Prospective Randomized Controlled Trial to Evaluate Locally Implemented Large Language Models for Simplifying Patient Communication in Hematology and Oncology

What this Trial Is About

Researchers are evaluating how well patients with blood cancers or other cancers understand their medical information when it is rewritten in simpler language by an artificial intelligence (AI) system. This study focuses on patients discharged from the hospital who receive a medical letter summarizing their diagnosis, treatment, and next steps, which is often written in technical language. The trial compares the standard letter with one simplified by a large language model (LLM) running securely on hospital servers, with physician review before delivering the simplified version. Participants are randomly assigned to two groups: one receives both the standard discharge letter and the AI-simplified version, and the other receives only the standard letter. A separate non-randomized group includes patients with limited German language skills who receive a simplified and translated letter. The LLM simplifies specific sections of the discharge letter, including Current Status, Medical History, Epicrisis, and Further Management. Study physicians review the simplified letters to ensure accuracy before they are given to patients. During the study, participants read their assigned letter(s) and complete a short questionnaire measuring their understanding using a 5-item comprehension scale. Additional assessments include satisfaction with information, uncertainty reduction, format preference, and physician review metrics such as time and correction rate. All data are securely stored on hospital servers with no external data transmission. The study takes place at TUM University Hospital in Munich, Germany, and involves about 180 patients, including a translation arm for non-German speakers.

LLM-Generated Plain-Language Patient Synopses to Improve Comprehension in Hematology and Oncology (oncOPAL)