ABSTRACT
OBJECTIVES (1) To examine ‘race’ in acute and chronic kidney disease under similar conditions, and (2) to encourage further study of implied cofactors and an international approach to end misuse of ‘race’.
METHODS We re-analyzed data from United States veterans, (1) comparing outcomes after removing ‘race correction’ from legacy estimated glomerular filtration rates and (2) graphing data to explore more subtle relationships. We hypothesized factors confounding the link between kidney disease and apolipoprotein L1 gene variants, including cofactors implied and controlled by veteran status.
RESULTS Studies of veterans minimized ‘racial’ disparity in chronic kidney disease, suggesting an effect of starting conditions, including Armed Services entry requirements and equitable access to healthcare. Apolipoprotein L1 “high-risk” and N264K+ gene variants may be proxies for ‘race’, with gradients of effect due to colorism.
CONCLUSIONS Under equitable conditions, comparable kidney disease outcomes should be the expected norm for all, regardless of ‘race’, ethnicity, or nationality. Discouraging misuse of ‘race’ in medical research and healthcare is an actionable Population Health initiative with potentially high impact but low effort and cost.
1. INTRODUCTION
Unchallenged for ‘racial’ inconsistency, and selectively including or overlooking its influence, medical research can unwittingly promote ‘racial’ illusions that advance the myth of ‘race’. Hunt et al noted,…current health literature encourages clinicians to see their patients through a racial lens…. the systematic inequalities that have been observed in the clinical care that racialized patients receive (Smedley et al. 2003) likely reflect more than structural inequalities alone. They may also be attributable to this systematic authorization of race-based care. The arbitrary and contradictory ways the clinicians in this study understand and apply the concept of race raise concern that not only does using race to inform diagnostic and treatment choices lack scientific rigor, it may also put patients at risk of receiving nonstandard care (Acquaviva and Mintz 2010) [1].
The persistence of ‘race’ in US healthcare may seem unbelievable to decent, well-meaning international colleagues, and that denial creates a stubborn obstacle to ending misuse of ‘race’. Before diving into our Results and Discussion, we offer this Introduction (with US examples) to orient international readers to the illusions of ‘race’ and ‘ethnicity’ that inspire bad science (BS) in ‘racialized’ countries.
1.1 ‘Race’ and ethnicity
‘Race’ and ‘ethnicity’ are terms commonly used to describe population groups, but their meaning has changed over time and between countries, and hidden assumptions may limit understanding by international colleagues.
In “The meaning of racial or ethnic origin in EU law”, a report prepared for the European Commission, Farkas noted:In EU law, the Racial Equality Directive (RED) prohibits discrimination on the grounds of racial or ethnic origin without defining them…. In Europe, most groups targeted by racism categorically reject race as a label. Still, racialisation as ‘a process that ascribes physical and cultural differences to individuals and groups’ is thriving….
Without real physical differences, people can be racialised on the basis of their perceived religiosity: the dress they wear or the way they cover their hair. Theories of Christianity’s superiority over the Jewish and Muslim faiths can be traced back over centuries of European civilisation, and they laid the ground for anti-Semitism and Islamophobia.
Poverty, too, can be portrayed as having its own culture, language and value system, thus poverty can also be racialised – particularly if a visible proportion of poor people belong to racial minorities….
The traumatic experiences of the Holocaust and the urge to respond to apartheid provoked the scientification of race in legal terminology, which trapped racial in the confines of biological racism – the essence of the ideologies it originally set out to combat. Scientification disassociated racial from nation(al minority) and consequently severed cultural, religious and linguistic from it. Reducing the meaning of racial to ‘biological’ inevitably paved the way to essentialisation, while insulating it from the influence of social sciences….
The invention of a supercategory with the intention of ensuring terminological unity and dispelling scientific misconceptions of race and racial origin at the international level had an unintended side effect in Europe. Here, the prism of nationalism resists the smooth transfer of terminological reform that followed WWII. While in France, ethnic origin is perceived as equivalent to race, in the UK it is separately defined in case law. In the rest of the EU, correlations between ethnic and racial occupy a central place in academic discussions and judicial interpretation [2].
Because everyone is confident in the local meaning of ‘race’, agreement on ‘race’ between co-located subject and observer reflects ‘shared illusion’ rather than scientific “validity” [3,4,5,6].
In 2022, Lu et al summarized contradictions and uncertainties in ‘race’ and ethnicity [7]. They noted (1) lack of “consensus definition of race or ethnicity”, (2) inconsistency between subject and observer identification for other than Black or White ‘race’, and (3) “fluidity” of self-identification. Their aspirational recommendations called for “specific definitions of” and “justification for collecting and analyzing” ‘race’, ethnicity, or ancestry data in the context of health research.
Subsequently, in a 2023 report on “Using Population Descriptors in Genetics and Genomics Research”, the National Academies of Sciences, Engineering, and Medicine (NASEM) stated:Race is neither useful nor scientifically valid as a measure of the structure of human genetic variation….
Over the past several decades, a number of professional societies have developed statements or guidelines on race, ethnicity, human diversity, and multicultural practices…. In every case, the guidelines have been aspirational and intended to encourage their professional colleagues to become educated and informed….
