Hidden in Plain Sight - Reconsidering The Use of Race Correction in Clinical Algorithms

The n e w e ng l a n d j o u r na l of m e dic i n e
Medicine a nd So cie t y
Debra Malina, Ph.D., Editor
Hidden in Plain Sight — Reconsidering the Use

of Race Correction in Clinical Algorithms
Darshali A. Vyas, M.D., Leo G. Eisenstein, M.D., and David S. Jones, M.D., Ph.D.
Physicians still lack consensus on the meaning subtle insertion of race into medicine involves
of race. When the Journal took up the topic in diagnostic algorithms and practice guidelines
2003 with a debate about the role of race in that adjust or “correct” their outputs on the basis
medicine, one side argued that racial and ethnic of a patient’s race or ethnicity. Physicians use these
categories reflected underlying population ge- algorithms to individualize risk assessment and
netics and could be clinically useful.1 Others guide clinical decisions. By embedding race into
held that any small benefit was outweighed by the basic data and decisions of health care, these
potential harms that arose from the long, rotten algorithms propagate race-based medicine. Many
history of racism in medicine.2 Weighing the of these race-adjusted algorithms guide decisions
two sides, the accompanying Perspective article in ways that may direct more attention or re-
concluded that though the concept of race was sources to white patients than to members of ra-
“fraught with sensitivities and fueled by past cial and ethnic minorities.
abuses and the potential for future abuses,” To illustrate the potential dangers of such
race-based medicine still had potential: “it seems practices, we have compiled a partial list of race-
unwise to abandon the practice of recording race adjusted algorithms (Table 1). We explore several
when we have barely begun to understand the of them in detail here. Given their potential to
architecture of the human genome.”3 perpetuate or even amplify race-based health
The next year, a randomized trial showed that inequities, they merit thorough scrutiny.
a combination of hydralazine and isosorbide
dinitrate reduced mortality due to heart failure C ardiolo gy
among patients who identified themselves as
black. The Food and Drug Administration grant- The American Heart Association (AHA) Get
ed a race-specific indication for that product, with the Guidelines–Heart Failure Risk Score pre-
BiDil, in 2005.4 Even though BiDil’s ultimate dicts the risk of death in patients admitted to the
commercial failure cast doubt on race-based hospital.9 It assigns three additional points to any
medicine, it did not lay the approach to rest. patient identified as “nonblack,” thereby catego-
Prominent geneticists have repeatedly called on rizing all black patients as being at lower risk.
physicians to take race seriously,5,6 while distin- The AHA does not provide a rationale for this
guished social scientists vehemently contest these adjustment. Clinicians are advised to use this risk
calls.7,8 score to guide decisions about referral to cardi-
Our understanding of race and human genet- ology and allocation of health care resources.
ics has advanced considerably since 2003, yet Since “black” is equated with lower risk, follow-
these insights have not led to clear guidelines on ing the guidelines could direct care away from
the use of race in medicine. The result is ongoing black patients. A 2019 study found that race may
conflict between the latest insights from popula- influence decisions in heart-failure management,
tion genetics and the clinical implementation of with measurable consequences: black and Latinx
race. For example, despite mounting evidence that patients who presented to a Boston emergency
race is not a reliable proxy for genetic difference, department with heart failure were less likely
the belief that it is has become embedded, some- than white patients to be admitted to the cardi-
times insidiously, within medical practice. One ology service.24
874 n engl j med 383;9 nejm.org August 27, 2020
The New England Journal of Medicine

