LARGE LANGUAGE MODEL-ENABLED MEDICAL CONSULTATION IN CHINA: A DISCRETE CHOICE EXPERIMENT
Author(s)
XINYANG MA, BSc, MPH, PhD1, Taoran Liu, PhD2, Wu cai lin, PhD3, jing Yu, PhD2, YANGYANG GAO, PhD4, Ginenus Fekadu, PhD5, Wai-kit Ming, MBA, MPH, PhD, MD6.
1RA, City University of Hong Kong, Hong Kong, China, 2City University of Hong Kong, Hong Kong, China, 3City University of Hong Kong, Shenzhen, China, 4City University of Hong Kong, HONG KONG, China, 5Wollega University, Nekemte, Ethiopia, 6City University of Hong Kong, City University of Hong Kong, China.
1RA, City University of Hong Kong, Hong Kong, China, 2City University of Hong Kong, Hong Kong, China, 3City University of Hong Kong, Shenzhen, China, 4City University of Hong Kong, HONG KONG, China, 5Wollega University, Nekemte, Ethiopia, 6City University of Hong Kong, City University of Hong Kong, China.
OBJECTIVES: To quantify how the Chinese public and health-care professionals (HCPs) trade off clinician oversight, accuracy, waiting time, cost, and privacy when evaluating LLM-enabled consultation services across clinical contexts.
METHODS: We conducted parallel anonymous cross-sectional stated-preference surveys in mainland China (Oct 12-Nov 17, 2025) among public respondents and HCPs. Each survey included a DCE in four contexts: chronic disease management, acute illness, cancer-related consultation, and sexually transmitted infection consultation. In each context, participants completed eight random choice tasks and one fixed task, choosing between two hypothetical consultation services and opt-out. Context-specific conditional logit models were estimated by respondent group. Prespecified sensitivity analyses used all-completer samples and response-quality diagnostics; heterogeneity, willingness-to-pay, mixed-logit, and choice-simulation analyses were exploratory.
RESULTS: The analytic sample included 1034 public respondents and 826 HCPs. Both groups disfavoured standalone LLM chatbots across contexts. Public respondents preferred hybrid LLM pre-advice with clinician review over standalone chatbots; however, in-person physician consultation remained the highest-utility provider model when other attributes were held constant. HCPs placed greater value on in-person physician consultation, especially for acute illness and cancer-related consultation. In cancer-related consultation, the in-person physician coefficient was 0.358 for HCPs versus 0.172 for public respondents. Waiting time and out-of-pocket cost influenced public choices. Opt-out was higher among HCPs, peaking in acute illness (8.7%) and cancer-related consultation (7.8%), while public opt-out varied little (4.3-4.8%).
CONCLUSIONS: Preferences for LLM-enabled consultation varied by user group and clinical context. Clinician-reviewed hybrid services may be more acceptable than standalone chatbots, but not necessarily preferred to in-person physician consultation. Governance, trust, privacy safeguards, and affordability are central to uptake.
METHODS: We conducted parallel anonymous cross-sectional stated-preference surveys in mainland China (Oct 12-Nov 17, 2025) among public respondents and HCPs. Each survey included a DCE in four contexts: chronic disease management, acute illness, cancer-related consultation, and sexually transmitted infection consultation. In each context, participants completed eight random choice tasks and one fixed task, choosing between two hypothetical consultation services and opt-out. Context-specific conditional logit models were estimated by respondent group. Prespecified sensitivity analyses used all-completer samples and response-quality diagnostics; heterogeneity, willingness-to-pay, mixed-logit, and choice-simulation analyses were exploratory.
RESULTS: The analytic sample included 1034 public respondents and 826 HCPs. Both groups disfavoured standalone LLM chatbots across contexts. Public respondents preferred hybrid LLM pre-advice with clinician review over standalone chatbots; however, in-person physician consultation remained the highest-utility provider model when other attributes were held constant. HCPs placed greater value on in-person physician consultation, especially for acute illness and cancer-related consultation. In cancer-related consultation, the in-person physician coefficient was 0.358 for HCPs versus 0.172 for public respondents. Waiting time and out-of-pocket cost influenced public choices. Opt-out was higher among HCPs, peaking in acute illness (8.7%) and cancer-related consultation (7.8%), while public opt-out varied little (4.3-4.8%).
CONCLUSIONS: Preferences for LLM-enabled consultation varied by user group and clinical context. Clinician-reviewed hybrid services may be more acceptable than standalone chatbots, but not necessarily preferred to in-person physician consultation. Governance, trust, privacy safeguards, and affordability are central to uptake.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HSD128
Topic
Epidemiology & Public Health, Health Service Delivery & Process of Care, Patient-Centered Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas