Use of Race and Ethnicity in Biomedical Publication

In a 1992 JAMA commentary, Osborne and Feit challenged journal editors to "do no harm" in publishing studies of racial differences.¹ Invoking the Hippocratic oath, they urged authors to write about race in a way that does not perpetuate racism.

Some researchers and policy analysts have argued eloquently against the use of racial/ethnic categories in biomedicine, suggesting that classifying people by race/ethnicity tacitly reinforces the racial/ethnic division of society.^{2 - 6} Others have presented compelling arguments for the continued importance of documenting health disparities.^{7 - 19} Advocates for underserved communities have stressed the need for more and better data, especially data on subpopulations that are typically aggregated into larger population groups.^{14 - 16 ,20}

The current focus on documenting health disparities makes this an opportune time for careful consideration of how and why information on race/ethnicity is collected and presented. For the discussion of disparities to be precise and meaningful, equal attention must be given to the way in which race and ethnicity are conceptualized and described and the rationale for reporting racial/ethnic differences. Journal editors have an important role to play in maintaining the quality of research and policy literature and can help establish standards for the use of race/ethnicity that reflect thoughtful attention to these important issues.^{21 - 29} Most biomedical journals have de facto guidelines for writing about race/ethnicity, which are often imbedded in copyediting standards. However, so far only a few journals have published explicit policy statements or guidelines, including the British Medical Journal,^{30 - 32} Paediatric and Perinatal Epidemiology,^{33 - 34} Nature Genetics,³⁵ and the Archives of Pediatrics & Adolescent Medicine.^{17 - 18 ,36 - 39}

In this article, we suggest some guidelines for the use of race/ethnicity in biomedical publication. To provide the context for these guidelines, we first outline 3 challenges faced by researchers, clinicians, and policy makers in writing about race/ethnicity.

The complexity of individual identity, the lack of clear-cut boundaries between categories used to capture socially defined constructs of race and ethnicity, and the lack of consistency across studies and data sets in the selection and definition of categories make it difficult to write about race/ethnicity with precision. The first major challenge in writing about race/ethnicity is to avoid implying that every individual has a fixed, unchangeable racial/ethnic identity that can be easily determined or that these identities sort themselves into mutually exclusive categories to which individuals can be validly and reliably assigned.

Identity Is Not Always Fixed or Easily Determined. Racial/ethnic self-identification can be complex and multilayered. While many individuals identify with a single racial/ethnic category, others identify themselves as biracial or multiracial, affirming their connection to more than 1 group. Self-identification can evolve over time^{40 - 44} as an individual's self-image changes and/or as categories take on new social or political meanings for specific populations or the larger society.⁴²

Concepts and Categories Are Inherently Imprecise. Racial/ethnic categories are at best approximations of societally defined groupings to which individuals are assigned based largely on skin color,⁷ country of origin or ancestry, and language or dialect spoken. The racial/ethnic categories commonly used in biomedical sciences and epidemiology are broad and overlapping. Individuals do not fit neatly into these categories, and these broad groupings can obscure significant within-group heterogeneity.^{9 ,27 ,40 ,42 ,45 - 53}

The terms "race" and "ethnicity" themselves are problematic. The word "race" suggests a biological basis for socially constructed categories and implies genetic homogeneity within broadly defined, heterogeneous population groups.^{54 - 57} "Ethnicity" means different things in different contexts. While in common usage, the term "ethnicity" refers to membership in a wide range of groups defined by culture, heritage, or national origin, ethnicity is often conceptualized more narrowly for research or surveillance purposes. Federal guidelines, for example, define ethnicity solely as "Hispanic or Latino" and "not Hispanic or Latino."⁵⁸ Some observers have recommended using the term "ethnicity" to encompass what are usually thought of as race and ethnicity,^{43 ,56 ,59 - 62} while others disagree with this recommendation.^{63 - 66}

