PROJECT ABSTRACT: Health care algorithms make decisions that impact the lives of millions of individuals, yet few are rigorously evaluated for systemic harms after they are implemented. Unfortunately, these oversights have led to the perpetuation of health care disparities. In an effort to leap ahead of the current state of the field and protect marginalized groups, our research would build a framework for assessing the social impact of health care algorithms before they are deployed, which is a novel line of investigation. This proposal is inherently interdisciplinary, spanning machine learning, ethical AI, health economics, public health, decision sciences, statistics, health policy, and qualitative research. Mathematical decision science microsimulation models will be developed, positing an underlying complex causal network of the health care system. In order to initialize these microsimulation models, a diverse collection of data sources will be used, including health care billing claims, policy intervention information, and community- based qualitative feedback. A main priority is to prevent harm to racial and ethnic minorities, individuals with disabilities, and older adults, among other marginalized groups. Primary outcomes under consideration center health care access, quality, and costs. Robustness and rigor are central to our work. Despite no comparator framework existing, we will additionally develop simpler causal network models along with Markov cohort models to provide a basis for comparison. Outputs produced will also be freely shared with the research and stakeholder communities in the form of open-source code and open-access preprints featuring transparent descriptions of all models and assumptions. This work would involve a substantial shift in focus for the Principal Investigator; pivoting to research that leverages mathematical decision science modeling and qualitative approaches while merging them with her expertise in machine learning for health care. Importantly, this framework has the potential to influence a new set of standards and guidelines for AI algorithms, establishing a blueprint where tools are routinely evaluated for social impact before deployment. Thus, creating this first-of-its-kind social impact framework could transform the development and application of algorithms in health care, representing a substantial paradigm shift with broad impact.

PROJECT NARRATIVE: Algorithms continue to proliferate in the health care system, often with dire consequences for marginalized groups. This research proposes building a framework for evaluating the social impact of algorithms in health care before they are deployed to avoid such harms using mathematical decision science modeling. Developing this open-source framework will have the potential for broad impact in reducing health care disparities with a flexible design to allow adaptation to many health settings.