Similar examples can be found throughout medicine. Some algorithm developers offer no explanation as to why racial or ethnic differences exist. Others offer rationale, but when traced back to their origins, they lead to old or questionable ethnology or to biased statements.22,30,31 In the cases discussed here, the researchers followed defensible empirical reasoning. They examined data sets of clinical outcomes and patient characteristics and then performed regression analyzes to identify patient factors that were significantly associated with relevant outcomes. Because minority patients routinely have different health outcomes than white patients, race and ethnicity are often associated with outcomes of interest. The researchers then decided that it was appropriate—and even necessary—to adapt race to their model.
These decisions are the crux of the problem. When compiling descriptive statistics, it may be appropriate to record data by race and ethnicity and to study their associations. But if race appears to correlate with clinical outcomes, does that justify its inclusion in diagnostic or predictive tools? The answer should depend on how race is understood to influence the outcome.30 Coming to such an understanding is not easy: the relationships between race and health reflect overlapping social and biological pathways.32 Epidemiologists continue to debate how to responsibly make causal inferences based on race.33 Due to this complexity, it is not sufficient to translate the data signal into a race modification without specifying which race it might represent in a particular context. Most race corrections work implicitly, if not explicitly, on the assumption that genetic variation reliably tracks with race. If the empirical differences between racial groups are indeed due to genetic differences, then ethnicity modification may be justified: different coefficients for different bodies.
Such cases, however, are extremely unlikely. Studies of the genetic makeup of human populations continue to find more differences within racial groups than between them.34, 35 Moreover, racial differences found in large data sets often reflect the effects of racism—the experience of being black in America rather than being black per se—such as toxic stress and its physiological consequences.32 In such cases, race adjustment will do nothing to address the cause of the disparity. Alternatively, if the amendments prevent clinicians from providing clinical services to specific patients, they risk introducing inequality into the system.
This risk was demonstrated in 2019 when researchers exposed algorithmic bias in medical artificial intelligence.36 A widely used clinical tool that takes past health care costs into account when predicting clinical risk. Because the health care system spends more money, on average, on white patients than on black patients, the tool returned higher risk scores for white patients compared to black patients. These scores may have led to more referrals of white patients to specialized services, perpetuating both spending inconsistencies and racial bias in health care.
A second problem arises from the ways in which racial and ethnic categories are operationalized. Physicians and medical researchers typically use the categories recommended by the Office of Management and Budget: five races and two races. But these categories are not reliable agents of genetic differences and fail to understand the complexity of patients’ racial and ethnic backgrounds.34, 35 Thus, correcting the sweat forces doctors to do ridiculous reduction exercises. For example, should a doctor use a double-correction in the VBAC calculator for a pregnant person from the Dominican Republic who identifies as black and Hispanic? Should the variable GFR be adjusted for race for a patient of a white mother and a black father? The guidelines are silent on such issues – a sign of their inadequacy.
Researchers are aware of this dangerous terrain. The Society of Thoracic Surgeons acknowledged concerns raised by clinicians and policy makers “that including SES factors in risk models may ‘neutralise’ disparities in quality of care”. However, he set out to consider “all preoperative factors that independently and significantly correlate with outcomes”: “Ethnicity has an empirical association with outcomes and has the potential to confound interpretation of hospital outcomes, although we do not know the underlying mechanism (eg. Genetic factors, differential efficacy of some drugs, rates of some associated diseases such as diabetes and hypertension, and possibly [socioeconomic status] for some outcomes such as readmission). “10 This decision reflects a hypothetical assumption in medicine: It is acceptable to use race modulation even without understanding what race represents in a particular context.
To be clear, we don’t think doctors should ignore race. Doing so would blind us to the ways in which race and racism build our society.37-39 However, when clinicians incorporate race into their instruments, they risk interpreting racial disparities as hard facts rather than an injustice that requires intervention. Researchers and clinicians must distinguish between the use of race in descriptive statistics, where it plays a vital role in epidemiological analyzes, and in descriptive clinical guidelines, where it can exacerbate inequality.
This problem is not limited to medicine. The criminal justice system, for example, uses recidivism prediction tools to guide decisions about bond amounts and prison sentences. One tool, COMPAS (Correctional Offenders Management’s Alternative Punishment Management Profile), while not using race per se, uses several factors that correlate with race and returns higher risk scores for black defendants.40 The tool’s creators explained that their design simply reflects the experimental data.41 But if the underlying data reflects racist social structures, then its use in predictive tools anchors racism in practice and policy. When these tools influence high-risk decisions, whether in the clinic or the courtroom, they are spreading inequality into our future.
In 2003, Kaplan and Bennett asked researchers to be careful when they invoked race in medical research: When researchers publish a finding based on race or ethnicity, they must follow seven guidelines, including justifying their use of race and ethnicity, and describing how to assign people. For each category, carefully consider other factors—particularly socioeconomic status—that may influence the results.42 We suggest adapting these guidelines for assessment of race correction in clinical settings. When developing or applying clinical algorithms, clinicians should ask three questions: Is the need for race correction based on solid evidence and statistical analyzes (eg, taking into account internal and external validity, potential confounders, and bias)? Is there a plausible causal mechanism of racial difference that justifies a race correction? Does the implementation of this racial correction mitigate or exacerbate health inequalities?
If clinicians and clinical educators thoroughly analyze the algorithms that include race correction, they can judge, with fresh eyes, whether the use of race or ethnicity is appropriate. In many cases, this assessment will require further research into the complex interactions between ancestry, race, racism, socioeconomic status, and environment. Much of the burden of this work falls on the researchers proposing race modification and on the institutions (such as professional societies and clinical laboratories) that validate and implement clinical algorithms. But doctors can be thoughtful and thoughtful users. They can discern whether a correction is likely to alleviate or exacerbate the inequality. If the latter, clinicians should examine whether the correction is justified. Some tools, including the eGFR and VBAC calculator, have already been challenged; Physicians have successfully advocated their institutions to remove race mod.43,44 Other algorithms may be subject to similar scrutiny.45 A full account would require medical professionals to critically evaluate their instruments and review them when indicated.
Our understanding of race has advanced greatly in the past two decades. The clinical tools we use daily must reflect these new ideas to remain scientifically rigorous. Equally important is the project to make medicine a more anti-racist field.46 This includes reconsidering how clinicians initially perceived race. One step in this process is to reconsider race correction in order to ensure that our clinical practices do not perpetuate the very inequalities we aim to correct.