Feasibility of large language Model–Generated thematic analysis in healthcare qualitative research: A comparative study

Background Qualitative research is critical in transplantation science for understanding stakeholder perspectives and informing implementation, ethics, and operations. However, qualitative analysis is time-intensive and requires multiple trained team members. Large language models (LLMs) such as ChatGPT have emerged as potential tools for qualitative analysis, but their performance relative to trained human coders remains unclear. The study compared LLM-generated and human thematic analyses across three transplant studies. Methods The Naeem et al. six-step thematic analysis process was applied to three existing Datasets: (1) semi-structured interviews with United States (US) Organ Procurement Organization executives about normothermic regional perfusion (NRP) practices and ethical concerns; (2) open-ended survey responses from parents of pediatric heart transplant recipients addressing daily concerns, quality of life, and transition to adult care; and (3) semi-structured interviews with infectious disease providers exploring attitudes toward living donation from donors with human immunodeficiency virus (HIV) to recipients with HIV. These datasets had previously been analyzed by humans trained in traditional qualitative analysis. A Health Insurance Portability and Accountability Act (HIPAA)-compliant instance of OpenAI’s Generative Pre-trained Transformer 5.4.(GPT-5.4) generated thematic outputs for comparison. Results Across all three studies, LLM-Generated themes largely overlapped with those identified by humans. However, interpretive depth and contextualization of findings still required human expertise. Conclusions GPT-5.4 generated themes that showed conceptual overlap with previously completed human-led analyses across three transplantation datasets, supporting a hybrid model in which AI may assist with early analytic tasks while human researchers provide interpretation and rigor.