DOI: 10.56150/tjhsl.1881554 ISSN: 2687-2730

The Potential of Large Language Models in Family Medicine Practice: Measuring the Artificial Intelligence Anxiety Among Family Medicine Residents

Giray Kolcu, Nebahat Bilge Balım, Funda Yıldırım Baş, Gökçe İşcan
Show PDF Cite
Large language models (LLMs) are increasingly being proposed as decision-support and administrative tools in primary healthcare. However, their integration into family medicine, a discipline grounded in continuity, trust, and holistic decision-making, may be shaped by clinicians’ emotional, ethical, and professional concerns. This study aimed to quantify artificial intelligence (AI) anxiety among family medicine residents in Turkey and compare anxiety levels across two specialty training pathways. A cross-sectional, descriptive quantitative study was conducted among all residents enrolled in a university family medicine department (N=82), of whom 76 participated voluntarily (response rate 92.6%). No exclusion criteria were applied, and ethical approval was obtained (12.05.2025, no:10). Data were collected online (02–20.06.2025) using the 21-item Artificial Intelligence Anxiety Scale (Learning, Job Replacement, Sociotechnical Blindness, AI Structuring; 5-point Likert; no cutoff), and analyzed in Jamovi/SPSS with normality testing (Shapiro–Wilk), appropriate confirmatory tests, and p<0.05. The participants’ mean age was 32.14±6.74 years (range, 25–53 years), 57% were male, and 55% were in the Family Medicine Specialty program versus 45% in the Contracted program; only 11% reported prior AI-related training. Item-level concerns were greatest for fears of the malicious use of AI (5.42±1.61) and AI increasing laziness (4.58±1.69), alongside elevated concerns about autonomy, loss of control, and humanoid threats. In contrast, anxiety about learning (e.g., reading manuals or taking courses) was the lowest (both 2.18). The subscale means were highest for Learning (24.82±12.72), followed by Job Replacement (20.32±7.65), Sociotechnical Blindness (17.34±5.87), and AI Structuring (10.82±5.15), with a total score of 73.29±24.91. Reliability was high (overall Cronbach’s α=0.94), subscales were internally consistent (α=0.87–0.93), and subdimensions were significantly intercorrelated (p<0.001), strongest between Job Replacement and Sociotechnical Blindness (r=0.84); no significant anxiety differences emerged between training programs. These findings indicate that residents’ AI anxiety is driven more by ethical risk, autonomy, and professional displacement than by fear of acquiring knowledge, underscoring the need to embed structured AI literacy and ethics-focused education within residency and continuing medical education to support the safe and acceptable adoption of LLM-enabled tools in primary care, while acknowledging the limitations of voluntary participation and a single, limited population.

More from our Archive