ACCEPTANCE OF ARTIFICIAL INTELLIGENCE IN MEDICAL CONSULTATIONS: EVIDENCE FROM A DISCRETE CHOICE EXPERIMENT
Author(s)
Lai Yim, PhD1, MingChun Chung, MBBS2, Yan HE, PhD3, Ginenus Fekadu, PhD4, Wai-kit Ming, MBA, MPH, PhD, MD5.
1City University of Hong Kong, Kowloon, Hong Kong, 2The University of Hong Kong, Hong Kong, Hong Kong, 3City University of Hong Kong, Kowlong, China, 4City University of Hong Kong, Hong Kong, Hong Kong, 5City University of Hong Kong, City University of Hong Kong, China.
1City University of Hong Kong, Kowloon, Hong Kong, 2The University of Hong Kong, Hong Kong, Hong Kong, 3City University of Hong Kong, Kowlong, China, 4City University of Hong Kong, Hong Kong, Hong Kong, 5City University of Hong Kong, City University of Hong Kong, China.
OBJECTIVES: As artificial intelligence (AI) becomes increasingly integrated into healthcare delivery, patient acceptance remains a key barrier to implementation. This study examined preferences for AI, human, and hybrid consultation mode across clinical scenario with varying perceived risk, focusing on diagnostic performance, trust, and AI literacy.
METHODS: A discrete choice experiment was conducted among 713 adults in Hong Kong. Respondents completed hypothetical choice tasks comparing consultation alternatives defined by clinician modality, waiting time, consultation duration, diagnostic accuracy, follow-up services, and out-of-pocket cost. Preferences were estimated using mixed logit models, with willingness-to-pay derived from coefficients. Latent class analysis identified heterogeneity, and multinomial logit models examined associations with AI literacy.
RESULTS: Consultation modality and diagnostic accuracy were key determinants of preferences. Respondents showed significant disutility toward AI-only consultations compared with human clinicians across specialties (β range: −0.56 to −0.82; all p<0.001), with stronger aversion in higher-risk contexts. Hybrid AI-human models were more acceptable than fully automated AI-led care, though less preferred than human-only consultations. Diagnostic accuracy was the most influential attribute. Willingness-to-pay estimates indicated substantial negative valuation for AI-only consultations relative to human clinicians, ranging from −US$66.9 to −US$157.9. Three preference segments were identified: pragmatic, performance-oriented, and AI-averse groups. Higher AI literacy was not associated with greater acceptance of AI-based care; instead, AI literacy was highest in the AI-averse segment.
CONCLUSIONS: Patient preferences for AI-enabled healthcare were shaped not only by diagnostic performance, but also by perceived risk, trust, and the value placed on human clinician involvement. Hybrid AI-human consultation models were more acceptable in higher-risk clinical scenario. The finding that higher AI literacy did not correspond to greater AI acceptance challenges the assumption that education alone will promote adoption. These results suggest that patient-centred integration of AI into healthcare will require attention to preference heterogeneity, perceived risk, and the preservation of trust through continued clinician involvement.
METHODS: A discrete choice experiment was conducted among 713 adults in Hong Kong. Respondents completed hypothetical choice tasks comparing consultation alternatives defined by clinician modality, waiting time, consultation duration, diagnostic accuracy, follow-up services, and out-of-pocket cost. Preferences were estimated using mixed logit models, with willingness-to-pay derived from coefficients. Latent class analysis identified heterogeneity, and multinomial logit models examined associations with AI literacy.
RESULTS: Consultation modality and diagnostic accuracy were key determinants of preferences. Respondents showed significant disutility toward AI-only consultations compared with human clinicians across specialties (β range: −0.56 to −0.82; all p<0.001), with stronger aversion in higher-risk contexts. Hybrid AI-human models were more acceptable than fully automated AI-led care, though less preferred than human-only consultations. Diagnostic accuracy was the most influential attribute. Willingness-to-pay estimates indicated substantial negative valuation for AI-only consultations relative to human clinicians, ranging from −US$66.9 to −US$157.9. Three preference segments were identified: pragmatic, performance-oriented, and AI-averse groups. Higher AI literacy was not associated with greater acceptance of AI-based care; instead, AI literacy was highest in the AI-averse segment.
CONCLUSIONS: Patient preferences for AI-enabled healthcare were shaped not only by diagnostic performance, but also by perceived risk, trust, and the value placed on human clinician involvement. Hybrid AI-human consultation models were more acceptable in higher-risk clinical scenario. The finding that higher AI literacy did not correspond to greater AI acceptance challenges the assumption that education alone will promote adoption. These results suggest that patient-centred integration of AI into healthcare will require attention to preference heterogeneity, perceived risk, and the preservation of trust through continued clinician involvement.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
PCR214
Topic
Health Service Delivery & Process of Care, Medical Technologies, Patient-Centered Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas