Pathways From Socioeconomic Factors to Major Cardiovascular Events Among Postmenopausal Veteran and Nonveteran Women: Findings From the Women's Health Initiative

METHODS

RESULTS

Among 145 521 WHI participants with nonmissing data on veteran status, an “unknown/not reported” category was created for 2742 (1.88%) who did not report their occupation and 9607 (6.60%) who did not report their household income. Furthermore, 129 340 (88.9%) of 145 521 WHI participants with no missing data on veteran status also had no missing data on socioeconomic characteristics, with 144 585 having nonmissing data on education and 130 158 having nonmissing data on NSES. Of those, 122 811 (94.9%) had nonmissing data on demographic, lifestyle, and health characteristics. Of those, 1525 women (45 veteran and 1480 nonveteran) were followed up for <12 months, experienced a major CVD event, or had missing data on major CVD events within 12 months postenrollment, and were therefore excluded. The final analytic sample consisted of 121 286 women (3091 veterans and 118 195 nonveterans) with a follow‐up time ≥12 months, yielding 16 108 major CVD events (560 veteran and 15 548 nonveteran). Of those, 4821 (126 veteran and 4695 nonveteran) had a myocardial infarction, 5559 (191 veteran and 5368 nonveteran) had a stroke, and 12 536 (477 veteran and 12 059 nonveteran) experienced a CVD‐related death, whereas 105 178 (2531 veteran and 102 647 nonveteran) did not experience a major CVD event between follow‐up time ≥12 months and December 31, 2022 (Figure S1). Of note, 29 656 women (987 veterans and 28 669 nonveterans) died from other causes besides a major CVD event between follow‐up time ≥12 months and December 31, 2022. Significant differences in distribution by WHI component, region of residence, race, and ethnicity were observed between 121 286 WHI participants with nonmissing and 6391 WHI participants with missing data on covariates (Table S2). Similarly, significant differences in distribution by region of residence, race, and ethnicity were observed between 121 286 WHI participants in the analytic sample and 145 521 WHI participants with nonmissing data on veteran status (Table S3), with the incidence of major CVD events being slightly less in the analytic sample versus those with nonmissing data on veteran status (HR, 0.98 [95% CI, 0.96–0.99).

Table 1 presents demographic, socioeconomic, lifestyle, and health characteristics of women who were enrolled between 1993 and 1998 (mean±SD age, 63.4±7.2 years; 86.3% white and 94.9% non‐Hispanic) by veteran status. Except for WHI component, NSES, BMI, cardiometabolic risk factors, self‐rated health, depressive symptoms, and current health care provider, the distribution by these baseline characteristics differed significantly between veteran and nonveteran groups, with veterans being older and more frequently highly educated, White, non‐Hispanic, single/divorced/widowed, not working/retired/disabled, ever‐smokers, or consumers of alcohol. Although they had a greater number of comorbidities, veterans had better diet quality and physical activity than their nonveteran counterparts.

A total of 16 108 (560 of 3091 [18.1%] veteran and 15 548 of 118 195 [13.2%] nonveteran) women experienced a major CVD event over a mean±SD follow‐up time of 16.98±8.14 years (range, 1–28 years) and a median of 18.43 years (interquartile range, 8.81–24.98 years). Based on examination of Kaplan–Meier curves (Figures S2 though S4) and the weak correlation between Schoenfeld residuals and follow‐up time among women who experienced a major CVD event in the context of multivariable Cox regression models (Table S4), the proportional hazards assumption was not violated. However, given the large sample size, the proportional hazards assumption test was statistically significant (P<0.0001) for hypothesized relationships. Accordingly, we also presented multivariable Cox regression models for these relationships, after stratifying by median follow‐up time.

Univariate analyses suggested that veterans were at higher risk for experiencing a major CVD event, with an unadjusted HR of 1.55 (95% CI, 1.43–1.69). However, this relationship became nonsignificant after sequentially controlling for demographic, socioeconomic, lifestyle, and health characteristics, suggesting that veteran women were not more likely to develop CVD as compared with their nonveteran counterparts, after adjusting for these characteristics (Table 2). Similar results were obtained with cause‐specific hazard models, whereby a death from another cause was considered a competing risk (Table S5) and after exclusion of women with <2 years or <5 years of follow‐up (Table S6). Stratified analyses by median follow‐up time yielded similar results to the overall analysis (Table S7).