Despite the many recommendations, guidelines, and strategies… there has been relatively little change in how any entities within the genetics and genomics research ecosystem use these descriptors or require them to be used [8].
1.2 Racialized US healthcare
Though often providing ‘the best healthcare in the world’, US healthcare ranked last among high-income countries [9] and 48th of 50 nations for kidney-related mortality [10], Fig 1.
Guaranteed funding for CKD and KF [11] represents 3% of federal spending, with Medicare costs doubling over a decade to $122 billion in 2020 [12], Fig 2. Fragmented US healthcare spends more per capita than any other nation [13], yet “low-income minorities with bad health had 68% less odds of being insured than high-income Whites with good health” [14]. There are numerous “racially segregated US hospital markets” [15], and at some academic centers, Black patients are more likely to be deemed “teaching cases” for training the medical students, residents, and fellows. In 2003, the National Academy of Medicine reported on “Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care” [16], but 20 years later, “There hasn’t been a lot of progress…. We are still largely seeing what some would call medical apartheid” [17].
‘Racial’ profiling, hypertension, kidney disease
In the US, chronic kidney disease (CKD) and kidney failure (KF) represent the eighth leading cause of death, with lifetime risk varying by ‘race’ and ethnicity [18,19], Fig 3— “one of the starkest examples of racial/ethnic disparities in health” [20]. Although 12% of the general population, Black people represent 35% of patients on dialysis. US kidney disease mortality rates per 100,000 across the 30 most populous US cities, between 2014 and 2018, averaged 12.4 for White and 28.2 for Black patients, but with marked geographic disparities: for White patients, from 2.0 in San Diego, California, to 18.2 in Louisville, Kentucky, and for Black patients, from 7.9 in New York City, to 45.4 in Charlotte, North Carolina [21].
Solid and striped bars show kidney failure (KF) death rates by US ‘race’ and ethnicity (nH = non-Hispanic). Clear bars are rates for other causes (combined sexes).
Eliminating ‘racial’ disparity in kidney disease for Black patients alone could redirect 20% of funds from late-stage CKD to prevention [22], Fig 2.
Data collected to monitor and remedy ‘racial’ disparities can worsen them, for example, by misuse in clinical algorithms [23] and popular point-of-care clinical decision-making tools [24,25]. Hunt and Kreiner studied the effect on primary care and noted:…that racial/ethnic group differences are so readily equated with presumed genetic differences, and that the idea of “Personalized Medicine” can so ironically be converted into carte blanche for practicing racialized medicine [26].
US treatment guidelines for hypertension may add to ‘racial’ disparities by favoring thiazide diuretics for Black patients [27,28], resulting in usage ratios of thiazides to loop diuretics of 1.06 for White and 1.65 for Black subjects (in a study that cautioned against diagnosing KF while on thiazide diuretics but did not report KF by ‘race’) [29]. A study promoting reduced cardiovascular mortality through tight blood-pressure control (to systolic blood-pressure less than 120 mm Hg) noted lack of benefit in Black subjects [30] but did not report their 100% increase in acute kidney injury (AKI) until years later [31], even though dehydration causes more AKI in low-income areas [32], increases CKD risk [33], and is prevalent among Black people [34], possibly interacting with thiazide use.
Despite ‘racialized’ treatment of Black patients and lingering ‘racial’ segregation in housing and education, US clinical studies collecting ‘race’ to monitor ‘racial’ disparities may not report outcomes by ‘race’. For example, a study recommending chlorthalidone for poorly controlled hypertension in advanced CKD noted the number of subjects suffering AKI tripled (and average AKI episodes per subject increased by 50%) but despite preferential use of thiazides in Black patients did not report AKI by ‘race’ [35].
‘Race correction’
In 1999, the Modification of Diet in Renal Disease (MDRD) equation for estimating glomerular filtration rate (GFR) introduced the first ‘race correction’ for estimated GFR (eGFR) [36]. More eGFR ‘corrections’ followed—for ethnicity in the US [37] and nationalities in Asia [38,39,40,41]. Laboratory reporting that some eGFR results were ‘race corrected’ led to questioning of the practice [42,43], and the “majority of studies found that removal of race adjustment improved bias, accuracy, and precision of eGFR equations for Black adults” [44].
Eneanya et al discussed misuse of ‘race’ in nephrology, distinguished mistreating individuals versus monitoring the effects of racism on a population, noted that “race should not be used to make any biological inferences about individuals”, but conceded, “the notion that the use of a Black race coefficient in eGFR equations contributes to health inequity for Black patients has not been universally accepted” [45].
In 2023, two years after creation of “race-free” equations, some researchers described opposition to eGFR ‘race correction’ “through the lens of clashing values…. [T]he activists had a strong moral case…. However, there was lack of acknowledgment for data showing that considering race does improve GFR estimation” [46]. Reports that the ostensibly “race-free” eGFR equations “underestimated measured GFR in Black participants… and overestimated measured GFR in non-Black participants” [47] merely shifted misuse of ‘race’ from the equations to conscious and unconscious pre-test “population weighting” [48,49], which is much harder to detect and stop.
