Acceptance of generative AI-assisted medical decision-making among Chinese physicians and patients and its ethical determinants: a cross-sectional survey.

OBJECTIVE: Generative artificial intelligence (GAI) is increasingly being integrated into medical decision-making, yet ethical challenges continue to hinder its widespread clinical implementation. Existing research has paid limited attention to the ethical determinants of acceptance and differences in perceptions among key stakeholder groups. This study examined acceptance of GAI-assisted medical decision-making among Chinese physicians, patients, and other healthcare stakeholders, identified its ethical determinants, and explored differences across stakeholder groups to inform ethical governance. METHODS: A cross-sectional survey was conducted using a multistage sampling strategy across healthcare institutions in Henan and Guangdong provinces, China, between December 2025 and January 2026. A total of 533 participants, including healthcare professionals, patients, and other stakeholders, were recruited. A self-developed four-dimensional scale assessing perceived functional value, perceived ethical risk, acceptance intention, and ethical governance expectations was developed through literature review, expert consultation, and pilot testing. Multivariable linear regression analysis was performed using SPSS version 26.0 to identify factors associated with acceptance intention. RESULTS: Participants reported a moderate level of acceptance of GAI-assisted medical decision-making (3.68 ± 0.70), while ethical governance expectations received the highest mean score (4.03 ± 0.76). Multivariable regression analysis showed that perceived functional value (β = 0.528, p < 0.001) and ethical governance expectations (β = 0.392, p < 0.001) were significant positive predictors of acceptance intention, whereas perceived ethical risk showed a statistically significant but relatively weak negative association. Healthcare professionals reported higher perceived ethical risk than patients (3.31 ± 0.75 vs. 3.23 ± 0.91, p < 0.05). Participants with postgraduate education reported the strongest ethical governance expectations (4.33 ± 0.59), whereas patients with chronic diseases perceived lower data privacy risk than those without chronic diseases (2.71 ± 1.08 vs. 3.57 ± 0.99, p < 0.001). CONCLUSION: Acceptance of GAI-assisted medical decision-making among Chinese stakeholders is primarily driven by perceived functional value and ethical governance expectations, with significant differences in ethical perceptions across stakeholder groups. These findings provide empirical evidence for developing targeted ethical governance frameworks to facilitate responsible GAI integration into clinical practice.