Abstract
Tables, figures, and listings (TFLs) are essential tools for summarizingclinical trial data. Creation of TFLs for reporting activities is often atime-consuming task encountered routinely during the execution of clinicaltrials. This study explored the use of large language models (LLMs) to automatethe generation of TFLs through prompt engineering and few-shot transferlearning. Using public clinical trial data in ADaM format, our resultsdemonstrated that LLMs can efficiently generate TFLs with prompt instructions,showcasing their potential in this domain. Furthermore, we developed aconservational agent named Clinical Trial TFL Generation Agent: An app thatmatches user queries to predefined prompts that produce customized programs togenerate specific predefined TFLs.