Motivation and outcomes
Kuhn presented ‘science’ as inherently social, demonstrating how the solutions of today (or the past) become the problems of tomorrow (or today) [50]. Articles discussing socially constructed ‘race’ as a proxy for social determinants of health are often labeled ‘Commentary’ [51]. Articles touting the objectivity of numbers and promoting ‘race’ as providing “the best medical treatment for all patients” [52] miss how ‘Confirmation Bias’ influences what numbers we collect, how we understand that data, our confidence in what we are doing now, and which articles get published.
‘Racialized’ practices—ongoing despite 100 years of scholarly articles in Anthropology that ‘race’ is a myth—inspired our data re-analysis (see Results) and examination of ‘race’ as a proxy for implied cofactors (see Discussion). To end misuse of ‘race’ is an overdue Kuhnian paradigm shift [50].
“With malice toward none, with charity for all” [53], we re-examined ‘race’ in kidney disease. Removing ‘race correction’ from GFR estimates of Black and White veterans showed comparable results. Re-analyzing two studies of gene variants associated with kidney risk in Black veterans—both published before NASEM guidance [8]—gave insights into the influence of ‘race’.
2. RESULTS
2.1 Removing ‘race correction’
Removing ‘race correction’ from GFR estimates of an impressive study of KF versus eGFR in 1.70 million White veterans (4% female) and 311,000 Black veterans (6.1% female) [54] tested our hypothesis that differences between arbitrary groups would be less after comparable conditions of selection and opportunity.
Equal outcomes despite ‘race’
Fig 4 shows baseline characteristics and renal comorbidities (all significant to p<0.001). Compared to White veterans, Black veterans were more likely to be younger, female, living in lower-socioeconomic-status zip codes, hypertensive, diabetic, and in the highest and lowest eGFR ranges, Fig 4. Home zip codes suggested lower socioeconomic status for Black than White veterans, Fig 4, shown graphically in Fig 5.
Mathematically removing the MDRD eGFR ‘race correction’ for Black veterans (dividing by 1.21) shifted rates for KF risk and eGFR prevalence to lower GFRs, Fig 6, as displayed graphically in Fig 7. The x-axis ‘race corrections’ (horizontal arrows) sometimes had significant y-axis effects (vertical arrows). The revised KF outcome curves (increasing to the left) overlapped for Black and White veterans (inset). The revised eGFR-prevalence curves (increasing to the right) showed higher prevalence of lower eGFRs in Black veterans. Under similar conditions, the paradox of low prevalence of early CKD yet faster progression at late stages appeared to be a mathematical artifact of ‘race correction’.
Dividing the upper limit of eGFR range for Black patients by the 1.21 ‘race correction’ numerically removed ‘race correction’, shifting their results to lower numbers on the x axis. In a table, the result is not striking. Note: “end-stage renal disease” (ESRD) is the US equivalent of kidney failure (KF).
On these graphs, the 1.21 multiplier in the Modification of Diet in Renal Disease (MDRD) eGFR equation had shifted results for Black patients to the right (solid black curves).
Removing ‘race correction’ (i.e., dividing by 1.21) shifted the results back to the left (blue curves) with no effect on y-axis values of those data points but revealed the effective x-axis and y-axis shifts indicated by the horizontal and vertical black arrows. Curves increasing to the left show that removing ‘race correction’ shifted kidney failure (KF) rates for Black veterans to within the 95% confidence interval of the curve for White veterans (inset). Curves increasing to the right show that ‘race correction’ for Black veterans falsely lowered the cumulated percentages below each eGFR ‘limit’ (y axis), the difference between the solid blue and black curves. Light blue arrows indicate the x-axis and y-axis effects of ‘race correction’.
This visual examination was supported analytically. When estimating magnitude of associations between inverse eGFR and KF rates with and without ‘race correction’, the ratio was exactly 1.21, as expected (‘race corrected’: beta = 11700.74; correction removed, beta = 9670.03; ratio = 1.21). Additionally, as hypothesized, White and Black patients had more similar associations between serum creatinine and GFR after removing the ‘race correction’ (t (10) = 0.91, p = 0.3860) than when ‘race correction’ was maintained (t (10) = 2.36, p = 0.0402). The sample size was small—only 15 data points—because continuous raw data was broken (dichotomized) into five segments, yielding five data points each for White veterans and Black veterans with and without ‘race correction’—therefore, the precision of this result may not hold up with replication. However, it validates concerns raised about an artifact of arithmetic resulting in differences in kidney disease diagnosis. It contradicts notions that Black Americans have less early CKD and need less early care.
2.2 APOL1, AKI, and COVID
Hung et al studied the effects of apolipoprotein L1 (APOL1) risk variant (RV) alleles among 990 participants in the Department of Veterans Affairs Million Veteran Program with “African ancestry” and hospitalized with coronavirus disease 2019 (COVID-19) [55]. Fig 8 shows AKI rates with and without various cofactors, ordered by odds ratios (ORs). They found the relative risk (RR) of AKI was 1.35 times higher for APOL1 “high-risk” (HR) genotype (APOL1 HR = 2 RV copies) than “low-risk” (LR) genotype (APOL1 LR = 0 or 1 RV copy). However, RR of AKI was 2.4 for mechanical ventilation and 2.63 for vasopressor use, both more common in the APOL1 HR group.
