AI Based Clinical Decision-Making Tool for Neurologists in the Emergency Department.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Neurology

Background:

• Emergency departments (EDs) face challenges in timely neurological consultation and decision-making.
• Advanced machine learning (ML) and large language models (LLMs) offer potential for improved clinical decision support.

Purpose of the Study:

• To evaluate an ensemble AI framework for predicting clinical outcomes in emergency department neurology patients.
• To assess the AI's accuracy in predicting hospital admission and mortality.
• To determine the AI's alignment with senior neurologist clinical judgment.

Main Methods:

• Engineered an ensemble AI framework using the Gemini 1.5-pro-002 LLM, prompt engineering, and retrieval-augmented generation (RAG).
• Combined eXtreme Gradient Boosting (XGBoost) and logistic regression for predictive modeling.
• Retrospectively analyzed 1368 ED neurological consultation cases, assessing clinical features and outcomes. Validated AI predictions against expert consensus on 100 cases.

Main Results:

• The AI model achieved an area under the curve (AUC) of 0.88 for general admission prediction and 0.86 for neurology department admissions.
• The model demonstrated high accuracy in predicting mortality risk (AUC 0.93 for long-term, 1.0 for 48-hour).
• AI predictions showed strong correlation with expert consensus (Pearson correlation 0.79, p < 0.001).

Conclusions:

• The developed Neuro AI model accurately predicts hospital admissions and neurological department admissions in the ED.
• The AI's strong performance and alignment with expert judgment indicate its utility in enhancing clinical decision-making for neurologists.
• This AI framework shows promise for improving patient management and outcomes in acute neurological care settings.