We examined HRs from Cox regression models to understand the relationship of socioeconomic characteristics with major CVD risk, resulting in the high‐risk versus low‐risk SES indicators to be evaluated in subsequent analyses (Table S8). Table 3 shows Cox regression models that adjusted for WHI component, region of residence, age, race, ethnicity, and marital status. These models suggest that SES indicators are not significantly related to major CVD events among veteran women. In contrast, among nonveteran women, having some college education or less, household income <$100 000, or NSES below the third tertile are associated with 13% to 35% greater risk for a major CVD event. Whereas interaction terms of occupation×veteran status (P=0.45) and veteran status × NSES (P=0.08) in adjusted models were not statistically significant, interaction term of education × veteran status (P=0.02) and household income×veteran status (P=0.03) in adjusted models were statistically significant. Similar results were obtained with cause‐specific hazard models, whereby a death from another cause was considered a competing risk (Table S5), as well as after excluding women with <2 years or <5 years of follow‐up (Table S6). Stratified analyses by median follow‐up time yielded similar results to the overall analysis, but also suggested that high‐risk SES was associated with major CVD events among nonveterans after ≥18 years of follow‐up (Table S9).

After examining bivariate relationships of demographic, lifestyle, and health characteristics with high‐risk versus low‐risk SES indicators (Table S10) and major CVD events (Table S11), we selected the key mediators from the conceptual framework described in Figure S5. Likelihood ratio tests for group invariance suggested that none of the GSEM estimates differed by veteran status. Accordingly, we performed GSEM to examine the pathway from each dichotomized SES indicator to major CVD event through ever‐smoker status, which may influence BMI, number of comorbidities, cardiometabolic risk factors, and self‐rated health, in both the overall sample (Table 4) and after stratifying by veteran status (Table 5). Of note, alcohol consumption, diet quality, physical activity, and depressive symptoms were not consistently related to SES indicators and major CVD event, and, therefore, did not meet the criteria for mediation between exposure and outcome variables. Consistent with Cox regression models, the total effects of socioeconomic characteristics were not significant for veterans, whereas the total effects of education, household income, and NSES were significant among nonveterans. Among veterans, education was associated with major CVD events only through BMI, comorbidities, cardiometabolic risk factors, and self‐rated health, with no mediating effect for smoking history. Similarly, among veterans, BMI mediated the relationship between household income and CVD events, whereas BMI, cardiometabolic risk factors, and self‐rated health mediated the relationship between NSES and CVD events, with no mediating effect of smoking history. Among nonveterans, SES indicators were associated with CVD events through BMI, comorbidities, cardiometabolic risk factors, and self‐rated health, and smoking history partly mediated the relationships of education, household income, and NSES with major CVD events through comorbidities and self‐rated health.

DISCUSSION

In this prospective cohort study, we followed 121 286 postmenopausal women, aged 50 to 79 years at enrollment (1993–1998), over an average of 17 years. Our results suggest that after controlling for demographic, socioeconomic, lifestyle, and health characteristics, there were no differences in CVD risk between veteran and nonveteran postmenopausal women. Among veterans, the total effects of socioeconomic characteristics on risk of major CVD events were not statistically significant, with socioeconomic characteristics associated with major CVD events only indirectly through BMI, comorbidities, cardiometabolic risk factors, and self‐rated health. In contrast, among nonveterans, lower education, household income, and NSES were positively related to major CVD events, and these total effects were partly mediated by indirect effects of BMI (education: 20%; household income: 12.5%, NSES: 15%), comorbidities (education: 20%; household income: 12.5%, NSES: 15%), cardiometabolic risk factors (education: 15%; household income: 10%, NSES: 10%), and self‐rated health (education: 15%; household income: 7.5%; NSES: 10%). Similarly, smoking history served as one of the mediators within pathways between SES indicators and major CVD events only among nonveterans.

The finding of a stronger relationship between SES and major CVD events and more complex pathways linking SES to major CVD events among nonveteran versus veteran women can be best interpreted through the lens of the marginalization‐related diminished returns theory, whereby veteran women can be conceptualized as a marginalized group in a similar fashion to other minority groups defined based on race, ethnicity, sexual orientation, and immigrant status, and for whom SES exerts a weaker health effect compared with more mainstream groups. ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ For instance, the pathway between education and major CVD through BMI was found to be weaker for the marginalized veteran group compared with the more privileged nonveteran group. This finding that, compared with nonveterans, veteran women face additional barriers possibly linked to posttraumatic stress, stigma, and other forms of social disadvantage, which can diminish the protective effect of SES on their health, contributes to the body of evidence on marginalization‐related diminished returns. It highlights veteran status, one of the least studied types of marginalization, as a key social determinant of health. Conversely, the double jeopardy hypothesis, ⁶⁷ , ⁶⁸ , ⁶⁹ which posits that marginalized groups experience compounded negative effects from multiple sources of disadvantage, may not fully explain these findings. According to this hypothesis, we would expect a stronger relationship between SES and major CVD outcomes among veterans versus nonveterans, which is not supported by the data.