Possible direct or indirect cofactors for AKI, ordered by odds ratios (ORs). Note the biggest differences for those indicating poor health.
ORs make the comparison more explicit, Fig 9. We also included data from a similar study of COVID-19 and AKI in the Brazilian general population [56]. Although recalculated confidence intervals are much wider for the Brazilian sample, the ORs between Brazil and the US are close for many cofactors.
Forest plot of AKI ORs (with 95% confidence intervals) suggests similarity of cofactors between US veterans and Brazilian general population despite their different genetic origins and different conceptions of ‘race’.
Many covariates showed statistically “small” effects (ORs below 1.68), some small enough to suggest random chance. Signs of critical illness showed “large” effects (OR above 6.71). Urine characteristics had ORs primarily toward the upper range, highlighting their importance. The low OR for “dipstick hematuria, year prior” [rows 49 and 61] might reflect two sub-populations: women (some menstruating) and men (hematuria always abnormal).
Bias correction
Mechanical ventilation and vasopressors showed the largest RRs, consistent with findings that AKI in veterans with COVID-19 varied with severity of illness [57]. Verbeek et al recommended correcting bias in observed RR using the “reference confounder” with the largest effect [58]. Unconfounding the APOL1 RR of AKI with vasopressors alone reduced the RR to 1.09, Fig 10.
Because vasopressors and mechanical ventilation had significant APOL1 exposures and had the largest effects on AKI outcomes, their effect could be unconfounded from the raw data for relative risk (RR) of APOL1 genotypes (high-risk versus low-risk). Although overlapping, the correlation between Mechanical Ventilation and Vasopressors was not 100 percent, so adjustment for both would move unconfounded APOL1 RR of AKI even closer to unity.
2.3 APOL1 and N264K
A second important study by Hung et al of 121,000 from the Million Veteran Program with “African ancestry” (13.8% female) compared non-diabetic CKD in patients with APOL1 HR versus LR and with and without N264K (a second-site APOL1 variant) [59]. N264K reportedly “modified the association between APOL1 HR genotypes (two copies of G1/G2) and CKD”. Some of their data came from smaller studies in the general population, but our re-analysis used only their data from veterans.
Fig 11 shows the percentage of subjects by kidney outcomes for the four combinations of APOL1 HR or LR and N264K alleles present (+) or absent or (-) [59]. Black bars, representing APOL1 high-risk genotype without N264K (APOL1 HR, N264K-), show the highest rate for every kidney outcome. Blue bars represent the largest group, APOL1 low-risk genotype without N264K (APOL1 LR, N264K-). Note that presence of the N264K allele (N264K+) in just 3.8% of APOL1 LR and 0.5% of APOL1 HR patients means green and orange bars combined represent less than 5% of subjects, with small absolute numbers (population times rate) in their CKD or outcome groups.
Ordered by odds ratio (OR) of kidney outcome between the two large groups, APOL1 genotype (low-risk versus high-risk) without N264K alleles (N264K-). The data show marked improvement in outcomes after enrollment in the Million Veteran Program. Note that combined subgroups positive for N264K allele (N264K+) represent less than 5% of all subjects.
The chart is ordered by OR between the two largest groups without N264K (N264K-), representing 95% of the veteran subjects and differing only by APOL1 HR vs LR. Note for APOL1 HR N264K- the association of lower rates of KF and CKD after than before Million Veteran Program “enrollment” [60] and that, after enrollment, rates of CKD were the same for all four groups.
3. DISCUSSION
Bioethical oversight of ‘race’ falls under a principle defined after decades of misuse of ‘race’ by the US Public Health Service [61,62], Fig 12. Yet, misuse of ‘race’ as a conscious or unconscious input for clinical care persists, even among those aware of its scientific invalidity [63,64]. The Food and Drug Administration’s approval of race-specific medicines “has not been challenged as a Fourteenth Amendment violation” [65,66,26], and voluntary corrective guidelines “have had little effect on how these concepts are deployed” [67,68]. Even the comprehensive, scholarly NASEM recommendations are voluntary [8].
In the US, ‘race’ is collected to monitor progress in addressing ‘racial’ disparities in healthcare outcomes but often misused in clinical care. Despite vulnerabilities and inequities related to ‘race’, there has been no oversight for Justice.
A study of 1,000 randomly selected Cochrane Reviews, published between 2000 and 2018, showed that only 14 (1.4%) had planned to include subgroup analysis by ‘race’ [69], and in those 14, there was only a single ‘racial’ finding of statistical significance. Despite centuries of claims of biological differences, ‘race’ was absent under the high standards of the Cochrane Reviews.
To avoid perpetuating the false notion of biological ‘race’ [70,71,72], we always put ‘race’ or ‘racial’ in quotation marks [73,74].
