Abstract
Background
In prior studies of cumulative risk factor exposure, self-identified race was independently associated with incident cardiovascular disease (CVD). A recent study suggests clinical, demographic, and socioeconomic factors explain racial differences. We used propensity score matching to study race as an independent incident CVD risk factor.
Objectives
The purpose of this study was to assess race as an independent risk factor for incident CVD.
Methods
We analyzed CARDIA (Coronary Artery Risk Development in Young Adults) study data using propensity score matching of White and Black women, and, separately, White and Black men, with respect to known CVD risk factors.
Results
Black men (n = 487), compared to White men (n = 487), had higher risk of CVD (HR: 2.30; 95% CI: 1.36-3.89; P = 0.0014), stroke (HR: 5.00; 95% CI: 1.45-17.3; P = 0.0047), and congestive heart failure (CHF) (HR: 3.60; 95% CI: 1.34-9.70; P = 0.0067). Black women (n = 640), compared to White women (n = 640), had higher CVD risk (HR: 2.36; 95% CI: 1.17-4.78; P = 0.014) and stroke risk (HR: 2.80; 95% CI: 1.01-7.77; P = 0.039) and borderline significantly higher CHF risk (HR: 3.50; 95% CI: 0.73-16.9; P = 0.096). Risk of coronary heart disease did not differ significantly by race in either sex. Multivariable analyses showed racial differences in the associations of multiple risk factors with incident CVD events independent of other known CVD risk factors.
Conclusions
Propensity score matching analyses demonstrate that race is an independent risk factor for incident CVD and its components, CHF, and stroke. Multivariable analyses suggest racial differences in Black vs White risk factor impact as the possible cause. Reasons for these differences remain to be explored.
Key words: cardiovascular disease, primary prevention, race, risk factors
Central Illustration
Cardiovascular disease (CVD) is the most common cause of death in the world with an incidence that is influenced by multiple risk factors, some, but not all, of which are potentially alterable.1 We have previously examined the impact of cumulative low-density lipoprotein cholesterol (LDL-C) exposure and cumulative exposure to all CVD risk factors, operating simultaneously, to develop statistical models that relate cumulative risk factor exposure to the risk of incident CVD and, separately, to the incidence of each of its components, coronary heart disease (CHD), congestive heart failure (CHF), and stroke.2,3 These analyses identified self-declared Black race as a CVD risk factor that is quantitatively important and operates independently of other known risk factors that are significant contributors to the predictive models. Other studies have demonstrated racial differences in CVD risk factors and drug response, including the lesser therapeutic value of angiotensin-converting enzyme inhibitors and beta-blockers in Black patients compared to White patients with systolic heart failure,4,5 and in the distribution of cardiovascular risk factors and CVD incidence.6, 7, 8, 9, 10 The mechanism of these differences remains unexplained. However, the importance of self-declared race as a crude, but independent, phenotypic risk factor for CVD, lies in the implications for designing personalized, as well as population-based, prevention and treatment regimens and in the design of primary prevention clinical trials.
In a recent paper, Shah et al11 used sex-stratified, multivariable-adjusted Cox proportional hazards models to study racial differences in CVD incidence in CARDIA (Coronary Artery Risk Development in Young Adults) study participants. They reported that the greater risk of premature CVD in Black vs White participants was explained by adjustment for risk factors, with clinical and neighborhood factors being the most important in women and clinical and socioeconomic risk factors being most important in men. Shah et al11 did not identify race as a statistically significant independent risk factor for CVD. Inoue et al have addressed the problem of heterogeneity in the association of risk factors with incident CVD.12 They point out that traditional statistical analysis is limited in failing to address heterogeneity in the association across the individual levels. These limitations may be overcome with the use of propensity score matching.
In this paper, we further examine race as a possible independent risk factor for CVD and for each of its components, CHD, CHF, and stroke, using propensity score matching of CARDIA participants to generate comparison groups that are comparable in known risk variables except race, thus overcoming the limitations of conventional multivariable analysis. Compared to the study by Shah et al which was based on the entire unmatched CARDIA cohort,11 our analysis focused on the propensity score-matched subsamples of the CARDIA study, ensuring that the clinical, lifestyle, depression, socioeconomic, and neighborhood factors were balanced between the 2 racial groups. Our study differs in design, statistical methods, and the covariates used. Unlike Shah et al’s analysis that modeled the risk of CVD over time since enrollment using multivariable Cox models with time-dependent covariates,11 our study results used a landmark survival analysis to assess and predict the CVD risk over time after age 40 years following adjustment for known risk factors as time-fixed covariates, including cumulative exposure to multiple risk factors previously identified.2,3 Specifically, CVD incidence and the incidence of CVD components are compared using propensity score-matched Black women and White women and, separately, in matched Black men and White men. We then utilized multivariable analyses that assessed the heterogeneity of risk factor associations in the various race/sex groups.
