Measuring Race and Ethnicity: Why and How?

Race and ethnicity are constantly evolving concepts, deceptively easy to measure and used ubiquitously in the biomedical literature, yet slippery to pinpoint as definitive individual characteristics. A current dictionary definition of race is “a family, tribe, people, or nation belonging to the same common stock, or a class or kind of people unified by shared interests, habits, or characteristics.”¹ For 154 years, the US government has defined race for its census takers, and for many years census takers then defined it for US residents. The terms used reflect the nation’s changing demographics and increasing recognition of human diversity. The 1850 enumerators used a form that assumed a default race of white, with a checkmark indicating nonwhites as black or mulatto, with additional indications for free or slave.² Indian was added as a category in 1860. Since 1960, individuals have been able to specify their own race and ethnicity, and by 2000 the census enumerated 126 racial and ethnic categories.³

Medical definitions of race have lagged behind, although thankfully the former Medical Subject Headings (MeSH) terms such as Caucasoid, Mongoloid, Negroid, and Australoid rarely appear in biomedical literature.⁴ Given that the connotations and definitions of race and ethnicity are constantly evolving, the use of the terms and concepts of race and ethnicity in the biomedical literature deserves examination.

Study results that associate characteristics and outcomes with race, often defined by an investigator as skin color, are largely a function of other less easily measured variables. From the “experiments” chronicled in the Mismeasure of Man⁵ to the (mis)interpretations in The Bell Curve,⁶ outcomes more directly linked to socioeconomic status and deprivation and poorer health and health care have been attributed to race. Such attribution can be a convenient explanation to avoid disrupting the status quo. If racial handicap is believed to equate to genetic defect, and if genetics implies that the resulting characteristics are predestined, societal interventions for individuals with smaller skulls (and worse nutrition)⁵ or worse cognitive ability (and worse education) would be futile.⁶ Biomedical research is not usually guilty of such extreme and egregious misinterpretations, but nevertheless race sometimes is used as a proxy for the more accurate and informative variables of socioeconomic status or education.

Despite the dictionary definition and common misunderstanding of race as equating to some common genetic stock, abundant research illustrates the genetic diversity present within racial groups. Wilson et al,⁷ in a study of drug-metabolizing enzymes and genotyping in 8 populations around the world, found that genotypes clustered into 4 groups, but these 4 groups in no way corresponded to the populations from which they were drawn. Wilson et al concluded that “commonly used ethnic labels are both insufficient and inaccurate representations of the inferred genetic clusters . . . [and that] drug-metabolizing profiles . . . differ significantly among the clusters.” The Human Genome Project has found that the human population has 99.9% of its DNA in common.⁸ Genetic similarity cannot be inferred simply based on racial categories.⁹

The use of race as a proxy for unmeasured confounders, such as cultural, social, and environmental influences, is commonplace, but race is a poor proxy for these measures. The life experience and cultural milieu of US immigrants may be completely different from those who grew up in the United States, despite being assigned to similar racial or ethnic categories.¹⁰ Socioeconomic status, not race, is likely the greater determinant of health and health-related qualities.¹¹ Therefore, race is not a substitute for carefully assessed social and cultural characteristics.

On the other hand, race can be an important indicator of health disparities and health care delivery. An American College of Physicians position paper attests to “ . . . ample evidence illustrating that minorities do not always receive the same quality of health care, do not have the same access to health care, are less represented in the health professions, and have poorer overall health status than nonminorities.”¹² While race is just a departure point when evaluating such disparities, the article by Bradley et al¹³ in this issue of JAMA illustrates how race can be used along with specifically defined characteristics to begin to explore some of the reasons behind health disparities. In this retrospective, observational study of inpatients from the US-based National Registry of Myocardial Infarction, who were hospitalized during 1999 through 2002 with ST-segment elevation or myocardial infarction or left bundle-branch block and receiving acute reperfusion therapy, Bradley et al assessed time from hospital arrival to acute reperfusion therapy. As previous studies have shown, nonwhites had longer times from hospital entry to reperfusion therapy, as much as 7.3 minutes longer for blacks receiving thrombolytic therapy and 18.9 minutes longer for blacks receiving percutaneous transluminal coronary angioplasty.

Given that prompt reperfusion reduces mortality, such delays could be significant. However, when the researchers control for other variables in a series of models, interesting patterns emerge that indicate that race is not the whole story. First, the authors control for average door-to-treatment time for the patient’s hospital. Controlling for this variable plus calendar time and random effects reduced the delay for blacks to 6.3 minutes for thrombolytic therapy and 12.7 minutes for percutaneous transluminal coronary angioplasty. Controlling for a host of other variables including insurance status, clinical characteristics, and time parameters further reduced the time discrepancy. Interestingly, the addition of hospital characteristics other than door-to-treatment time had little effect on the models.¹³ These data suggest that rather than focusing on the possibility that some individuals receive worse treatment within a given hospital due to overt or subconscious racial bias, misperception, or patient choice, the problem may lie with the hospitals to which the patients are admitted. In fact, the discrepancies may be more a result of the hospital than the individuals within the hospitals, requiring reworking systems in addition to retraining individuals.