3.1 Removing ‘race correction’
Our re-analysis removed ‘race correction’ from eGFR values and found Black and White veterans share the same curve for KF, suggesting something associated with veteran status minimized ‘racial’ disparities, Fig 7.
Comparable conditions
Implied (but typically unstated) in studies of US veterans are the educational and physical selection standards of the US Armed Forces [75]. Long before veterans enrolled in research, servicemembers met written requirements that screen out most of the widespread illiteracy, Fig 13, and physical requirements (including a urine test) for comparable access to military training and jobs. With rare exception, servicemembers earned access to Veterans Administration (VA) Healthcare by completing basic training, entering active duty, and serving honorably.
While Farkas noted that cumulated racial or ethnic discrimination “can lead to a cycle of disadvantage which is frequently passed from one generation to the next” [2], US military communities are integrated, and Black service members fared better than Black civilians on numerous metrics [76]. Black and White veterans had similar employment rates [77] and more similar socioeconomic status, Fig 5, in contrast to the relative advantage/disadvantage of US Black and White neighborhoods [78,79], Fig 14. Actual economics of Black and White veterans might be even closer because ‘race’ can influence residency options: for example, one can (1) simultaneously hold Adichie’s admiration for “the sacrifices of African Americans…. that there’s a context” [80] and Smith’s awareness of social disapproval for not “behaving in ways coded Black” [81], while another can (2) hold “implicit bias in perceptions of crime and disorder”, which Sampson and Bean noted, may be “one of the underappreciated causes of continued racial segregation” [82].
Until confounding factors are characterized and controlled for, analyzing veterans separately from the general population may avoid “Simpson’s paradox”. Peterson et al showed higher baseline comorbidities and illness severity yet lower mortality among Black veterans in VA healthcare in all but six diseases [83] (or five, after our re-analysis). Despite a higher COVID-19 infection rate among minority women veterans, Tsai et al found equal complications of mortality, cardiovascular events, or onset of heart disease regardless of ‘race’ [84].
3.2 ‘Race’ and genetics
3.2.1 Inconsistency of ‘race’
‘Race’, ‘mixed race’, and ‘multiracial’ are all scientifically invalid inputs for healthcare [8,67]. The mechanism of action of ‘race’ and APOL1 genotypes in CKD must account for inconsistencies in folk taxonomies of ‘race’ [85,86,87,88], for example in places like South Africa [89], Nigeria [80] and Brazil:…there appears to be no racial descent rule operational in Brazil and it is possible for two siblings differing in Color to belong to completely diverse racial categories…. asked about their origins (the question admitted multiple responses) <10% of Brazilian black individuals gave Africa as one of their answers [90,91].
Classification in the US dates to a 1662 statute [92] reflected in this segregation-era Census instruction: “A person of mixed white and Negro blood should be returned as a Negro, no matter how small the percentage of Negro blood” [93,94].
Colorism and APOL1
…the most effective way to segregate black people was to “use the dark skin slaves vs. the light skin slaves and the light skin slaves vs. the dark skin slaves.”.… Subsequently, colorism became an effective strategy in colonizing and dividing people [95].
“Colorism” varied the abuse of racism to stratify and stabilize ‘racial’ hierarchies, correlating increasing pigment with worse deprivation.
The odds of inheriting two, one, or zero copies of APOL1 RV alleles vary, for example, with the number of great-grandparents of West African origin. With eight West African great-grandparents, all with APOL1 HR genotype (i.e., two RV alleles), a subject is essentially guaranteed to inherit two RV alleles. With only one high-risk great-grandparent, the subject “usually looks white or almost so” [96], but the odds of inheriting RV alleles are 75% for zero copies, 25% for one, and zero for two copies.
In contrast, N264K+ is more common in Europeans, found in less than 5% of Black patients, and “mutually exclusive with the APOL1 G1 allele…. the presence of a single copy of the APOL1 N264K mutation mitigated the increased risk conferred by HR APOL1 genotypes” [97].
With APOL1 RVs more common in West Africa and N264K+ more common in Europe, people with one or the other regional distribution marker likely have different physical appearance and lived experiences. Thus, geographically localized “ancestry markers” can be proxies for ‘race’, and colorism can be a mechanism stratifying subjects even in ostensibly ‘single-race’ studies.
3.2.2 Confusing ‘race’, ethnicity, ancestry, and geography
Commonly used terms for population groups are ethnicity in Europe and ‘race’ in the US. ‘Race’, ethnicity and ancestry are different (albeit similar and overlapping) concepts and all are subject to misuse.
NASEM recommended against conflating ‘race’ with ancestry:Conclusion 4-2. Using socially constructed groupings indiscriminately in human genetics research can be harmful. Their use reinforces the misconception that differences in social inequities or other factors are caused by innate biological differences and diverts attention from addressing the root causes of those social differences, which compromises the rigor and potential positive effect of the research. Recommendation 1. Researchers should not use race as a proxy for human genetic variation. In particular, researchers should not assign genetic ancestry group labels to individuals or sets of individuals based on their race, whether self-identified or not [8].
