In this issue of Epidemiology, Goldstein and colleagues1 highlight the staggering racial disparities in HIV infection that exist in the United States. In particular, the authors note that among men who have sex with men, a group at an elevated risk for HIV infection, Blacks/African Americans (AAs) are disproportionately infected with HIV. The aforementioned racial disparity among men who have sex with men is reflective of broader HIV racial/ethnic disparities, where AAs are the racial/ethnic group most impacted by the domestic HIV epidemic.2,3 Addressing such disparities has been and continues to be a top priority of the US government.4–6
To observe reductions in racial/ethnic disparities in the domestic HIV epidemic, factors that contribute to or have the potential to lessen the disparity must first be identified. Next, interventions targeted at identified factors must be developed and validly evaluated before widespread implementation. After widespread implementation, national and other data on HIV-related outcomes must be accurately monitored for improvements in racial/ethnic disparities both across and within other groups including men who have sex with men. Achieving the aforementioned identification, evaluation, and monitoring requires “good data.”
In the context of this commentary, good data are those where all relevant information has been measured with negligible error. Although obtaining good data as previously defined is unrealistic, as epidemiologists and, more generally, public health researchers, we must make every effort to measure all relevant information with as minimal error as time and costs allows. Therefore, during the study design phase, first, we must spend considerable time identifying the most important information that needs to be collected. In the context of HIV racial/ethnic disparities research, such information may often include data on early life circumstances, neighborhoods, socioeconomic position, stress, racism, and discrimination.7–11 Second, we must determine the best way to collect/measure the relevant information. Third, we must develop new measures when no appropriate measures are available.
When designing a study, we must keep in mind that reliable and valid measures developed and tested in racial/ethnic nonminority populations may not be generalizable to racial/ethnic minority populations. Therefore, assessing whether a new measure is needed may entail evaluating the appropriateness of using an existing measure in a racial/ethnic minority population.12 In addition, developing new measures may require that we either undergo additional training in measure development and evaluation or seek out assistance from researchers with relevant expertise as we assemble our study research teams.
Good data are critical to observing reductions in HIV racial/ethnic disparities for several reasons. First, correctly identifying factors that contribute to or have the potential to lessen racial/ethnic disparities will often require that the exchangeability assumption hold.13,14 For the exchangeability assumption to hold, all sources of confounding and selection bias must be accurately measured. Second, identification also requires that potential factors under consideration and the outcome of interest are not measured with considerable error.15 Third, valid evaluation of developed interventions targeted at identified factors similarly requires that the exchangeability assumption hold, as well as accurate measurement of relevant information. Fourth, as demonstrated in the study by Goldstein and colleagues in this issue of Epidemiology, accurate measurement of relevant information is also required to appropriately document and monitor HIV racial/ethnic disparities.
Specifically, the publication by Goldstein and colleagues examined racial differences in the association between the exposure, being a man who has sex with men, and the outcome, HIV infection. The study by these authors nicely illustrates how potential measurement error and unmeasured information on potential confounders may hinder an investigator’s ability to accurately determine whether an HIV racial/ethnic disparity, in this case regarding the effect of an exposure, in fact exists, as well as the extent and direction of that disparity. Accurately determining the existence, extent, and direction of a disparity is critical to appropriately mobilizing resources to lessen the disparity if one should exist, as well as avoiding unnecessary targeting of resources based on race/ethnicity in the absence of a disparity.
Important to note is that the independent impacts of measurement error and unmeasured confounding on the observed disparity are likely more pronounced as the degree of the measurement error and unmeasured confounding becomes more dependent on race/ethnicity. Furthermore, even if the disparity is largely due to inaccurately measured or unmeasured confounders, interventions based on race/ethnicity that are targeted at the confounders may still be warranted. Alternatively, if the disparity is not due to systematic bias and instead due to some underlying biologic process that impacts health and varies by race/ethnicity, then interventions targeted at that biologic process would be warranted and likely more beneficial in specific racial/ethnic groups.
