Announcement
Thank you for attending THSNA 2026. The virtual meeting is now closed.
Thank you for attending THSNA 2026. The virtual meeting is now closed.
Presentation Details
| AI-Driven Dynamic Patient Avatars with Recurrent Thrombosis and Inherited Thrombophilia: A Scalable Simulation Tool for Physician Training and Handoff Anvi Agarwal1, Asmi Agarwal1, Elijah Eldrim2, Andrew Primous2, Yanhui Guo3, Ruchika Goel4, 5, Richard Selinfreund6. 1McNeese P4 (Physician Preparatory Pathway Program), SIU School of Medicine, Springfield, IL, USA.2Medical Microbiology and Immunology, SIU School of Medicine, Springfield, IL, USA.3Department of Computer Science, University of Illinois Springfield, Springfield, IL, USA.4Division of Hematology/Oncology, Simmons Cancer Institute, SIU School of Medicine, Springfield, IL, USA.5Department of Pathology, Johns Hopkins University, Baltimore, MD, USA.6Departments of Medical Microbiology, Cell Biology and Immunology, and Medical Education, Springfield, IL, USA |
Abstract
Background:
Rapid recognition and management of high-risk venous thromboembolism requires swift clinical assessment and multidisciplinary coordination. Traditional simulation-based education is resource-intensive and limited in scalability. We developed a novel AI-driven dynamic patient avatar, designed to emulate real-time clinical deterioration, variable communication patterns, and emotional nuances. It uses the case of a 36-year-old male presenting with recurrent Venous Thromboembolism (VTE), including bilateral deep venous thrombosis and hemodynamically unstable saddle PE with a history of inherited thrombophilia, to realistically train clinicians in acute decision-making and structured handoffs. Methods:
A multimodal avatar was created using generative AI, voice-to-animation engines, and a rules-based behavioral model. The avatar incorporates patient vital signs, physical exam findings, laboratory and imaging results (including bilateral proximal deep venous thrombosis and saddle PE (right heart strain and CT-confirmed PE), as well as a layered disclosure model which reflected realistic patient behaviors such as question deflection, anxiety, and symptom minimization. The system was tested in simulated sessions emphasizing relevant and focused history-taking, ordering imaging and laboratory investigations, including the details and timing of thrombophilia testing, initiation of anticoagulation therapy, and referral to specialist consultations, including hematology and vascular medicine. Medical trainees could interact with the avatar using natural language, triggering physiologic and conversational responses that evolved in real time with the clinical scenario. (Image 1: Click on QR code for sample AI patient avatar; Image 2: Still of AI patient avatar). Results:
This first-of-a-kind AI-based avatar provides a scalable, high-fidelity solution for simulation training in VTE management. By integrating dynamic physiology with natural language interaction and emotionally informed behavior, it addresses critical gaps in current educational methods. Unlike static case scenarios, the avatar fosters adaptive clinical reasoning and responsive communication, both essential in high-risk clinical cases. A scoring schema has been designed to objectively assess participants on improved ability to integrate diagnostic findings and prioritize clinical actions. These key markers are included in trainee reports following simulation exposure. Pilot assessments demonstrated earlier initiation of anticoagulation, more accurate PE risk stratification, and more complete ICU handoffs, including hemodynamic status, right-heart strain, thrombolysis need, and consultation triggers. Discussion/Conclusion:
This AI-based avatar provides a scalable, high-fidelity platform for teaching physicians and medical trainees acute PE management and handoff communication. By integrating clinical data, emotional realism, and dynamic physiologic modeling, it addresses critical gaps in current simulation training. This AI-driven patient avatar demonstrates a powerful new paradigm for education and clinical readiness in thrombosis and hemostasis and meaningfully enhances competency in diagnosing and managing acute venous thromboembolism. Its natural-language interaction model strengthens clinical reasoning, interdisciplinary coordination, and improves handoff communication, all of which are critical to improving outcomes in thrombotic emergencies. This approach can be generalized to broader spectrums of thrombosis and hemostasis scenarios, including cancer-associated thrombosis, HIT, massive PE requiring advanced therapies, unusual-site thrombosis, bleeding complications from anticoagulation, and perioperative coagulation management, representing a transformative direction for hematology and acute-care education. This proof-of-concept lays the groundwork for multicenter deployment, research on behavioral and cognitive outcomes, and integration into hemostasis thrombosis stewardship programs.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
Rapid recognition and management of high-risk venous thromboembolism requires swift clinical assessment and multidisciplinary coordination. Traditional simulation-based education is resource-intensive and limited in scalability. We developed a novel AI-driven dynamic patient avatar, designed to emulate real-time clinical deterioration, variable communication patterns, and emotional nuances. It uses the case of a 36-year-old male presenting with recurrent Venous Thromboembolism (VTE), including bilateral deep venous thrombosis and hemodynamically unstable saddle PE with a history of inherited thrombophilia, to realistically train clinicians in acute decision-making and structured handoffs. Methods:
A multimodal avatar was created using generative AI, voice-to-animation engines, and a rules-based behavioral model. The avatar incorporates patient vital signs, physical exam findings, laboratory and imaging results (including bilateral proximal deep venous thrombosis and saddle PE (right heart strain and CT-confirmed PE), as well as a layered disclosure model which reflected realistic patient behaviors such as question deflection, anxiety, and symptom minimization. The system was tested in simulated sessions emphasizing relevant and focused history-taking, ordering imaging and laboratory investigations, including the details and timing of thrombophilia testing, initiation of anticoagulation therapy, and referral to specialist consultations, including hematology and vascular medicine. Medical trainees could interact with the avatar using natural language, triggering physiologic and conversational responses that evolved in real time with the clinical scenario. (Image 1: Click on QR code for sample AI patient avatar; Image 2: Still of AI patient avatar). Results:
This first-of-a-kind AI-based avatar provides a scalable, high-fidelity solution for simulation training in VTE management. By integrating dynamic physiology with natural language interaction and emotionally informed behavior, it addresses critical gaps in current educational methods. Unlike static case scenarios, the avatar fosters adaptive clinical reasoning and responsive communication, both essential in high-risk clinical cases. A scoring schema has been designed to objectively assess participants on improved ability to integrate diagnostic findings and prioritize clinical actions. These key markers are included in trainee reports following simulation exposure. Pilot assessments demonstrated earlier initiation of anticoagulation, more accurate PE risk stratification, and more complete ICU handoffs, including hemodynamic status, right-heart strain, thrombolysis need, and consultation triggers. Discussion/Conclusion:
This AI-based avatar provides a scalable, high-fidelity platform for teaching physicians and medical trainees acute PE management and handoff communication. By integrating clinical data, emotional realism, and dynamic physiologic modeling, it addresses critical gaps in current simulation training. This AI-driven patient avatar demonstrates a powerful new paradigm for education and clinical readiness in thrombosis and hemostasis and meaningfully enhances competency in diagnosing and managing acute venous thromboembolism. Its natural-language interaction model strengthens clinical reasoning, interdisciplinary coordination, and improves handoff communication, all of which are critical to improving outcomes in thrombotic emergencies. This approach can be generalized to broader spectrums of thrombosis and hemostasis scenarios, including cancer-associated thrombosis, HIT, massive PE requiring advanced therapies, unusual-site thrombosis, bleeding complications from anticoagulation, and perioperative coagulation management, representing a transformative direction for hematology and acute-care education. This proof-of-concept lays the groundwork for multicenter deployment, research on behavioral and cognitive outcomes, and integration into hemostasis thrombosis stewardship programs.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.