These data are particularly interesting in light of the recent study by Bach et al,¹⁴ in which black and white Medicare patients were found to be treated by different types of physicians. According to Bach et al, the physicians treating black patients were less likely to be board certified (77.4% vs 86.%) and reported being less able to access high-quality specialists, high-quality diagnostic imaging, nonemergency hospital admissions, adequate number of inpatient days, and high-quality ancillary services. Thus, rather than simply stating that blacks and whites should receive equal care once admitted, significant problems of health delivery may exist at the hospitals to which patients are admitted and within the continuity of care between physician and hospital.

Why race is measured is important, but how race is measured must be defined as well. Just as diagnostic criteria for myocardial infarction or assessment of dietary intake are described as part of a study’s methodology, defining race is essential to interpret individual studies and to facilitate comparisons. According to Bradley and colleagues,¹³ patients’ racial/ethnic groups were abstracted from the medical records using the following categories: white, African American/black, Hispanic, Asian/Pacific Islander, American Indian or Alaska native, and other or unknown race/ethnicity. In the study by Bradley and colleagues, “Admission or triage staff recorded race/ethnicity as the patient was registered, using hospital-defined race/ethnicity; in [National Registry of Myocardial Infarction], patients were assigned to only 1 race/ethnicity category.”

Generally speaking, individuals should self-designate race to ensure that the designation most closely matches what they believe reflects their personal and cultural background.² Since the 1960 census, individuals have been able to self-designate race; an option for Hispanic identity was added in 1970, and beginning in 2000 respondents could self-designate more than 1 race category. Self-designated race is particularly important when race is measured to act as a proxy for a social or cultural milieu. Bradley et al¹³ suggest that having admissions or triage staff designate the racial identity of participants for purposes of the study reflects how those providing care in the hospital view the individual. If that view explained the differences in health care delivery across categories of race and ethnicity, that could be a useful approach. However, the results of Bradley et al support a larger construct of hospital availability and selection, so it is likely that self-designated race and ethnicity would be the more accurate for evaluating attitude, behavior, and access of groups to hospitals by the community.

A second issue is how race is categorized. While race and ethnicity now have 126 categories in the United States alone, depending on a study’s size some categories may be too small to render the group meaningful from a statistical standpoint. In addition, in the 2000 census, 1 in 4 individuals in some regions of the country checked more than 1 box indicating race and ethnicity.¹⁵ Researchers will need to make choices about the number of categories to use and how to group them, and convey their choices to readers. The National Institutes of Health Web site for grants cites the 1997 Office of Management and Budget categories of 5 racial groups: American Indian or Alaska Native, Asian, black or African American, Native Hawaiian or other Pacific Islander, and white.³ Hispanic status is recorded separately. These categories likely will continue to evolve over time, and how these groups are categorized can have ramifications for health policy, for estimates of the uninsured, and potentially for health screening and interventions.² Given the changing nature of the field, the most important directive for researchers is to report fully what they have done to make the process transparent to others so that valid comparisons can be drawn across studies.

Given the importance of clearly defining race and ethnicity in medical research, JAMA is elaborating on the instructions given by the International Committee of Medical Journal Editors: “When authors use variables such as race or ethnicity, they should define how they measured the variables and justify their relevance”¹⁶ to make them even more explicit. When reporting race, ethnicity, or both, authors should describe who designated race and/or ethnicity for an individual; self-designation generally is preferred. Authors should indicate whether the options for designation were closed or open. If the options were closed, authors are asked to provide what the options were, whether categories were combined, and, if so, how. Open-ended options allow for more individually accurate description, but categorization for a study might be a challenge; authors should endeavor to make the process of coding transparent.

Finally, authors should indicate why race and/or ethnicity is believed to be relevant to the particular study. As Rivara and Finberg¹⁷ note, “analysis by race and ethnicity has become a knee jerk reflex, accompanying every table that examines demographic differences, such as age and sex.” While it may be appropriate to analyze race and ethnicity, the fact that race was assessed is not sufficient reason to analyze outcomes by racial categories. JAMA recognizes that authors of studies funded by the National Institutes of Health are required to address race and sex.¹⁸ Nevertheless, authors should state the relevance of race in the study, based on past literature or authors’ hypotheses, to facilitate critical evaluation of race and ethnicity as constructs within the study. If race, ethnicity, or both are being used as a proxy measure for other more difficult-to-measure variables, the rationale for doing so should be stated. Researchers should attempt to measure as many variables as possible directly, such as socioeconomic status, education, urban vs rural location, or income region by ZIP code. By doing so, researchers can begin to sort out whether an outcome is truly related to race (as defined in the study) or to other factors with a closer relationship to the causal pathway.

Determining race can be an important initial step in assessing quality of care delivery and outcomes, as the study by Bradley and colleagues illustrates. However, it is just a first step. By reporting race and ethnicity transparently and beginning to explore other important and related characteristics, biomedical research can move beyond race as a social construct in itself and explore other tangible components that can be affected to improve the public’s health.