Algorithmic Bias Initiative

The Playbook

Is your organization using biased algorithms? How would you know? What would you do if so? This playbook describes 4 steps your organization can take to answer these questions. It distills insights from our years of applied work helping others diagnose and mitigate bias in live algorithms.

The Research

Recently, our researchers published an article in Science that showed significant racial bias in a widely deployed health-care algorithm that selects patients for a managed care plan. Were that bias to be eliminated, the number of African American patients selected would nearly double. Working with the developer of the algorithm, we found solutions to dramatically reduce the bias and improve the care of many patients.

In addition, a paper we published in Nature Medicine helps show how algorithms can help physicians more equitably treat diverse populations who experience chronic pain differently. By carefully selecting training data based on patient reporting, we showed that the algorithm was more accurate than the typical methodology clinicians use and was able to significantly reduce racial disparity at every level of pain measurement.

Now, we're working with organizations across the country to help their teams understand how bias arises in healthcare algorithms and what they can do to mitigate those problems with the goal of providing more equitable healthcare for everyone.