Correcting medicine's use of race in clinical algorithms

To understand the problem with using race in clinical algorithms, consider Barack Obama's kidneys.

President Obama's mother was a White woman from Kansas, and his father was a Black man from Kenya. People generally consider him to be a Black man, and if he were admitted to the hospital, his medical record would likely describe him as such.

Suppose then that a physician wanted to know the former president's estimated glomerular filtration rate (eGFR), the calculation of which is based on serum creatinine level, age, sex, and, usually, race. “There are thousands of hospitals in the United States. All but a small number would race-correct Obama's eGFR,” said David Jones, MD, PhD, professor of the culture of medicine at Harvard Medical School in Boston.

The race correction in the eGFR algorithms makes the result higher in Black patients, based on data that Black patients had higher GFRs at a given creatinine level than patients of other races. There's no evidence that this would be more likely for someone with President Obama's genetics, though.

“If there is anything to race and ancestry in eGFR, it relates to APOL1 alleles that you see most often in West Africans. And Barack Obama has zero West African ancestry,” said Dr. Jones. “He self-identifies as Black for cultural reasons. His self-identification then leads us to assume that those are the alleles that his muscles and kidneys have. … It makes no sense.”

A growing number of experts and hospital leaders agree. A dozen or so prominent hospitals have so far announced plans to stop race-correcting eGFR. Recent efforts, led in many cases by medical students and residents, have pushed institutions to reconsider the use of race not just in estimating kidney function, but in a variety of algorithms and scores developed and applied by a range of specialties.

Questioning race-based scores

Clinicians encounter many race-based treatment algorithms and risk scores, said Jessica Cerdeña, MPhil, an MD/PhD student at Yale University in New Haven, Conn., and lead author of a Viewpoint published in The Lancet on Oct. 10, 2020, that urged clinicians to move from race-based to race-conscious medicine.

“I recall very vividly being on the wards, and we were talking about the Eighth Joint National Committee [JNC8] algorithm for hypertension treatment, and my supervising resident said we weren't going to give an [angiotensin-converting enzyme] inhibitor to one of our patients who was Black,” she said. “I was like, ‘Why? It's a very well-tolerated drug. It would make sense in this patient.’ He just said, ‘Well, the data say that we shouldn't.’”

For students and trainees like Ms. Cerdeña, the resident's answer may seem insufficient after years spent learning about historical examples of racism, such as eugenics, and current racial disparities in health outcomes.

“In the classroom, our social medicine curriculum [taught] about race as a social construct that lacks genetic basis,” said Darshali Vyas, MD, a resident physician at Massachusetts General Hospital (MGH) in Boston. “Entering the hospital and seeing that race was inserted into many different types of tools as an input variable” raised questions for her and some of her peers.

Dr. Vyas and Dr. Jones (along with Leo G. Eisenstein, MD) listed a number of such questionable uses of race in algorithms in an article published by the New England Journal of Medicine on Aug. 27, 2020. In addition to eGFR, they called out scores used to stratify risks during heart failure hospitalizations and thoracic surgery, predictors of the success of kidney donation and vaginal birth, an assessment of the likelihood of ureteral stones, two tools each for evaluating breast cancer and osteoporosis risk, and the interpretation of spirometry in pulmonary function testing.

“Many of these tools rely on very simple race categories, not even the usual five race and ethnic categories, but on a distinction between Black and non-Black, as if you can divide the world's people into those categories,” said Dr. Jones.

The tools also often rely on epidemiologic data, and when researchers have recently dug into the underlying evidence, they've identified issues with how developers interpreted the statistics to create the scores and algorithms. For example, Dr. Vyas began her work in this field studying the Vaginal Birth After Cesarean (VBAC) calculator in 2017.

“These tools are often built after large data sets [are used to] examine who is most likely to have a successful outcome, such as a successful vaginal delivery. They look at all the factors that correlate significantly with the outcome of interest and some of them get incorporated into the final model,” she explained.

In the case of the VBAC score, the developers found that Black and Hispanic women were less likely to have a successful VBAC than White women, so they included correctors that reduce the projected likelihood of success for women classified as Black or Hispanic.

But an analysis of the original data, published by Dr. Vyas and colleagues in Women's Health Issues on May 6, 2019, found that other statistically significant factors were missing from the VBAC score. Specifically, research had shown that women who were married or had insurance were more likely to have successful vaginal deliveries than those who were not married or without insurance.

“There's a willingness to believe that race is biological, even when we're confronted side by side with factors like marital status and insurance that were very clearly excluded from the model because they were understood to be socially mediated,” she said.

