2

History of Race-Based Clinical Algorithms and Their Impact on Achieving Health Equity

Aletha Maybank, chief health equity officer of the American Medical Association (AMA), opened the workshop and explained that the AMA in 2020 called for ridding the nation’s health care system of racial essentialism. This policy statement recognized race as a social construct; called for supporting the elimination of race as a proxy for ancestry, genetics, and biology in medicine, medical education, research, and clinical practice; and moving to race-conscious rather than race-based medicine.

Dorothy Roberts, the George A. Weiss university professor of law and sociology at the University of Pennsylvania, summarized the historical and ideological origins of race-based clinical algorithms and how they became embedded in medicine. She said understanding the context of the roots of race correction¹ and race-based algorithms requires examining the concept of race and acknowledging it as a socially invented division of humans with no biological backing. Roberts further explained, this invention of race occurred at the onset of European conquests and the enslavement of African people facilitated the pretense that humans are divided into races and belief that some people were innately enslavable from the moment of conception and throughout their entire lives. At the same time, dominant Christian theology justified enslaving African people, conquering and disposing of Indigenous people, and dominating

___________________

¹ Race correction or race adjustment in medicine refers to the use of race to influence how an individual is treated.

them based on the notion that God divided humans by race and created White people in God’s image (Roberts, 2011), Roberts said.

In the eighteenth and nineteenth centuries, scientific arguments also supported and justified the notion of race as a biological construct, with many notable scientists of the day arguing that nature created the distinctions between races, thus excusing racial oppression and White supremacy as natural. Scientists even used supposed anatomical differences across races to rank the relative value of each race, with White people at the top of this ranking and Black people at the bottom, Roberts explained. Again, this invented hierarchy was used to justify social inequality among the races. A prominent physician named Samuel Cartwright promoted a race-based concept of disease: the idea that people from different races have different diseases and experience common diseases differently (Cartwright, 1851; Roberts, 2011; Willoughby, 2018). The racial concept of disease, Roberts said, is still what most U.S. medical schools teach their students.

Cartwright also promoted the idea in 1851 that slavery was justified because Black people have an innately lower lung capacity than White people and therefore had to be forced to work by White people to be healthy. This idea, that enslaving humans of a particular race was actually good for their health, was a powerful justification for enslaving Black people (Cartwright, 1851; Roberts, 2011). Cartwright used the spirometer, which measures lung capacity, to support his claims. Race, said Roberts, still plays a role in interpreting spirometry readings today, when measurements are “corrected” to account for the presumed lower lung capacity among Black patients. A 2020 paper listed many ways race correction still plays a role in clinical medicine across multiple fields, including cardiology, nephrology, obstetrics, and oncology (Vyas et al., 2020), which Roberts said shows how embedded and widespread the biological race concept that began to justify slavery continues to influence clinical algorithms today.

A more contemporary example of racial bias in medical technologies emerged during the COVID-19 pandemic when researchers found that pulse oximetry was less likely to detect low blood oxygen levels among Black patients than among White patients. As a result, Black patients experienced increased risk of receiving inadequate supplemental oxygen (Chesley et al., 2022; Sjoding et al., 2020). Roberts said the reason was not that pulse oximeters were inherently racially biased but that pulse oximeters were calibrated on people with lighter skin. “It is bias based on skin color, which cuts across racial divisions,” said Roberts. “It is a problem of using Whiteness as a standard and excluding people with a diversity of skin colors.” Roberts said the question is not whether racist intent underpins the use of technology but whether a false understanding of race is embedded within it (Roberts, 2021).

Recently, a National Academies of Sciences, Engineering, and Medicine consensus report concluded that race is not useful or scientifically valid as a measure of human genetic variation and recommended that researchers should not use race as a proxy for it (NASEM, 2023). Roberts framed the issue: “Race is not a biological category that naturally produces health disparities because of genetic differences or other innate biological differences. Race is a political category that does have staggering biological consequences because of the impact of social inequality, including structural racism, on people’s health.” Recent research on the effect of race correction in algorithms used within race-based diagnostic tools has identified multiple harms, especially to Black patients. Roberts said race correction results from failing to understand the meaning of race and its connection to past and present racism, and “the persistence of race correction shows an unwillingness from some in medicine to change.”