Hung et al were not clear on what they meant by ‘race’ but expressed confidence in ‘race’ and in their use of mathematics and big assumptions about the relationships between ‘race’ and personal ancestry [55,59]. They described selecting a subset from the Million Veteran Program databank “…with genetic information available, of African ancestry” and used the method of “genetically inferred ancestry” described by Fang et al, in 2019 [98].
NASEM subsequently noted:There is a pervasive misconception and belief that humans can be grouped into discrete innate categories…. The illusion of discontinuity between racialized groups has supported a history of typological and hierarchical thinking…. These modes of thinking often spill over to other descent-associated population descriptors such as ethnicity and ancestry (Byeon et al., 2021; Fang et al., 2019). The structure of human genetic variation, though, is the result of human population movement and mixing and so is more related to geography than to any racial or ethnic classification [8].
‘Race’ in research design
NASEM noted,
Conclusion 4-8. In the absence of measured environmental factors, researchers often wrongly attribute unexplained phenotypic variance between populations to unmeasured genetic differences. Recommendation 4. Researchers conducting human genetics studies should directly evaluate the environmental factors or exposures that are of potential relevance to their studies, rather than rely on population descriptors as proxies [8].
The data from Hung et al are valuable but needed to be explored [55,59]. The ‘single-race’ study design presupposed an association between genetic biology and ‘race’, made analytic assumptions, then unconsciously overlooked more direct causes of AKI in the data. Selective reporting on APOL1 genotypes appeared to justify the myth of ‘race’ as biology.
The directed acyclic graph (DAG) [99] in Fig 15 is a causal framework the study might have imagined, in modeling their choices, but with our modifications to show potential “back door paths” that could explain the outcomes better than a direct effect of APOL1. The grey oval represents their focus on biology with no attempt to consider possible differential effects of racism between APOL1 HR and LR genotypes. Their meta-analysis models included some demographic variables, but none related to education, employment, colorism, socioeconomic status, or any other social determinants of health (SDOH), ignoring all the context our re-analysis of veterans suggests is important. Nor did they justify choosing any variables for the analysis, which is important because adding more to a model can introduce selection bias.
One back door path from APOL1 to kidney disease is via West African ancestry (i.e., West African ancestry increases odds a person has APOL1 high-risk copies but also alters physical appearance, and physical appearance feeds discrimination). The study assumed the 10 ancestry principal components (plus other variables in other models) analytically controlled for this to justify focusing on relationships in the shaded area. First, we disagree that ancestry principal components controlled for the back door path via skin tone. Second, even if principal components addressed the confound of skin tone, that does not account for the history of discrimination that is its own confound, including generational wealth as a significant factor. Third, given their regional dependence and low frequencies, the inclusion of both N264K and APOL1 in one analysis is problematic. This approach leads to redundancy and collinearity, resulting in uninterpretable coefficients.
We found several potential back-door paths. First, darker skin tone and colorism: regional ancestry often has aspects of physical appearance, and our re-analysis suggests back door paths attributable to physical appearance and SDOH that undermine analysis focused solely on biology and genetics. Second, generational wealth: even comparing patients who can all ‘pass’ as White, Black Americans denied generational wealth had limited opportunities, which the research design did not address (i.e., the problem is more than colorism alone, which also was not addressed). Third, West African ancestry and the N264K gene: regional distribution of N264K variant, with quite low frequencies in African-American veterans, also likely correlates with physical traits, and because few subjects showed this combination of traits, we do not know whether all combinations truly exist in nature.
3.2.3 Equity and APOL1
Roberts noted, “The issue is not whether genes affect health—of course they do— but whether genetic difference explains racial disparities in health” [100].
The APOL1 mechanism that explains ‘racial’ differences in KF is unknown [101]. Higher RVs of the APOL1 gene were identified in all haplotype backgrounds, but the magnitude of their effect was dependent on expression levels. Equity is a confounding factor in all these findings.
‘Race’ affects COVID outcomes through misdiagnosed lung disease [102,103,104], mismeasured blood oxygen levels [105,106], and inequitable care [107]. The different ‘racial’ systems in Brazil and the US converge in colorism. However, because links between ‘race’, skin color, and ancestry varied under different ‘racial’ systems, a broader range of physical appearances and SDOH might weaken APOL1 associations with ‘race’ in the Brazilian population (i.e., more noise in the data) [108]. Given the wide confidence intervals (recalculated from p value and z-statistic) in the Brazilian data [56] and a general caution with accepting the null hypothesis of no difference, we cannot say the ORs are significantly different when comparing the same risk factors between the two samples. However, the remarkably similar ORs between US veterans and Brazilians for a variety of non-genetic covariates supports other aspects of personal circumstance as more important than the APOL1 gene.
Chen et al adjusted for age, sex, and baseline eGFR in a general population study and eliminated correlation between KF and APOL1 [109]. Grams et al found that ‘racial’ disparities (e.g., in income and education) explained AKI disparities better than APOL1 risk genotypes [110]. Grams et al also noted, “the majority of blacks with the high-risk genotype experience eGFR decline similar to blacks without the high-risk genotype” [111].
