Medical Algorithms Have a Race Problem

When Vanessa Grubbs, MD, first met Eli about a decade ago, he was a muscular man in his 50s. Grubbs, a nephrologist at Zuckerberg San Francisco General Hospital, was treating him for membranous nephropathy, a kidney condition that can lead to organ damage and failure. To take his vitals, she'd get out an extra-large blood pressure cuff.

Eli is not his real name. Grubbs refers to her patient as "Book of Eli" because when they first met, he looked like Denzel Washington's brawny character in the 2010 action movie by that name. But over the next five years, as his condition progressed, Eli slowly lost some of his bulk. Grubbs switched to a regular-sized cuff to take his blood pressure.

Eventually, the condition left Eli's kidneys so damaged that it was time to consider an organ transplant. But kidneys are in short supply: About 23,400 transplants took place last year, and more than 92,000 people are on a national waitlist.

So when Grubbs ordered a GFR test for Eli, she got two numbers back. The report said his GFR was estimated at 20 "if not African American," and 24 "if African American." There were no other racial categories.

If Eli had been white, his blood test result would have qualified him for a spot on the transplant waitlist. Because he is Black, he did not appear to make the cut. The use of two different numbers, one for Black patients and another for everyone else, dates to a 1999 study on kidney function. Similar race adjustments (also called race corrections) crop up in all sorts of clinical algorithms in medicine.

Some of the algorithms help doctors decipher test results like Eli's. Others combine medical and demographic information to recommend a specific diagnostic test, or produce a risk score that helps determine whether a patient is a good candidate for a particular treatment. Algorithms like these sometimes adjust for age, sex, and other factors that can help account for broad physiological differences among patients. But the race adjustments are more controversial.

Grubbs, who has long been skeptical of race-adjusted formulas, didn't stop with Eli's initial GFR test results. "I didn't believe that just because he was Black he had higher kidney function," she says.

Weight loss like Eli's can affect some GFR estimates, and he was very close to the transplant cutoff. So Grubbs ordered a more elaborate GFR test, one that she considers to be more precise. The test is onerous, though, requiring a patient to collect urine samples over a period of 24 hours, in addition to having blood drawn. It doesn't have a race adjustment factor.

The results came back with a single value: 20. Grubbs referred Eli to a transplant center, which entered his name on the kidney transplant waitlist. Now, after suffering through other health setbacks, Eli is nearing the front of the transplant line, Grubbs says, and he could get a new kidney within a year.

And in nephrology, race-adjusted risk scores can affect patients long before they need a transplant. Because a higher GFR estimate can make Black patients' kidneys appear healthier than non-Black patients', they could be referred to a specialist later, or get prescribed medications at doses that are too high for people with impaired kidney function. 

These worries hit close to home for Nwamaka Eneanya, MD, a nephrologist and an assistant professor at the University of Pennsylvania.

Several years ago, one of Eneanya's relatives sought care for kidney disease. For a long time, he was not referred for a kidney transplant evaluation because his race-adjusted GFR value was above 20. (Like Eneanya, he is Black.) Eneanya persuaded him to switch to a new doctor, who ordered more testing, and eventually he got on the waitlist.

She has this advice for people in similar situations: "If you have kidney disease and are Black, ask your doctor, ‘What does this mean for my current care? If I am unable to receive certain specialty referrals or treatments, are there alternative measurements of kidney function that can be used that do not include race?'"

Last year, Eneanya co-authored an influential essay for the Journal of the American Medical Association (JAMA) that laid out some of the ways that race-adjusted GFR equations can hurt Black patients. The delay in getting an accurate GFR estimate had real consequences for Eneanya's family member.

"Those few years of back-and-forth" could have been spent on the transplant wait list, she tells CR. "Several years were lost because of this one number."

Critics like Roberts, Eneanya, and Grubbs say that race is given too much weight in medical decisions. "It's promoting the idea that Black people as a race are distinguishable biologically—just because of their race—from other human beings," Roberts says. That idea has been debunked, experts say: There's no genetic test or biological marker that places someone in one race or another. Instead, races are loosely defined social categories that change over time.

