APSR2026-329

Yes
Late-Breaking
Submitted
Oral Presentation
AO11-005
 
 
https://storage.unitedwebnetwork.com/files/1285/1218206-46903-APSR26_Oral+admissionAC_433882.pptx
Format: PPTX Only
 
Format: PDF Only
15. Pulmonary Circulation
5

No more than 15 authors can be listed.
Please ensure the authors are listed in the right order.
If there are multiple affiliations, please use the "+" button next to "Select Country/Region" to add additional affiliation fields.

Guohui Fan * China-Japan Friendship Hospital Department of Clinical research and Data management, Center of Respiratory Medicine Beijing China
Dingyi Wang - China-Japan Friendship Hospital Department of Clinical research and Data management, Center of Respiratory Medicine Beijing China
Jing Ma - China-Japan Friendship Hospital Center of Respiratory Medicine Beijing China
Feiya Xu - China-Japan Friendship Hospital Center of Respiratory Medicine Beijing China
Zhenguo Zhai - China-Japan Friendship Hospital Center of Respiratory Medicine Beijing China
 
 
 
 
 
 
 
 
 
 
Presenting Author Details
Guohui
Fan
fanguohui08@163.com
 
31-40
China
Original Research
Differences in Treatment Strategies for Intermediate-high to High Pulmonary Embolism: A Comparison of Human Decision-Making and AI-driven MDT Recommendations on 30-Day Outcomes Using Real-World Data
Maximum 25 words
Abstract Content
Intermediate-high to high pulmonary embolism (PE) is fatal but with controversial treatment strategies at acute phase. Large language models (LLMs) have shown considerable abilities in diagnostic reasoning yet are influenced by effective prompts and reasoning depth during application, making single-pass reasoning often insufficient for providing reliable assistance in medical decision-making. Here, we present MARS-PE (Multi-disciplinary Agent-based Resolution Simulator for PE), a framework designed to explore digital approaches for enhancing decision support in acute PE treatment.
Data from the multi-center China Pulmonary Thromboembolism Registry Study (CURES) were analyzed, which enrolled PE patients hospitalized from 2016 to 2021, and prospectively followed up for at least 1 year. Clinical features, temporal lab examinations and medications were recorded by pre-designed electronic data capture (EDC) system. MARS-PE was developed based on Qwen3-next-80B and simulated multidisciplinary team (MDT) by selecting pre-designed built-in medical specialty agents (pulmonologists, cardiologist, intensivist, pharmacist, emergency physician) to compare with actual treatment strategies (anticoagulation, systemic thrombolysis, or interventional therapy) chosen by physicians. Differences in treatment strategies and corresponding 30-day bleeding and all-cause mortality between MARS-PE and physicians were analyzed.
A total of 1522 intermediate-high to high PE patients were analyzed. The AI-human consistency rate in anticoagulant choice at discharge was 67.0% (1022 consistent vs 502 inconsistent). Among those inconsistent, 93.5% were prescribed with anticoagulants-only by physicians but were recommended by MARS-PE to thrombolysis (56.9%) and interventional therapy (36.9%), respectively. After being adjusted for Sex, BMI, age, PE risk level, hemoglobin and active cancer, COX regression showed human-AI consistent group contributed to significantly lower 30-day all-cause mortality (6.9% vs 13.6%, HR 0.638, 95% CI 0.436-0.932, P=0.020) compared with inconsistent group. No difference was observed for 30-day bleeding event.
MARS-PE as LLM-based MDT tool may provide effective decision support for treatment strategies during hospitalization for PE patients.
 
 
 
 
 
No
Yes
https://storage.unitedwebnetwork.com/files/1285/1218206-37089-InhosTreat_376496.png
JPG or JPEG or PNG format ONLY
AI-MDT, treatment strategies during hospitalization, PE management
maximum 3 keywords and no abbreviations, separated by commas
none
304