Presentation
Primary Care Physicians’ Perspectives on the Adoption of Large Language Models in Practice: A Qualitative Study Among American and Dutch Physicians
SessionPoster Session 1
DescriptionIntroduction
Primary care systems worldwide are facing a growing shortage of physicians (Lawson, 2023). In the United States alone, by 2036 there is an expected shortage of between 20,200 and 40,400 primary care physicians (PCP) (AAMC, 2024). PCPs, also known as family doctors or general practitioners, are physicians who provide comprehensive, continuous, and first-contact medical care for patients, addressing a wide range of health needs, coordinating specialist care when necessary, and focusing on prevention, diagnosis, and management of common illnesses (Russo et al., 2023). The shortage in PCPs contributes to rising workloads, increased burnout, and challenges in delivering accessible, high-quality care (Lawson, 2023; Russo et al., 2023). Large Language Models (LLMs) have emerged as a promising technology with potential applications in primary care, ranging from clinical documentation and decision support to patient communication and workflow optimization (Andrew, 2024). Despite their potential, the successful integration of LLMs into healthcare depends not only on their technical capabilities but also on their acceptance and adoption by PCPs (Weik et al., 2024). Understanding how primary care physicians perceive LLM adoption, its barriers, influencing factors and opportunities, is crucial to fostering the effective and sustainable adoption of LLMs (Weik et al., 2024). This study explores the perspectives of primary care physicians in the United States and the Netherlands to identify the factors shaping their attitudes toward adopting LLMs in clinical practice and on a wider scale in primary care. The insights from this study could be used for tailoring LLMs to the specific needs of PCPs and promote effective adoption strategies for LLMs on a wider scale.
Methods
We conducted a qualitative study using semi-structured Zoom interviews with primary care physicians in the United States and the Netherlands who had experience using large language model tools in practice. Participants were recruited through purposive and snowball sampling via email, LinkedIn, and professional networks. Eligibility required current practice in primary care, fluency in English, and prior experience with LLMs. An incentive of $30 (€30 for Dutch participants) was provided for each participant.
Interviews were conducted between February and June 2025, audio-recorded, and transcribed automatically, and manually checked for inconsistencies. An interview guide was used to ensure consistency across interviews while allowing flexibility to follow up on relevant themes that emerged during the conversation. Interviews lasted 32 minutes on average. Data were analyzed thematically following Braun and Clarke’s framework (Braun & Clarke, 2022). Coding was conducted manually in Excel, and themes were iteratively developed through affinity diagramming, reflexive discussions, and peer debriefing. Data collection continued until thematic saturation was reached (N=15). Ethical approval was obtained from the Stevens Institute of Technology IRB (ID 2024-070 (N)), and informed consent was secured from all participants prior to data collection.
Results
In total, our study included 15 primary care physicians, of which 8 Dutch (53.3%) and 7 American (46.7%). Of the physicians, there were 6 females (40%) and 9 males (60%). There were 3 physicians with 0 to 5 years of experience (20%), 3 physicians with 6 to 10 years of experience (20%), 2 physicians with 11 to 15 years of experience (13.3%), 4 physicians with 16 to 20 years of experience (26.7%), 2 physicians with 20+ years of experience (13.3%), and one physician for which the number of years of experience is unknown (6.7%).
Our semi-structured interviews study with primary care physicians yielded a set of rich perspectives on the opportunities and barriers of adopting large language models (LLMs) in clinical practice. Our analysis produced four overarching themes.
Theme 1: Factors influencing adoption. Physicians highlighted multiple considerations shaping whether and how LLMs might be adopted in primary care. These included patient expectations for personalized care, broader system-level drivers (e.g., geopolitics and governmental influences), and the need for targeted practitioner education. Barriers such as generational differences, difficulty keeping pace with rapid technological change, concerns about environmental impact, and financial costs were also noted.
Theme 2: User enthusiasm. Many participants expressed optimism about LLMs’ potential to improve aspects of primary care and anticipated rapid adoption once reliable tools are available. Enthusiasm was driven by perceptions of LLMs as augmenting efficiency, supporting clinicians, and offering meaningful value to patients.
Theme 3: User skepticism. In contrast, PCPs also voiced significant reservations. Physicians described selective trust in specific LLM capabilities and stressed the necessity of physician oversight in any clinical application. Concerns included the possibility of LLMs replacing or devaluing medical knowledge, the risk of unnecessary use without clear benefit, and limitations related to the immaturity of current systems.
Theme 4: Potential future applications. Participants envisioned several promising use cases where LLMs could reduce workload and enhance care. These included chart summarization, support for differential diagnosis and decision-making, answering routine patient questions, automating calculations, generating animated educational handouts, and tailoring information to individual patients.
