Comparing Humans and Large Language Models in Filling Clinical Questionnaires

Abstract

Filling clinical questionnaires to perform retrospective studies is a time-consuming task that requires strong expertise in specific domains. We exploit prompt engineering techniques to optimize the completion of clinical questionnaires through Large Language Models (LLMs), aiming to compare their performance with respect to human experts. Despite challenges related to limited access to input data, our preliminary experimental results demonstrate the potential of LLMs to streamline clinical data collection, greatly reducing the manual workload for healthcare professionals. However, human validation remains essential to ensure accuracy and reliability in real-world applications.