Given all of the aforementioned, I applaud Goldstein and colleagues for their use of more sophisticated quantitative techniques to discern whether a racial disparity in the effect of being a man who has sex with other men was due to measurement error or unmeasured confounders. Although not implied by the authors in their study, it is important to note that the Bayesian techniques that they applied and other approaches are not a substitute for thorough and sound data collection, especially because the validity of such approaches often relies on assumptions that may not be met, including good validation data.16 Instead, such approaches should be used only after reasonable efforts to thoroughly collect sound data have been made. Thorough and sound data collection also lessens the amount of systematic bias that needs to be accounted for in sensitivity analyses, such as the analyses performed by Goldstein and colleagues. Having to account for fewer sources of systematic bias will likely lead to simpler and perhaps more informative sensitivity analyses where we potentially may need to account for only one source of bias.
The publication by Goldstein and colleagues also highlights the fact that obtaining good data when studying HIV racial/ethnic disparities may be hindered by the sensitive nature of the information that often needs to be collected. Such sensitive information may include sexual (e.g., being a man who has sex with men), alcohol use, and drug use behaviors, as well as health conditions (e.g., HIV status). The need for sensitive information coupled with the tendency for individuals to provide more socially desirable responses particularly in the context of HIV/AIDS and sexual behavioral research can lead to measurement error.17,18 To facilitate more accurate data collection in this context, the best approaches to lessen the impact of social desirability need to be identified in the study design phase and implemented whenever possible during the conduct of the study.
The best approaches to lessen the impact of social desirability may vary by the population under study, as well as the specific information that is being reported and will likely include techniques that maximize participant privacy and anonymity, such as collecting data via self-administered questionnaires rather than face-to-face interviews.19 To further limit the potential for social desirability bias, whenever feasible and appropriate, biologic tests should be used to supplement self-reported behavioral information (e.g., condom use).19–21 Biologic tests should also be used instead of self-reports when ascertaining biologic information (e.g., HIV infection), again whenever feasible and appropriate. In addition, research efforts should continue to identify how best to further limit the tendency for participants to provide more socially desirable responses especially as newer modes including settings for ascertaining self-reported data are developed and implemented.22 The aforementioned research may require validation data based on biologic tests19,20 and needs to be conducted with participants from communities at highest risk for the outcome (e.g., HIV infection) such as AA men who have sex with men.
Given that social desirability bias may be more likely to occur when reporting on highly stigmatizing conditions and behaviors,19 research focused on how best to lessen the degree or impact of stigma may eventually serve to limit the tendency for individuals to provide more socially desirable responses and in turn limit measurement error. In terms of the substantive question that was the focus of the analysis by Goldstein and colleagues, such research could aim to identify how best to lessen the degree or impact of stigma associated with being a man who has sex with men and HIV infection, particularly in communities of color. As such, research could specifically aim to identify factors that buffer against the adverse effects of stigma on the tendency to provide socially desirable responses. Such factors may include social support and specific coping strategies given prior theoretical and/or empirical work that indicate that social support and specific coping strategies may serve as resources to offset the adverse effects of stigma on HIV-related outcomes, such as HIV symptoms.23,24
In summary, good data are critical to establish, monitor, and reduce HIV racial/ethnic disparities. Although sophisticated quantitative techniques such as the one employed by Goldstein and colleagues are potentially useful for correcting data that are not completely observed, sophisticated quantitative techniques are not a substitute for sound and thorough data collection. Going forward, collecting good data in the context of HIV racial/ethnic disparities research will likely require developing and evaluating new measures, as well as the continued development and implementation of strategies to elicit the most accurate information despite the often sensitive nature of the necessary information.
ABOUT THE AUTHOR
CHANELLE J. HOWE is an Assistant Professor of Epidemiology at the Brown University School of Public Health. She is an epidemiologic methodologist specializing in quantitative methods to enhance causal inference particularly in the area of HIV racial/ethnic disparities. She is especially interested in using advanced quantitative techniques to identify intervention targets to considerably reduce persistent HIV racial/ethnic disparities.
The author thanks Drs. Stephen Cole and Akilah Dulin-Keita for helpful feedback on an earlier draft of this commentary. Additional thanks to Drs. Christopher Kahler and Joseph Fava for their expert advice.