Neither the validity nor the reliability of racial/ethnic assignment can be assumed.^{43 ,67 - 79} Individuals are frequently assigned to racial/ethnic categories by researchers, interviewers, hospital admitting clerks, clinicians, undertakers, and medical examiners.^{43 ,76 ,79} These assignments are made according to the assumptions and biases of the observer, based on factors such as appearance, surname, language spoken, and neighborhood of residence. Boehmer et al⁷⁷ found high rates of discordance between race/ethnicity as self-perceived and as recorded in clinical records for a large sample of Veterans Administration dental outpatients. Among 12 444 survey respondents who selected a single race/ethnicity, 70.5% of those who selected American Indian or Alaska Native were inconsistently classified in the clinical database, as were 14.1% of those who chose Spanish, Hispanic, or Latino, 13.8% of those who chose Asian, 5.0% of those who chose black or African American, and 1.5% of those who chose white.⁷⁸

Boehmer et al⁷⁷ used self-perception as the criterion standard for identification. Self-report is often suggested as the optimal method for collecting racial/ethnic data.^{40 ,45 ,58 ,80 - 81} However, questions have been raised about the validity and reliability of self-reported data.^{43 ,82} Self-report may not fully capture the effects of discrimination, which is more likely to be based on observers' perceptions than on self-perception. Williams suggests that the optimal way to assess race/ethnicity depends on the purpose for which data are being collected (written communication, David R. Williams, PhD, MPH, Institute for Social Research, University of Michigan, Ann Arbor, August 30, 2002). In contexts in which the focus is on the assessment of discrimination, observer report is likely to be more important than self-report (written communication, David R. Williams, PhD, MPH).

Choice of Categories Varies Across Studies and Data Systems. In writing about race/ethnicity, authors need to be careful not to imply that terms such as "black," "Hispanic," or "white" have fixed or uniform definitions. Although the ways in which race and ethnicity are conceptualized and categorized can have important programmatic and policy implications,^{7 ,13 - 16 ,20 ,44 ,60 ,83 - 85} there is no consensus among researchers on how categories should be defined or how individuals should be assigned to them.^{7 ,86} Categories vary from study to study and from data set to data set,^{19 ,23 ,53 ,82 ,87} and the criteria by which individuals are assigned to categories are typically implied rather than explicitly stated.^{22 - 23}

When racial/ethnic identification is self-reported, individuals may not identify with the response choices offered to them,⁷⁹ and the response choices offered in one context may be different from those used in another, making it difficult to compare data across studies or data systems.

Research findings, such as incidence and prevalence rates, may be difficult to calculate or compare when different classification systems are used to generate numerator and denominator data.⁸² The US Census Bureau is the most common source of denominator data, while numerator data can be drawn from many other sources. Race and ethnicity are often coded very differently in state vital statistics data, medical records data, or other data systems from the way they are coded in the US Census data.⁵³

The Office of Management and Budget (OMB) sets standards for racial/ethnic classification in federal data, including US Census data.^{80 - 81 ,84 ,88} In 1977, the OMB specified a minimum set of categories for race/ethnicity data that included 4 race categories (American Indian or Alaskan Native, Asian or Pacific Islander, black, and white) and 2 ethnicity categories (Hispanic origin and not of Hispanic origin). In doing so, the OMB acknowledged the lack of scientific basis for these categories.^{59 ,80 - 81 ,84 ,88}

The OMB announced revisions to these standards in October 1997 following a lengthy review process.⁵⁸ The revised standards require federal agencies to allow individuals to select 1 or more races whenever self-identification is allowed, although the OMB rejected the suggestion to include a single multiracial category. Five minimum race categories are now specified: American Indian or Alaska Native, Asian, black or African American, Native Hawaiian or Other Pacific Islander, and white.⁵⁸ The new standards use Hispanic or Latino and not Hispanic or Latino to define ethnicity, representing a change in wording but not a broadening of concept.

The limitations of the federal categories are apparent. First, the specific categories used, which to some extent reflect political compromises,^{45 ,88} have changed over time, making it difficult to measure trends.⁸⁹ Second, like other attempts to codify the social constructions of race and ethnicity, this classification system fails to capture the full range of ways in which people identify themselves.⁷⁹ Finally, the decision not to allow a biracial or multiracial identity eliminates an important way in which people self-identify, while the decision to allow people to choose multiple races creates a range of statistical and reporting problems.^{44 ,88}

This is not to say that standardizing the collection of race/ethnicity data would address all of the limitations of these data. As Gimenez points out, "The problem facing social scientists and public health specialists in trying to make sense of the [racial/ethnic] data collected by federal, state, and other agencies is a problem not only of comparability but of meaning."⁴⁷ Any standardized set of categories would at best approximate societally defined race/ethnicity, with each category representing a heterogeneous population. Investigators would have to be careful not to imply that individuals fit neatly into these categories or that each category captures a homogeneous grouping.

The distinction between risk factor and risk marker, as articulated by Osborne and Feit,¹ highlights the complex interplay of socially defined race/ethnicity, racism, discrimination, socioeconomic status (SES), social class, and genetics.

Membership in a given racial/ethnic group may be a risk marker for a particular medical condition if the incidence or prevalence of the condition is higher in that group than in other racial/ethnic groups. However, membership in that group is not necessarily a risk factor for the condition. The likelihood of developing the condition may vary considerably among members of the group, and those who are actually at risk may share relevant characteristics with people in other racial/ethnic groups.

In exploring health-related differences between population groups, it is important to determine how race/ethnicity contributes to these differences. Although race and ethnicity as commonly conceptualized in this society have powerful consequences for individuals and communities, their relationship to health is not fully understood. A statistically significant association between race or ethnicity and an undesirable health outcome does not by itself establish causality.^{29 ,90 - 91} Health outcomes usually have multiple causes that can be either direct or indirect and are often interrelated and interactive.⁹ Race/ethnicity and social class influence health through complex pathways.^{8 - 9 ,48 ,92 - 100} Racism and discrimination are important factors. Racial/ethnic health disparities may reflect differences in access to health care and other resources, in the quality of health care,¹⁰¹ or in exposure to occupational or environmental hazards, among other factors.

To write about race/ethnicity so that it does not stigmatize and does not imply a we/they dichotomy between health professionals and populations of color is challenging. Language both reflects and shapes belief and understanding. Authors need to be thoughtful and deliberate in writing about race/ethnicity to avoid reinforcing stereotypes about racial/ethnic groups or assumptions about differences between groups.

This is not simply a matter of choosing between terms such as "African American" and "black," "Hispanic" and "Latina(o)," "Native American" and "American Indian," although these choices are not trivial. Authors must also take care not to convey unintended messages by implication; for example, that all people of color are poor.

Health professionals need to find a way to write about race/ethnicity that does not imply that each individual has a fixed racial/ethnic identity; does not imply a biological basis for the categories used; allows for comparisons across studies and data sets; does not confuse correlation or association with causation; is not stigmatizing; and does not imply a we/they dichotomy between health professionals and populations of color. The goal must be to find the least imprecise way to approximate socially defined concepts of race and ethnicity for descriptive and statistical purposes.

Considerable attention has been devoted to these issues in the biomedical literature. A consensus has emerged, at least within a subset of the literature, that the reason for using race/ethnicity should be specified when race/ethnicity data are presented,^{1 ,21 - 22 ,25 ,32 ,34 ,36 ,40 - 41 ,43 ,59 ,102 - 104} that categories should be described and justified,^{11 ,21 - 23 ,31 ,34 - 35 ,40 ,86 ,103 ,105 - 107} and that all relevant variables, including social class and SES, should be considered in data analyses.^{5 ,8 - 9 ,11 ,21 - 22 ,29 ,31 ,34 ,40 - 41 ,46 ,48 ,51 ,83 ,102 ,107 - 111} For example, participants in a 1993 workshop sponsored by the Centers for Disease Control and Prevention and the Agency for Toxic Substances and Disease Registry concluded that

In all reports and other uses of surveillance data, the reasons for analyzing race and/or ethnicity should be given, approaches to measurement of race and ethnicity should be specified, and findings should be interpreted.⁴⁰

These 3 recommendations form the conceptual framework for our suggested guidelines.

1. When race/ethnicity is used as a study variable, the reason for its use should be specified. Hundreds of biomedical journals have signed the Uniform Requirements for Manuscripts Submitted to Biomedical Journals.¹⁰⁵ In May 2000, language was added to the Uniform Requirements offering the following guidance to authors:

Because the relevance of such variables as . . . ethnicity to the object of research is not always clear, authors should explicitly justify them when they are included in a study report. The guiding principle should be clarity about how and why a study was done in a particular way.¹⁰⁵

Osborne and Feit observed that categorization by race/ethnicity can convey the "misleading implication" that race/ethnicity is a "more important determinant of disease" than health behaviors, SES, or social class.¹ Using race/ethnicity routinely as a descriptor may create an impression or reinforce a belief that health disparities are directly due to race/ethnicity and that specific causal mechanisms need not be hypothesized for demonstrated relationships between race/ethnicity and particular health outcomes.^{8 ,29}

In reporting racial/ethnic differences, authors face the danger of reinforcing these differences. When researchers use race/ethnicity as a variable without explanation, they imply that race/ethnicity is "a primary, natural, and neutral means of grouping humans."²⁵ Using race/ethnicity without specifying a reason for doing so can also reinforce perceptions that the categories used are mutually exclusive and exhaustive and that individuals can be validly and reliably assigned to categories. As noted by Hahn and Stroup, "scientific use of a social category may be interpreted as endorsement of its validity."⁵⁹

2. In citing race/ethnicity data from any source, authors should describe the way in which individuals were assigned to racial/ethnic categories.^{11 ,21 - 23 ,25 ,31 - 32 ,34 - 35 ,41 ,43 ,59 ,102 - 103 ,105 ,107 ,112} If racial/ethnic identification was self-reported, authors should specify whether individuals answered an open-ended question or chose from a fixed set of categories.^{22 ,106 ,113}

In describing their own or others' data, authors should be as precise as possible in specifying the way categories are defined and the way individuals are assigned to them. For example, instead of writing "Snow et al found a higher prevalence of diabetes among African American adults than among white adults," it would be more informative to write "In a sample of adults ages 18 to 64 years, Snow et al found a higher prevalence of diabetes among individuals identified as African American in their medical records than among individuals identified as white."

When citing an article in which archaic terms such as "Caucasian" are used, authors can make clear that the terminology is not theirs by putting these terms in quotes and specifying their source.

3. Race/ethnicity should not be used as a proxy for genetic variation.^{1 ,29 ,31 ,50 ,54 - 55 ,114 - 115} Statements about genetic differences should be supported by evidence from gene studies.^{35 ,58 ,116 - 117} Genetic hypotheses should be firmly grounded in existing evidence, clearly stated, and rigorously tested.

DNA evidence reveals that more genetic variation exists within socially defined racial groups than between groups.^{54 - 57} In its 1998 Statement on "Race," the American Anthropological Association concluded that "no matter how racial groups are defined, two people from the same racial group are about as different from each other as two people from any two different racial groups."⁵⁵ Given the heterogeneity within broadly defined population groups, there is no justification for concluding, in the absence of genetic evidence, that differences between populations are, or are likely to be, genetic in nature. Observed differences may be due to a wide range of influences, including the effects of poverty and racism.

Whether there is conclusive evidence of medically significant genetic differences between broad population groups is a hotly debated issue that is outside the scope of the present article. It is important to note, however, that genetic researchers are not immune to the challenges outlined above. Apparent genetic differences are difficult to interpret when the underlying populations have mixed geographic ancestries, as do the Asian, black, Native American, and white populations in the United States. To understand the implications of a given study, readers need to know how individuals in the study sample were assigned to racial/ethnic categories and how those categories were defined.