Together, these findings illustrate both excitement and caution regarding LLM adoption in primary care. While physicians see clear opportunities for efficiency and patient support, they also emphasize the importance of safeguards, education, and careful evaluation to ensure LLMs complement rather than undermine clinical expertise.
Conclusion
Our research highlights what barriers, influencing factors and opportunities are perceived by primary care physicians regarding the adoption of large language models in clinical practice. The barriers range from lower individual level reservations, like recognizing the risk of unnecessary use without adding much value or fearing it might replace primary care physicians’ knowledge, to high level system-based constraints, like geopolitics, environmental impact and rising healthcare costs. The results also revealed considerable enthusiasm for LLM adoption, with participants anticipating substantial improvements in primary care delivery and projecting a trajectory of rapid, widespread integration within the field. At the same time, many were open to future applications such as chart summarization, decision-support, and automated patient communication. Overall, these findings suggest that successful adoption of LLMs in primary care will require balancing innovation with caution. Targeted practitioner education, thoughtful system-level policies, and designs that prioritize physician oversight will be essential. By attending to both the barriers and facilitators identified here, developers, policymakers, and research professionals can better align emerging LLM technologies with the realities of clinical practice and physician needs.
References
AAMC. (2024). The Complexities of Physician Supply and Demand: Projections From 2021 to 2036.
Andrew, A. (2024). Potential applications and implications of large language models in primary care. Fam Med Community Health, 12(Suppl 1). https://doi.org/10.1136/fmch-2023-002602
Braun, V., & Clarke, V. (2022). Conceptual and design thinking for thematic analysis. Qualitative Psychology, 9(1), 3-26. https://doi.org/10.1037/qup0000196
Lawson, E. (2023). The Global Primary Care Crisis. Br J Gen Pract, 73(726), 3. https://doi.org/10.3399/bjgp23X731469
Russo, G., Perelman, J., Zapata, T., & Santric-Milicevic, M. (2023). The layered crisis of the primary care medical workforce in the European region: what evidence do we need to identify causes and solutions? Hum Resour Health, 21(1), 55. https://doi.org/10.1186/s12960-023-00842-4
Weik, L., Fehring, L., Mortsiefer, A., & Meister, S. (2024). Understanding inherent influencing factors to digital health adoption in general practices through a mixed-methods analysis. NPJ Digit Med, 7(1), 47. https://doi.org/10.1038/s41746-024-01049-0
Primary care systems worldwide are facing a growing shortage of physicians (Lawson, 2023). In the United States alone, by 2036 there is an expected shortage of between 20,200 and 40,400 primary care physicians (PCP) (AAMC, 2024). PCPs, also known as family doctors or general practitioners, are physicians who provide comprehensive, continuous, and first-contact medical care for patients, addressing a wide range of health needs, coordinating specialist care when necessary, and focusing on prevention, diagnosis, and management of common illnesses (Russo et al., 2023). The shortage in PCPs contributes to rising workloads, increased burnout, and challenges in delivering accessible, high-quality care (Lawson, 2023; Russo et al., 2023). Large Language Models (LLMs) have emerged as a promising technology with potential applications in primary care, ranging from clinical documentation and decision support to patient communication and workflow optimization (Andrew, 2024). Despite their potential, the successful integration of LLMs into healthcare depends not only on their technical capabilities but also on their acceptance and adoption by PCPs (Weik et al., 2024). Understanding how primary care physicians perceive LLM adoption, its barriers, influencing factors and opportunities, is crucial to fostering the effective and sustainable adoption of LLMs (Weik et al., 2024). This study explores the perspectives of primary care physicians in the United States and the Netherlands to identify the factors shaping their attitudes toward adopting LLMs in clinical practice and on a wider scale in primary care. The insights from this study could be used for tailoring LLMs to the specific needs of PCPs and promote effective adoption strategies for LLMs on a wider scale.
Methods
We conducted a qualitative study using semi-structured Zoom interviews with primary care physicians in the United States and the Netherlands who had experience using large language model tools in practice. Participants were recruited through purposive and snowball sampling via email, LinkedIn, and professional networks. Eligibility required current practice in primary care, fluency in English, and prior experience with LLMs. An incentive of $30 (€30 for Dutch participants) was provided for each participant.
Interviews were conducted between February and June 2025, audio-recorded, and transcribed automatically, and manually checked for inconsistencies. An interview guide was used to ensure consistency across interviews while allowing flexibility to follow up on relevant themes that emerged during the conversation. Interviews lasted 32 minutes on average. Data were analyzed thematically following Braun and Clarke’s framework (Braun & Clarke, 2022). Coding was conducted manually in Excel, and themes were iteratively developed through affinity diagramming, reflexive discussions, and peer debriefing. Data collection continued until thematic saturation was reached (N=15). Ethical approval was obtained from the Stevens Institute of Technology IRB (ID 2024-070 (N)), and informed consent was secured from all participants prior to data collection.
