How Scientists Are Subtracting Race From Medical Risk Calculators

To pediatrician Nader Shaikh, the rhythm of treating babies running high fevers is familiar. After ruling out the obvious colds and other common viruses, he must often thread a catheter into a months-old baby to draw a urine sample and check for a urinary tract infection (UTI). {snip}

UTIs, although relatively rare in children under age 2, carry a high risk of kidney damage in this group if left untreated. Often, the only symptom is a high fever. But high fevers can also signal a brain or blood infection, or a dozen other illnesses that can be diagnosed without a urine sample. To help clinicians avoid the unnecessary pain and expense of catheterizing a shrieking infant, Shaikh and his colleagues developed an equation that gauges a child’s risk of a UTI based on age, fever, circumcision status, gender, and other factors—including whether the child is Black or white. Race is part of the equation because previous studies found that—for reasons that aren’t clear—UTIs are far less common in Black children than in white ones.

The UTI algorithm is only one of several risk calculators that factor in race, which doctors routinely use to make decisions about patients’ care. Some help them decide what tests to perform next or which patients to refer to a specialist. Others help gauge a patient’s lung health, their ability to donate a liver or kidney, or which diabetes medicines they need.

In the past few years, however, U.S. doctors and students reckoning with racism in medicine have questioned the use of algorithms that include race as a variable. Their efforts gained momentum thanks to the Black Lives Matter movement. In August 2020, a commentary published in The New England Journal of Medicine (NEJM) highlighted the use of race in calculators as a problem “hidden in plain sight.” It’s widely agreed that race is a classification system designed by humans that lacks a genetic basis, says Darshali Vyas, a medical resident at Massachusetts General Hospital and co-author on the paper. {snip}

Vyas and others warn that using race to adjust risk calculators may also widen existing health disparities. Black Americans are generally diagnosed with kidney disease later than white Americans, which delays treatment and puts them at greater risk of developing kidney failure—yet an equation widely used to measure kidney function tends to estimate better function for Black patients relative to non-Black patients. Osteoporosis is underdiagnosed and undertreated in Black women, but a common bone fracture risk calculator places them, along with Asian and Hispanic women, at lower risk than white women. “We know these disparities exist, yet the calculators tell us that we don’t need to worry about this population,” says epidemiologist Anjum Hajat of the University of Washington, Seattle.

Some of these calculations are rooted in racist assumptions. Others emerged out of an effort to improve predictions across racial groups. The challenge of defining “normal” versus “diseased” and capturing these qualities accurately in a simple test led scientists to grasp whatever data they could to make their tools more accurate. And at a population scale, race often does correlate with medical outcomes, in part because it acts as a proxy for the influence of other socioeconomic factors on health.

But even if racial trends sharpen predictions, using them to make decisions about an individual’s treatment is problematic, Hajat says. “Even if a calculator is not causing disparities, it is maintaining and perpetuating them,” she says. For some, applying a different standard to Black patients than to white ones recalls a long history of neglect and discrimination in medicine. “I don’t think people had bad intentions when they were creating these calculators,” Hajat says. “But we have to be aware that biomedical research has really contributed to upholding white supremacy, which is why we’re reexamining the calculators now.”

The questions are already spurring change. In March, a task force from the American Society of Nephrology and the National Kidney Foundation recommended removing race as a variable in the kidney function calculator, known as the estimated glomerular filtration rate (eGFR) equation. The University of Washington, Beth Israel Deaconess Medical Center, and others have already dropped race from their eGFR calculations.

But similar efforts met resistance at other institutions. To some researchers and clinicians, the use of calculators that incorporate race seems not just appropriate, but a crucial measure to avoid unnecessary medication or invasive treatments, such as a catheter in a 6-month-old baby. Shaikh sees the UTI equation’s use of race as an effort to achieve equity, not worsen disparities. “It sounds weird to use race to pick patients, and it doesn’t look good on the surface,” he says. “But which one is worse: catheterizing kids who don’t need it or using race in an algorithm? It’s more complicated than it seems.”

{snip}

At least two modern-day risk calculators have been accused of having similarly racist logic: One, which estimates a woman’s odds of successful vaginal birth after cesarean section (VBAC), falsely assumes that women’s pelvis shapes differ based on race, making this form of childbirth riskier for Black and Hispanic women compared with white women. Another equation estimates lung function by gauging the maximum amount of air a person can exhale forcefully into an instrument called a spirometer. Lower measurements are considered normal for Black and Asian people, based on the disputed assumption that their lung capacity is lower. “The spirometer was built on anti-Black racism,” says Lundy Braun of Brown University, who studies the history of racial health disparities. The VBAC calculator was updated to remove race in May, but spirometers still include a race adjustment. The American Thoracic Society (ATS) has begun to examine its use, Braun says.

In other calculators, race has been added to bring measurements in line with the best available data. The eGFR equation, developed in 1999, estimates how well a person’s kidneys function based on urinary levels of a compound called creatinine, which builds up in blood when kidney filtration declines. Because the equation doesn’t test kidney function directly, its developers compared its results with kidney filtration rates measured using a more definitive test, based on a radioactive tracer, that is too complex to perform routinely. They found the eGFR equation consistently underestimated kidney function in Black patients, so they used a common statistical method called curve fitting to adjust the estimates according to race.

Other risk calculators have added race in an effort to better match epidemiological data. In 1992, the World Health Organization recognized an epidemic of osteoporosis and funded research to develop a tool that could assess a person’s risk for fractures based on the brittleness of their bones. Researchers developed several country-specific versions of the tool for the United States, Canada, South Africa, and others, which incorporated race-specific prevalence where data were available.

When adapting the equation to U.S. populations, the researchers included a race correction to account for the lower reported occurrence of osteoporosis in Black women. The goal was to avoid medicating people who didn’t need it, and the correction brought fracture predictions in line with official rates of disease.

And in this case, the differences may have a physiological underpinning, says epidemiologist Nicole Wright of the University of Alabama, Birmingham, who studies disparities in bone health. “Genetically, people of African descent have higher bone mass than others, so you need to account for that,” she says. {snip}

{snip}

Replacing race with a different metric is not always easy. Recent studies have attempted to use ZIP codes, income or education levels, or a measure of socioeconomic status called the area deprivation index instead of race to capture conditions that influence health. Precisely how they’d be implemented isn’t clear, and few have been put to work in clinics or endorsed by professional societies of clinicians.

{snip}