Needs assessment in primary care physicians on the use of artificial intelligence in geriatrics and palliative care education and clinical practice

Abstract

Abstract Background: Artificial intelligence (AI), including large language model (LLM)–based tools, is increasingly available to clinicians, yet real-world adoption in primary care— particularly for geriatrics, and palliative care remains limited and poorly characterized. We conducted a multi-site needs assessment to understand primary care physicians’ attitudes, desired use cases, and concerns regarding AI in clinical practice, with attention to applications that could support geriatrics and palliative care education. Methods: We performed a qualitative descriptive study using individual semi-structured interviews with practicing primary care physicians in the United States (Stanford; n=15) and Germany (Erlangen region; n=9). Interviews were conducted via Zoom, audio-recorded, transcribed verbatim (and translated/back-translated for German interviews). We used a hybrid inductive-deductive coding strategy. Two coders independently coded all transcripts in MAXQDA and reconciled discrepancies through consensus. We then conducted codebook thematic analysis to synthesize the data by gropuing related codes into broader themes. Results: Physicians in both countries expressed cautious optimism about AI, primarily valuing it for efficiency, administrative support, and rapid retrieval of targeted information in complex clinical scenarios. Commonly desired functions included documentation support (notes, letters, visit summarization), drafting patient messages, chart summarization for medically complex older adults, and assistance with medication management and drug–drug/supplement interactions. Adoption was strongly conditioned on trustworthiness: participants emphasized accuracy, transparency, and traceable citations, often comparing desired AI behavior to trusted resources (e.g., UpToDate in the US, AMBOSS in Germany). Major concerns included hallucinations, bias, and privacy/security, with data protection and regulatory compliance particularly salient in Germany. Clinicians prioritized low-friction workflow integration (ideally embedded in the electronic health record). For goals of care discussions, participants viewed AI as potentially useful for preparation—such as presenting prognostic information in patient- friendly formats, suggesting culturally sensitive communication approaches, and providing relevant legal context—while emphasizing that nuanced judgment, empathy, and accountability must remain with clinicians. Conclusions: Primary care physicians in the US and Germany perceive substantial potential for AI to reduce workload and support clinical reasoning, including in geriatrics and palliative care–related clinical care and education. However, successful implementation requires trustworthy, privacy-preserving tools with transparent sourcing and seamless workflow integration, positioned as clinician-support rather than autonomous decision-making systems.