Methods
The CARDIA study design has been published previously.13,14 The methods of the CARDIA study and the method used in developing the propensity score-matched sample used in this article are briefly summarized below. Details of the statistical approach are presented in the online Supplemental Material.
CARDIA study participants
A total of 5,115 Black and White, male and female participants from 4 cities in the United States (Birmingham, Alabama; Oakland, California; Chicago, Illinois; and Minneapolis, Minnesota) were entered into the CARDIA study from 1985 to 1986 (baseline; Y0). The study protocols were approved by the Institutional Review Boards of the study sites, and all participants gave written informed consent. Participants were balanced by race, sex, education (less or more than high school) status, and age (<24 or ≥25 years) at year 0 for each site. Over the course of the study, CARDIA participants have had 10 in-person examinations: at Y0, Y2, Y5, Y7, Y10, Y15, Y20, Y25, Y30, and Y35. Among surviving participants, CARDIA retention rates at the in-person examinations were 91%, 86%, 81%, 79%, 74%, 72%, 72%, 71%, and 67% (during the COVID-19 pandemic), respectively. Contact continues to be maintained with participants via telephone, mail, or e-mail every 6 months, and with annual interim medical history ascertainment. In the last 5 years, >90% of surviving participants have responded, and vital status ascertainment is virtually complete through contacts and intermittent queries of the National Death Index.
Study outcome and ascertainment
The primary outcome of this study was a composite of incident CVD events that included the first occurrence of nonfatal CHD (myocardial infarction, nonmyocardial infarction acute coronary syndrome, and coronary revascularization), stroke, transient ischemic attack, hospitalization for heart failure, intervention for peripheral arterial disease, or death from cardiovascular causes. Specific components of CVD included CHD, CHF, and stroke. New cardiovascular and cerebrovascular events were recorded from the baseline examination through August 2020. During scheduled study examinations and yearly telephone interviews, participants or their designated proxy were questioned regarding interim hospital admissions, outpatient procedures, and deaths. For participants who had been hospitalized or who had received an outpatient revascularization procedure, medical records were requested. Vital status was assessed every 6 months; medical and related death records for deceased individuals were requested with consent of the next of kin. Two physician members of the Endpoints Surveillance and Adjudication Subcommittee independently reviewed medical records and recorded information to adjudicate possible cardiovascular or cerebrovascular events or underlying causes of death using specific definitions from the manual of operations: (https://www.cardia.dopm.uab.edu). Disagreements between the 2 physician reviewers were adjudicated by the full committee.
Covariates and risk exposure
Medication usage, physical measurement data, lifestyle factors, medical and family history, and laboratory data at baseline and follow-up examinations were obtained according to standard protocols.13,14 Lipid measurements using fasting blood samples were collected at baseline and at all follow-up examinations. Total cholesterol and triglyceride (TG) levels were measured enzymatically, high-density lipoprotein cholesterol (HDL-C) was determined after precipitation with dextran sulfate/magnesium chloride, and LDL-C was calculated by the Friedewald