Article Text

PDF

Ethnicity and epidemiological research: not so black and white
  1. C C Tam1,2,
  2. S J Lee1,
  3. L C Rodrigues1
  1. 1Department of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, UK
  2. 2Gastrointestinal Diseases Department, Health Protection Agency Communicable Disease Surveillance Centre, London, UK
  1. Correspondence to:
 Mr C C Tam
 Infectious Disease Epidemiology Unit, Department of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK; clarence.tamlshtm.ac.uk
  1. J Ahern,
  2. K E Pickett
  1. New York Academy of Medicine, 1216 5th Avenue, Room 553, New York 10029, USA; jahernnyam.org

    Statistics from Altmetric.com

    The analysis by Ahern et al of risk factors for preterm birth among African American and white women in San Francisco1 concluded that pregnant African American smokers are more prone to preterm delivery than white pregnant smokers. This conclusion is misleading. Firstly, the evidence of interaction between smoking and ethnicity was unconvincing—the difference in the odds ratios (ORs) was modest, and confidence intervals (CI) overlapped considerably (African American women: OR 1.77, 95% CI 1.12 to 2.79; white women: OR 1.25, 95% CI 1.01 to 1.55). Secondly, the authors did not consider residual confounding by factors such as maternal infection and previous preterm birth, which differ by ethnic group.2,3 Their assumption that the smoking-preterm birth association is linear seems biologically unlikely and problematic, as African American women in their study population smoked more than white women.

    Such analyses raise a more fundamental issue: the limitations of using ethnicity/race in epidemiological studies of causality. There is no consensus as to what “ethnicity” means. To some, “ethnicity” describes cultural differences between populations definable by phenotype, while “race” signifies differences strictly under genetic influence. This separation is imperfect—“ethnic groups” differ in genetic mix, culture, and socioeconomic situation. The “catch all” quality of ethnicity makes its use in epidemiological research attractive, but is an obstacle to causal inference. Ethnic groups have qualitative and contextual differences that do not make them directly comparable. Ethnic differences in biological effects will be difficult to uncover, as cultural, social, and economic factors that act as causal intermediates are too numerous and divergent to be adequately controlled.

    Finally, ethnicity is not a singular exposure that can be turned on and off. The observation of higher preterm birth rates among black than among white women smokers is not simply the answer to the question “what would rates of preterm birth among white women who smoke be, if they were black?” Social inequity and deprivation, however, are amenable to change in a way that ethnicity is not. In the paper by Ahern et al, African American women delivering preterm clearly were more often single, working class, receiving public insurance, and had lower level of education. Yet their analysis stratified by ethnic group does little to explain these great disparities between African American and white women, concentrating instead on variations within ethnic groups. An analysis of their data comparing preterm birth between social strata rather than ethnicity would shed some light on the role of ethnicity in research: useful tool or epidemiological distraction.

    References

    Authors’ reply

    In their letter, Tam et al present several critiques of our analysis of smoking and preterm delivery in African American and white women and then broadly criticise the use of race and ethnicity as factors in epidemiological research. We address each issue in turn.

    Firstly, Tam et al object to the fact that the odds ratios (ORs) for the effects of smoking on preterm delivery for African American women and white women in our analysis were only modestly different and that their confidence intervals overlap. However, even a small difference in risk can translate to a very large difference in the number of people affected on the population level. In addition, the ORs presented in our paper were for 10 cigarettes per day. The ORs for a pack of 20 cigarettes per day were 3.13 for African American women and 1.55 for white women, a more substantial difference. More importantly, the key difference worth examining is that of the parameter estimates on the log odds scale. The change in the log odds of preterm delivery for every cigarette smoked per day was 0.057 for African American women, while it is only 0.022 among white women—a difference of more than twofold. Ultimately, while there was some overlap in the confidence intervals presented, this observation is not equivalent to a statistical test of the difference between the values. Such a test was not used because of the stratified nature of our sample;we chose a stratified sample for methodological reasons that we address below. We believe the interpretation of point estimates is still appropriate in our study.

    Next, the authors suggest our effect estimates may be subject to residual confounding. This clearly may be the case with any observational study. However, in this particular instance, the factors that Tam et al discuss as omissions on our part are potentially on the causal pathway between our exposures and outcomes of interest, and thus, should not be adjusted for in our models. If prior cigarette smoking contributed to an earlier preterm delivery, adjusting for prior preterm delivery might have washed out the very effect we were interested in studying. The issue of maternal infections was explicitly discussed in our paper as a factor that may interact with smoking in its relation to preterm delivery. While we were unable to examine this in our analysis, as we lacked data on maternal infections, we hope that this work and our discussion will encourage others to look at this question.

    Tam et al also question the linearity of the relation between smoking and preterm delivery. In our analysis, we found no evidence of non-linearity in the relation between smoking and preterm delivery. The fact that African American women smoked more on average does not affect the validity of the risk estimates, as cigarette smoking was modelled as cigarettes per day and the overall range of values was similar for African American and white women.

    Finally, Tam et al criticise our decision to examine effects of neighbourhood and individual risk factors for preterm delivery within, rather than across, racial groups. Our decision to do so was inspired by the lack of such prior research and the fact that an expert conference on the problem of racial disparities in preterm delivery had called for just such an analysis.1

    We agree with Tam et al that race encompasses social, economic, and cultural factors. Of course this makes the study of race complex. However, complexity has never before been a reason to abandon the study of an issue. In the context of the United States, the history of racial oppression and strife makes this a crucial issue to study in the hope of understanding its effects on people’s lives. The differences in the rates of preterm delivery between African American and white women are independent of socioeconomic status2,3 making racial differences even more important to understand. The complexity of the question, far from scaring us away, should draw us in. While race is not amenable to change, there may be underlying reasons for a racial disparity in health that may in fact be amenable to change. A differential effect of smoking on preterm delivery by race, suggested in our analysis, may be one such underlying factor contributing to racial disparities. We hope our analysis will lead to studies that help us intervene on one aspect of this important health problem and cause of racial disparities in health.

    References

    View Abstract

    Request permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.