4. In stating hypotheses and describing study results, authors should distinguish between race/ethnicity as a risk factor and race/ethnicity as a risk marker.^{1 ,40 ,117} If a researcher finds a higher prevalence of a given medical condition among individuals identified as white than among individuals assigned to other racial/ethnic groups, it should not be assumed, in the absence of evidence, that white people have some quality or characteristic that makes them more susceptible to the particular condition. Describing the racial/ethnic variations found in a study sample as due to inherent, underlying differences between racial/ethnic populations is circular reasoning.^{29 ,112}

5. In the interpretation of racial/ethnic differences, all conceptually relevant factors should be considered, including racism and discrimination,^{9 - 10 ,20 ,48 ,52 ,93 - 94 ,107 ,118} SES,^{20 ,40 ,47 - 48 ,93 - 94 ,100 ,109 - 111 ,119 - 120} social class,^{11 ,47 ,121 - 123} personal or family wealth,^{100 ,121 ,123} environmental exposures, insurance status,¹⁰⁸ age,⁴⁷ diet and nutrition, health beliefs and practices,^{9 ,48} educational level,^{40 ,48 ,51 ,119 ,123} language spoken, religion,³¹ tribal affiliation,¹⁴ country of birth,^{9 ,16 ,46 - 47 ,51 ,102} parents' country of birth, length of time in the country of residence,^{47 ,51} and place of residence.^{5 ,48 ,95 ,99 ,109}

6. Because lack of adjustment for SES or social class is the most important potential source of bias in studies of racial/ethnic differences,^{109 - 116} researchers should make every effort to adjust for conceptually relevant measures of SES or social class when comparing racial/ethnic groups.^{31 ,42 ,48 ,51 ,110 - 111 ,119} Unadjusted findings should be clearly labeled as such, and in general they should be reported in conjunction with adjusted findings for comparison purposes.

Demonstrating a crude association between race/ethnicity and an adverse health outcome, for example, documenting a difference between racial/ethnic groups in the incidence or prevalence of a disease, can be important in assessing service needs or generating research hypotheses but such an association cannot always fully explain causation. Many apparent racial/ethnic health disparities are more meaningfully understood as SES or class disparities; thus, Krieger and Fee¹²³ described adjustment for SES or social class as "a necessary first step" in exploring racial/ethnic disparities. Race/ethnicity can have effects on health that cut across SES or class categories or that interact with SES or class.⁴² However, these effects, often attributable to racism or discrimination, cannot usually be fully demonstrated without adjustment for relevant SES or social class indicators.¹¹¹

The appropriate study design depends on the research question and the causal pathways under investigation in a particular study. It is important not to ignore the existence of SES or class diversity within racial/ethnic groups. Race/ethnicity is not an appropriate proxy for SES or social class.^{20 ,109} Ungrounded, and often unconscious, assumptions are made that all poor people are people of color and that all people of color are examples of the ecological fallacy, that is, assigning the attributes of a group to each individual member of the group.

Study designs should take into account the evidence that SES or social class can change over the lifecourse,^{42 ,120} that different SES or class measures can yield different results,^{111 ,119} and that SES and class indicators may not be comparable across population groups.^{42 ,48 ,51 ,100 ,109 ,120 - 121} At a given income or education level, for example, racial/ethnic groups may differ in terms of wealth,^{100 ,121 ,123} buying power,¹²⁰ living conditions,¹²⁰ or access to resources. Braveman et al¹¹¹ recommend the use of at least 2 SES measures per study.

7. In describing racial/ethnic groups, authors should use terminology that is not stigmatizing, does not reflect unscientific classification systems, and does not imply that race/ethnicity is an inherent, immutable attribute of an individual. Because race and ethnicity are socially constructed, the language used to describe them is both politically charged and constantly evolving. Journal editors should take into account current community preferences in developing and/or periodically updating guidelines for authors and copy editors on terminology related to race/ethnicity. Guidelines might specify unacceptable, acceptable, and preferred terminology, and the journal's preferred spelling and hyphenation of acceptable terms. Thoughtful discussions of specific terms have been published.^{27 ,46 - 47 ,56 ,86 ,89 ,124 - 128}

At a minimum, journals should follow the guidance of the Council of Science Editors in prohibiting use of the term "Caucasian," which, like "Caucasoid," "Mongoloid," and "Negroid," is "based on an outmoded theory of racial distinction," and requiring use of the term "Asian" instead of "Oriental" or "Asiatic."¹⁰² Such terms may be overly broad and ambiguous in many contexts. For example, as Bhopal and others point out,^{124 - 125} "Asian" is an overly broad and ambiguous term, often used in the United Kingdom to refer to individuals from the Indian subcontinent, and in the United States, it is used to refer to individuals with origins in East or Southeast Asia.¹²⁴ "South Asian" is somewhat more precise, but it still includes a diverse population. Residents of the Indian subcontinent and their descendants tend not to self-identify as either Asian or South Asian but rather in more specific geographic, linguistic, and religious terms.^{124 - 125} Similar ambiguity exists for terms such as "Hispanic."

