Id: 4347362 From Prediction to Intervention: Machine Learning Identifies Modifiable Drivers of Af Progression for Clinical Action
Year: 2025
Type: article
Study design: Clinical Trial
Abstract: Background: Patient education to enable timely, guidelineconcordant atrial fibrillation (AF) management remains a challenge.Artificial intelligence (AI) agents built on large language models could enhance patient education, if concerns like acceptability, empathy and trust are addressed.Objectives: To perform a pilot study to evaluate feasibility of using an AI agent for AF education.Primary outcome was patient perceived degree of empathy in agent responses.Secondary outcomes were AF knowledge questionnaire score, and trust in the healthcare provider (HCP) before and after agent interaction.Methods: The agent was built on ChatGPT 4o, via physicianguided prompt engineering and retrieval augmented generation using proprietary text and videos from Heart Rhythm Society's patient education website (UpBeat.org) to constrain responses and minimize hallucinations.Following HCP encounter, a textbased Q-and-A interaction between patient and agent occurred, ending with a link to UpBeat.Org AF videos (Fig 1).Two surveys were conducted pre-/post-agent interaction: a Likert score based quantitative survey about patients' AF experience (scale 1-7, higher score implying more trust, empathy) with one qualitative query about AF education; and the Jessa AF knowledge questionnaire (score range 0%-100%).Results: We piloted the agent at four hospitals (community and quaternary care centers) within a large academic health system.We recruited 40 patients (20 inpatient, 20 outpatient) with AF (57% female, 97.5% White, median age 69.5 yrs [IQR 9.5]), of whom 42.5% had paroxysmal AF, 57.5% persistent.Mean patient perceived empathy score for patient-agent interaction was numerically higher than patient-HCP interaction (6.361.7 vs 5.860.9;CI -0.1-1.1;P50.10).There was a statistically significant increase in mean AF knowledge score after agent interaction (76.7%616.2%vs 69.3%616.2%;CI 0.2-14.6;P50.04), and there was no change in patients' trust in their HCP (6.561.3 vs 6.561; CI -0.5-0.5;P51.00) (Fig 2).Qualitative theme analysis revealed desire for more AF education, positive agent experience and significant trust in HCPs.Conclusion: AF education using a physician-guided AI agent is feasible and well accepted by patients.Perceived degree of agent empathy was non-inferior to HCP.Agent interaction improved patients' AF knowledge without diminishing trust in HCP.Further studies testing AI agents for patient education, engagement and behavioral change are warranted.
Source: Heart Rhythm O2
Authors: Mohammad Montaser Atasi, Yishi Jia, Chanho Lim, Shea Hubbard, Ghassan Bidaoui, Mayana Bsoul, Hadi Younes, Abboud Hassan, Michel Abou Khalil, Christian M. Massad, Yara Menassa, Yingshuo Liu, Duo Yu, Eli Tsakiris, Joe Abi-Rached, Hongyu Miao, Xiang Li, Estelle Taki, Charbel Noujaim, Charles Campbell, Amitabh C. Pandey, Swati Rao, Omar Kreidieh, Han Feng, Nassir F. Marrouche
Language: English
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,150
OpenAlex ID: W4413966812