Downloaded from nejm.org at BAYLOR UNIVERSITY MEDICAL CENTER on May 11, 2023. For personal use only. No other uses without permission.
Copyright © 2020 Massachusetts Medical Society. All rights reserved.
Medicine and Society
Cardiac surgeons also consider race. The So- As long as uncertainty persists about the cause
ciety of Thoracic Surgeons produces elaborate of racial differences in serum creatinine levels,
calculators to estimate the risk of death and we should favor practices that may alleviate health
other complications during surgery.10 The calcu- inequities over those that may exacerbate them.
lators include race and ethnicity because of ob- Similar adjustment practices affect kidney
served differences in surgical outcomes among transplantation. The Kidney Donor Risk Index
racial and ethnic groups; the authors acknowl- (KDRI), implemented by the national Kidney Al-
edge that the mechanism underlying these dif- location System in 2014, uses donor characteris-
ferences is not known. An isolated coronary ar- tics, including race, to predict the risk that a
tery bypass in a low-risk white patient carries an kidney graft will fail.12 The race adjustment is
estimated risk of death of 0.492%. Changing the based on an empirical finding that black donors’
race to “black/African American” increases the kidneys perform worse than nonblack donors’
risk by nearly 20%, to 0.586%. Changing to any kidneys, regardless of the recipient’s race.28 The
other race or ethnicity does not increase the es- developers of the KDRI do not provide possible
timated risk of death as compared with a white explanations for this difference.12 If the potential
patient, but it does change the risk of renal fail- donor is identified as black, the KDRI returns a
ure, stroke, or prolonged ventilation. When used higher risk of graft failure, marking the candi-
preoperatively to assess risk, these calculations date as a less suitable donor. Meanwhile, black
could steer minority patients, deemed to be at patients in the United States still have longer
higher risk, away from surgery. wait times for kidney transplants than nonblack
patients.29 Since black patients are more likely to
receive kidneys from black donors, anything that
Nephr olo gy
reduces the likelihood of donation from black
Since it is cumbersome to measure kidney func- people could contribute to the wait-time dispar-
tion directly, researchers have developed equations ity.29 Use of the KDRI may do just that. Mindful
that determine the estimated glomerular filtra- of this limitation of the KDRI, some observers
tion rate (eGFR) from an accessible measure, the have proposed replacing “the vagaries associated
serum creatinine level. These algorithms result with inclusion of a variable termed ‘race’” with a
in higher reported eGFR values (which suggest more specific, ancestry-associated risk factor,
better kidney function) for anyone identified as such as APOL1 genotype.28
black.11,25 The algorithm developers justified these
outcomes with evidence of higher average serum O b s te tric s
creatinine concentrations among black people
than among white people. Explanations that have The Vaginal Birth after Cesarean (VBAC) algo-
been given for this finding include the notion that rithm predicts the risk posed by a trial of labor
black people release more creatinine into their for someone who has previously undergone ce-
blood at baseline, in part because they are re- sarean section. It predicts a lower likelihood of
portedly more muscular.11,25 Analyses have cast success for anyone identified as African American
doubt on this claim,26 but the “race-corrected” or Hispanic.13 The study used to produce the al-
eGFR remains the standard. Proponents of the gorithm found that other variables, such as mari-
equations have acknowledged that race adjust- tal status and insurance type, also correlated
ment “is problematic because race is a social with VBAC success.14 Those variables, however,
rather than a biological construct” but warn that were not incorporated into the algorithm. The
ending race adjustment of eGFR might lead to health benefits of successful vaginal deliveries
overdiagnosis and overtreatment of black pa- are well known, including lower rates of surgical
tients.27 Conversely, race adjustments that yield complications, faster recovery time, and fewer
higher estimates of kidney function in black complications during subsequent pregnancies.
patients might delay their referral for specialist Nonwhite U.S. women continue to have higher
care or transplantation and lead to worse out- rates of cesarean section than white U.S. women.
comes, while black people already have higher Use of a calculator that lowers the estimate of
rates of end-stage kidney disease and death due VBAC success for people of color could exacerbate
to kidney failure than the overall population.25 these disparities. This dynamic is particularly
n engl j med 383;9 nejm.org August 27, 2020 875

Table 1. Examples of Race Correction in Clinical Medicine.*
876
Tool and Clinical Utility Input Variables Use of Race Equity Concern
Cardiology
The American Heart Association’s Get with the Systolic blood pressure Adds 3 points to the risk score if the patient The original study envisioned using this score
Guidelines–Heart Failure9 (https://www Blood urea nitrogen is identified as nonblack. This addition to “increase the use of recommended
.mdcalc.com/gwtg-heart-failure-risk-score) Sodium increases the estimated probability of medical therapy in high-risk patients and
Age death (higher scores predict higher reduce resource utilization in those at low
Predicts in-hospital mortality in patients with Heart rate mortality). risk.”9 The race correction regards black
acute heart failure. Clinicians are advised to use History of COPD patients as lower risk and may raise the
this risk stratification to guide decisions regarding Race: black or nonblack threshold for using clinical resources for
initiating medical therapy. black patients.
Cardiac surgery
The Society of Thoracic Surgeons Short Term Operation type The risk score for operative mortality and When used preoperatively to assess a patient’s
Risk Calculator10 (http://riskcalc.sts.org/ Age and sex major complications increases (in some risk, these calculations could steer minority
stswebriskcalc/calculate) Race: black/African American, Asian, cases, by 20%) if a patient is identified patients, deemed higher risk, away from
American Indian/Alaskan Native, as black. Identification as another non- these procedures.
The
Calculates a patient’s risks of complications Native Hawaiian/Pacific Islander, or white race or ethnicity does not increase
and death with the most common cardiac sur “Hispanic, Latino or Spanish ethnic- the risk score for death, but it does
geries. Considers >60 variables, some of which ity”; white race is the default setting. change the risk score for major compli-
are listed here. BMI cations such as renal failure, stroke, and
prolonged ventilation.
Nephrology
Estimated glomerular filtration rate (eGFR) Serum creatinine The MDRD equation reports a higher eGFR Both equations report higher eGFR values
n engl j med 383;9

