From One-time to Revocable Consent: The Process of Granular Consent and Data Sharing with Artificial Intelligence (AI) in Healthcare: Cross-Section Survey Study (Preprint)

Abstract

<sec> <title>BACKGROUND</title> Artificial intelligence (AI) shifts the focus of individual data owners’ data control, from one time to revocable consent with the growing needs of data sharing and training for AI. Without explicit incentives and undefined secondary uses of personal health information, tension can arise between data owners’ privacy concerns and their actual data-sharing behavior. Particularly in healthcare, while granting selective sharing and allowing data owners to withdraw personal health data, termed granular consent, is expected to mitigate such tension and promote data owners’ control of health data sharing, limited empirical evidence exists whether and how such a granular consent option motivates health data owners democratize their data to AI. </sec> <sec> <title>OBJECTIVE</title> The objective of this study is to examine the mechanisms and boundary conditions of health data owners’ data sharing decision when they are flexible to change their consent in the data sharing process. We hypothesize that health data sharing at individual level is a simultaneous act of benefiting oneself and others, and that the altruistic or reciprocal motivations of health data owners can disproportionately influence the relationship between micro-consent, perceived privacy control, and data sharing decision with AI. </sec> <sec> <title>METHODS</title> A total of 389 responses were collected via a cross-section online survey. A conditional process analysis was employed to qualitatively explore the boundary conditions of health data owners’ granular consent mechanisms with AI using SAS 9.4 PROCESS macro program. </sec> <sec> <title>RESULTS</title> Results from the conditional process models show that (1) health consumers’ data sharing with AI follows a staged process such that there is a positive mediation between micro-consent, privacy control (β=0.31, p&lt;0.001), and data sharing with AI (β=0.32, p&lt;0.001); and (2) while there is no moderated mediation effect of altruism (β=0.01, p=0.67), reciprocity is likely to enhance the path between privacy control and data sharing intention with AI (β=0.07, p=0.047). We further retested our models with a granular verbatim of micro-consent (e.g., options to change or withdraw), finding that 1) all three forms of granular consent (broad, change/control, cancel/withdraw) is mediated by privacy control; 2) the verbatim of cancel/withdraw has the highest proportion of mediation (20%); and 3)the moderated mediation effect of altruism is present when health consumers can “cancel” or “withdraw” their granular consent in the data sharing process (β=0.12, p&lt;0.001). </sec> <sec> <title>CONCLUSIONS</title> The findings of our study can shed light on how to motivate health data owners with a granular consent option and intrinsic incentives (reciprocity) visible in the health AI application development and design. Health policy makers and health AI designers need to consider delicately devising verbatim of the granular consent option and embedding reciprocal features/capabilities so that individual health data owners can perceive intrinsic benefits coming from other participants’ information sharing in exchange for sharing their own personal data. </sec>

Ähnliche Arbeiten