Predictive Technologies in Healthcare: Public Perspectives and Health System Governance in the Context of Structural Inequity

The health data ecosystem is increasingly focused on the design and implementation predictions in the form of AI-enabled clinical decision support, risk calculation, and resource allocation. This system of prediction in healthcare is developing rapidly in the context of limited regulation and structural inequity. The stakes for patients and health systems are high as predictive models are deployed more widely, affecting multiple aspects of care from appointment wait times to treatment for sepsis. Risks of racism, bias, and other inequities in the data used to build these models are increasingly recognized. However, public perspectives and values related to predictive modeling in healthcare have not yet been studied at the national level. It is also unclear how health systems are currently governing prediction, especially in the context of structural inequity. In this dissertation, I analyze an original national survey of the public to understand their perspectives on prediction in healthcare. I also analyze qualitative in-depth interviews with health system leadership to examine their governance strategies for predictive models. This approach treats both health system leadership and members of the public as key stakeholders engaged in and affected by the sociotechnical system of prediction. In the first study, I analyze public comfort with data use for prediction using survey responses from a national sample of US adults. I identify that the public differentiates between the use of various data types for prediction and observe higher comfort among 1) white respondents and 2) those who have not experienced discrimination while seeking healthcare. In the second study of a national sample of US adults, I identify misalignment between public perspectives and current regulatory frameworks. Analyzing original survey measures of comfort with six specific predictive models in healthcare, I find that the public is less comfortable with administrative applications of prediction (e.g., predicting missed appointments) than with clinical applications (e.g., predicting stroke). The third study presents findings from qualitative interviews with leadership from academic medical centers across the country about how they manage and design governance processes. This project focuses on understanding how predictive models are currently governed, how regulation shapes that governance, and whether equity is a consideration in health system governance processes. I identify variation among academic medical centers in their governance structures and the degree to which they consider equity when evaluating predictive models. I also find that current regulation is ambiguous for these decision-makers and could be strengthened to provide important guidance for health system policy. As patients are increasingly exposed to predictive technologies and healthcare systems are expected to govern them, there is a critical need for empirical evidence on both stakeholders’ needs, perspectives, and expectations. Policymakers, model developers, and health system leadership have roles to play in leveraging this evidence to design more responsive and equitable predictive systems in healthcare.PhDHealth Services Organization & PolicyUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/177847/1/ptassie_1.pd