Photo courtesy: Pixabay/vitalworks
A national, U.S.-based algorithm used by health insurers to make decisions for millions of patients demonstrates “significant racial bias,” according to a study by a team of researchers.
The authors are working with developers to reduce bias.
According to a statement by the American Association for the Advancement of Science (AAAS), the tool “underestimates the health needs of black patients [and] determines that black patients are healthier than equally sick whites, thus reducing the number of black patients who are identified as requiring extra care.”
The findings may have widespread implications, given how heavily health care institutions rely on algorithms to identify risks and plot treatments.
The algorithm analyzed was created by Optum and guides decision-making for millions of people, but The Washington Post reports the issue could be present in other tools used by the health care industry, which collectively manages 200 million people annually.
VIDEO: Biased algorithms can cause significant harm
Don’t blame the algorithm
“We shouldn’t be blaming the algorithm,” lead author Dr. Ziad Obermeyer, a health researcher and professor at UC Berkely, said in an interview with the LA Times.
“We should be blaming ourselves because the algorithm is just learning from the data we give it.”
He said his team “stumbled” across the bias nearly by accident.
Obermeyer and his team were recruited to analyze the app after it was purchased by an academic hospital. The institution planned to use it to identify patients who meet the criteria for a program providing vulnerable individuals with extra care.
Back-end access to the program enabled the team to examine health data from 6,079 black patients and 43,539 white patients.
That’s when it became apparent that black patients who were assigned the same risk score as white patients were more likely to experience health deterioration over the following year.
“At a given level of risk as seen by the algorithm, black patients ended up getting much sicker than white patients,” Obermeyer told LA Times, adding the findings prompted the team to zoom in on the discrepancy.
The authors determined that racial disparities arise because the algorithm uses “health costs as a proxy for health needs,” which could be perceived as a racially-based metric:
“Blacks incur lower costs than whites because of a lack of access … as well as due to systemic racism,” reads a statement by AAAS.
“Obermeyer et al. show that reformulating the algorithm using a different proxy led to an 84% reduction in racial bias.”
The full paper can be found in the journal Science.