Performance of an artificial intelligence model for evaluation of unnecessary central lines, Northern California 2025

Key Points

• This research aims to assess the performance of an AI model in identifying unnecessary central lines within electronic health records.
• Integrated a large language model into the electronic health record system
• Evaluated the performance based on sensitivity and specificity metrics
• Achieved 16% sensitivity for detecting unnecessary central lines
• Revealed 99% specificity, indicating a low rate of false positives
• Suggests the model could be useful where human review is challenging