UChicago releases playbook to eliminate AI bias in healthcare


Researchers from the University of Chicago released a playbook with four steps to diminish biases in artificial intelligence in healthcare, according to a June report by the Center for Applied AI at Chicago Booth.

Subtle technical choices can make the difference between biased and unbiased algorithms in healthcare. Biased algorithms in hospitals can influence clinical care, operational workflow and policy. 

To develop these four steps, University of Chicago worked with Boston-based Brigham & Women's Hospital, Wellesley, Mass.-based Harvard Pilgrim Health Care, Blue Cross Blue Shield North Carolina and many others.

Four steps to reduce AI bias in healthcare:

  1. Inventory: Hospitals should list all algorithms being used in their organization. Select a team member who is responsible for keeping algorithm inventory updated.

  2. Screen: Conduct a screening process for each algorithm for bias. Calculate what the algorithm is supposed to be flagging and what it actually is flagging. Discern what causes these discrepancies.

  3. Retrain: Suspend using these algorithms until they can be improved. Consider using alternative processes until an algorithm is launched again.

  4. Prevent: Develop the groundwork needed so future algorithms cannot be implemented if they contain implicit bias. Conduct regular audits of algorithms to ensure they are up to standard.

© Copyright ASC COMMUNICATIONS 2021. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.


Featured Whitepapers

Featured Webinars