New Research Finds "Significant Racial Bias" in Commonly-Used Healthcare Algorithm

The U.S. healthcare system uses common commercial algorithms to inform treatment decisions for millions of patients. But one algorithm, widely used for the care of millions of U.S. patients, may lose its license to practice.

On Friday, Science published a study in which researchers found "significant racial bias" in the algorithm, which consistently and dramatically undervalued the healthcare needs of black patients. The study doesn't name the algorithm provider—the WaPo says it's the health services company Optum.

• The problem: The algorithm uses healthcare costs as a proxy for health needs.
• The solution: It needs to use more accurate variables, like "active chronic conditions."

Specifically, the algorithm gave sicker black patients the same risk scores as it did to comparatively healthier white patients. Without the algorithm's bias, the percentage of black patients receiving extra help would jump from 17.7% to 46.5%.

Big picture: This research shows the real-world consequences of algorithmic bias. These systems are meant to increase the efficient allocation of resources and definitely shouldn't be restacking them along racial lines.