Sussing out social factors

Many of the correlations between race and outcomes that were used in the development of algorithms could be due to social factors, Dr. Jones said. He offered the Society of Thoracic Surgeons' risk score for coronary artery bypass grafting as an example.

“I don't think anyone believes that there's an allele that makes Black people die from cardiac surgery 19% of the time more than White people do. Is it because on average these people are poor? Is it because on average they have different pre-existing conditions? It is because on average they go to a different set of hospitals?” he said.

If a pre-existing condition were responsible for the disparity, physicians could advise affected patients accordingly, or if the variation were in hospital quality, everyone treated at those facilities should be aware of that, Dr. Jones noted. Without such differentiation of the causes, identifying a difference between White and Black patients is generally not very helpful to clinical care, the experts agreed.

And when the tools push racial divides in the direction of existing health inequities, they are potentially harmful. “In this country, there are rampant racial disparities in maternal morbidity and mortality around pregnancy, so any tool that discounts or reduces the likelihood of successful vaginal delivery among women of color is concerning because vaginal births often have fewer health risks,” said Dr. Vyas. “The concern is that the tool could be guiding clinicians to disproportionately recommend cesarean sections to women of color.”

There are some formal studies of the effects of the eGFR race corrector on Black patients. A recent cross-sectional study used a registry of 56,845 patients treated for chronic kidney disease (CKD) in the Partners HealthCare System, 2,225 of them Black, and recalculated their eGFRs without race correction.

“When we started pursuing this effort and we started talking to some other nephrologists about it, some of them said, ‘Maybe don't waste your time, because we don't think that it's going to make that much of a difference,’” said study author Mallika L. Mendu, MD, MBA, an assistant professor at Harvard Medical School and a nephrologist at Brigham and Women's Hospital in Boston.

The study found, however, that a third (33.4%) of the Black patients would be reclassified to a more severe CKD stage if the race multiplier were removed, according to results published by the Journal of General Internal Medicine (JGIM) on Oct. 15, 2020. Of these, 64 patients would see their eGFRs drop below the 20 mL/min/1.73 m² cutoff, which would qualify them for the list for kidney transplant priority. None of those patients had been referred, evaluated, or wait-listed for transplant.

“That was really striking to us,” said Dr. Mendu. “Based on that small, incremental change in eGFR, that's having a profound impact on whether or not they're referred to transplant, and that's so important because we know there are major disparities when it comes to transplantation for African-American patients.”

A subsequent study, which used data from the National Health and Nutrition Examination Survey to look at 9,522 Black patients, had similar findings, published by JAMA on Dec. 2, 2020. Researchers evaluated how removing race from the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation would affect the patients' eGFR and found that the median eGFR decreased from 102.9 to 88.8 mL/min/1.73 m² and the prevalence of CKD among Black adults increased from 14.9% to 18.4%, along with increases in the proportions qualifying for transplant and other interventions.

An accompanying editorial comment noted that removing the race modifier from equations to estimate kidney function could cause some patients to lose eligibility for medications known to have clinical benefit in later disease but concluded that the change “could help guard against multiple potential harms of delayed diagnoses, suboptimal prevention, and late specialty care for a large population of Black individuals.”

Taking action

The JGIM study results led MGH and Brigham and Women's Hospital to be among the early facilities to give up race correction in their eGFR calculations. Another Boston hospital, Beth Israel Deaconess Medical Center, was the first in the U.S. to do so in 2017, followed by the University of Washington and the University of California, San Francisco.

These race-corrected algorithms are almost entirely an American phenomenon, according to Dr. Jones. “My understanding is that eGFR is race-corrected in the U.S. and the U.K., but nowhere else,” he said. “If you wanted to figure out how to practice race-blind medicine, just look at what people do in Canada, Latin America, Africa, Europe, Asia.”

Canada uses a version of the VBAC calculator that is similar to that in the U.S. except that it does not include race, and there are variations on the Fracture Risk Assessment Tool for dozens of countries, Dr. Jones noted. “Only three of those tools race-correct,” he said.

He also pointed out that pediatric nephrologists in the U.S. do not race-correct eGFR, so Black children have their kidney function calculated the same way as those of other races. “The day you turn 18, most hospitals in this country would give you a new eGFR,” Dr. Jones said.

That situation may change, however. In July 2020, the American Society of Nephrology and the National Kidney Foundation announced the formation of a task force, which includes Dr. Mendu, to focus on the use of race in eGFR. The task force's final report was scheduled to be released in the spring, but the leaders of both organizations released a statement in March. It said that they “agree that 1) race modifiers should not be included in equations to estimate kidney function and 2) current race-based equations should be replaced by a suitable approach that is accurate, inclusive, and standardized in every laboratory in the United States.” An interim report was published in the Journal of the American Society of Nephrology in April.

