Development and validation of the Oxford Attitudes to Artificial Intelligence in Medicine Scale

Objective Public willingness to accept medical artificial intelligence (AI) tools affect the potential real-world impact of these evolving technologies. We therefore developed and validated the Oxford Attitudes to Artificial Intelligence in Medicine Scale (OAAIMS), an instrument designed to assess public and patient attitudes towards AI use in healthcare. Methods Candidate domains and items were generated through literature review, expert panels, and open-ended surveys, then administered to demographically representative samples (the United Kingdom n = 491; the United States n = 500). Exploratory factor analysis using polychoric correlations was conducted on the UK sample; confirmatory factor analysis was used to test the model on the US sample. Internal consistency (McDonald's ω), convergent validity (correlation with the attitudes toward AI scale), discriminant validity (correlation with the Oxford Vaccine Hesitancy Scale), and criterion validity (logistic regression predicting preference for human-only care) were assessed. Results Exploratory factor analysis supported a 20-item, 4-factor structure with domains of Benefit (9 items), Trust (5 items), Task-specific Comfort (3 items), and Concerns (3 items). Confirmatory factor analysis showed good fit (root mean squared error of approximation=0.055; standardized root mean squared residual=0.037; comparative fit index=0.963). Internal consistency was high for the total scale (ω = 0.95 UK, 0.96 US) and subscales (ω = 0.81-0.94). OAAIMS correlated with general AI attitudes (r = 0.70-0.71) and minimally with vaccine hesitancy (|r| ≤ 0.14). Each 1-point increase in OAAIMS score reduced the odds of preferring AI-free care by 8% to 9% (UK odds ratio of association [OR] 0.91, US OR 0.92; both P < .001). Discussion The OAAIMS is a brief, psychometrically robust scale capturing 4 distinct domains of public attitudes toward medical AI, developed using demographically representative samples. Conclusion The OAAIMS can serve as a baseline descriptor, outcome measure, or stratification variable in trials and implementation studies, permitting quantifiable tracking of patient-centered barriers and allowing for de-risking nascent technologies. OAAIMS is an informatics-ready instrument for baselining, stratifying, and longitudinally monitoring public acceptance of medical AI in trials and health-system implementations.