iiinova — where your digital self keeps evolving

Perspective by I. Nenadic et al.: From Technical Performance to Clinical Readiness: A Phase-Based Framework for Evidence Standards in Clinical Artificial Intelligence nejm.ai/4yIi4CK #AIinMedicine

000

NEJM AI @ai.nejm.org · 28/09/2026

Policy Corner by A. Wong et al.: How Repeal of the NTAP Alternative Pathway will Impact U.S. Clinical AI Innovation nejm.ai/4r7fjbd #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 25/09/2026

Original Article by L. Poursoltan et al.: Physician Edits to AI-Drafted Patient Messages and Their Impact on Clinical Workload nejm.ai/4c7M3Lv #ArtificialIntelligence #AIinMedicine

A diagram showing an AI-assisted clinical messaging workflow within EHR system.

A bar chart showing the distribution of clinician modification categories by frequency and extent of editing.

A pyramid chart showing the four-tier stratification of clinician workload for the categories of edits observed for AI-drafted replies.

000

NEJM AI @ai.nejm.org · 25/09/2026

Perspective by Susan Michie, DPhil (@susanmichie.bsky.social), Robert West, PhD (@robertjwest.bsky.social), and Janna Hastings, PhD (@jannahastings.bsky.social): Building an AI-Ready Evidence Base for Behavior Change nejm.ai/4wJFmX8 #ArtificialIntelligence #AIinMedicine

The image is page 1 of the Perspective "Building an AI-Ready Evidence Base for Behavior Change" published in NEJM AI. A dark banner at the bottom encourages readers to access the full Perspective at ai.nejm.org.

000

NEJM AI @ai.nejm.org · 24/09/2026

Case Study by A.M. Kerr et al.: Parental Perspectives on Using Large Language Model–Powered Chatbots in Rare Disease Care nejm.ai/4zChqYG #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 23/09/2026

On NEJM AI Grand Rounds, @xiaoliu.bsky.social says fewer than 5% of papers in their review met that bar, and the qualifying models performed at best equivalently to radiologists. The signal matters. So does separating it from the noise. Listen to the full episode: nejm.ai/ep46

010

NEJM AI @ai.nejm.org · 22/09/2026

In a new study on value-informed proxy decision support, Nolan and colleagues examine whether LLMs can improve alignment with patient preferences, extract values from clinical notes, and maintain performance across different reading levels. Full results: nejm.ai/3UfoOc9

A bar chart showing the average large language model–patient agreement (blue bars) compared with average human proxy–patient agreement (horizontal brown and black dotted lines) across 10 repetition trials (±95% confidence intervals).

020

NEJM AI @ai.nejm.org · 21/09/2026

Perspective by B. Sheng et al.: A Classification of Safety Risks in Medical AI nejm.ai/4wo9oQ4 #ArtificialIntelligence #AIinMedicine

The image is page 1 of the Perspective "A Classification of Safety Risks in Medical AI" published in NEJM AI. A dark blue banner at the bottom encourages readers to access the full Perspective at ai.nejm.org.

110

NEJM AI @ai.nejm.org · 18/09/2026

A new Perspective proposes a five-phase framework for evaluating the evidentiary maturity of clinical AI to help distinguish technical performance from clinical readiness and align claims with evidence needed to support safe, effective use. Learn more: nejm.ai/4yIi4CK

A table showing phase-based framework for evidentiary maturity in clinical artificial intelligence. The table includes five phases: Model development and representation learning, Internal validation, External validation, Decision impact and implementation readiness, and Postdeployment monitoring. Each phase details core questions, evidence expected, and example claims supported, from technical feasibility to durable clinical effectiveness.

000

NEJM AI @ai.nejm.org · 18/09/2026

A new Policy Corner analyzes how CMS’s repeal of the NTAP alternative pathway may exacerbate an existing imbalance in the U.S. market by stifling competition and driving health systems toward inferior but built-in EHR vendor–developed AI. Learn more: nejm.ai/4r7fjbd

A figure showing the New Technology Add-On Payment (NTAP) policy timeline and approvals by fiscal year. The annual number of NTAP approvals through the traditional (blue) and alternative (reddish brown) pathways is graphed alongside a timeline of major NTAP-related policy events. The number of NTAP technologies approved through the alternative pathway has steadily increased since the pathway was first established in 2020.

000

NEJM AI @ai.nejm.org · 16/09/2026

The promise of medical AI is not simply better performance — it is better care grounded in evidence that can be trusted. On AI Grand Rounds, @xiaoliu.bsky.social discusses what rigorous evaluation looks like as technologies move from publications into practice. Listen now: nejm.ai/ep46

Promotional image for episode 46 of the NEJM AI Grand Rounds podcast featuring Dr. Xiao Liu. The left side includes the title: "Beyond the Hype: Dr. Xiao Liu on Evaluating Medical AI." A photo of Dr. Liu is on the right.

011

NEJM AI @ai.nejm.org · 15/09/2026

Perspective by Lars Masanneck, MD, MSc, and Sibylle C. Mellinghoff, MD: Agentic AI Teammates in Medical Research — From Tools to Collaborators — and the Accelerating Digital Divide nejm.ai/4xIiMiR #ArtificialIntelligence #AIinMedicine

010

NEJM AI @ai.nejm.org · 14/09/2026

Drugs. Devices. And a third modality: AI-based interventions. On NEJM AI Grand Rounds, Dr. Suchi Saria envisions software protocols that identify precisely when to act, what to do, and which patient needs it — while making delivery easier at scale. Hear more from Dr. Saria: nejm.ai/ep45

000

NEJM AI @ai.nejm.org · 11/09/2026

A new study examines parents’ perceptions of a secure LLM chatbot for pediatric cancer and vascular anomalies, highlighting opportunities for caregiver support alongside challenges related to trust, overreliance, emotional readiness, and implementation. Learn more: nejm.ai/4zChqYG

A table showing parents’ recommendations for chatbot developers. It includes two columns: "Theme" and "Recommendations." The themes listed are "Tailored information," "Response format," "Queries," and "Chat history." Recommendations include ensuring accuracy, providing simplified information, offering summaries, following up responses, allowing search within the chatbot, and letting users delete chat histories.

001

NEJM AI @ai.nejm.org · 11/09/2026

Perspective by Alfredo Madrid-García, PhD, Beatriz Merino-Barbancho, PhD, and Miguel Rujas, MSc: When the Chatbot Leaks: Securing Patient-Facing Medical AI in the Age of Dual-Use Large Language Models nejm.ai/45py8fS #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 10/09/2026

A new study analyzing the impact of clinicians modifying AI-generated drafts within an electronic health record showed that complex edits require the most additional time while common administrative edits contribute the greatest workload. See how: nejm.ai/4c7M3Lv

A table showing the cumulative clinician time spent editing responses by modification category, April 2024–August 2025. Categories include scheduling, lifestyle advice, and laboratory results, among others. The table details the percentage increase per message, number of messages, and cumulative increase in response time (hours per 1000 physicians).

010

NEJM AI @ai.nejm.org · 09/09/2026

Documentation is not the holy grail. Better care is. On NEJM AI Grand Rounds, Dr. Suchi Saria argues that clinical AI must move into the center of the encounter, where rigorous tools can help clinicians recognize risk and act sooner. Listen to the full episode: nejm.ai/ep45

020

NEJM AI @ai.nejm.org · 08/09/2026

The authors of a new Perspective argue that realizing the potential of AI to improve behavior change interventions will require a fundamental transformation of behavioral science. Read the full Perspective: nejm.ai/4wJFmX8

000

NEJM AI @ai.nejm.org · 07/09/2026

A new study evaluates a value-informed LLM framework designed to support surrogate decision-making for patients who lose decisional capacity by comparing LLM-generated treatment recommendations with choices made by patients and their proxies. Learn more: nejm.ai/3UfoOc9

000

NEJM AI @ai.nejm.org · 04/09/2026

On NEJM AI Grand Rounds, Dr. Suchi Saria explains why health systems need evidence that AI can improve outcomes, reduce utilization, and support a sustainable path to implementation. Listen to the full episode: nejm.ai/ep45

000

NEJM AI @ai.nejm.org · 02/09/2026

Current medical AI frameworks do not sufficiently address patient safety because harms are often delayed, difficult to attribute, and irreversible. A new Perspective proposes a four-tier classification of medical AI safety. Learn more: nejm.ai/4wo9oQ4

100

NEJM AI @ai.nejm.org · 01/09/2026

A new Perspective examines how agentic AI may transform biomedical research by shifting bottlenecks from analysis itself to the data infrastructure, governance, and institutional capabilities required to deploy AI effectively. Learn more: nejm.ai/4xIiMiR

This image is a flowchart titled "Worked Example of an Agentic Research Workflow from Clinical Question to Auditable Output." A clinician-scientist’s question — illustrated here by a multiple sclerosis example — can be translated into a multistep agentic workflow that defines a computable cohort, searches a secure data environment, maps concepts to available data, retrieves external knowledge, drafts an analysis plan, generates code for supervised execution, produces an auditable report, and routes results to human review.

020

NEJM AI @ai.nejm.org · 31/08/2026

What changes when an abstract research problem becomes a family tragedy? On NEJM AI Grand Rounds, Dr. Suchi Saria explains how losing her nephew to sepsis led her to build Bayesian and pursue earlier, actionable recognition at the point of care. Listen to the full episode: nejm.ai/ep45

020

NEJM AI @ai.nejm.org · 28/08/2026

Sign up for our next event exploring how AI is reshaping guideline development and clinical decision-making. Leading experts will discuss how to harness the strengths of LLMs while preserving clinician autonomy, minimizing bias, and ensuring evidence continues to guide care. nejm.ai/4zKA5S4

011

NEJM AI @ai.nejm.org · 28/08/2026

A new Perspective describes an anonymized security assessment of a publicly deployed patient-facing retrieval-augmented generation chatbot and argues that patient-facing health AI systems must be evaluated as secure software deployments. Learn more: nejm.ai/45py8fS

A table showing minimum security and governance expectations for patient-facing retrieval-augmented generation chatbots. It has columns for Domain, Key Question, and Minimum Expectation. Domains include Configuration, Access control, Data stewardship, Knowledge base, Response minimization, Monitoring, Independent audit, and Dual-use readiness. Key questions address aspects like system prompts and RAG parameters, administrative and data end points, and storage of patient conversations. Minimum Expectation provides requirements such as verifying access, assuming redactions, and having audits before launch.

020

NEJM AI @ai.nejm.org · 27/08/2026

Editorial by Todd Hollon, MD: Benchmarking the Brain’s Blood Vessels — Why We Need Specialized Models in Medical AI nejm.ai/44Nz5hT #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 26/08/2026

Original Article by T.A. Nahass et al.: Implementation of an AI-Triggered Rapid Response — Association with Mortality nejm.ai/3TsbwbS #ArtificialIntelligence #AIinMedicine

The image is Figure 1, which is titled "Visual Displays of the Epic Deterioration Index and Implementation of Escalating Responses." Panel A shows the patient list with varying levels of severity for the Epic Deterioration Index along with a hover report showing the trend in a patient’s score over time. Panel B shows the storyboard alert that is continually present if a patient has had a red alert in the past 48 hours. Panel C is an example of the mobile push notification, with key factors contributing to the score.

000

NEJM AI @ai.nejm.org · 25/08/2026

What makes medicine such a compelling test for AI? The data are messy, noisy, and biased. On NEJM AI Grand Rounds, Dr. Suchi Saria argues that this complexity is exactly where well-designed machine learning can create meaningful clinical value. Hear more from Dr. Saria: nejm.ai/ep45

010

Reposted by NEJM AI

David Shaywitz @dshaywitz.bsky.social · 20/08/2026

Perspective @ai.nejm.org from @susanmichie.bsky.social & colleagues-who have devoted careers to substantial structural challenges around study of behav health-offers impt perspective on what needs to change to accelerate meaningful learning in this vitally important area. ai.nejm.org/doi/full/10....

ai.nejm.org

Building an AI-Ready Evidence Base for Behavior Change

AI has begun to make an impact on the development of interventions to promote healthier behaviors and assist with disease management, primarily through apps and chatbots, but also in health communi...

023

NEJM AI @ai.nejm.org · 24/08/2026

Datasets, Benchmarks, and Protocols article by K. Yang et al.: The TopCoW Challenge — Topology-Aware Circle of Willis Segmentation for CT and MR Angiography nejm.ai/4fnSMTs #ArtificialIntelligence #AIinMedicine

The image is a composite diagram showing the process and results of annotating the Circle of Willis (CoW) in medical imaging. It includes 3D visualizations, CT and MR scans, voxel annotations, and graphs. Panel A shows data annotation using virtual reality, tailored settings, and 13 anatomical labels for the CoW anatomy. In Panel B, the Topology-Aware Anatomical Segmentation of the Circle of Willis for Computed Tomography Angiography and Magnetic Resonance Angiography (TopCoW) dataset provides paired images from two modalities, computed tomography angiography and magnetic resonance angiography, for each patient. In Panel C, the CoW segmentation mask is converted into a CoW variant graph annotation. Panel D shows interrater agreement for CoW variant classification on 40 TopCoW CTA test cases.