Results
In total, our study included 15 primary care physicians, of which 8 Dutch (53.3%) and 7 American (46.7%). Of the physicians, there were 6 females (40%) and 9 males (60%). There were 3 physicians with 0 to 5 years of experience (20%), 3 physicians with 6 to 10 years of experience (20%), 2 physicians with 11 to 15 years of experience (13.3%), 4 physicians with 16 to 20 years of experience (26.7%), 2 physicians with 20+ years of experience (13.3%), and one physician for which the number of years of experience is unknown (6.7%).
Our semi-structured interviews study with primary care physicians yielded a set of rich perspectives on the opportunities and barriers of adopting large language models (LLMs) in clinical practice. Our analysis produced four overarching themes.
Theme 1: Factors influencing adoption. Physicians highlighted multiple considerations shaping whether and how LLMs might be adopted in primary care. These included patient expectations for personalized care, broader system-level drivers (e.g., geopolitics and governmental influences), and the need for targeted practitioner education. Barriers such as generational differences, difficulty keeping pace with rapid technological change, concerns about environmental impact, and financial costs were also noted.
Theme 2: User enthusiasm. Many participants expressed optimism about LLMs’ potential to improve aspects of primary care and anticipated rapid adoption once reliable tools are available. Enthusiasm was driven by perceptions of LLMs as augmenting efficiency, supporting clinicians, and offering meaningful value to patients.
Theme 3: User skepticism. In contrast, PCPs also voiced significant reservations. Physicians described selective trust in specific LLM capabilities and stressed the necessity of physician oversight in any clinical application. Concerns included the possibility of LLMs replacing or devaluing medical knowledge, the risk of unnecessary use without clear benefit, and limitations related to the immaturity of current systems.
Theme 4: Potential future applications. Participants envisioned several promising use cases where LLMs could reduce workload and enhance care. These included chart summarization, support for differential diagnosis and decision-making, answering routine patient questions, automating calculations, generating animated educational handouts, and tailoring information to individual patients.
Together, these findings illustrate both excitement and caution regarding LLM adoption in primary care. While physicians see clear opportunities for efficiency and patient support, they also emphasize the importance of safeguards, education, and careful evaluation to ensure LLMs complement rather than undermine clinical expertise.
Conclusion
Our research highlights what barriers, influencing factors and opportunities are perceived by primary care physicians regarding the adoption of large language models in clinical practice. The barriers range from lower individual level reservations, like recognizing the risk of unnecessary use without adding much value or fearing it might replace primary care physicians’ knowledge, to high level system-based constraints, like geopolitics, environmental impact and rising healthcare costs. The results also revealed considerable enthusiasm for LLM adoption, with participants anticipating substantial improvements in primary care delivery and projecting a trajectory of rapid, widespread integration within the field. At the same time, many were open to future applications such as chart summarization, decision-support, and automated patient communication. Overall, these findings suggest that successful adoption of LLMs in primary care will require balancing innovation with caution. Targeted practitioner education, thoughtful system-level policies, and designs that prioritize physician oversight will be essential. By attending to both the barriers and facilitators identified here, developers, policymakers, and research professionals can better align emerging LLM technologies with the realities of clinical practice and physician needs.
References
AAMC. (2024). The Complexities of Physician Supply and Demand: Projections From 2021 to 2036.
Andrew, A. (2024). Potential applications and implications of large language models in primary care. Fam Med Community Health, 12(Suppl 1). https://doi.org/10.1136/fmch-2023-002602
Braun, V., & Clarke, V. (2022). Conceptual and design thinking for thematic analysis. Qualitative Psychology, 9(1), 3-26. https://doi.org/10.1037/qup0000196
Lawson, E. (2023). The Global Primary Care Crisis. Br J Gen Pract, 73(726), 3. https://doi.org/10.3399/bjgp23X731469
Russo, G., Perelman, J., Zapata, T., & Santric-Milicevic, M. (2023). The layered crisis of the primary care medical workforce in the European region: what evidence do we need to identify causes and solutions? Hum Resour Health, 21(1), 55. https://doi.org/10.1186/s12960-023-00842-4
Weik, L., Fehring, L., Mortsiefer, A., & Meister, S. (2024). Understanding inherent influencing factors to digital health adoption in general practices through a mixed-methods analysis. NPJ Digit Med, 7(1), 47. https://doi.org/10.1038/s41746-024-01049-0
Event Type
Poster Presentation
TimeMonday, March 234:45pm - 6:15pm EDT
LocationRhinelander Gallery
Digital Health