RACE IN PREDICTION MODELS: RECONSIDERING RACE CORRECTION IN CLINICAL ALGORITHMS

Darshali Vyas, resident physician in internal medicine and instructor in medicine at Massachusetts General Hospital, explained the origins of her interest in this topic: her medical school experience. Although social science and genetics faculty stated that race is a social construct and not a reliable proxy for genetic difference, she saw a clear contradiction to this message in the clinic, where she saw myriad ways providers accounted for race by using diagnostic algorithms with outputs adjusted according to their patient’s race or ethnicity (Vyas et al., 2020). Vyas said, “As physicians, we use these algorithms daily to risk assess our patients and guide our clinical decisions, but by embedding race into the basic data in health care decisions, these algorithms propagate race-based medicine.”

One result of this practice, Vyas noted, is that race-based algorithms may exacerbate existing inequities by influencing decisions in ways that direct more attention or resources to White patients than to patients from racial and ethnic minoritized populations. For example, the race-based algorithm physicians use to approximate kidney function, and to determine how to manage kidney disease, produces both African American and non-African American values for a patient’s estimated glomerular filtration rate (eGFR), a measure of how well the kidneys remove toxins from the blood. The race-based eGFR calculation produces higher estimates of kidney function in Black individuals, which can result in delays in the treatment of kidney disease (e.g., referrals for specialist care planning for dialysis and transplantation). One study found that removing race-based eGFR adjustments would result in reassigning one-third of

Black patients to a more advanced stage of chronic kidney disease, and an additional 3 percent of Black patients would meet the criterion for accumulating kidney transplant priority (Ahmed et al., 2021).

Vyas noted she and her fellow medical students advocated successfully in 2017 for the Beth Israel Deaconess Medical Center laboratory to discontinue race-based reporting of eGFR. Since then, the National Kidney Foundation and the American Society of Nephrology Task Force recommended removing race-based corrections for eGFR calculations and released a race-neutral equation (Delgado et al., 2021, 2022; Inker et al., 2021). Vyas said, “Despite having a national recommendation from our professional societies, most labs across the country still use race in calculating eGFR.” She added that changing this situation requires consistent advocacy by physicians and medical students.

Another commonly used race-based algorithm produces recommendations for ascertaining which women can have a vaginal birth after a prior cesarean section (VBAC). This tool, Vyas explained, includes an adjustment factor that predicts a lower likelihood of successful VBAC for women identified as African American or Hispanic. The adjustment results in poorer birth outcomes for African American women than for White women (Vyas et al., 2019). Since African American women already experience maternal mortality rates more than three times higher than White women, this algorithm could exacerbate existing inequity. Vyas noted that other variables that also predict VBAC success (e.g., marital status and insurance type) identified by investigators in the study used to produce this algorithm were not incorporated in the final algorithm. However, a validated race-neutral VBAC algorithm replaced the race-based calculator in 2021 (Grobman et al., 2021).

Vyas also said many algorithm developers fail to provide explanations for why racial or ethnic differences might exist, and others offer rationales based on outdated and suspect anatomical and physiological data. While racial and ethnic differences in health outcomes exist, they reflect complex, interrelated social and biological pathways. Vyas argued that translating a data signal into a race adjustment without determining what race represents in a specific context is insufficient (Vyas et al., 2020). She and her collaborators have developed a framework for reevaluating race correction in algorithms and addressing one algorithm at a time (Vyas et al., 2020).

CONCEPTUALIZATIONS OF RACE AND INEQUITY IN MEDICAL EDUCATION, RESEARCH, AND PRACTICE

Jennifer Tsai, an emergency department physician in New Haven, CT, said, “The ways we imagine and use definitions of race are linked to our

understanding of racial disparities, health equity, and the tools we have to create solutions to those very big problems.” She noted the National Institutes of Health Revitalization Act of 1993 mandated that all clinical research include data on racial minorities and women. However, while this legislation requires research to include racial categories, few scientists and physicians have studied the complexities of racial politics in the United States and are ill-equipped to understand racial formation theories despite being major players in shaping the public’s understanding of race and its definitions, Tsai said.

