The Need for Interpretable and Fair Algorithms in Health Care and Policy

• Activity Number: 395 - The Need for Interpretable and Fair Algorithms in Health Policy
• Type: Invited
• Date/Time: Wednesday, August 5, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Marzyeh Ghassemi* and Melody Goodman* and Sherri Rose* and Julius Adebayo*
• Companies: University of Toronto and NYU and Harvard Medical School and MIT
• Keywords: fairness; interpretable algorithms; machine learning; health care; health policy

Abstract

This invited hybrid panel session will focus on engaging talks and panel discussion on the topic of interpretable and fair algorithmic approaches for health care and policy. Interpretability and fairness have particular salience in health applications, given the stakes involved. The broad construct for this session is drawn from the larger theme of creating accountable algorithms. We specifically focus on how in health care and policy we can create tools that: engage key stakeholders, are explainable to communities, and avoid discriminatory or unjust impacts. Each panelist will give a 15 minute presentation followed by a structured discussion among the panelists incorporating audience Q&A. Topics include: (1) machine learning from our mistakes; (2) development and implementation of a computer adaptive algorithm to reduce participant burden; (3) simplifying risk adjustment for interpretability and fairness; and (4) protected attributes and fair representations in machine learning. Our panel will therefore provide contemporary insights on the timely topic of contributing to the social good with empirical evidence to ultimately improve human health.