Public preferences for health data sharing governance: a discrete choice experiment in China

Public acceptance is central to legitimate and effective health data sharing governance, yet evidence on how people evaluate multidimensional arrangements across data-sensitivity contexts remains limited. Existing studies often examine institutional attributes in isolation. We investigated public preferences for health data sharing governance in China and examined whether data sensitivity, trust in government organizations, and individual heterogeneity shaped these preferences.

We conducted a discrete choice experiment with adult Chinese citizens using a D-efficient design. Fifteen choice sets were divided into three blocks, and each respondent was randomly assigned to one block. The experiment varied five attributes: data custodian, purpose of use, rights protection, incentive type, and data type. Using 7,405 valid choice observations from 1,481 respondents, we estimated average marginal component effects with ordinary least squares and respondent-clustered standard errors. Interaction models examined moderation by data type and by trust in government organizations, and a supplementary latent class analysis examined preference heterogeneity.

Within this design, data type had the largest average marginal component effect and moderated the effects of the other governance arrangements. For sensitive health data, respondents preferred strong rights protection and public hospital custodianship. For general health data, government-affiliated neutral third-party custodianship and non-cash incentives increased the probability of choosing a sharing profile, whereas the same arrangements had negative effects for sensitive data. The moderating role of trust in government organizations was conditional. After correction for multiple comparisons, higher trust strengthened the preference for government-affiliated custodians and weakened the effect of health points, and no other attribute effect varied with trust. A supplementary penalized latent class analysis suggested three exploratory preference profiles.

Preferences in this online sample were context-dependent and are unlikely to be accommodated by uniform governance frameworks. Risk-stratified governance should combine stronger rights protection and accountable custodianship for sensitive data with streamlined procedures and calibrated incentives for general data. Trust in government organizations conditioned selected governance preferences, and the social usefulness of health information depends on the conditions under which it is collected, accessed, and shared.