A figure showing the progression of TopCoW submissions and winning strategies. Panel A shows the performance of the five teams that participated in both years on the same 34 patients in each year’s test sets. The metrics shown are class-average Dice scores, variant-balanced accuracy of the anterior and posterior circle of Willis variant classifications, and average F1 score for detection of anterior communicating arteries, posterior communicating arteries, and third A2 arteries. Panel B shows key characteristics of the segmentation algorithms from the top six teams from both tracks in alphabetical team name order. Panel C shows two methodological breakthroughs for segmentation in 2023 that got picked up by many more teams in the following year.

010

NEJM AI @ai.nejm.org · 21/08/2026

Case Study by C.R. Cleland et al.: Will Artificial Intelligence Augment Global Health Inequalities? nejm.ai/4aPRgqG #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 21/08/2026

Editorial by A.A. Agarwal et al.: After Clearance — Continuous Monitoring as the Foundation of Clinical AI Oversight nejm.ai/4wffVgQ #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 19/08/2026

A patient can deteriorate while warning signs remain buried in an electronic health record. Dr. Suchi Saria has spent her career trying to surface those signals sooner. On NEJM AI Grand Rounds, she explains why a strong model is only the starting point. Listen now: nejm.ai/ep45

Promotional image for episode 45 of the NEJM AI Grand Rounds podcast featuring Dr. Suchi Saria discussing building AI that changes care. Includes a photo of Dr. Saria.

011

NEJM AI @ai.nejm.org · 18/08/2026

Original Article by @juliecachia.bsky.social et al.: AI for Proactive Mental Health: A Multi-Institutional, Longitudinal Randomized Controlled Trial nejm.ai/3TJ9iVy #ArtificialIntelligence #AIinMedicine

The image is a high-level summary of the behavioral loop of the Flourish app intervention. It features three main sections: Real-Time Support, Science-Based Activities, and Weekly Insights. Each section contains app screenshots displaying cheerful graphics, icons, and user interface elements, such as messages and charts. Arrows connect the sections, indicating a progression flow.

000

NEJM AI @ai.nejm.org · 17/08/2026

TopCoW is the first public benchmark and paired computed tomography angiography and magnetic resonance angiography dataset for topology-aware anatomical segmentation of the circle of Willis. Learn more: nejm.ai/4fnSMTs

000

NEJM AI @ai.nejm.org · 15/08/2026

AI alone won’t solve biology. In the latest episode of the NEJM AI Grand Rounds podcast, Brandon Rice of Weave argues that trust and domain expertise remain essential. Listen to the full episode: nejm.ai/ep44

000

NEJM AI @ai.nejm.org · 14/08/2026

Perspective by David Blumenthal, MD, MPP, and Meredith B. Rosenthal, PhD: How the Impact of Artificial Intelligence on Health Care Costs Will Be Shaped by Policy and Management Choices nejm.ai/4f3PfcX #ArtificialIntelligence #AIinMedicine

The image is page 1 of the Perspective "How the Impact of Artificial Intelligence on Health Care Costs Will Be Shaped by Policy and Management Choices" published in NEJM AI. A dark blue banner at the bottom encourages readers to access the full Perspective at ai.nejm.org.

010

NEJM AI @ai.nejm.org · 14/08/2026

Perspective by Fang-Yi Su, MD, PhD, et al.: Bridging the Gap — Translating AI in Pathology into Clinical Impact nejm.ai/44xQhI1 #ArtificialIntelligence #AIinMedicine

The image is page 1 of the Perspective "Bridging the Gap — Translating AI in Pathology into
Clinical Impact" published in NEJM AI. An orange banner at the bottom encourages readers to access the full Perspective at ai.nejm.org.