Not everyone agrees with the recent clinical changes. Neil Powe, MD, chief of medicine at Zuckerberg San Francisco General Hospital, says there isn't enough evidence yet for dropping race-corrected algorithms. In a July essay published in JAMA titled "Black Kidney Function Matters," Powe says the race adjustment found in the most common GFR equations makes them more accurate for Black patients, and can help make sure they get the right care.

The disagreement over GFR numbers comes down to some decades-old research. The 1999 study on kidney function that forms the basis of one widely used algorithm included 197 Black participants and 1,304 white participants. It found that the Black people in the study had somewhat higher concentrations of creatinine than white people whose kidneys were filtering blood at the same rate.

The authors proposed that a Black person's test result be multiplied by a particular number to adjust their score. Doing that, they concluded, would give a better picture of a Black patient's real kidney function. Another study published in 2009 included more Black participants and found similar differences, but with a smaller gap.

No one knows exactly what caused the differences between Black people and others in the two studies. But Powe says the data can still be useful. And he worries that removing race adjustments without more research could hurt Black patients. What if their kidneys start to look worse than they really are? That could lead to overdiagnosis of kidney disease, potentially causing patients with other medical conditions to be denied adequate pain medication or chemotherapy. It could decrease access to potentially lifesaving clinical trials, he says, and even result in higher life insurance premiums. 

Lesley Inker, MD, a nephrologist at Tufts University, helped develop the GFR algorithms that include race adjustments. She argues that tossing out the race variable would set GFR estimation back 30 years. "For people who have Black ancestry, [this algorithm] is the best guess I've got."

Inker's research group is currently analyzing how removing the race coefficient would affect patients. In the meantime, she emphasizes that the race-adjusted GFR evaluation is meant as "a first-line test." In high-stakes situations like those involving medicines with high toxicity levels, doctors should do additional testing, she says.

One thing Powe, Grubbs, and others in the debate agree on: Americans' health status does vary significantly by race—on average, Black Americans are more likely than white Americans to have any of an array of health problems, including diabetes, stroke, high blood pressure, and childhood asthma. The question is why, and how doctors should use that information. For instance, beyond race, asthma is strongly associated with environmental factors such as poverty, outdoor air pollution from sources including diesel exhaust, and smoking—does it make more sense to note whether a child with asthma is Black, or what their living conditions are like?

The U.S. has a long and often ugly history of race-based medicine, starting long before the infamous Tuskegee syphilis study, in which hundreds of African-American men were kept in the dark about their diagnosis and went untreated for decades, while researchers observed the devastating progression of the disease. In one example from before the Civil War, a Southern physician named Samuel Cartwright used a device called a spirometer to compare the lung capacities of enslaved Black people and white people. Cartwright concluded that Black people have less lung capacity than white people—a "deficiency" that he argued made them less mentally alert, and unfit for freedom.

The idea that there are innate racial differences in lung function was latched onto by pro-slavery scientists and, later, by eugenicists who argued that certain people should be sterilized based on their disabilities, race, or ethnicity, according to Lundy Braun, PhD, a Brown University professor who chronicles the history of race and the spirometer in her book "Breathing Race Into the Machine."

Some of that history may have been forgotten by 1974, when race-based differences in lung capacity were codified into what may have been the first clinical algorithm with a race factor. "That was a significant moment in race correction," Braun says.

Spirometers are still used today to measure lung, or "vital," capacity. For healthy people, the measurement varies mainly with height, age, and sex, but modern studies find that the measurement also tends to differ by race. Vital capacity for a Black person is typically lower than for a similar white patient, and test results often include a race adjustment: Some readings that seem abnormally low for a white person might fall in what's considered a normal range for a Black person. That can affect how patients are diagnosed and treated.

"We've really never stopped doing these Black-white comparisons and just assuming that Black people are inherently different," Grubbs says.

Removing race from GFR estimates didn't come as easily at some other institutions. At UCSF, a group of physicians that included Eli's doctor, Vanessa Grubbs, began organizing in the summer of 2019 to ask for the race adjustment to be dropped. The request set off a yearlong tug-of-war.

