Ensuring Equity in the Development of Health AI Algorithms

The burgeoning field of bioinformatics promises to lead healthcare into an age of seemingly endless possibility, especially when incorporated with the powerful developing techniques of artificial intelligence and its subdiscipline machine learning. Biotechnology companies and hospitals have improved diagnosis and treatment of diseases via predictive bioinformatics tools that help clinicians understand a patient’s disease and devise individualized treatment strategies. AI algorithms can provide great utility to healthcare providers in these ways, but such algorithms also spark concerns about equity and fairness. Researchers developing health AI algorithms, such as myself, must study how and why health AI algorithms become inequitable so we can avoid making the same mistakes in our own work. To seek these much-needed answers, my STS thesis investigates the following primary research question: how do we ensure that healthcare ML algorithms serve all kinds of people equitably, minimizing rather than exacerbating disparity in healthcare? In this paper, I give an equity analysis of three case studies of health AI implementations using the Social Construction of Technology framework. These case studies include the Impact Pro health AI for recommending additional care, the Sepsis Watch health AI for diagnosing sepsis, and a health AI created by researchers at the University of Florida for diagnosing bacterial vaginosis. Finally, I consolidate these findings and add my own insights to provide a set of guidelines for researchers to ensure equitable health AI moving forward.

School of Engineering and Applied Science

Bachelor of Science in Biomedical Engineering

STS Advisor: Joshua Earle