equation.13 Because TG measurements were skewed, we transformed the TG measurements using the natural logarithmic transformation, log-TG = log (TG). Body mass index was calculated as weight (kilograms) divided by height in meters squared (m2). Diabetes mellitus was defined by fasting glucose ≥126 mg/dL (all 9 examinations), or use of diabetic medication, or 2-hour glucose ≥200 mg/dL (years 10, 20, and 25), or hemoglobin A1C ≥6.5% (years 20 and 25) when available. Following a 5-minute period of sitting quietly, systolic blood pressure (SBP) and diastolic blood pressure (DBP) were calculated as the average of the last 2 of 3 determinations. Hypertension was defined by SBP ≥140 or DBP ≥90 mm Hg or use of antihypertensive medication. Pulse pressure (PP) was calculated as the difference between SBP and DBP and mean arterial pressure (MAP) was the weighted average of SBP and DBP, MAP = SBP/3+ (2 × DBP)/3. The estimated glomerular filtration rate was calculated based on serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration equation.15 To assure the comparability of the same risk factor measurements across the 30 years of follow-up time, calibration studies of glucose, creatinine, and blood pressure were conducted,16,17 and the recalibrated measures were used in our analysis. We also included the participants’ lifestyle (smoking status, alcohol consumption, physical activity), depression symptom score, socioeconomic (education, employment) and neighborhood factors (residential racial composition, neighborhood-level poverty) described previously.11
Participant-specific cumulative areas of risk factors
In addition to the original LDL-C, HDL-C, log-TG, MAP, and PP measures, we computed new measurements characterizing the trajectories of each measure over age as the cumulative exposure measurements (cumulative area under the risk factor’s trajectory curve vs age). The participant-specific trajectories of LDL-C, HDL-C, log-TG, MAP, and PP levels over age were estimated using the nonparametric cubic spline mixed-effects models (Supplemental Appendix).2,3
Propensity score matched samples
Since the Black and White participants were different in many characteristics (Supplemental Table 1), evaluating the association of race with risks of various CVDs using standard regression models was subject to potential extrapolation drawbacks due to lack of overlap in covariate distributions across the Black and White groups.18,19 The propensity score matching method can correct this potential sample selection bias and adjust for confounding by identifying the comparable subgroups within the CARDIA study with similar observed characteristics of the Black and White groups.12,18,19 Two propensity-matched samples were created, one for men and the other for women, using 1:1 ratio between Black and White participants without replacement with a caliper of 0.2 SD of the logit propensity score.20 The variables used for propensity score matching are shown in Table 1. These variables include the cumulative exposure to risk factors, diabetic status, medication use, lifestyle, socioeconomic, depression, and neighborhood variables that were considered in earlier publications of the CARDIA study.3,11
Table 1.
Characteristics of the Participants After Propensity Score Matching by Race and Sex at Age 40 Years
| Men |
Women |
|||
|---|---|---|---|---|
| Black (n = 487) | White (n = 487) | Black (n = 640) | White (n = 640) | |
| Median follow-up time after age 40 y | 19.2 (3.8) | 19.5 (3.5) | 19.3 (4.0) | 19.6 (3.8) |
| Incident cardiovascular disease events per 1,000 person-years | 5.6 | 4.2 | 2.3 | 1.5 |