Female veterans in the United States represent a vulnerable population with unique cardiovascular care needs and disparities. ⁶ , ¹⁶ , ²⁰ , ⁷⁰ , ⁷¹ They may experience a higher CVD risk due to increased adverse childhood experiences, combat exposure, military sexual trauma, intimate partner violence, suboptimal CVD literacy, risky health behaviors, chronic medical conditions, mental health disorders (eg, depression, posttraumatic stress disorder), less regular exercise, lower social support, greater risk of homelessness, and poorer overall health, compared with civilians and male veterans. ¹ , ⁵ , ⁶ , ¹⁴ , ¹⁵ , ²⁰ , ⁷² Contrary to our findings, CVD risk factors were previously found to be more prevalent among female veterans than their civilian counterparts, with higher prevalence rates of obesity, obesity‐related chronic conditions (eg, hypertension, diabetes, and hyperlipidemia), and smoking among veterans. ¹ Compared with male veterans, female veterans often face disparities in CVD prevention, including lower likelihood of receiving nicotine replacement therapy and achieving adequate cholesterol control. ¹ Whereas the majority of veterans receive their health care at other facilities, the US Department of Veterans Affairs (VA) has made significant progress in providing comprehensive and sex‐specific health care over the past few decades, including the establishment of the VA Women's Health Services national program office in the early 1990s. ⁵ Regional and local clinics have improved women's primary care, outperforming private sector quality metrics and reducing VA disparities among the sexes. ⁵ However, gaps remain in diabetes, lipid control, and ischemic heart disease, with health literacy among female veterans remaining suboptimal. ⁵ , ⁶ , ¹⁶ In this study, we found that postmenopausal women in the WHI classified as veterans were 55% more likely to experience a major CVD event compared with their nonveteran counterparts. However, this relationship was confounded by demographic, socioeconomic, lifestyle, and health characteristics. Since nearly 50% of veteran women were70 years and older compared with only 22% of nonveteran women within this age range, it is plausible that age is the main driver for differences in major CVD risk between veteran and nonveteran women in the WHI. Further analyses suggest that the age‐adjusted HR for the relationship between veteran status and major CVD risk was 1.04 (95% CI, 0.96–1.14).

While traditional CVD risk factors have been extensively investigated, socioeconomic characteristics at different levels of organization such as individual‐ and neighborhood‐level SES, are established social determinants of health, and these may directly or indirectly influence cardiovascular health, but have received less attention. ⁴ , ⁵ , ⁶ , ¹⁵ , ⁷⁰ For instance, lower educational attainment has been linked to higher CVD risks and was associated with increased all‐cause and CVD mortality rates, potentially through lower health literacy, more risky health behaviors, and a greater need for health care services. ¹ , ³ , ⁴ , ⁵ , ⁶ , ¹⁶ , ²⁰ In contrast, higher educational attainment and income can lead to better access to health care services and lower mortality rates, whereas racial and ethnic minorities may experience socioeconomic challenges in obtaining higher education, insurance, and access to health care. ⁴ , ¹⁴ , ¹⁶ , ⁷⁰ , ⁷¹ , ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ Although the simultaneous examination of the NSES index with individual‐level SES characteristics has the potential for Berkson measurement error, NSES differs from education, occupation, and income, because of its relationship to the built environment, which can impact CVD risk through resource distribution, physical activity, and general well‐being. At a broader level, CVD mortality rates for nonelderly adults in the United States have remained stagnant over the past decade. This could be explained by economic trends, especially with rising income inequality, income fluctuation, and job instability, and the decline in social cohesiveness. ² , ⁷⁶ , ⁷⁷ , ⁷⁸ This study identified education, household income, and NSES as being more strongly related to major CVD events than occupation. It also highlighted a more salient role for socioeconomic characteristics in major CVD events among nonveterans versus veterans. It is important to recognize that veterans may have unique protective factors such as access to health care and case management supports through the VA. ⁷⁹ , ⁸⁰ There may be heterogeneity within the group of women veterans according to their level of access to VA health care services, ⁸¹ necessitating further evaluation. Although a question regarding health insurance was asked of WHI participants at enrollment, it cannot be used to ascertain their lifetime access to VA health care services. Therefore, in‐depth studies are needed that can explain why SES is more relevant to major CVD events among nonveteran versus veteran women, and how SES can potentially influence smoking, BMI, comorbidities, cardiovascular risk factors, and self‐rated health, as mediators for major CVD events.