1. Goldstein ND, Burstyn I, Welles SL. The racial disparity conundrum of HIV risk among men who have sex with men: Bayesian approaches for correcting misclassification and residual confounding. Epidemiology. 2016. In press.
4. Office of National AIDS Policy. National HIV/AIDS Strategy for the United States. 2010.
5. Office of National AIDS Policy. National HIV/AIDS Strategy for the United States: Update of 2014 Federal Actions to Achieve National Goals and Improve Outcomes Along the HIV Care Continuum. 2014.
6. Office of National AIDS Policy. National HIV/AIDS Strategy for the United States: Updated to 2020. 2015.
7. Ferraro KF, Shippee TP. Aging and cumulative inequality: how does inequality get under the skin? Gerontologist. 2009;49:333–343.
8. Geronimus AT. Understanding and eliminating racial inequalities in women’s health in the United States: the role of the weathering conceptual framework. J Am Med Womens Assoc. 2001;56:133–6, 149.
9. VanderWeele TJ, Robinson WR. On the causal interpretation of race in regressions adjusting for confounding and mediating variables. Epidemiology. 2014;25:473–484.
10. Latkin CA, German D, Vlahov D, Galea S. Neighborhoods and HIV: a social ecological approach to prevention and care. Am Psychol. 2013;68:210–224.
11. Jones CP. Invited commentary: “race,” racism, and the practice of epidemiology. Am J Epidemiol. 2001;154:299–304; discussion 305.
12. Gregorich SE. Do self-report instruments allow meaningful comparisons across diverse population groups? Testing measurement invariance using the confirmatory factor analysis framework. Med Care. 2006;44(11 suppl 3):S78–S94.
13. Hernán MA, Robins J. Causal Inference Book. 2016.Boca Raton, FL: Chapman & Hall/CRC; Forthcoming.
14. VanderWeele TJ. Marginal structural models for the estimation of direct and indirect effects. Epidemiology. 2009;20:18–26.
15. Howe CJ, Cain LE, Hogan JW. Are all biases missing data problems? Curr Epidemiol Rep. 2015;2:162–171.
16. Funk MJ, Landi SN. Misclassification in administrative claims data: quantifying the impact on treatment effect estimates. Curr Epidemiol Rep. 2014;1:175–185.
17. Catania JA, Gibson DR, Chitwood DD, Coates TJ. Methodological problems in AIDS behavioral research: influences on measurement error and participation bias in studies of sexual behavior. Psychol Bull. 1990;108:339–362.
18. Fenton KA, Johnson AM, McManus S, Erens B. Measuring sexual behaviour: methodological challenges in survey research. Sex Transm Infect. 2001;77:84–92.
19. Phillips AE, Gomez GB, Boily MC, Garnett GP. A systematic review and meta-analysis of quantitative interviewing tools to investigate self-reported HIV and STI associated behaviours in low- and middle-income countries. Int J Epidemiol. 2010;39:1541–1555.
20. Siegfried N, Mathews C. Commentary: All is not what it seems: a systematic review and meta-analysis of quantitative interviewing tools to investigate self-reported HIV and STI-associated behaviours in low- and middle-income countries. Int J Epidemiol. 2010;39:1556–1557.
21. Zimmerman RS, Morisky DE, Harrison L, Mark HD. Validity of behavioral measures as proxies for HIV-related outcomes. J Acquir Immune Defic Syndr. 2014;66(suppl 3):S285–S292.
22. Kozak MS, Mugavero MJ, Ye J, et al. Patient reported outcomes in routine care: advancing data capture for HIV cohort research. Clin Infect Dis. 2012;54:141–147.
23. Earnshaw VA, Bogart LM, Dovidio JF, Williams DR. Stigma and racial/ethnic HIV disparities: moving toward resilience. Am Psychol. 2013;68:225–236.
24. Earnshaw VA, Lang SM, Lippitt M, Jin H, Chaudoir SR. HIV stigma and physical health symptoms: do social support, adaptive coping, and/or identity centrality act as resilience resources? AIDS Behav. 2015;19:41–49.