Applications of large language model in general practice: a qualitative interview study with professionals and patients
Year: 2026
Type: preprint
Abstract: Background: Workload is increasing in general practice. Large language models (LLMs) are a promising tool that could relieve part of this workload. However, existing studies offer little insight into how professionals and patients want specific LLM applications to be implemented in practice. Objectives: This study aimed to prioritize a broad range of LLM applications in general practice and identify the benefits, risks, and requirements of each application. Methods: We conducted semi-structured one-on-one interviews with 12 general practitioners (GPs), 7 GP assistants, 6 practice nurses, and 6 patient representatives in the Netherlands between October and December 2025. To guide the interviews, we developed a list of 21 LLM applications derived from literature, existing tools, and GP input. Application were divided into 5 categories: administrative tasks, pre-consultation and triage, clinical decision support, management of (chronic) diseases, and patient interaction and communication. Participants discussed and prioritized each application. Transcripts were analyzed with a combined deductive and inductive approach. Results: Priorities for LLM applications differed by group: GPs favored clinical decision support, assistants favored triage, and practice nurses and patient representatives favored chronic disease management. The most prioritized applications were summarizing consultations and triage and urgency classification (high priority 18/31; 58%), followed by clarification of medical information around the consultation (14/31; 45%). The qualitative analysis revealed three recurring considerations when applying LLMs for different tasks. First, participants questioned whether an LLM is needed, as simpler or existing tools may suffice. Second, they questioned whether LLMs relieve workload or create extra work. Third, a possible benefit on workload has to explicitly be weighed against risks such as mistakes, reduced personal care and accessibility, and negative effects on professionals. Conclusions: Professionals and patients see potential for LLM support across many areas of general practice. Realizing this potential requires three steps: critically assessing whether an LLM is the right tool, establishing that it relieves workload, and designing applications so that their benefits outweigh the risks. Our application-level prioritization and analysis provide a starting point for assessment and design.
Source: medRxiv
Institutions: Erasmus MC, Delft University of Technology
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funder: Health~Holland
OpenAlex ID: W7225039235