Focal segmental glomerulosclerosis
Our re-analyses did not conclude that APOL1 HR genotype has no effect on AKI and CKD. Hung et al showed that APOL1 HR genotype modestly increased risk of focal segmental glomerulosclerosis (FSGS) in veterans [59], which is consistent with other recent reports [112,113,114,115], including low penetrance [116]. Kallash et al summarized, “the development of kidney disease is rare, suggesting the need for a second hit” [117], which may represent cumulative effects of multiple cofactors. Veteran status appears to partially mitigate the FSGS risk for ORs within the block of modest ORs, Fig 9, small absolute percentage rates, and relatively few FSGS and KF events, Fig 11.
The risk of stereotyping dark skin as a sign of APOL1 HR genotype follows trans-Atlantic enslavement of West African people (i.e., descendants of formerly enslaved people—both still living in the US and recently immigrated from Caribbean countries). However, mass migration, globalization, and more recent US immigration fueled by military and economic conflicts in Central and East African countries continue to diversify African ancestral histories. Even if the likelihood were increased by visual examination, who and when to test for APOL1 HR genotype may hinge on understanding reduced disparity in veterans and whether nonspecific Population Health measures can mitigate accumulated “second hit(s)” from SDOH—even before any component cofactors are fully characterized.
3.3 Future directions
Commercial interests drive innovation, but there is also profit-potential in misuse of ‘race’ [118]. Research into low-cost preventive measures and SDOH usually requires government funding, so we highlight two such areas for further research: two cofactors implied by outcome differences between veterans and the general population, and an approach that might discourage misuse of ‘race’.
3.3.1 Proxy for illiteracy and deprivation
Illiteracy does not cause but is associated with CKD [119,120], implicating broad consequences of reading failure (e.g., difficulty obtaining jobs with health benefits). In contrast, literacy diminished the association between long-term illness and Black ‘race’ and removed the predictive power of “education” and being African American [121]. The rate of functional illiteracy varies by district, Fig 13, affecting up to 90% of high school graduates [122]. Adult literacy programs did not improve CKD [123]. Further research might show whether early intervention [124,125,126] or literacy testing in medical research [127,128] might prevent ‘race’ from being a proxy for illiteracy.
Deprivation is associated with CKD [129,130]. The origin of US housing segregation “has now largely been forgotten” [131,132], but patients from disadvantaged neighborhoods, Fig 14 [133,79], may suffer health effects of childhood poisonings [134,135,136] and prenatal exposures [137,138] that increase CKD risk [139,140,141]. Further research (e.g., in zip codes of the few white and black areas with similar demographics) might show whether early and ongoing exposures contribute to the “senile nephrosclerosis” found in one-third of young (but absent in 10% of elderly) kidney donors [142] and to the purported “racial” disparities in GFR and CKD.
3.3.2 International bioethics
NASEM noted:…studies that are poorly designed to answer research questions are scientifically invalid and unethical….
What is necessary is an understanding of the underlying issues during study design and long before data analyses: the moment of publication is far too late [8].
Although studies perpetuate the ‘race’ myth in ways that are “largely unintentional” [143], Hunt and Megyesi found, in 2008, that most genetics researchers (1) understood that ‘race’ is invalid, (2) used ‘race’ as a proxy for biological cofactors without bothering to explore further (simply assuming cofactors would be difficult to identify and measure), (3) could list serious potential negative consequences of misusing ‘race’, but (4) did so anyway [63]. Some advocate continued publishing of BS [144], favoring “the tranquilizing drug of gradualism” [145] and silence [146].
American bioethics favors autonomy almost to the exclusion of its other three founding principles, but as Fins noted, we should “…move beyond narrow questions of patient choice, particularly when the disenfranchised are not in a position to exercise that choice” [147]. Fins added, “…that while rights are necessary to avoid exploitation, they are not sufficient and that instead there should be a focus on capabilities that promote human flourishing” [personal comm]. In this case, it is about identifying nascent kidney disease with metrics that are accurate and accountable to scientific and ethical norms. Further research might show whether an appeal to international bioethics could end misuse of ‘race’ in medical research.
4. CONCLUSION
Selection and training of servicemembers creates comparable starting conditions for veterans. Those in VA healthcare showed comparable social determinants of health and comparable kidney outcomes, which should be the expected norm for all under equitable conditions—regardless of ‘race’, ethnicity, or nationality. Further research to identify essential veteran cofactors may improve general population health, reduce ‘racial’ disparities, and redirect some resources to prevention.
Apoprotein L1 high-risk variants increase risk of focal segmental glomerulosclerosis, but the risk appears too small to justify treatment at the population level. Broad application of nonspecific Population Health measures may further mitigate risk.
Despite scholarly consensus and 100 years of scholarly articles that ‘race’ is scientifically invalid, it persists in medical research, suggesting an hypothesis for further study: to end ‘racial’ systems that were self-sustaining for hundreds of years, further research might show whether (1) working together, (2) discouraging standing in the doorway, and (3) appealing to principles of international bioethics could end misuse of ‘race’ in research and healthcare.
