Effects of large language model-generated, patient-oriented discharge summaries on patient activation: a single-centre, single-blind, randomised controlled trial in Germany

Background: The transition from hospital to home represents an important period in patient care. Evidence suggests that improving patient activation (knowledge, skills, and confidence in self-care) reduces adverse post-discharge outcomes. Patient-oriented discharge summaries (PODS) show promise in increasing patient activation, but are laborious and time-consuming to generate manually. We investigated whether patient-oriented discharge summaries generated by a large language model (LLM) could enhance patient activation compared with standard discharge summaries. Methods: In this single-centre, single-blind, randomised controlled trial in the Department of Internal Medicine, Helios University Hospital Wuppertal (Wuppertal, Germany), adults aged 18 years and older were randomly allocated (1:1) to receive either a standard discharge summary (control) or an LLM-generated PODS (intervention). The LLM was GPT-4o (version 2024-11-20). Anonymised standard discharge summaries were input into the LLM, which was prompted to produce the PODS. LLM outputs were reviewed by the treating physician. The primary outcome was patient activation, assessed using the Patient Activation Measure (PAM-13). This study is registered with the German Clinical Trials Register, DRKS00035739, and is completed. Findings: Between Jan 7 and Feb 5, 2025, 128 participants were recruited and randomly assigned, 64 to the intervention group and 64 to the control group. 68 (53%) participants self-identified as men, 57 (45%) as women, and two (2%) as diverse; gender data were missing for one participant. At discharge, the intervention group had significantly higher PAM-13 scores than controls (Hodges–Lehmann median of differences 9·6 points [95% CI 4·3–15·1], p=0·0009). Interpretation: LLM-generated PODS improved patient activation, suggesting that they can facilitate more patient-oriented care transitions. Despite promising results, further research on quality assurance, data privacy, and regulation is needed before routine clinical adoption can be considered. Funding: School of Medicine, Faculty of Health, Witten/Herdecke University.