MDRD and CKD-EPI equations11 (https:// Age and sex (by a factor of 1.210) if the patient is (given the same creatinine measurement)
ukidney.com/nephrology-resources/egfr Race: black vs. white or other identified as black. This adjustment is for patients identified as black, suggesting
-calculator) similar in magnitude to the correction better kidney function. These higher eGFR
for sex (0.742 if female). values may delay referral to specialist care
nejm.org
Estimates glomerular filtration rate on the basis The CKD-EPI equation (which included a or listing for kidney transplantation.
of a measurement of serum creatinine. larger number of black patients in the
n e w e ng l a n d j o u r na l

study population), proposes a more
of
modest race correction (by a factor

of 1.159) if the patient is identified as
black. This correction is larger than the
correction for sex (1.018 if female).
August 27, 2020

Organ Procurement and Transplantation Network: Age Increases the predicted risk of kidney graft Use of this tool may reduce the pool of African-

m e dic i n e
Kidney Donor Risk Index (KDRI)12 (https:// Hypertension, diabetes failure if the potential donor is identified American kidney donors in the United
optn.transplant.hrsa.gov/resources/allocation Serum creatinine level as African American (coefficient, 0.179), States. Since African-American patients are
-calculators/kdpi-calculator/) Cause of death (e.g., cerebrovascular a risk adjustment intermediate between more likely to receive kidneys from African-
accident) those for hypertension (0.126) and American donors, by reducing the pool of
Estimates predicted risk of donor kidney graft Donation after cardiac death diabetes (0.130) and that for elevated available kidneys, the KDRI could exacer-
failure, which is used to predict viability of poten- Hepatitis C creatinine (0.209–0.220). bate this racial inequity in access to kidneys
tial kidney donor.† Height and weight for transplantation.
HLA matching
Cold ischemia
En bloc transplantation
Double kidney transplantation
Race: African American
Obstetrics
Vaginal Birth after Cesarean (VBAC) Risk Age The African-American and Hispanic correc- The VBAC score predicts a lower chance of
Calculator13,14 (https://mfmunetwork.bsc.gwu BMI tion factors subtract from the estimated success if the person is identified as black
.edu/PublicBSC/MFMU/VGBirthCalc/vagbirth Prior vaginal delivery success rate for any person identified or Hispanic. These lower estimates may
.html) Prior VBAC as black or Hispanic. The decrement dissuade clinicians from offering trials of
Recurring indication for cesarean for black (0.671) or Hispanic (0.680) is labor to people of color.
Estimates the probability of successful vaginal section almost as large as the benefit from prior
birth after prior cesarean section. Clinicians can African-American race vaginal delivery (0.888) or prior VBAC
use this estimate to counsel people who have to Hispanic ethnicity (1.003).
decide whether to attempt a trial of labor rather
than undergo a repeat cesarean section.
Urology
STONE Score15,16 Sex Produces a score on a 13-point scale, with By systematically reporting lower risk for black
Acute onset of pain a higher score indicating a higher risk of patients than for all nonblack patients, this
Predicts the risk of a ureteral stone in patients Race: black or nonblack a ureteral stone; 3 points are added for calculator may steer clinicians away from
who present with flank pain Nausea or vomiting nonblack race. This adjustment is the aggressive evaluations of black patients.
n engl j med 383;9

Hematuria same magnitude as for hematuria.
Urinary tract infection (UTI) calculator17 (https:// Age <12 months Assigns a lower likelihood of UTI if the child By systematically reporting lower risk for black
uticalc.pitt.edu/) Maximum temperature >39°C is black (i.e., reports a roughly 2.5-times children than for all nonblack children, this
Race: Describes self as black (fully or increased risk in patients who do not calculator may deter clinicians from pursu-
nejm.org
Estimates the risk of UTI in children 2–23 mo partially) describe themselves as black). ing definitive diagnostic testing for black
of age to guide decisions about when to pursue Female or uncircumcised male children presenting with symptoms of UTI.
urine testing for definitive diagnosis Other fever source
Oncology
Rectal Cancer Survival Calculator18 (http:// Age and sex White patients are assigned a regression The calculator predicts that black patients will
August 27, 2020

www3.mdanderson.org/app/medcalc/index Race: white, black, other coefficient of 1, with higher coefficients have shorter cancer-specific survival from
.cfm?pagename=rectumcancer) Grade (depending on stage) assigned to black rectal cancer than white patients. Clinicians
Stage patients (1.18–1.72). might be more or less likely to offer inter-
Estimates conditional survival 1–5 yr after diag- Surgical history ventions to patients with lower predicted
nosis with rectal cancer survival rates.

