Invited Commentary: Sex and Race Differences in Diabetes and Cardiovascular Disease—Achieving the Promise of Sex and Race Subgroup Analyses in Epidemiologic Research

Abstract

Diabetes confers a higher risk of cardiovascular disease on women than on men. The reasons for these sex differences, such as poorer cardiovascular risk factor profiles, have received considerable attention. However, a recent report on sex × diabetes interactions on cardiovascular disease identified that few if any prior studies have confirmed these sex differences in black individuals, despite known diabetes-related disparities. In this issue of the Journal, George et al. (Am J Epidemiol. 2018;187(3):403–410.) found marginally significant multiplicative sex × diabetes interactions in black but not white study participants after adjustments for traditional and behavioral risk factors, competing risk, and change in diabetes status over time. This study is notable for its attempt to fill an important literature gap, and it elegantly addressed multiple statistical considerations in assessing sex × diabetes interactions according to race strata. The findings also highlighted several important considerations for conducting race and sex subgroup analyses.

Research has long shown that diabetes confers a higher risk of cardiovascular disease (CVD) and all-cause mortality among women than among men (1, 2). Several mechanisms have been proposed, including the stronger impact of CVD risk factors among women than among men, as well as hormonal, physiologic, behavioral, and treatment factors (3, 4). To date, much of the literature on sex differences in the relationship between diabetes and CVD has been in predominantly white cohorts, with few if any prior trials assessing potential sex differences in other racial/ethnic minorities (4).

This is an especially striking “investigational disparity” for two reasons. While the age-adjusted rates of incident diabetes per 100 US civilians increased across all sex/race strata over the past 4 decades, the rates in black women have remained higher than in any other demographic, with a 30% (7.6–9.9 per 100 US civilians) increase during 1997–2014 alone (5). Secondly, despite the aforementioned sex × diabetes interactions among white persons, among persons with diabetes, the overall CVD prevalence remains higher among white men than among white women, while black women appear to have higher overall rates of CVD than do black men (6).

To augment existing literature on the known sex × diabetes interaction on CVD in white persons and answer calls to address the dearth of research in racial/ethnic subgroups (4), George et al. (7) have presented an elegant, rigorous, and detailed analysis of sex × diabetes interactions in black and white individuals in this issue of the Journal. Using data from the Atherosclerosis Risk in Communities (ARIC) Study, a large biracial cohort with many years of follow-up, the authors demonstrated a synergistic effect of being a black woman and having diabetes on incident CVD (defined here as peripheral artery disease, revascularization, stroke, and coronary heart disease events/deaths) (in women, hazard ratio (HR) = 2.3 and adjusted HR = 2.0; in men, HR = 1.8 and adjusted HR = 1.8), with similar findings confirmed in white ARIC participants (7). This study is notable for several reasons. First, it confirms prior research regarding diabetes increasing the risk of CVD particularly among women and extends these findings to a black cohort using more sophisticated analyses. Second, while the analyses cannot provide insight into mechanisms as to why sex (and race) alter the relationship between diabetes and CVD, they prompt discussion of underlying mechanisms and highlight the need to focus further clinical attention on women and blacks with diabetes. Finally, the authors demonstrate the importance of simultaneously considering race and sex in carefully designed analyses. Given calls to more systematically consider subgroups of sex and race in trials and other research, this is a timely and instructive study. In this invited commentary, we examine and expand upon these points.

First, while the authors confirm a known sex × diabetes interaction, their work enhances prior analyses in several important ways. George et al. (7) conducted a series of well-designed sensitivity analyses adjusting for competing non-CVD risk and time-varying diabetes status. The authors addressed the limited acknowledgment of competing non-CVD mortality risk in prior sex × diabetes studies, particularly important in black diabetic individuals, who have higher rates of microvascular and renal disease (8). They also accounted for the increased incidence rates of diabetes over time by updating diabetes status during follow-up. Interestingly, they found attenuated but persistent sex differences in the risk of diabetes on incident CVD among black participants, but nonsignificant interactions among white participants (7). While time-varying analyses likely reduced the CVD risk given the inclusion of newly diagnosed, lower-risk, more rigorously managed individuals with shorter follow-up times (7), the competing-risk results are compelling (adjusted HR = 1.9 in black women; 1.4 in black men and white women and men) (7), and suggest greater sex differences among black participants than among white participants.

Similar to prior literature, George et al. (7) also confirmed that despite sex × diabetes interactions, white men with diabetes continued to display greater absolute rates of incident CVD than white women. What is particularly interesting is their finding of more similar CVD incidence rates in black men and women with diabetes (7), suggesting that unlike their white counterparts, black women with diabetes may see an attenuation in their cardiovascular risk protection compared with men (6).

Despite these compelling findings, the authors noted several limitations, including power concerns in assessing sex × diabetes interactions in blacks, use of self-reported diabetes follow-up measures, and poor description of chronicity of diabetes status (7). Other unmentioned limitations should also be noted. There was no data on the use of lipid-lowering medications, such as statins, a key factor in CVD incidence. Interim glycemic control (e.g., glycated hemoglobin) and treatment (e.g., insulin use), both markers of diabetes severity (9), were also not described. Further, the authors were unable to control for time-varying covariates other than diabetes status, which might be particularly important in women whose excess risk of CVD appears to be attributable to CVD risk factors (4). Nonetheless, these findings lend support to prior research suggesting that the sex × diabetes interaction is modest and explained by CVD risk factors in mostly white cohorts (1, 10). Importantly, the multiplicative sex × diabetes interaction in black participants, even after adjustment for CVD risk factors and competing risk analyses, suggests a residual, albeit small effect in black cohorts.