| Age at enrollment, y | 24.0 ± 3.7 | 25.5 ± 3.4 | 24.0 ± 3.8 | 25.7 ± 3.4 |
| Site: Birmingham, Alabama | 115 (23.6) | 115 (23.6) | 138 (21.6) | 114 (17.8) |
| Chicago, Illinois | 96 (19.7) | 86 (17.7) | 122 (19.1) | 114 (17.8) |
| Minneapolis, Minnesota | 135 (27.7) | 163 (33.5) | 162 (25.3) | 216 (33.8) |
| Oakland, California | 141 (29.0) | 123 (25.3) | 218 (34.1) | 196 (30.6) |
| Education: never went to college (grade 12 or less) | 108 (22.2) | 87 (17.9) | 112 (17.5) | 97 (15.2) |
| Employment: employed full-time | 470 (96.5) | 473 (97.1) | 584 (91.3) | 598 (93.4) |
| Smoking | ||||
| Never | 290 (59.5) | 287 (58.9) | 370 (57.8) | 357 (55.8) |
| Former | 64 (13.1) | 72 (14.8) | 117 (18.3) | 135 (21.1) |
| Current | 133 (27.3) | 128 (26.3) | 153 (23.9) | 148 (23.1) |
| Alcohol intake | ||||
| None | 192 (39.4) | 171 (35.1) | 335 (52.3) | 310 (48.4) |
| Moderate | 234 (48.0) | 254 (52.2) | 224 (35.0) | 245 (38.3) |
| Heavy | 61 (12.5) | 171 (12.7) | 81 (12.7) | 85 (13.3) |
| Physical activity, exercise units | 438.8 ± 327.3 | 431.4 ± 300.8 | 275.4 ± 259.3 | 291.2 ± 239.8 |
| Family history of coronary heart disease | 128 (26.3) | 117 (24.0) | 166 (25.9) | 160 (25.0) |
| Diabetes mellitus | 24 (4.9) | 23 (4.7) | 26 (4.1) | 25 (3.9) |
| Hypertension | 144 (29.6) | 140 (28.7) | 148 (23.1) | 141 (22.0) |
| Body mass index, kg/m2 | 28.0 ± 5.0 | 27.7 ± 5.4 | 28.2 ± 6.2 | 27.6 ± 7.1 |
| Waist circumference, cm | 91.1 ± 12.2 | 91.4 ± 12.6 | 82.9 ± 13.4 | 81.7 ± 14.8 |
| Fasting glucose, mg/dL | 95.6 ± 23.2 | 95.2 ± 18.7 | 90.4 ± 21.6 | 90.9 ± 19.0 |
| Creatinine, mg/dL | 1.1 ± 0.6 | 0.9 ± 0.1 | 0.8 ± 0.1 | 0.7 ± 0.1 |
| eGFR, mL/min/1.73 m2 | 107.4 ± 20.1 | 103.4 ± 13.3 | 114.1 ± 17.9 | 102.0 ± 14.2 |
| Systolic blood pressure, mm Hg | 116.0 ± 12.5 | 114.9 ± 12.1 | 108.7 ± 10.8 | 107.5 ± 11.7 |
| Diastolic blood pressure, mm Hg | 75.4 ± 10.9 | 74.6 ± 10.0 | 71.0 ± 9.2 | 70.2 ± 9.8 |
| Total cholesterol, mg/dL | 187.1 ± 39.3 | 187.0 ± 35.9 | 178.8 ± 32.6 | 180.4 ± 32.0 |
| LDL cholesterol, mg/dL | 117.5 ± 34.5 | 117.5 ± 32.3 | 106.3 ± 30.8 | 107.1 ± 29.4 |
| HDL cholesterol, mg/dL | 46.5 ± 12.8 | 46.0 ± 14.7 | 56.3 ± 14.9 | 55.2 ± 14.3 |
| TG, mg/dL | 115.4 ± 85.4 | 118.2 ± 98.8 | 79.3 ± 44.5 | 89.4 ± 61.6 |
| LDL-AUC over age 18–40 y, mg/dL × ya | 2,513.1 ± 612.6 | 2,520.0 ± 550.6 | 2,353.6 ± 526.3 | 2,352.3 ± 510.7 |
| HDL-AUC over age 18–40 y, mg/dL × ya | 1,088.4 ± 217.5 | 1,068.9 ± 240.2 | 1,238.8 ± 227.0 | 1,242.0 ± 231.8 |
| log-TG-AUC over age 18–40 y, log-mg/dL × ya | 95.4 ± 7.5 | 95.6 ± 9.3 | 91.1 ± 6.8 | 91.7 ± 7.6 |
| MAP-AUC over age 18–40 y, mm Hg × ya | 1874.8 ± 131.7 | 1867.5 ± 129.6 | 1774.0 ± 114.9 | 1761.7 ± 124.6 |
| PP-AUC over age 18–40 y, mm Hg × ya | 918.4 ± 102.8 | 912.7 ± 106.8 | 848.8 ± 95.5 | 846.6 ± 93.1 |
| Center for Epidemiological Studies-Depression score | 9.5 ± 7.2 | 8.9 ± 7.2 | 10.6 ± 8.9 | 10.1 ± 8.6 |
| Neighborhood: G-statistic of Black segregation | 1.1 ± 2.2 | 1.0 ± 3.0 | 0.9 ± 2.2 | 0.8 ± 2.6 |
| Percentage of census tract living in poverty (%) | 12.3 ± 10.7 | 12.2 ± 11.4 | 11.3 ± 9.9 | 10.8 ± 10.0 |
Values are mean ± SD or n (%).
AUC = area under the curve; eGFR = estimated glomerular filtration rate, HDL = high-density lipoprotein; LDL = low-density lipoprotein; MAP = mean arterial pressure; PP = pulse pressure; TG = triglyceride.
The area under the curve (AUC) of LDL-C (HDL-C, TG, MAP, and PP) vs age. LDL-AUC (HDL-AUC, TG-AUC, MAP-AUC, and PP-AUC) was calculated in the unit of mg/dL× years (integrating LDL measured in mg/dL over age in years) to describe the cumulative LDL-C (HDL-C, TG, MAP, PP) exposure.
