Domain-Specific Large Language Model Assistance for Emergency Neurological Diagnosis and Treatment（DEMAND）

Domain-Specific Large Language Model Assistance for Emergency Neurological Diagnosis and Treatment: A Randomized Controlled Clinical Trial

This study will evaluate whether Xuanwu-NeuroAid 2.0, a large language model for emergency neurology, can improve 30-day diagnostic quality in adults with acute neurological symptoms. Physicians will be randomly assigned to AI-assisted care or usual care. In the AI-assisted group, the model will provide diagnostic and management suggestions, while physicians will make all final clinical decisions. The usual-care group will receive standard emergency neurology care without large language model assistance.

This multicenter, prospective, cluster randomized trial will evaluate whether Xuanwu-NeuroAid 2.0, a domain-specific large language model for emergency neurology, can improve diagnostic quality in adults presenting with acute neurological symptoms. Physicians will be randomized to AI-assisted care or usual care. In the AI-assisted group, the model will provide diagnostic and management suggestions based on available clinical information, while physicians will remain responsible for all final clinical decisions. the model's recommendations could be disregarded when they were considered inappropriate. The usual-care group will receive standard emergency neurology care without large language model assistance.

Inclusion Criteria:

• Age ≥18 years;
• Presentation to the emergency neurology service with acute neurological symptoms;
• Written informed consent provided by the patient or a legally authorized representative.

Exclusion Criteria:

• Presentation primarily for trauma;
• Pregnancy;
• Requiring immediate life-saving interventions;
• Estimated life expectancy of less than 30 days;
• Participation in another clinical trial within the previous 30 days or in a trial that could interfere with the study or outcome assessment;
• Any condition that, in the opinion of the investigators, would interfere with the conduct of the trial or the interpretation of the results.