Can we use LLMs to generate data-driven patient narratives? A multi-method exploration in the oncology domain - Data and analysis code

A multi-method exploration in the oncology domain," authored by Nele Albers, Linwei He, Anouk Teunissen, Kelly de Ligt, Nadine Bol, and Emiel Krahmer. This includes: the LLM prompts and the real PROM profiles we used to generate narratives the narratives and LLM-as-judge accuracy evaluations together with any code used to generate them the data from the quality evaluation study with human participants data preprocessing code analysis code for 1) the textual properties of the generated narratives, 2) the descriptive accuracy analyses, 3) the quality variation across prompt paraphrases, 4) our statistical analyses, 5) our descriptive quality analyses, and for 6) computing the overall best and worst narratives