Statistical methods
We analyzed the propensity score-matched samples for women and men separately in a landmark survival analysis. The event time was defined to be the time from the landmark age of 40 years to the first CVD event.21, 22, 23 We used age 18 to 40 years as the “observation window” for the cumulative risk factor burden and used the landmark analysis models to characterize the risk exposure with the development of CVD events during a follow-up period up to 26 years (“prediction window” ages 40-66 years). Since the propensity score-matched White and Black participants were well-balanced in the individual risk factors and cumulative exposures to risk factors (LDL-C, HDL-C, log-TG, MAP, and PP), we fit the stratified Cox cause-specific proportional hazards model for the composite CVD events and CVD components (CHD, CHF, and stroke) to assess the relationship of race with these CVD events.24,25
Next, for participants of each race (White and Black) and sex (men and women), the multivariable Cox proportional hazards model were used to assess the relationship of cumulative LDL- C, HDL-C, log-TG, MAP, and PP risk exposures to the time to a first CVD event occurring after age 40 years, adjusted for other traditional cardiovascular risk factors measured around age 40 years, including education level, smoking history, physical activity, family history of CHD, body mass index, glucose, estimated glomerular filtration rate, and the presence of diabetes mellitus. For smoking and medical history, data collected from the first to the last clinical visits prior to age 40 years were examined. We excluded patients who died or had a nonfatal incident CVD event before age 40 years or who were missing covariates used in the multivariable analyses. Stepwise variable selection methods were used to determine the covariates that were significantly associated with the outcomes. Performance of these models was measured using the C-statistics and the modified Hosmer-Lemeshow test. The proportional hazards assumptions were evaluated based on scaled Schoenfeld residuals and the score test. Our test did not reject the proportionality assumption in the Cox models presented. A 2-sided P value <0.05 was considered statistically significant. Statistical analyses were performed using the R version 4.4.3 (R Foundation for Statistical Computing) and propensity score-matched analysis using the R package MatchIt.
Results
Propensity matched comparisons
Figure 1 shows the study design. Of total of 5,115 CARDIA participants aged 18 to 30 years at entry, 156 were excluded from this analysis because of 140 early deaths and 16 nonfatal CVD events before age 40 years, 86 were excluded due to missing information on cardiovascular risk factors used for matching, and one was excluded because of withdrawn consent. The remaining 4,872 CARDIA participants, 2,697 women and 2,175 men, were included in the analysis. Black women 1,419 and White women 1,278 were propensity score matched, with the final analytical sample composed of 640 Black women and 640 White women. Similar propensity matching of Black and White men yielded a final analysis sample of 487 Black men propensity matched with 487 White men.
Figure 1.
Study Design
CVD = cardiovascular disease; CARDIA = Coronary Artery Risk Development in Young Adults Study.
Table 1 shows the characteristics of the comparison groups, Black men compared to White men and Black women compared to White women after propensity score matching. Comparison groups were very closely matched for known CVD risk factors, including those previously identified important socioeconomic covariates as independent risk factors, and also with respect to cumulative risk factor exposures over time.
The Central Illustration and Table 2 show the results of the propensity score-matched comparison of overall CVD risk and risk of its components, CHD, CHF, and stroke, in Black men compared to White men and in Black women compared to White women. After propensity score matching for all known covariates of risk, Black men had a higher risk of CVD (HR: 2.30; 95% CI: 1.36-3.89; P = 0.0014), stroke (HR: 5.00; 95% CI: 1.45-17.3; P = 0.0047), and CHF (HR: 3.60; 95% CI: 1.34-9.70; P = 0.0067) than White men. Following propensity score matching for known CVD risk factors, compared to White women, Black women had a significantly higher CVD risk (HR: 2.36; 95% CI: 1.17-4.78; P = 0.014) and stroke risk (HR: 2.80; 95% CI: 1.01-7.77; P = 0.039), and a borderline significantly higher CHF risk (HR: 3.50; 95% CI: 0.73-16.9; P = 0.096). The CHD incidences for Blacks and Whites were not statistically significantly different for either men or women after propensity score matching.
Central Illustration.
Cardiovascular Disease Event Rates for Matched White and Black Men and Women
(A) Cardiovascular disease; (B) coronary heart disease; (C) congestive heart failure; and (D) stroke.
Table 2.
