DeLiriuMAgents: A Large Language Model-Driven Multi-Disciplinary AI Agent System for Predicting Delirium in Emergency Critically Ill Patients
Research demonstration of DeLiriuMAgents for ED-ICU delirium risk assessment. The current version supports running only after a demo case is loaded. ⚠️ For research use only. Not for clinical use.
Task: delirium_agents

Inputs

Please first select an example case from the list below, then click Load to populate the form. After the case data are loaded, click Run at the bottom to generate the prediction results. In the current demo version, running is supported only for preloaded example cases.
Example case
The example cases used in this demo are derived from the MIMIC-IV database. Please select one case and click Load to populate the form before running the system.
Developed by
Peking University and Peking University Third Hospital
Technical contact
Tongyue Shi, Peking University, shitongyue@hsc.pku.edu.cn
References
[1] Wen Shang, Tongyue Shi, Qingbian Ma*, Guilan Kong*. DeLiriuMAgents: A Large Language Model-Driven Multi-Disciplinary AI Agent System for Predicting Delirium in Emergency Critically Ill Patients. 2026.
[2] Tongyue Shi, Jun Ma, Zihan Yu, Haowei Xu, Rongxin Yang, Minqi Xiong, Meirong Xiao, Yilin Li, Huiying Zhao, Guilan Kong*. Large Language Models in Critical Care Medicine: Scoping Review. JMIR Medical Informatics. 2025;13:e76326.
[3] Tongyue Shi, Meirong Xiao, Haowei Xu, Huiying Zhao, Guilan Kong*. AKI-Detector: A Multi-Agent Framework by Integrating Machine Learning and Large Language Models for Early Prediction of Acute Kidney Injury in ICU. Proceedings of AMIA 2025 Annual Symposium. 2025.
[4] Tongyue Shi, Liya Guo, Zeyuan Shen, Guilan Kong*. ERTool: A Python Package for Efficient Implementation of the Evidential Reasoning Approach for Multi-Source Evidence Fusion. Health Data Science. 2024;4:0128.

Outputs

The output panel presents the reports generated by ModelPhysicianAgent, EmergencyDoctorAgent, NeurologyDoctorAgent, PsychiatryDoctorAgent, together with the final integrated report of DeLiriuMAgents. In the four individual agent report boxes, users can scroll vertically to read the complete report content. A red background denotes a prediction of delirium occurrence, whereas a green background denotes a prediction that delirium will not occur.
Report of ModelPhysicianAgent
None
Report of EmergencyDoctorAgent
None
Report of NeurologyDoctorAgent
None
Report of PsychiatryDoctorAgent
None
Final prediction report by DeLiriuMAgents
None