000

NEJM AI @ai.nejm.org · 12/08/2026

A new study evaluated the implementation of a machine learning–based clinical deterioration prediction model integrated with automated rapid response team notifications across 11 hospitals in a large regional health system. Full study results: nejm.ai/3TsbwbS

A sequence of illustrations depicting a patient's progression from walking independently to using a walker, and finally being assisted in a wheelchair. The background includes health chart graphics and age milestones. Color-coded symbols for prescription, medical care, and caution appear at different stages of the timeline.

001

NEJM AI @ai.nejm.org · 11/08/2026

The future of medical AI will require clinically grounded datasets, shared benchmarks, multimodal learning, and annotation systems that make expert knowledge easier to capture. Read the editorial by Todd Hollon, MD: nejm.ai/44Nz5hT

020

NEJM AI @ai.nejm.org · 10/08/2026

The first public benchmark and paired CT and MR angiography dataset for topology-aware circle of Willis segmentation demonstrates that AI models can delineate cerebrovascular anatomy across diverse imaging modalities and institutions. Learn more: nejm.ai/4fnSMTs

000

NEJM AI @ai.nejm.org · 07/08/2026

Don’t build for today’s model. Build for tomorrow’s. On the NEJM AI Grand Rounds podcast, Brandon Rice explains why Weave treats AI as the engine — not the vehicle. Listen to latest episode: nejm.ai/ep44

000

NEJM AI @ai.nejm.org · 07/08/2026

A new editorial examines the limitations of current regulatory approaches for clinical AI and argues for continuous postdeployment oversight as part of an evolving regulatory framework. Full editorial: nejm.ai/4wffVgQ #ArtificialIntelligence #AIinMedicine

Figure 1 illustrates the current versus proposed regulatory pathways for clinical AI. Panel A: The current pathway, which relies on retrospective evidence and includes no systematic postmarket evaluation. Panel B: The proposed pathway, which adds multisite validation and continuous monitoring with a feedback loop for reevaluation when performance thresholds are not met.

031

NEJM AI @ai.nejm.org · 05/08/2026

A new Case Study maps ophthalmology AI research against workforce capacity and blindness burden, showing how concentrated AI development in high-income countries could worsen global health inequities. Learn more: nejm.ai/4aPRgqG #ArtificialIntelligence #AIinMedicine

Three density-weighted cartograms display global distributions. Panel A presents age-adjusted prevalence of blindness (%) in 2020; Panel B presents number of ophthalmologists per million population; and Panel C presents number of AI publications in ophthalmology from 2008 to 2025.

000

NEJM AI @ai.nejm.org · 03/08/2026

A new study evaluates Flourish, a mobile app using generative AI to deliver personalized, strengths-based well-being support to young adults through an AI well-being coach, in a 6-week randomized trial across three U.S. campuses. Learn more: nejm.ai/3TJ9iVy

Illustration of three individuals walking across large, stacked smartphone screens. A heading below reads: "AI for Proactive Mental Health: A Multi-Institutional, Longitudinal Randomized Controlled Trial."

020

NEJM AI @ai.nejm.org · 03/08/2026

Before a drug reaches patients, it has to clear a critical gate. In the latest episode of the NEJM AI Grand Rounds podcast, Brandon Rice of Weave explains why the Investigational New Drug (IND) process shapes everything that follows. Learn more: nejm.ai/ep44

000

NEJM AI @ai.nejm.org · 31/07/2026

A new Perspective examines the impact of AI on national health care expenditures, arguing that near-term cost reductions will depend heavily on policy and management choices rather than the technology alone. Learn more: nejm.ai/4f3PfcX #ArtificialIntelligence #AIinMedicine

000

NEJM AI @ai.nejm.org · 29/07/2026

Drug development has embraced cutting-edge science. The paperwork often hasn’t. In the latest episode of the NEJM AI Grand Rounds podcast, Brandon Rice of Weave explains why regulatory workflows are a major opportunity for AI. Full episode: nejm.ai/ep44

001

NEJM AI @ai.nejm.org · 29/07/2026

A new Perspective examines why pathology, despite being foundational to modern diagnostics, has seen limited clinical adoption of AI-enabled imaging tools. It outlines barriers and pathways for AI-assisted workflows and clinical trial applications. Learn more: nejm.ai/44xQhI1

000

NEJM AI @ai.nejm.org · 24/07/2026

Curiosity came first. Entrepreneurship came later. In the latest episode of the NEJM AI Grand Rounds podcast, Brandon Rice traces the path from building computers as a kid to building Weave. Listen to the full episode: nejm.ai/ep44

000