Association of Race With CVD, CHD, CHF, and Stroke Using Stratified Univariable Cox Models
| Men (n = 974) |
Women (n = 1,280) |
|||
|---|---|---|---|---|
| Black vs White HR (95% CI) |
P Value | Black vs White HR (95% CI) |
P Value | |
| CVD | 2.30 (1.36-3.89) | 0.0014 | 2.36 (1.17-4.78) | 0.014 |
| CHD | 1.40 (0.72-2.72) | 0.32 | 1.60 (0.52-4.89) | 0.41 |
| CHF | 3.60 (1.34-9.70) | 0.0067 | 3.50 (0.73-16.9) | 0.096 |
| Stroke | 5.00 (1.45-17.3) | 0.0047 | 2.80 (1.01-7.77) | 0.039 |
CHD = coronary heart disease; CHF = congestive heart failure; CVD = cardiovascular disease.
Multivariable analyses by race
The above findings show that even after propensity score matching for known cardiovascular risk factors and cumulative exposures to multiple risk factors, there were racial differences in CVD outcomes in White vs Black participants. In order to better understand these differences, we performed multivariable analyses to assess for differences in the impact of known CVD risk factors in all Black vs White study participants of each sex. These analyses, described below, suggest substantial racial heterogeneity in the associations of various risk factors with CVD events.
Cardiovascular disease
Table 3 shows the results of Cox models for the association of risk factors with time to first CVD event in men, demonstrating significant differences for Black vs White men. While cumulative LDL-C exposure was significantly and independently associated with incident CVD in White men, it was significant only in univariable analysis of Black men, suggesting the greater relative importance of other risk factors in the latter. In contrast, cumulative MAP was significantly and independently associated with incident CVD in Black men but achieved significance only on univariable analysis in White men, suggesting that other risk factors were more important than cumulative MAP exposure in White men. The results for Black vs White men are fully tabulated in Table 3.
Table 3.
Multivariable Cox Model for Time-to-CVD Event as a Function of Cumulative Exposure and Risk Factors at Year 40 for Men
| Black (n = 1,057) |
White (n = 1,118) |
|||||||
|---|---|---|---|---|---|---|---|---|
| Univariable Model |
Multivariable Model |
Univariable Model |
Multivariable Model |
|||||
| HR (95% CI) | P Value | HR (95% CI) | P Value | HR (95% CI) | P Value | HR (95% CI) | P Value | |
| Cumulative exposure: AUC (18-40 y) | ||||||||
| LDL-C AUC[18-40] (per 100 mg/dL ×y) | 1.044 (1.015-1.073) | 0.003 | 1.025 (0.995-1.056) | 0.107 | 1.122 (1.089-1.157) | <0.001 | 1.097 (1.061-1.136) | <0.001 |
| HDL-C AUC[18-40] (per 100 mg/dL ×y) | 0.925 (0.851-1.005) | 0.066 | - | - | 0.757 (0.672-0.853) | <0.001 | 0.852 (0.734-0.989) | 0.036 |
| log (TG) AUC[18-40] (mg/dL ×y) | 1.048 (1.023-1.074) | <0.001 | 1.027 (1.000-1.055) | 0.052 | 1.062 (1.041-1.083) | <0.001 | 1.014 (0.986-1.044) | 0.331 |
| MAP AUC[18-40] (per 100 mm Hg×y) | 1.426 (1.256-1.620) | <0.001 | 1.339 (1.171-1.531) | <0.001 | 1.339 (1.155-1.552) | <0.001 | - | - |
| PP AUC[18-40] (per 100 mm Hg×y) | 1.137 (0.967-1.337) | 0.12 | - | - | 0.940 (0.773-1.144) | 0.537 | - | - |
| Predictors at age 40 y | ||||||||
| MAP | 1.037 (1.022-1.052) | <0.001 | - | - | 1.046 (1.026-1.065) | <0.001 | 1.020 (0.999-1.040) | 0.057 |
| PP | 1.029 (1.011-1.047) | 0.002 | 1.025 (1.007-1.043) | 0.007 | 0.998 (0.975-1.022) | 0.889 | - | - |
| log (fasting glucose) | 4.013 (1.982-8.127) | <0.001 | 1.904 (0.860-4.215) | 0.112 | 7.615 (3.144-18.44) | <0.001 | 3.358 (1.157-9.751) | 0.026 |
| Ever smoked | 1.332 (0.908-1.954) | 0.142 | - | - | 1.411 (0.948-2.100) | 0.089 | - | - |
| Never attend college (grade 12 or less) | 1.016 (0.686-1.505) | 0.938 | - | - | 2.103 (1.327-3.335) | 0.002 | 1.572 (0.983-2.514) | 0.059 |
| Family history of CHD | 0.958 (0.620-1.481) | 0.847 | - | - | 2.707 (1.821-4.025) | <0.001 | 2.160 (1.442-3.236) | <0.001 |
Similar heterogeneity was seen in CVD risk factor impact by race in women (Table 4). In White women, cumulative LDL-C and cumulative TG exposures were independently associated with increased CVD risk whereas for Black women, there was only a weak trend. However, cumulative MAP exposure was independently associated with incident CVD in Black women but not White women. Table 4 provides a full tabulation of racial difference in CVD risk factors after adjusting for all other factors. Figure 2 displays these findings as a forest plot. A central insight is the heterogeneity of risk factor importance as a function of race (Tables 3 and 4, Figure 2). For the components of CVD, there were also important differences in risk factor impact as shown below.
