Very Little Is Keeping Doctors From Using Racist Health Formulas
Yet while medicine has roiled over the issue of race in equations, academic research—the birthplace of these calculators—has remained relatively untouched. Walker and his colleagues receive daily requests from researchers asking for their equations to be added to MDCalc. The MDCalc team vets tools closely and flags ones that don’t meet certain standards, but their criteria are not a general requirement for equations in the scientific literature.
No one markets these calculators, and no one gains financially from their use, at least at first. Once they’re in the research literature, professional societies—much like the ones that made the recent announcement about the kidney function equation—may occasionally endorse certain equations. They’re incorporated in apps and tools for clinicians. As they grow popular and become part of routine care, some equations find their way onto drug labels and into medical instruments and electronic health record systems, or become available as web-based versions. The consequences can be severe: In one study last year, researchers trying to gauge the impact of the race-based kidney function equation found that at one hospital, one-third of Black patients would be reclassified to a more severe stage of disease—and receive speedier referrals for dialysis and transplants—if the race multiplier were removed.
Risk equations aren’t regulated by the FDA because “simple calculations routinely used in clinical practice are not generally the focus of our regulatory oversight,” an agency spokesperson said via email. “The FDA does not regulate the practice of medicine.” Perhaps, suggests Walker, who has spoken with FDA officials in the past, this is because they’re all based on publicly available research. “It might be a little different if they were more of a black box, like AI-based algorithms, where you don’t actually know why the computer suggests a certain drug or test,” Walker adds. But even if they are publicly available, it’s uncertain how many doctors examine the data sources and statistical methods used to develop a calculator before using it.
Without guardrails, there’s little to determine how equations get used once they’re publicly available. Many are used to inform public health decisions, such as suggesting when someone should get a cancer screening test, receive preventive care, or to prioritize communities or individuals for vaccinations. For example, a county public health official could theoretically use Covid risk calculators to “protect” a community that’s at high risk of death—and the data would suggest their residents are largely people of color—with lockdowns and curfews rather than making vaccines or masks available sooner.
Researchers can’t stop collecting data on disparities, nor ignore the evidence that some biological outcomes—such as dying of cancer or Covid-19—appear to affect people of certain races disproportionately. But they must find new ways to figure out why without evoking race and ethnicity. To do so, academia must see demographic columns in decades-old databases for what they are: clues to a problem, not part of the solution.
To build better risk equations, researchers will have to swap out race variables with numbers linked to more objective, biological explanations, such as whether someone had a viral infection that spikes their risk of cancer. For instance, cervical cancer typically occurs more often and causes more deaths in nonwhite populations. “There was a strong relationship between race, ethnicity, and the disease, but it wasn’t necessarily biological,” says clinical genetics researcher Nicolas Wentzensen of the National Cancer Institute. Insufficient screening and follow-up care were key contributors, but were not represented in equations. Once researchers identified the link between HPV infection and cervical cancer and included the viral infection into risk equations, they found demographic variables didn’t matter anymore. “Once we understand the biology of the disease, we can take out some of these factors,” Wentzensen says. “There’s no race in any of the models because all the risk is explained by when you were exposed to the virus.”