The laboratory at the UCSF-affiliated Zuckerberg San Francisco General Hospital removed race from its GFR estimates late last year, then prepared to add it back in response to internal blowback. Eventually, after a petition calling on UCSF and SF General Hospital to report GFR without using race garnered several hundred signatures, the hospital said it would keep race out of its estimates. Now, UCSF is making plans to roll out a race-free GFR equation across all its Bay Area hospitals and clinics.

Powe thinks this is a mistake, one that's likely to hurt Black people seeking care. "To be crude, I believe what's happening is a knee-jerk response," he tells CR. "We need to slow down as a community of physicians to figure out how best to do this."

Stephen Richmond, MD, who recently finished his residency training at UCSF, was one of the petition's co-organizers. He says he wasn't surprised at the resistance his group faced. "We didn't expect these dominoes to fall with ease."

In recent weeks, UCSF has become much more enthusiastic about removing race from GFR, Richmond says. He and other young advocates who spoke with CR attributed their recent successes in part to this year's surge in Black Lives Matter protests. "This is a political and social issue that requires advocacy just as much as it is a scientific one," he says.

Several major institutions are making changes to the way they use race in algorithms. Massachusetts General Hospital and Brigham and Women's Hospital, as well as hospitals affiliated with Vanderbilt University, Brown University, and the University of Colorado, have removed race factors from the GFR algorithms they use. (In 2017, Boston's Beth Israel Deaconess Medical Center stopped citing race in GFR reporting, but it still uses the adjustment factor to report a range of estimated GFR numbers for every patient.)

Groups at New York's Icahn School of Medicine at Mount Sinai, the University of Southern California, and the University of Nebraska, among others, are advocating for the removal of race adjustments at their institutions, too.

At the end of August, the National Kidney Foundation and the American Society of Nephrology announced a working group that will debate the use of race in GFR calculators. It includes Eneanya and Inker, and Powe is its co-chair. If the group recommends removing race, the decision could ripple across the U.S., reaching community clinics and rural hospitals that may be less likely to push for controversial changes than teaching hospitals in major cities. The group says its recommendations will be published later this year. 

But the activists aren't waiting. In August, Richmond and Grubbs spoke at a well-attended online workshop from the Institute for Healing and Justice in Medicine, designed to help students at medical schools and hospitals around the country push for an end to the race adjustment in GFR estimates. And they don't intend to stop with kidneys.

"We don't want this to be, ‘We fixed [the GFR test] and we fixed the whole thing,'" says Maddy Kane, one of the workshop's co-organizers and a student at a joint UC Berkeley–UCSF degree program that teaches medicine and public health. "We refuse to be a part of a system that's perpetuating harm."

For now, race-adjusted algorithms remain in wide use. That leaves some people who are facing serious health problems in limbo. Should patients who are Black worry that their lab results are being skewed because of their skin color?  

Doctors point out that race adjustments don't come up with every medical test and procedure. Patients with certain chronic conditions—kidney disease, heart disease, hypertension, and some forms of cancer—are the most likely to encounter race-adjusted algorithms, says David Jones, MD, a Harvard Medical School professor and co-author of the NEJM article on race adjustments. Patients have a right to ask about them, he says, just as they should feel free to inquire about any other aspect of their care.

"I'd likely ask the doctor, ‘I'm just curious: Do any of the tests or practice guidelines you are using take my race or ethnicity into consideration?'" Jones says. If the answer is yes, Jones counsels skepticism over how the doctor is classifying your race and using it to inform the care they deliver.

Outside of nephrology, the revisions to race-adjusted algorithms are just beginning. The calculator used to guide obstetricians deciding whether to recommend a C-section could be revised soon to eliminate the race multiplier. In other medical fields changes to race adjustments in algorithms could come slowly, as specialists in each discipline study how patients would be affected.

"No one is saying to throw away science," says Eneanya, from the University of Pennsylvania. "We just want to make sure that we are not causing harm to our patients."

Update: This article has been updated with additional information from Neil Powe, MD, on the potential ill effects of removing the race adjustment from common GFR tests, including his concern that overdiagnosis of kidney problems could lead to patients with other conditions receiving inadequate doses of important medicines. It was originally published Sept. 17, 2020.