5. METHODS
5.1 Ethics statement
5.1.1 ‘Race’ in research
NASEM advised, “Researchers should be as transparent as possible about the specific procedures used to name groups within their data sets” [8]. We did not present ‘race’ as an input to clinical decisions, a means of stratifying care, or an hypothesis for genetic difference without data. One primary source used ‘race’ designated in Medicare Denominator File or VA data sources [54]. We explored the approach to ‘race’ and ancestry in the other primary sources (see Discussion).
5.1.2 Human subjects
The primary sources used information collected for non-research activities in VA Healthcare or obtained informed consent from all participants after receiving IRB approval. Secondary re-analysis of publicly available data is not considered human subjects research and is exempt from IRB review [148].
5.1.3 Privacy
The study included no individually identifiable patient health information.
5.2 Re-analyses
Our three secondary re-analyses involved extracting data from tables and figures in previously published materials to reorganize the primary data into simple, accessible graphs. Our main concern was demonstrating that context can explain differences by ‘race’.
5.2.1 Re: Removing ‘race correction’
We repurposed data from the first and second tables of Choi et al [54], removing ‘race correction’ by dividing MDRD eGFRs by 1.21, and examined eGFR prevalence and rates of KF against eGFR results with and without ‘race’. Mathematically, transforming a variable by a constant has a one-to-one impact on the magnitude of the regression coefficient [149], and the estimated association will change proportional to that number, which may be useful if an actual difference could be overlooked. However, if no true ‘racial’ difference exists, the association between eGFR prevalence rates and KF rates for White and Black patients should be similar, and ‘race correction’ would inflate the association’s magnitude, implying more KF than found.
We estimated the association among White patients, Black patients with unadjusted eGFR rates, and the same Black patients with eGFR ‘race correction’ removed, hypothesizing that the ratio of associations for ‘race-corrected’ data and ‘not-race-corrected’ data would equal 1.21, the exact magnitude of ‘race correction’. Additionally, we hypothesized that there would not be a significant difference in associations between White and Black patients after removing ‘race correction’.
5.2.2 Re: APOL1, AKI, and COVID
We re-analyzed a study by Hung et al [55], calculating RR and OR from primary data in their first and second tables. We adjusted the APOL1 RR to remove the confounding of high-impact cofactors. For comparison, we included a similar study in the Brazilian general population [56], where ‘race’ is defined differently, and because some CIs were unusually asymmetric, we recalculated them from the given p values and z-statistics.
5.2.3 Re: APOL1 and N264K
We similarly re-analyzed another study by Hung et al [59] that examined associations between APOL1 LR and HR genotypes, N264K variants, and US taxonomy of ‘race’. We focused on their data from veterans—the bulk of their data—and created simple graphs to better show the relationships within the data. Because N264K+ is an APOL1 variant more common in Europe (occurring in less than 5% of US Black veterans), we further narrowed to the 95% of veterans in the two N264K- groups, calculating and ordering results by ORs between these two subgroups.
Data Availability
All data re-analyzed in the present work are contained in the manuscript or are publicly available in the original source articles.
ACKNOWLEDGMENTS
In memory of Rear Admiral W. Norman Johnson and many others who endured kidney failure after being prescribed nephrotoxic drugs that might have been avoided with early warning and caution, and of the Rev. Dr. Canon Cyril C. Burke, Sr., who taught ethics and whose final medical care was complicated by ‘race’.
The authors gratefully acknowledge Edward Feller, MD, FACP, FACG, Ruth Levy Guyer, PhD, James M. Gilchrist, Professor Emeritus of Neurology at Southern Illinois University School of Medicine, and anonymous colleagues for critical review of a draft of this article, and Joseph J. Fins, MD, MACP, FRCP, and John C. Kotelly, PhD, US Air Force mathematician (ret.), for their insights.
ABBREVIATIONS
- AKI
- acute kidney injury
- APOL1
- apolipoprotein L1
- BS
- bad science
- CKD
- chronic kidney disease
- COVID
- coronavirus disease
- eGFR
- estimated GFR
- FSGS
- focal segmental glomerulosclerosis
- GFR
- glomerular filtration rate
- HR
- high risk
- LR
- low risk
- IRB
- institutional review board
- KF
- kidney failure
- MDRD
- Modification of Diet in Renal Disease
- NASEM
- National Academies of Science Engineering and Medicine
- OR
- odds ratio
- RR
- relative risk
- RV
- risk variant
- SDOH
- social determinants of health
- VA
- Veterans Administration
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.↵
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.↵
- 118.↵
- 119.↵
- 120.↵
- 121.↵
- 122.↵
- 123.↵
- 124.↵
- 125.↵
- 126.↵
- 127.↵
- 128.↵
- 129.↵
- 130.↵
- 131.↵
- 132.↵
- 133.↵
- 134.↵
- 135.↵
- 136.↵
- 137.↵
- 138.↵
- 139.↵
- 140.↵
- 141.↵
- 142.↵
- 143.↵
- 144.↵
- 145.↵
- 146.↵
- 147.↵
- 148.↵
- 149.↵
