Evaluating Large Language Models for Extracting Clinical Recommendations from Practice Guidelines: A Preliminary...

CPGs (Clinical Practice Guidelines) contain the best care practices for clinicians to use and are created in many formats. The development of LLMs (Large Language Models) has led to their use in extracting or adapting CPG content to improve the ease of access to information for clinicians. In this paper, we investigate four different LLMs' ability to extract clinical recommendations from guidelines and apply basic categories to each recommendation, with one test being performed with an example of extracted recommendations and one test without this example. Of the LLMs used, DeepSeek and Grok created the best outputs, extracting the most recommendations and extracting them most correctly, achieving >90% accuracy. While this model does show that LLMs show promise in knowledge extraction, this preliminary evaluation highlights both potential and limitations of LLMs in automating knowledge extraction from clinical guidelines.