The American Medical Association's (AMA) Manual of Style calls for distinguishing between Asian and Asian American and, more specifically, between Chinese and Chinese American.¹⁰³ The AMA manual also specifies that, when possible, a more descriptive term should be substituted for Hispanic or Latino, such as Mexican American, Cuban American, or Puerto Rican.^{89 ,103}

Journal editors may also want to discourage the use of depersonalizing plural nouns such as "Asians," "blacks," "Hispanics," or "whites." Terms such as "Asian Americans" or "Asian/Pacific Islander Americans" are less stigmatizing. Authors can be encouraged to use even more precise and descriptive terms such as "Cambodian-born US residents," "enrolled members of the Rosebud Sioux Tribe," or "adults in the study sample who self-identified as white." The AMA style manual also suggests that authors should not use terms such as nonwhite in describing aggregated groups to avoid the implication that the white population is normative.¹⁰³

Journal editors may choose to require the use of quotation marks to set off the words "race" and "racial" to emphasize that racial categories are not biologically based.^{113 ,129} An exception can be made when authors explicitly acknowledge that these terms refer to socially constructed categories.

It is important to avoid conveying the impression that race/ethnicity is inherent or fixed. Descriptions such as "black respondents" or "white adults ages 40 to 49 years" imply that each individual included in the category has a fixed identity and that the writer has absolute knowledge of that identity. Thus, a statement such as "We found a higher prevalence of asthma among Native American study participants than among white participants" is meaningful only when the authors have clearly identified what it means to be Native American or white for the purposes of the particular study. The use of more accurate and descriptive wording can be especially important in citing published studies, to avoid incorrect inferences. When no information is provided in a research report about how individuals were assigned to racial/ethnic categories, descriptors such as "individuals assigned to the Japanese ancestry category" could be used. An alternative is to use quotation marks with the names of categories to signal the fact that they are quoted directly from the article cited.

For tables and figures to be as accurate and informative as possible, the terms "race," "ethnicity," or "race/ethnicity" should not be used to label columns, rows, or axes without appropriate caveats. Authors might be required to substitute more precise descriptions such as "self-reported race/ethnicity" or "racial/ethnic category." Journals in which tables and figures are expected to stand alone (ie, without reference to the text) could require authors to use quotation marks with the word race and racial and/or describe in footnotes to tables or figures how categories were defined and how individuals were assigned to them.

Researchers, clinicians, and policy makers face a number of challenges in writing about race/ethnicity. First, the complexity and fluidity of self-identification, the lack of clear-cut boundaries between socially defined categories, and the lack of consistency in the selection and definition of categories make it difficult to write about race/ethnicity with precision. For meaningful statements to be made about health disparities, careful consideration must be given to the way in which race and ethnicity are conceptualized, the choice and definition of categories, and the way in which individuals are assigned to categories. Second, the use of race/ethnicity as a variable can reinforce the simplistic view that race/ethnicity is the sole and/or direct cause of health disparities unless thoughtful attention is paid to the specific pathways through which race/ethnicity influences health outcomes. Finally, unwarranted categorization by race/ethnicity carries the risk of reinforcing stereotypes and contributing to the racial/ethnic divisions that undermine the health of individuals and societies.^{1 - 2 ,25 ,29 ,60 ,112}

To address these challenges, we have proposed a set of guidelines based on 3 broad recommendations drawn from the biomedical literature: that the reason for using race/ethnicity should be specified, that categories should be described and justified, and that all relevant variables should be considered in analyses.