National Cancer Institute Breast Cancer Risk Current age, age at menarche, and age The calculator returns lower risk estimates Though the model is intended to help concep-
Assessment Tool (https://bcrisktool.cancer at first live birth for women who are African American, tualize risk and guide screening decisions,
.gov/calculator.html) First-degree relatives with breast cancer Hispanic/Latina, or Asian American it may inappropriately discourage more ag-
Prior benign biopsies, atypical biopsies (e.g., Chinese). gressive screening among some groups of
Estimates 5-yr and lifetime risk of developing Race/ethnicity: white, African American, nonwhite women.
breast cancer, for women without prior history Hispanic/Latina, Asian American,
of breast cancer, DCIS, or LCIS. American Indian/Alaska Native,
unknown
877
Table 1. (Continued.)
878
Breast Cancer Surveillance Consortium Risk Age The coefficients rank the race/ethnicity Returns lower risk estimates for all nonwhite
Calculator19 (https://tools.bcsc-scc.org/ Race/ethnicity: white, black, Asian, categories in the following descending race/ethnicity categories, potentially reduc-
BC5yearRisk/calculator.htm) Native American, other/multiple order of risk: white, American Indian, ing the likelihood of close surveillance in
races, unknown black, Hispanic, Asian. these patients.
Estimates 5- and 10-yr risk of developing breast BIRADS breast density score
cancer in women with no previous diagnosis of First-degree relative with breast cancer
breast cancer, DCIS, prior breast augmentation, Pathology results from prior biopsies
or prior mastectomy
Endocrinology
Osteoporosis Risk SCORE (Simple Calculated Rheumatoid arthritis Assigns 5 additional points (maximum By systematically lowering the estimated risk
Osteoporosis Risk Estimation)20 (https://www History of fracture score of 50, indicating highest risk) if of osteoporosis in black patients, SCORE
.mdapp.co/osteoporosis-risk-score-calculator Age the patient is identified as nonblack may discourage clinicians from pursuing
-316/) Estrogen use further evaluation (e.g., DXA scan) in black
Weight patients, potentially delaying diagnosis and
Determines whether a woman is at low, moder- Race: black or not black intervention.
The
ate, or high risk for low bone density in order to

guide decisions about screening with DXA scan
Fracture Risk Assessment Tool (FRAX)21 (https:// Age and sex The U.S. calculator returns a lower fracture The calculator reports 10-yr risk of major osteo-
www.sheffield.ac.uk/FRAX/tool.aspx) Weight and height risk if a female patient is identified as porotic fracture for black women as less
Previous fracture black (by a factor of 0.43), Asian (0.50), than half that for white women with iden
Estimates 10-yr risk of a hip fracture or other Parent who had a hip fracture or Hispanic (0.53). Estimates are not tical risk factors. For Asian and Hispanic
major osteoporotic fracture on the basis of Current smoking provided for Native American patients women, risk is estimated at about half that
n engl j med 383;9

patient demographics and risk-factor profile. Glucocorticoid use or for multiracial patients. for white women. This lower risk reported
Calculators are country-specific.‡ Rheumatoid arthritis for nonwhite women may delay intervention
Secondary osteoporosis with osteoporosis therapy.
Alcohol use, ≥3 drinks per day
nejm.org
Femoral neck bone mineral density
n e w e ng l a n d j o u r na l
Pulmonology

of
Pulmonary-function tests22 Age and sex In the U.S., spirometers use correction Inaccurate estimates of lung function may
Height factors for persons labeled as black result in the misclassification of disease
Uses spirometry to measure lung volume and Race/ethnicity (10–15%) or Asian (4–6%). severity and impairment for racial/ethnic
August 27, 2020

the rate of flow through airways in order to minorities (e.g., in asthma and COPD).23
diagnose and monitor pulmonary disease

