Knowledge assortment and analysis workflow. Credit score: npj Digital Drugs (2025). DOI: 10.1038/s41746-025-01972-w
A brand new research from NYU Tandon, NYU Langone Well being, and the NYU Stern Faculty of Enterprise affords one of many first data-driven appears to be like at how generative AI would possibly assist well being care suppliers handle their message overload—and why many are hesitant to undertake the expertise.
Over a ten-month interval from October 2023 by way of August 2024, a group led by Morton L. Topfer Professor of Know-how Administration Oded Nov noticed greater than 55,000 affected person messages despatched to well being care suppliers by way of a safe on-line affected person portal. The system used an embedded generative AI instrument that mechanically generated draft replies for incoming affected person messages; well being care suppliers might select to start out with the draft, start a reply from scratch, or use their standard reply interface.
The analysis is revealed in npj Digital Drugs.
“This paper provides evidence that AI has the potential to make patient-provider communication more efficient and more responsive,” says Soumik Mandal, analysis scientist and lead writer of the analysis. “To unlock its full potential in the next phase, however, will require tailored implementation to ensure that AI tools meaningfully reduce clinician burden while enhancing care quality. The paper outlines some practical strategies to improve draft utilization and guide future implementation efforts as key next steps.”
Different authors embrace NYU Stern’s Batia M. Wiesenfeld, in addition to NYU Langone Well being’s Adam C. Szerencsy, William R. Small, Vincent Main, Safiya Richardson, Antoinette Schoenthaler, and Devin Mann.
Tendencies in utilization. Credit score: npj Digital Drugs (2025). DOI: 10.1038/s41746-025-01972-w
Based on the revealed outcomes, suppliers selected to “Start with Draft” in 19.4% of circumstances the place a draft was proven. Adoption rose modestly over the course of the research because the system’s prompting improved. Utilizing a draft shaved roughly 7% off response occasions, a median of 331 seconds versus 355 seconds when drafting from scratch, however in lots of circumstances, this time saved was offset by time spent reviewing, enhancing, or ignoring drafts.
“LLMs are a new technology that can help providers be more responsive, more effective and more efficient in their communication with their patients,” says Nov. “The more we understand who uses it and why, the better we can leverage it.”
By analyzing tens of hundreds of messages, the researchers discovered that sure qualities made drafts extra probably for use. Shorter, extra readable, and extra informative drafts tended to be most popular. Tone additionally mattered: messages that sounded barely extra human and empathetic had been extra prone to be adopted, although the best stability differed by function.
Physicians leaned towards concise, impartial textual content, whereas help employees had been extra receptive to messages with a hotter tone. These preferences trace at a future the place AI techniques might adapt their writing type based mostly on the person’s function or communication historical past.
Nonetheless, the research reveals how hesitant well being care suppliers stay to depend on AI-generated language in any respect. The authors recommend a number of potential causes, together with suboptimal alignment with medical workflows, and the cognitive price of reviewing a continuing stream of AI output, a lot of which can be irrelevant. Merely producing textual content for each message, they argue, can create muddle that undermines the very effectivity such instruments are supposed to present.
The researchers see ample alternative forward. Future techniques could must be taught every person’s type, selectively generate drafts just for messages prone to profit, and repeatedly adapt immediate methods.
Extra info:
Soumik Mandal et al, Utilization of Generative AI-drafted Responses for Managing Affected person-Supplier Communication, npj Digital Drugs (2025). DOI: 10.1038/s41746-025-01972-w
Offered by
NYU Tandon Faculty of Engineering
Quotation:
New analysis reveals uptake of AI-powered messaging in well being care settings (2025, October 16)
retrieved 16 October 2025
from https://medicalxpress.com/information/2025-10-reveals-uptake-ai-powered-messaging.html
This doc is topic to copyright. Aside from any truthful dealing for the aim of personal research or analysis, no
half could also be reproduced with out the written permission. The content material is supplied for info functions solely.