It is worth noting that this study examined depression as the only psychosocial factor that can mediate the effect of SES indicators on major CVD events, whereas previous WHI studies found that traits such as dispositional optimism and cynical hostility, which are related to SES but independent from depression, may be linked to incident CVD, all‐cause, and CVD‐ and cancer‐related mortality risks. ⁸² These trait‐like attitudes, which are formed early in life and are crucial for stress response, social interactions, and health behaviors, should be taken into consideration in future WHI studies of veteran versus nonveteran populations. Future studies should also explore alternatives to NSES, including urban–rural residence based on the Rural–Urban Commuting Areas system, which considers population density and how closely a community is linked socioeconomically to larger urban centers, as well as food security, which incorporates functional status and the HEI as another NSES component for CVD risk. Finally, future studies should explore the complex interrelationships among key mediators within pathways between SES indicators and CVD risk, especially since self‐rated health may be considered as an outcome of BMI, comorbidities, and cardiometabolic risk factors.

This study has several strengths, including its comprehensive data collection at enrollment, which enables easy assessment of associations and accounting for confounders. It also offers an acceptable generalizability to postmenopausal women from diverse geographical areas in the United States, although racial and ethnic minorities within the WHI have higher levels of education and income and may experience fewer psychosocial adversities when compared with the general US population. ⁸³ Our study also has several limitations. First, we performed secondary data analyses using a subsample of the original WHI participants, which could potentially result in selection bias. Second, information bias may have occurred because many of the exposure, mediator, covariate, and outcome measurements were self‐reported. Similarly, most variables were assessed at the enrollment visit (1993–1998), and despite a long‐term follow‐up of 17 years, on average, heterogeneities among different components of the WHI study precluded us from defining time‐varying exposures, mediators, and covariates. Furthermore, SES indicators assessed at the time of WHI enrollment do not necessarily coincide with many early life exposures that may also influence CVD outcomes. Since the WHI study collected data on household income as an ordinal variable and household size was not provided within the baseline questionnaire, we were not able to calculate an SES measure that combines household income with size and composition of the household. Although posttraumatic stress disorder may be a more relevant mediator between SES and major CVD events among veterans, depressive symptoms were evaluated as a proxy for mental health disorders. Third, residual confounding due to unmeasured or inadequately measured confounders remains a concern for observational study designs. Of note, a causal relationship between socioeconomic characteristics and major CVD events through the mediation of lifestyle and health characteristics can only be established in the context of an experimental design. Fourth, veteran status is a complex issue due to various factors such as being deployed to war zones, having psychiatric conditions, receiving health care at VA facilities, as well as eligibility for disability benefits and other opportunities not afforded to nonveterans. Veterans may also inaccurately report their household income because of concerns over eligibility for VA benefits. Fifth, the role of chance in analyses stratified by veteran status cannot be ruled out, given that the number of veterans is considerably less than that of nonveterans. Larger samples are also needed to further stratify analyses by branch of military service and history of CVD. Finally, participants in the WHI were volunteer postmenopausal women at clinical centers, restricting our ability to generalize study findings to younger women of diverse racial and ethnic background. Specifically, Black race is an important confounder that may underlie the greater CVD risk among female veterans. In the United States, CVD is overrepresented among racial and ethnic minorities, ⁸⁴ , ⁸⁵ , ⁸⁶ and Black women are overrepresented within the veteran population. ⁸ In fact, Black women represent 12.4% of the nonveteran but 19.3% of the veteran female population in the United States, ⁸ while Black nonveteran (7.7%) and veteran (6.6%) women are underrepresented in the WHI. As such, these study findings may not generalize to a more racially diverse population of female veterans, potentially biasing CVD morbidity and mortality risks to the null. In addition, these results may not generalize to military veterans in other countries besides the United States.

In conclusion, a stronger link exists between SES and major CVD events among nonveteran versus veteran postmenopausal women. Furthermore, nonveteran postmenopausal women exhibited more complex pathways between socioeconomic characteristics and major CVD events through lifestyle and health characteristics than their veteran counterparts. These findings inform the future design, conduct, and evaluation of preventive strategies that target CVD risk according to veteran status. This study expands the marginalization‐related diminished returns literature and is among the first to highlight the importance of considering the unique barriers faced by veteran women as a marginalized group, when examining the relationship between SES and CVD outcomes. The findings underscore the need for tailored interventions that address these barriers and help mitigate the diminished returns of SES in improving health for veteran women as a marginalized population.

Sources of Funding

The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, and the US Department of Health and Human Services through 75N92021D00001, 75N92021D00002, 75N92021D00003, 75N92021D00004, 75N92021D00005.

Disclosures

None.

Supporting information