Framing race as a biological construct is harmful, Tsai explained. Although medical discourse and research often treat racial groups as discrete, immutable, and genetically homogeneous, genetic variation is actually continuous and overlapping among populations. Racial categories, she noted, are dynamic, fluid, transient, and socially and politically constructed. This reality is clearly reflected in the U.S. Census; since 1790, its racial-category definitions have changed in each subsequent census. Moreover, studies have demonstrated that genetic differences are statistically higher within racial groups than between racial groups (Lewontin, 1972; Yu et al., 2002) and that physicians inaccurately classify the race of more than half their patients (Braun et al., 2007; Witzig and Dery, 2014).

However, Tsai noted that race cannot be entirely discounted. Although it is not written into DNA, race significantly affects people and their health, she said. That impact, though, originates in social inequities caused by a legacy of discriminatory institutions. Indeed, rates of heart disease, cancer, stroke, diabetes, disease severity, and mortality are higher among Black people than White people in the United States, and 100,000 additional Black men, women, and children die annually as a result. (Betancourt and King, 2003).

Tsai added that the pervasive framing of race as a meaningful biological variable deeply affects the ways medicine operationalizes biological assumptions about race. In a review of preclinical lectures for medical students, Tsai and her colleagues found that 96 percent of included slides framed race as explicitly or implicitly biological (Tsai et al., 2016). In addition, Tsai also discussed race and spirometry. She noted race correction within spirometry changes the threshold for disease diagnosis and results in lower eligibility rates for disability support among Black Americans. Data from the National Institute of Occupational Safety and Health showed only 81 percent of Black workers qualify for disability compensation based on race-adjusted spirometry data, but 94 percent qualify without race adjustment (McClure et al., 2020). The implications are stark. For a biracial individual, classification as Black or White related to race-adjusted algorithms can mean the difference between an eGFR level that qualifies them for the kidney transplant waitlist—and one that does not.

Similarly, Tsai highlighted, the National Football League’s use of race-based algorithms led to fewer Black former professional football players receiving compensation for neurological deficits stemming from concussions they experienced during their careers. Such algorithms are built on assumptions that Black former players initially possess lower cognitive functioning than White former players. Therefore, Tasi explained, when a Black former player and a White former player receive the exact same raw scores on a battery of tests designed to assess current cognitive functioning, the Black former player is presumed to have suffered less impairment and is less likely to qualify for compensation (Canada and Carter, 2021).

Tsai reasserted that notions of racial essentialism are not historical artifacts. A study in 2016 found that a significant number of medical students and physicians still believe in fundamental biological differences between Black and White people; for instance, they believe Black people have less sensitive nerve endings, thicker skins, and less efficient respiratory systems (Hoffman et al., 2016). “These beliefs are baked into medicine,” Tsai said. “They are pervasive and persist today.”

Notably, Tsai emphasized, clinical medicine has struggled to define race. This is unsurprising, she said, since race is not a biological construct. One review of genetics research studies published between 2001 and 2004 found that none of the articles stated how they defined race or ethnicity (Sankar et al., 2007). Tsai suggested this finding shows how normalized and unquestioned biologically-based notions of racial difference have become within medicine. “In what other contexts can a scientific paper fail to define a major research variable and still be published?” she asked.

Tsai provided several additional examples of how biological notions of race continue to factor into clinical decision making and research, which leads to poor science that leading journals publish. For example, Tsai noted published studies on genetic susceptibility for death associated with COVID-19 identified racial differences in biology as a major factor. She acknowledged that race plays a role but not a biological one: Black individuals are disproportionately affected by certain social determinants of health (e.g., higher levels of underlying health conditions, lower rates of insurance coverage, greater disease exposure due to disproportionate representation among frontline workers, and greater air-pollution exposure because of increased residential proximity to highways and toxic landfills).²

Roberts, Vyas, and Tsai agreed that reforming medical education is necessary to rid medicine of race-based algorithms and race-based clinical decisions. “We need to be deliberately antiracist in medical educa-

___________________

² https://www.sciencenews.org/article/coronavirus-why-african-americans-vulnerable-covid-19-health-race (accessed September 29, 2023).

tion, and it cannot just be a lecture at the beginning. This needs to be infused throughout the medical curriculum,” said Roberts. Roberts also called for requiring the people perpetuating race-based algorithms and other aspects of race-based medicine to confront both their historical roots and the existing rigorous, empirical, peer-reviewed research that demonstrates the harm of such practices—especially to Black patients.

This page intentionally left blank.