m e dic i n e
* BIRADS denotes Breast Imaging Reporting and Data System, BMI body-mass index (the weight in kilograms divided by the square of the height in meters), CKD-EPI Chronic Kidney
Disease Epidemiology Collaboration, COPD chronic obstructive pulmonary disease, DCIS ductal carcinoma in situ, DXA dual-energy x-ray absorptiometry, LCIS lobular carcinoma in
situ, and MDRD Modification of Diet in Renal Disease study.
† The current calculator uses Ethnicity/Race, with the following options: American Indian or Alaska Native, Asian, Black or African American, Hispanic/Latino, Native Hawaiian or Other
Pacific Islander, White, and Multiracial.
‡ Three countries’ calculators are further subcategorized by race, ethnicity, or location: China (Mainland China, Hong Kong), Singapore (Chinese, Malay, Indian), and the United States
(Caucasian, black, Hispanic, Asian).
troubling because black people already have high- that justify its inclusion in diagnostic or predic-
er rates of maternal mortality.30 tive tools? The answer should depend on how
race is understood to affect the outcome.30 Arriv-
ing at such an understanding is not a simple
Ur olo gy
matter: relationships between race and health
The STONE score predicts the likelihood of kid- reflect enmeshed social and biologic pathways.32
ney stones in patients who present to the emer- Epidemiologists continue to debate how to re-
gency department with flank pain. The “origin/ sponsibly make causal inferences based on race.33
race” factor adds 3 points (of a possible 13) for Given this complexity, it is insufficient to trans-
a patient identified as “nonblack.”15 By assigning late a data signal into a race adjustment without
a lower score to black patients, the STONE algo- determining what race might represent in the
rithm may steer clinicians away from thorough particular context. Most race corrections implic-
evaluation for kidney stones in black patients. itly, if not explicitly, operate on the assumption
The developers of the algorithm did not suggest that genetic difference tracks reliably with race.
why black patients would be less likely to have a If the empirical differences seen between racial
kidney stone. An effort to externally validate the groups were actually due to genetic differences,
STONE score determined that the origin/race then race adjustment might be justified: differ-
variable was not actually predictive of the risk of ent coefficients for different bodies.
kidney stones.16 In a parallel development, a new Such situations, however, are exceedingly un-
model for predicting urinary tract infection likely. Studies of the genetic structure of human
(UTI) in children similarly assigns lower risk to populations continue to find more variation within
children identified as “fully or partially black.”17 racial groups than between them.34,35 Moreover,
This tool echoes UTI testing guidelines released the racial differences found in large data sets
by the American Academy of Pediatrics in 2011 most likely often reflect effects of racism — that
that were recently criticized for categorizing is, the experience of being black in America
black children as low risk.31 rather than being black itself — such as toxic
stress and its physiological consequences.32 In
such cases, race adjustment would do nothing to
A sse ssment
address the cause of the disparity. Instead, if
Similar examples can be found throughout medi- adjustments deter clinicians from offering clini-
cine. Some algorithm developers offer no expla- cal services to certain patients, they risk baking
nation of why racial or ethnic differences might inequity into the system.
exist. Others offer rationales, but when these are This risk was demonstrated in 2019 when
traced to their origins, they lead to outdated, researchers revealed algorithmic bias in medical
suspect racial science or to biased data.22,30,31 In artificial intelligence.36 A widely used clinical
the cases discussed here, researchers followed a tool took past health care costs into consider-
defensible empirical logic. They examined data ation in predicting clinical risk. Since the health
sets of clinical outcomes and patient character- care system has spent more money, on average,
istics and then performed regression analyses to on white patients than on black patients, the
identify which patient factors correlated signifi- tool returned higher risk scores for white pa-
cantly with the relevant outcomes. Since minor- tients than for black patients. These scores may
ity patients routinely have different health out- well have led to more referrals for white patients
comes from white patients, race and ethnicity to specialty services, perpetuating both spend-
often correlated with the outcome of interest. ing discrepancies and race bias in health care.
Researchers then decided that it was appropriate A second problem arises from the ways in
— even essential — to adjust for race in their which racial and ethnic categories are operation-
model. alized. Clinicians and medical researchers typi-
These decisions are the crux of the problem. cally use the categories recommended by the
When compiling descriptive statistics, it may be Office of Management and Budget: five races and
appropriate to record data by race and ethnicity two ethnicities. But these categories are unreli-
and to study their associations. But if race does able proxies for genetic differences and fail to
appear to correlate with clinical outcomes, does capture the complexity of patients’ racial and