This leads to our second point of discussion. The consistent sex differences in black compared with white participants prompts several interesting and exciting questions regarding contributing factors. What is driving sex differences in blacks compared with whites? Researchers have historically attributed the sex × diabetes interaction to the greater number and severity of CVD risk factors (e.g., adiposity, cholesterol) in diabetic (vs. nondiabetic) women compared with men (10, 11), as well as less aggressive risk-factor management by clinicians among women than among men (10, 12). Here, the risk of diabetes on CVD appeared to be more attenuated among black women than among black men after adjustment for CVD risk factors (7). As such, it can be gleaned, that similar to the case among white persons, behavioral and traditional modifiable cardiovascular risk factors appear to explain some of the sex differences in black individuals. Racial/ethnic disparities may intensify these differences, however. For example, black women have higher rates of obesity (13) and hypertension (14) than do white women. Research has also shown that non-Hispanic black women appear to have less favorable CVD risk-factor profiles and higher CVD death rates, in general, than non-Hispanic white women (15, 16). In fact, 50% of the excess risk of diabetes incidence in black women is due to modifiable risk factors (17). Black women have also been reported to be less likely to be adherent to CVD regimens than have members of other racial groups (18).

Other previously described factors posited to explain sex differences in the effect of diabetes on CVD include sex hormones, endothelial/thrombotic factors, poorer glycemic control, suboptimal adherence rates, psychosocial stressors, other sex-specific factors (such as gestational diabetes), and differential responses to treatment and lifestyle interventions (e.g., exercise) (3, 4). Race may compound many of these sex-specific factors. For now, mechanisms to explain sex differences remain poorly understood among both black and white populations and warrant further study.

Finally, this study provides an opportunity to examine the state of sex and race/ethnicity subgroup analyses in research. The National Institutes of Health and the Food and Drug Administration both recommend the inclusion of race and sex in studies and in published research. These guidelines were instituted to combat the lack of representativeness according to sex and race in prior studies. Also, the Agency for Healthcare Research and Quality and the National Institutes of Health Office of Research on Women’s Health conducted several systematic reviews and meta-analyses and found a striking lack of studies stratified by sex and race/ethnicity from which to draw conclusions (10). Representation of minorities and women in investigational trials is fundamental to improving our underlying mechanistic understanding and to targeting interventions. With these recommendations have come improvements in subgroup analyses in diabetes-related research. The ACCORD (19) and Look AHEAD (20) trials did specifically assess race and sex interactions for primary results, although they found no evidence of differential effect of intervention according to race or sex. However, neither of these publications examined sex × race interactions.

This study from George et al. highlights the importance of considering subgroup analyses according to both sex and race. That being said, investigators need to move beyond examining differences solely due to literature gaps or even due to known racial disparities in chronic diseases prevalence rates. Hypothesis-driven analyses that expand and clarify the rationale for testing sex or race differences will be key to fully realizing the benefits of subgroup analyses. George et al. (7) acknowledge that there is no pathophysiologic reason to suggest that sex differences in the risk of diabetes on CVD would be any different among black versus white participants. However, as we’ve shown above, the complex relationship between race and sex warrants greater discussion, and their findings open the door to considering multiple ways in which sex and race interact to influence CVD in adults with diabetes.

Fully understanding the findings of a residual sex × diabetes interaction on CVD in black individuals will require moving beyond biologic or genetic perspectives to explore the ways in which race and ethnicity are often used as proxies for unmeasured cultural, social, and environmental influences and factors (21, 22). Similarly, sex differences are often used as proxies for sex and cultural norms (23). Improving the measurement and understanding of social determinants that are more amenable to public health interventions than race and sex will be key to finally making an impact on health disparities (21, 22). Race and sex—and/or their proxies—might interact through access to care and preventive health services, the built environment, and health behaviors (smoking, physical inactivity, comorbid depression, nonadherence/self-monitoring) to influence outcomes in patients with diabetes (24). In fact, there were notable environmental/geographic differences between white and black participants in ARIC that might have intensified sex differences. Researchers should also consider ways in which stress responses, discrimination, and other social determinants of health disproportionately affect women and have impact on the race-sex-diabetes interaction.

In addition to prompting the exploration of approaches to subgroup analyses, these findings serve as a reminder that despite decades of research, diabetes incidence rates have continued to climb, particularly among black women (5). Policy makers and public health officials should consider ways to improve preventive screening and treatment measures as well as educational campaigns for high-risk groups. Decades of effective interventions in behavioral change and treatment adherence still need to be leveraged, harnessed, and disseminated, particularly in partnership with community members. Researchers should additionally consider sex and sex × race differences in participant response to lifestyle and behavioral interventions through well-powered subgroup analyses when appropriate (4). Further research is also needed on sex differences in efficacy, pharmacodynamics, pharmacokinetics, and side effects for glucose-lowering therapies (3). Finally, providers should consider aggressive CVD risk-factor management in women, improve doctor-patient communication, and consider personalized approaches that account for sex/race differences in diabetes management, particularly as they relate to social determinants of health.

In summary, we commend the authors for the rigorous, comprehensive, and transparent way in which they tackled the dearth of research on sex × diabetes interactions on CVD among blacks. Their elegant approach, which included multiple, comprehensive ways of assessing the durability of the sex × diabetes interaction in black and white individuals, serves as reminder of complex ways race and sex may interact in diabetic patients. We need to do a better job of describing differences by race and sex simultaneously, and George et al. provide us with a compelling illustration in this issue.

Abbreviations

ARIC

Atherosclerosis Risk in Communities Study

CVD

cardiovascular disease

HR

hazard ratio