by Eleanor Ball
A recent study published in Science found a big issue with an algorithm used to assign health care to millions of patients a year in America. That issue? The algorithm is inherently biased against black people.
The study does not specify the name of the algorithm, but characterizes it as “a live, scaled algorithm deployed nationwide today” and “one of the largest and most typical examples” of tools used to allocate care to around 200 million Americans on a yearly basis. Large health care systems and insurers rely on these algorithms to apportion spots in “high-risk care management programs.” Specialized programs like this are designed to treat patients with complicated health needs, and they often require significant additional resources and funding. Due to these costs, health care systems rely primarily on algorithms--as opposed to providers, who are much more likely to introduce bias into the decision--to give out spots in these programs. The issue scientists found was that the algorithms had been designed in a way that leads to their decisions reflecting not just human error, but deep and systemic racism.
The algorithm works by assigning health risk scores to patients based on how chronically sick they are, then prioritizing the referral of those with higher scores (and, therefore, more pervasive and serious illness) to these specialized programs. The trouble comes with how those risk scores are determined. Risk scores are calculated based on the amount of health care costs the person has accrued in one year. Since higher health care costs are generally associated with greater a need for health care, this seems fair--especially considering the average black person and the average white person in the data set had the same amount of health care costs. However, a closer look at the data revealed the average black person had more health care needs than the average white person. Within black populations in the data set, there was a much higher incidence of conditions like diabetes, high blood pressure, and kidney failure than within white populations in the data set. Consequently, despite black people often needing more care than white people, they were being provided the same amount of care. And because they were being provided the same amount of care, the algorithm interpreted their health needs as being the same. Therefore, when spots in specialized programs were being allocated, a chronically ill black person who had been previously underprovided care would very often lose out to a white person who had been provided the right amount of care.
The discrepancy that results from this is shocking. The researchers calculated that currently, only 17.7% of the patients who receive extra care after being fed through the algorithm are black. If the algorithm were not racially biased, this number would be 46.5%.
Where does the racial disparity in providing health care come from? Experts believe there are many reasons black people are chronically underserved. Widespread distrust of health systems and providers, cultural barriers between patients and providers, and racial stereotyping or direct discrimination by providers are among the most prevelant. Though black populations are the focus of this study, these factors also contribute to the poor quality of health care experienced by immigrants and other people of color. All of these factors result from systemic racism.
Outside of the health care sector, this bias causes ripple effects. Black people are more likely to suffer in other areas of society, such as being victims of police brutality and having trouble finding a job. Poor access to health care--for example, being denied access to a program that provides necessary care to those suffering from complex combinations of mental illness, which may follow witnessing or being victimized by police violence--can compound the problems black people face in other areas of their lives.
Our health is the bedrock of our lives, and public health is the bedrock of society. When we don’t give people the ability to lead healthy lives, we take away so much more than a therapy session or bottle of pills. Luckily, the scientists behind this study have identified variables other than health care costs, such as the number of active chronic illnesses, that could lead to a substantially less biased algorithm. Hopefully, the resultant equity--not the bias and suffering too often experienced today--will be the hallmark of America’s future health care system.
Eleanor Ball is a GW Scope staff writer and junior humanities editor for the George Washington Undergraduate Review. She is currently researching 17th century Flemish art and the relationships between period innovations in science and aesthetics. A freshman studying international affairs and public health, she is also on the board of GW Shakespeare and a member of GlobeMed and the Politics & Values program.
Comments