The n e w e ng l a n d j o u r na l of m e dic i n e
ethnic backgrounds.34,35 Race correction therefore But if the underlying data reflect racist social
forces clinicians into absurdly reductionistic ex- structures, then their use in predictive tools ce-
ercises. For example, should a physician use a ments racism into practice and policy. When these
double correction in the VBAC calculator for a tools influence high-stakes decisions, whether in
pregnant person from the Dominican Republic the clinic or the courtroom, they propagate in-
who identifies as black and Hispanic? Should equity into our future.
eGFR be race-adjusted for a patient with a white In 2003, Kaplan and Bennet asked research-
mother and a black father? Guidelines are silent ers to exercise caution when they invoked race in
on such issues — an indication of their inade- medical research: whenever researchers publish
quacy. a finding based on race or ethnicity, they should
Researchers are aware of this dangerous ter- follow seven guidelines, including justifying their
rain. The Society of Thoracic Surgeons acknowl- use of race and ethnicity, describing how subjects
edged concerns raised by clinicians and policy- were assigned to each category, and carefully
makers “that inclusion of SES factors in risk considering other factors — especially socioeco-
models may ‘adjust away’ disparities in quality of nomic status — that might affect the results.42
care.” Nonetheless, it proceeded to consider “all We propose an adaptation of these guidelines to
preoperative factors that are independently and evaluate race correction in clinical settings. When
significantly associated with outcomes”: “Race developing or applying clinical algorithms, physi-
has an empiric association with outcomes and cians should ask three questions: Is the need for
has the potential to confound the interpretation race correction based on robust evidence and sta-
of a hospital’s outcomes, although we do not tistical analyses (e.g., with consideration of inter-
know the underlying mechanism (e.g., genetic nal and external validity, potential confounders,
factors, differential effectiveness of certain med- and bias)? Is there a plausible causal mechanism
ications, rates of certain associated diseases for the racial difference that justifies the race
such as diabetes and hypertension, and poten- correction? And would implementing this race
tially [socioeconomic status] for some outcomes correction relieve or exacerbate health inequities?
such as readmission).”10 This decision reflects a If doctors and clinical educators rigorously
default assumption in medicine: it is acceptable analyze algorithms that include race correction,
to use race adjustment even without understand- they can judge, with fresh eyes, whether the use
ing what race represents in a given context. of race or ethnicity is appropriate. In many cases,
To be clear, we do not believe that physicians this appraisal will require further research into
should ignore race. Doing so would blind us to the complex interactions among ancestry, race,
the ways in which race and racism structure our racism, socioeconomic status, and environment.
society.37-39 However, when clinicians insert race Much of the burden of this work falls on the
into their tools, they risk interpreting racial dis- researchers who propose race adjustment and on
parities as immutable facts rather than as injus- the institutions (e.g., professional societies, clin-
tices that require intervention. Researchers and ical laboratories) that endorse and implement
clinicians must distinguish between the use of clinical algorithms. But clinicians can be thought-
race in descriptive statistics, where it plays a vi- ful and deliberate users. They can discern wheth-
tal role in epidemiologic analyses, and in pre- er the correction is likely to relieve or exacerbate
scriptive clinical guidelines, where it can exacer- inequities. If the latter, then clinicians should
bate inequities. examine whether the correction is warranted.
This problem is not unique to medicine. The Some tools, including eGFR and the VBAC cal-
criminal justice system, for instance, uses recid- culator, have already been challenged; clinicians
ivism-prediction tools to guide decisions about have advocated successfully for their institutions
bond amounts and prison sentences. One tool, to remove the adjustment for race.43,44 Other al-
COMPAS (Correctional Offender Management Pro- gorithms may succumb to similar scrutiny.45 A
filing for Alternative Sanctions), while not using full reckoning will require medical specialties to
race per se, uses many factors that correlate with critically appraise their tools and revise them
race and returns higher risk scores for black when indicated.
defendants.40 The tool’s creators explained that Our understanding of race has advanced con-
their design simply reflected empirical data.41 siderably in the past two decades. The clinical

