AI.
This image is a table titled "Large Language Model Applications for Informed Consent (Examples, Evidence, Risks, and Mitigations)." It includes columns for Application Domain, Potential Use Cases, Key Resources and Models, Principal Risks, and Recommended Mitigation. Examples include technical term explanations, language assistance using GPT-4 and ChatGPT, and chatbots for consent processes, with risks and mitigations noted for each.
On the right, there's a photo of Dr. Travis Zack. On the left, there's a promo for "AI Grand Rounds, Episode 42: The OpenEvidence Episode: Dr. Travis Zack on the Future of Clinical Evidence." A digital background with binary code is present.
Quote about conceptual clarity in AI and health care, emphasizing effective implementation and regulation. The Perspective is titled "A Typology of Generative Health Care Artificial Intelligence — Definitions and Policy Implications" by David Blumenthal and Vivian S. Lee, published by NEJM AI.
Figure 2. Comparison of the Primary Outcome for Physicians in the Treatment Group with Physicians in the Control Group.
Table 1. Overview of the Health AI Partnership Artificial Intelligence Vendor Disclosure Framework.
Figure 1. A Framework for Selecting Human–AI Collaboration Strategies as a Function of Relative Accuracy (X-Axis: Human>AI on the Left, AI>Human on the Right) and Error Complementarity (Y-Axis: Same Errors at Bottom, Different Errors at Top).
Figure 2. Quality Evaluations of ChexGen-Generated Images.
Page 1 of the Perspective “I Hope You Are Doing Well — Will AI Widen or Close Health Care’s Disparity Gap?” Read the full Perspective at ai.nejm.org.
Figure 2. Comparison of the Primary Outcome for Physicians in the Treatment Group with Physicians in the Control Group.
Table 1. Overview of the Health AI Partnership Artificial Intelligence Vendor Disclosure Framework.
Figure 1. A Framework for Selecting Human–AI Collaboration Strategies as a Function of Relative Accuracy (X-Axis: Human>AI on the Left, AI>Human on the Right) and Error Complementarity (Y-Axis: Same Errors at Bottom, Different Errors at Top).
Illustration of X-ray images of a chest with medical terms about lungs and radiology overlaid in various colors on a blue background. Original Article A Generative Foundation Model for Chest Radiography Y. Ji and Others
“If clinicians and health care leaders committed to equity shape this transformation, AI could narrow disparity gaps while preserving the central role providers have held since Hippocrates.” Perspective “I Hope You Are Doing Well — Will AI Widen or Close Health Care’s Disparity Gap?” by Sam Joseph Amirfar, M.D., M.S.
Figure 1. Summary of the Contributions of this Article.