The creators of the VBAC calculator are reviewing this issue as well, experts said, and the Maternal Fetal Medicine Unit's VBAC tool website says that “a new calculator without race and ethnicity is now under development.” Legislators have also taken an interest in how race is being used in medicine. Michelle Morse, MD, MPH, a hospitalist and an author of the JGIM study, was serving on the House Ways and Means Committee throughout 2020 in her role as a Robert Wood Johnson Foundation health policy fellow in Washington, D.C.

In September 2020, the chair of the committee, Rep. Richard Neal (D-MA), sent letters to several medical societies requesting that they reexamine the use of race in clinical algorithms. In October, he asked CMS to look at racial bias in clinical decision-making tools and created a request for information from the broader medical community, and in November, he sent letters to an additional four professional societies. The societies have replied, and the committee has been reviewing the results, Dr. Morse reported.

One major point of debate is likely to be what evidence is needed before race is dropped from these algorithms. “I don't think anyone studied this carefully when they implemented race correction,” said Dr. Jones. “A much higher standard of proof is being demanded to stop race correction.”

Jasmine Marcelin, MD, FACP, agreed. “My argument would be that there shouldn't be any need for data to demonstrate that it's bogus, because race and genetic ancestry are completely different,” said Dr. Marcelin, an infectious diseases physician and assistant professor of internal medicine at the University of Nebraska in Omaha who focuses on diversity and equity.

However, even with data like the JGIM study as support, the push for change in the eGFR has encountered resistance, including on clinical grounds, Dr. Jones said. “The concern is that if you stop race-correcting African-American patients, some number of patients will no longer meet the threshold for safe metformin use” or use of other drugs with renal clearance cutoffs. He believes that risk is outweighed by the potential benefits of removing race correction.

Future directions

Advocates for eliminating race correction aren't sure whether the next steps in this project will involve additional subspecialty assessments or more overarching change.

“One avenue is having individual professional societies convene task forces that hopefully include physicians, epidemiologists, critical race theorists, public health experts to look at the use of race within each separate field and come up with new guidelines,” said Dr. Vyas. “Another interesting option is to have a broader statement of principles or guidelines for the field of medicine at large created by an overarching organization, like the American Medical Association.”

Shortly after she said that, the American Medical Association on Nov. 16, 2020, announced new policies recognizing race as a social construct, rather than an inherent biological trait, and calling for clinicians and researchers to focus on genetics and biology, the experience of racism, and social determinants of health when describing risk factors for disease.

However, the effort to remove race from algorithms is likely to encounter ongoing resistance, the experts agreed.

For one thing, it requires a significant change in thinking, said Ms. Cerdeña. “I find that a lot of my colleagues really have an embedded notion of race being meaningful, because it's what they've been taught and what they've learned. I always get questions like, ‘But what about sickle cell? What about cystic fibrosis? You can't deny that there are certain diseases that are more common in certain racial groups,’” she said. “What I would come back with is, ‘Sure, but if you're always looking for sickle cell disease in Black patients, you could miss it in a White patient or in a Middle Eastern patient or in other patients who are descendants from areas in which malaria is also highly prevalent.’”

Another challenge is that discussions about race and racism can stir emotions. “For many folks, it triggers an emotional response that leads to defensiveness,” said Dr. Morse. “If we can't get past the emotional reaction that is inherent in talking about things like racism, then we end up being stuck with the more comfortable response, which is to do nothing.”

Even physicians who are reluctant to take dramatic action should try to learn more about the issue, recommended Salman Ahmed, MD, MPH, lead author of the JGIM study and a nephrologist at Brigham and Women's. “While clinicians may want to wait for more information to change practice, they don't need to wait for any more information to have the conversation at an institutional level,” he said.

Those who are convinced that race should be taken out of algorithms can also alter practice on their own. “I decided that I'm going to not use a race-based adjustment of eGFR in my calculation for my individual patients,” said Dr. Marcelin, noting that while her hospital has not yet adopted that policy, the division of nephrology is planning to make that change.

One smaller change clinicians can make is being aware of and transparent with patients about the use of race in clinical algorithms, the experts advised. “You could know that what you're looking at is a race-corrected eGFR. And you could know that if the race correction is moving this patient back and forth across some relevant threshold, maybe you should look very carefully and make a decision that's based on more nuance than just the patient's race,” said Dr. Jones.

And there's no time like the present to start, said Dr. Mendu. “What we've heard a bit from folks is, ‘Why now? There's so much going on with COVID and there's so much going on with equity. Do we really need to address this now?’” Her answer is simple. “Anything we can do that could potentially alleviate disparities, we should be doing, especially if it's something that is as simple as stop using a variable in an equation.”