tools we use daily should reflect these new in- 16. Wang RC, Rodriguez RM, Moghadassi M, et al. External
validation of the STONE score, a clinical prediction rule for ure-
sights to remain scientifically rigorous. Equally teral stone: an observational multi-institutional study. Ann
important is the project of making medicine a Emerg Med 2016;67(4):423.e2-432.e2.
more antiracist field.46 This involves revisiting 17. Shaikh N, Hoberman A, Hum SW, et al. Development and
validation of a calculator for estimating the probability of uri-
how clinicians conceptualize race to begin with. nary tract infection in young febrile children. JAMA Pediatr
One step in this process is reconsidering race 2018;172:550-6.
correction in order to ensure that our clinical 18. Bowles TL, Hu C-Y, You NY, Skibber JM, Rodriguez-Bigas
MA, Chang GJ. An individualized conditional survival calculator
practices do not perpetuate the very inequities for patients with rectal cancer. Dis Colon Rectum 2013;56:551-9.
we aim to repair. 19. Tice JA, Miglioretti DL, Li C-S, Vachon CM, Gard CC, Ker-
Disclosure forms provided by the authors are available at likowske K. Breast density and benign breast disease: risk as-
NEJM.org. sessment to identify women at high risk of breast cancer. J Clin
Oncol 2015;33:3137-43.
From the Department of Medicine, Massachusetts General 20. Lydick E, Cook K, Turpin J, Melton M, Stine R, Byrnes C.
Hospital (D.A.V.), and the Department of Global Health and Development and validation of a simple questionnaire to facili-
Social Medicine, Harvard Medical School (D.S.J.) — both in tate identification of women likely to have low bone density. Am
Boston, the Department of the History of Science, Harvard J Manag Care 1998;4:37-48.
University, Cambridge, MA (D.S.J.), and the Department of 21. Kanis JA. Assessment of osteoporosis at the primary health
Medicine, NYU Langone Medical Center, New York (L.G.E.). care level. WHO Scientific Group technical report. Sheffield,
United Kingdom:World Health Organization Collaborating
This article was published on June 17, 2020, at NEJM.org. Centre for Metabolic Bone Diseases, University of Sheffield,
2007.
1. Burchard EG, Ziv E, Coyle N, et al. The importance of race 22. Braun L. Breathing race into the machine: The surprising
and ethnic background in biomedical research and clinical prac- career of the spirometer from planation to genetics. Minneapo-
tice. N Engl J Med 2003;348:1170-5. lis:University of Minnesota Press, 2014.
2. Cooper RS, Kaufman JS, Ward R. Race and genomics. 23. Kumar R, Seibold MA, Aldrich MC, et al. Genetic ancestry in
N Engl J Med 2003;348:1166-70. lung-function predictions. N Engl J Med 2010;363:321-30.
3. Phimister EG. Medicine and the racial divide. N Engl J Med 24. Eberly LA, Richterman A, Beckett AG, et al. Identification of
2003;348:1081-2. racial inequities in access to specialized inpatient heart failure
4. Jones DS, Dorr GM, Pollock A. Symposium: race, pharma- care at an academic medical center. Circ Heart Fail 2019;12(11):
ceuticals, and medical technology. J Law Med Ethics 2008;36: e006214.
439-545. 25. Eneanya ND, Yang W, Reese PP. Reconsidering the conse-
5. Leroi AM. A family tree in every gene. New York Times. quences of using race to estimate kidney function. JAMA 2019;
March 14, 2005:A21. 322:113-4.
6. Reich D. How genetics is changing our understanding of 26. Hsu J, Johansen KL, Hsu C-Y, Kaysen GA, Chertow GM.
‘race.’ New York Times. March 23, 2018:SR1. Higher serum creatinine concentrations in black patients with
7. Social Science Research Council. Is race “real”? June 2006 chronic kidney disease: beyond nutritional status and body com-
(http://raceandgenomics.ssrc.org). position. Clin J Am Soc Nephrol 2008;3:992-7.
8. Kahn J, Nelson A, Graves JL Jr, et al. How not to talk about 27. Levey AS, Tighiouart H, Titan SM, Inker LA. Estimation of
race and genetics. Buzzfeed News. March 30, 2018 (https://www glomerular filtration rate with vs without including patient race.
.buzzfeednews.com/article/bfopinion/race-genetics-david JAMA Intern Med 2020;180:793-5.
-reich). 28. Julian BA, Gaston RS, Brown WM, et al. Effect of replacing
9. Peterson PN, Rumsfeld JS, Liang L, et al. A validated risk race with apolipoprotein L1 genotype in calculation of Kidney
score for in-hospital mortality in patients with heart failure Donor Risk Index. Am J Transplant 2017;17:1540-8.
from the American Heart Association Get with the Guidelines 29. Cannon RM, Brock GN, Marvin MR, Slakey DP, Buell JF. The
program. Circ Cardiovasc Qual Outcomes 2010;3:25-32. contribution of donor quality to differential graft survival in
10. Shahian DM, Jacobs JP, Badhwar V, et al. The Society of Tho- African American and Caucasian renal transplant recipients. Am
racic Surgeons 2018 adult cardiac surgery risk models. 1. Back- J Transplant 2012;12:1776-83.
ground, design considerations, and model development. Ann 30. Vyas DA, Jones DS, Meadows AR, Diouf K, Nour NM,
Thorac Surg 2018;105:1411-8. Schantz-Dunn J. Challenging the use of race in the vaginal birth
11. Levey AS, Stevens LA, Schmid CH, et al. A new equation to after cesarean section calculator. Womens Health Issues 2019;
estimate glomerular filtration rate. Ann Intern Med 2009;150: 29:201-4.
604-12. 31. Kowalsky RH, Rondini AC, Platt SL. The case for removing
12. Rao PS, Schaubel DE, Guidinger MK, et al. A comprehensive race from the American Academy of Pediatrics clinical practice
risk quantification score for deceased donor kidneys: the kidney guideline for urinary tract infection in infants and young chil-
donor risk index. Transplantation 2009;88:231-6. dren with fever. JAMA Pediatr 2020;174:229-30.
13. Grobman WA, Lai Y, Landon MB, et al. Development of a 32. Krieger N. Methods for the scientific study of discrimina-
nomogram for prediction of vaginal birth after cesarean deliv- tion and health: an ecosocial approach. Am J Public Health
ery. Obstet Gynecol 2007;109:806-12. 2012;102:936-44.
14. Landon MB, Leindecker S, Spong CY, et al. The MFMU Ce- 33. VanderWeele TJ, Robinson WR. On the causal interpretation
sarean Registry: factors affecting the success of trial of labor of race in regressions adjusting for confounding and mediating
after previous cesarean delivery. Am J Obstet Gynecol 2005;193: variables. Epidemiology 2014;25:473-84.
1016-23. 34. Maglo KN, Mersha TB, Martin LJ. Population genomics and
15. Moore CL, Bomann S, Daniels B, et al. Derivation and valida- the statistical values of race: an interdisciplinary perspective on
tion of a clinical prediction rule for uncomplicated ureteral the biological classification of human populations and implica-
stone — the STONE score: retrospective and prospective obser- tions for clinical genetic epidemiological research. Front Genet
vational cohort studies. BMJ 2014;348:g2191. 2016;7:22.

