Factors for Patient Trust and Acceptance of Medical Artificial Intelligence

Abstract

Importance: Artificial intelligence (AI) is increasingly used in clinical care, but widespread adoption requires patient trust. Trust may be enhanced through systemic governance mechanisms or frontline clinicians providing a human in the loop for AI oversight. However, it is unclear how different approaches specifically influence patient trust in the use of medical AI. Objective: To determine the extent to which patient trust in and choice of medical scenarios involving AI are associated with governance mechanisms, clinician presence, performance, and data quality. Design, Setting, and Participants: This preregistered conjoint survey study was conducted online among a diverse national sample of English-speaking US adults with access to the internet between December 11, 2024, and January 1, 2025. Respondents were presented with hypothetical AI-assisted diagnosis scenarios and paired visits featuring 6 purely randomized attributes: the presence of a clinician, AI performance (relative to general practitioners and specialists), governance (US Food and Drug Administration approval, Mayo Clinic certification, local hospital certification), and AI data quality. Respondents chose their preferred visit, provided up to a single-sentence open-ended response explaining their choice, and then rated their trust in the diagnosis they would receive in each of the 2 visit choices presented to them. Respondents repeated the exercise 6 times, evaluating 12 hypothetical visits in total, yielding 36 000 observations (12 per respondent). Main Outcomes and Measures: The primary outcomes were patient choice of a hypothetical medical encounter and patient trust in that encounter, measured on a 1 (would not trust at all) to 5 (would trust a great deal) response scale. Average marginal component effects (AMCEs) were estimated using linear regression. Qualitative responses were coded to elucidate reasoning. Results: A total of 3000 participants completed the survey (1644 [54.8%] women; mean [SD] age, 48 [16] years), including 382 Black respondents (12.7%), 504 Hispanic respondents (16.8%), and 1855 White respondents (61.9%), with most respondents having some college or more (1989 respondents [66.3%]), and 1270 respondents (42.4%) having income between $50 000 and $99 000. The factor associated with the largest change in likelihood of patient choice was AI performance; performance at or above the specialist level was associated with increasing the probability of selecting a visit by 24.8% (95% CI, 23.4%-26.2%; P < .00025) and 32.5% (95% CI, 31.0%-33.9%; P < .00025), respectively. The presence of a clinician was associated with increasing the probability of selecting a visit by 18.4% (95% CI, 17.3%-19.5%; P < .00025). Respondents who received information on representative AI training data also were more likely to prefer that visit scenario. Respondents preferred all forms of AI governance compared with none. Qualitative responses emphasized AI performance and clinician presence as primary factors in choice of visit. Conclusions and Relevance: In this survey study of patient trust in and choice of medical AI, AI performance, clinician presence, disclosure of representative data, and systemic governance were associated with increased respondent trust in and preference for clinical encounters. These findings suggest that ensuring resource-appropriate combinations of these tools is an important step in helping AI achieve its transformative potential for the health system.

Ähnliche Arbeiten