Data Science for Health Equity: Perils and Promise of AI/ML in Healthcare

Abstract

Unequal access to healthcare and inequitable health outcomes are widely prevalent among underserved and marginalized populations. Today, policy makers acknowledge health equity as one of the world’s "grand challenges," that must be intentionally addressed. The renewed focus on health equity in public discourse is occurring contemporaneously with another major trend in society in general and healthcare in particular: a digital transformation of economic and social transactions. In this commentary we juxtapose these two trends to pose the question: what is the interaction between health equity and AI and machine learning (AI/ML) in healthcare, and what are the associated implications for data science? Our goal is to contribute to the critical and emerging area of healthcare equity, disparities, and AI/ML by unpacking the challenges that digitization, structural inequities, and algorithmic bias pose for the vision of health equity. We describe the phenomena of structural and systemic inequity and focus on the unique distinctions between AI/ML in health care and other fields, specifically addressing issues of scoping, centering, and bias correction. We examine two critical aspects of any ML project in healthcare: the data feeding the algorithms and the incorporation of fairness in modeling. We offer a call-to-action for data scientists working on healthcare AI/ML research, motivated by a need to proactively address a critical societal problem. We argue that data scientists and health care operations researchers need to proactively address the many algorithmic and operational challenges that arise in order to deliver on the promise of AI/ML in health care equitably for all populations.

Note:
Funding Information: There is no funding to report

Conflict of Interests: There are no competing interest to report.

Keywords: Algorithmic Fairness, Machine Learning Bias, Healthcare

JEL Classification: I18