Ensuring Algorithmic Fairness in Healthcare: Challenges, Implications, and Strategies

Presentation Time: 08:30 AM - 11:30 AM

Abstract Keywords: Artificial Intelligence, Machine Learning, Fairness and elimination of bias, Causal Inference, Racial disparities, Health Equity
Working Group: Health and Healthcare Equity Working Group
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

Algorithmic fairness is the ability of an algorithm to produce unbiased results where no group is systematically favored over another group. Though predictive ML models are being deployed in numerous healthcare settings, the fairness of these models is less frequently discussed than their predictive performance. This has consequences, including inappropriate recommendations for clinical care for certain groups, which can be seen as discriminatory. The workshop will begin by describing the concept of data equity and responsible usage of data and identify a framework that applies principles of data equity to a project life cycle. We will delve into specific case studies of algorithmic bias and then review different categories and measures of algorithmic fairness, tradeoffs between different measures, and tradeoffs between fairness and model accuracy. For sources of bias, we will review a macro societal model of how discrimination in the real world is reflected in data and how micro-programming sources of bias appear in the training and deployment of a predictive model. We will then review holistic views of mitigating bias and technical solutions. We will present Fairlearn, a Python package that produces fairness metrics for a given ML model and can rectify these biases. We will conclude by providing hands-on training for Fairlearn with a healthcare dataset.

Speakers:
Jodi Lapidus, PhD
Oregon Health Sciences University

Mohammad Adibuzzaman, PhD
Oregon Health & Science University

Katarina Pejcinovic, M.S.
Oregon Health & Science University

Rumel Mahmood, PhD
Oregon Health Sciences University

Authors:
Rumel Mahmood, PhD - Oregon Health Sciences University; Katarina Pejcinovic, M.S. - Oregon Health & Science University; Jodi Lapidus, PhD - Oregon Health Sciences University; Mohammad Adibuzzaman, PhD - Oregon Health & Science University;

---