35. American Association of Physical Anthropologists. AAPA 42. Kaplan JB, Bennett T. Use of race and ethnicity in biomedical
statement on race & racism. March 27, 2019 (https://physanth publication. JAMA 2003;289:2709-16.
.org/about/position-statements/aapa-statement-race-and-racism 43. Morris JE, Grubbs V, Hahn M, Richmond S. Abolish race-
-2019/). based medicine in kidney disease and beyond. San Francisco
36. Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissect- Examiner. November 27, 2019 (https://www .sfexaminer.com/
ing racial bias in an algorithm used to manage the health of opinion/abolish-race-based-medicine-in-kidney-disease-and
populations. Science 2019;366:447-53. -beyond/).
37. Roberts D. Fatal invention: how science, politics, and big 44. UW Medicine to exclude race from calculation of eGFR
business re-create race in the twenty-first century. New York: (measure of kidney function). University of Washington, Depart-
New Press, 2011. ment of Medicine, May 29, 2020 (https://medicine.uw.edu/news/
38. Yudell M, Roberts D, DeSalle R, Tishkoff S. Taking race out uw-medicine-exclude-race-calculation-egfr-measure-kidney
of human genetics. Science 2016;351:564-5. -f unction).
39. Benjamin R. Race after technology: abolitionist tools for the 45. Tsai J. What role should race play in medicine? Scientific
new Jim code. Medford, MA:Polity, 2019. American. September 12, 2018 (https://blogs.scientificamerican
40. Angwin J, Larson J, Mattu S, Kirchner L. Machine bias. Pro- .com/voices/what-role-should-race-play-in-medicine/).
Publica. May 23, 2016 (https://www.propublica.org/article/ma- 46. Jones CP. Toward the science and practice of anti-racism:
chine-bias-risk-assessments-in-criminal-sentencing). launching a national campaign against racism. Ethn Dis 2018;
41. Dieterich W, Mendoza C, Brennan T. COMPAS risk scales: 28:231-4.
demonstrating accuracy equity and predictive parity. North-
pointe. July 8, 2016 (http://go.volarisgroup.com/rs/430-MBX-989/ DOI: 10.1056/NEJMms2004740
images/ProPublica_Commentary_Final_070616.pdf). Copyright © 2020 Massachusetts Medical Society.
my nejm in the journal online

Individual subscribers can store articles and searches using a feature
on the Journal’s website (NEJM.org) called “My Account.”
Each article and search result links to this feature. Users can create
personal folders and move articles into them for convenient retrieval later.


Hidden in Plain Sight - Reconsidering The Use of Race Correction in Clinical Algorithms

Uploaded by

Copyright:

Available Formats

Hidden in Plain Sight - Reconsidering The Use of Race Correction in Clinical Algorithms

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hidden in Plain Sight - Reconsidering The Use of Race Correction in Clinical Algorithms

Uploaded by

Copyright:

Available Formats

The n e w e ng l a n d j o u r na l of m e dic i n e

Debra Malina, Ph.D., Editor

Hidden in Plain Sight — Reconsidering the Use

874 n engl j med 383;9 nejm.org August 27, 2020

The New England Journal of Medicine

n engl j med 383;9 nejm.org August 27, 2020 875

n engl j med 383;9

The New England Journal of Medicine

modest race correction (by a factor

August 27, 2020

Copyright © 2020 Massachusetts Medical Society. All rights reserved.

n engl j med 383;9

August 27, 2020

The New England Journal of Medicine

Copyright © 2020 Massachusetts Medical Society. All rights reserved.

ate, or high risk for low bone density in order to

n engl j med 383;9

The New England Journal of Medicine

August 27, 2020

Copyright © 2020 Massachusetts Medical Society. All rights reserved.

n engl j med 383;9 nejm.org August 27, 2020 879

880 n engl j med 383;9 nejm.org August 27, 2020

The New England Journal of Medicine

n engl j med 383;9 nejm.org August 27, 2020 881

my nejm in the journal online

882 n engl j med 383;9 nejm.org August 27, 2020

The New England Journal of Medicine

You might also like