Table 4.
Multivariable Cox Model for Time-to-CVD Event as a Function of Cumulative Exposure and Risk Factors at Year 40 for Women
| Black (n = 1,419) |
White (n = 1,278) |
|||||||
|---|---|---|---|---|---|---|---|---|
| Univariable Model |
Multivariable Model |
Univariable Model |
Multivariable Model |
|||||
| HR (95% CI) | P Value | HR (95% CI) | P Value | HR (95% CI) | P Value | HR (95% CI) | P Value | |
| Cumulative exposure: AUC[18-40] | ||||||||
| LDL-C AUC[18-40] (per 100 mg/dL ×y) | 1.018 (0.981-1.056) | 0.34 | - | - | 1.111 (1.055-1.170) | <0.001 | 1.065 (1.003-1.130) | 0.041 |
| HDL-C AUC[18-40] (per 100 mg/dL ×y) | 0.823 (0.747-0.908) | <0.001 | 0.895 (0.816-0.982) | 0.019 | 0.778 (0.667-0.907) | 0.001 | - | - |
| log (TG) AUC[18-40] (mg/dL ×y) | 1.067 (1.038-1.097) | <0.001 | - | - | 1.093 (1.053-1.135) | <0.001 | 1.061 (1.016-1.107) | 0.007 |
| MAP AUC[18-40] (per 100 mm Hg×y) | 1.564 (1.400-1.747) | <0.001 | 1.215 (1.009-1.463) | 0.040 | 1.290 (1.019-1.633) | 0.034 | - | - |
| PP AUC[18-40] (per 100 mm Hg×y) | 1.427 (1.217-1.673) | <0.001 | 1.182 (0.997-1.401) | 0.054 | 1.481 (1.101-1.991) | 0.009 | 1.270 (0.951-1.697) | 0.106 |
| Predictors at age 40 y | ||||||||
| MAP | 1.044 (1.033-1.054) | <0.001 | 1.020 (1.001-1.040) | 0.041 | 1.026 (0.994-1.060) | 0.108 | - | - |
| PP | 1.036 (1.019-1.052) | <0.001 | - | - | 1.034 (0.994-1.076) | 0.095 | - | - |
| log (fasting glucose) | 9.280 (5.165-16.67) | <0.001 | 5.965 (3.070-11.593) | <0.001 | 8.366 (1.827-38.31) | 0.006 | - | - |
| Ever smoked | 1.961 (1.295-2.968) | 0.001 | 1.818 (1.196-2.763) | 0.005 | 2.335 (1.162-4.690) | 0.017 | 1.757 (0.855-3.612) | 0.125 |
| Never attend college (grade 12 or less) | 1.661 (1.098-2.515) | 0.016 | - | - | 2.569 (1.279-5.161) | 0.008 | 1.775 (0.861-3.658) | 0.120 |
| Family history of CHD | 1.963 (1.308-2.946) | 0.001 | 1.649 (1.098-2.477) | 0.016 | 3.428 (1.819-6.460) | <0.001 | 2.891 (1.522-5.495) | 0.001 |
The C-statistic was 0.685 for the Black men and 0.781 for modeling the White men; the C-statistic was 0.751 for the black women and 0.761 for White women. For each continuous variable, the HR corresponds to HR per 1-U increment. Both fasting glucose and triglyceride were skewed and were log-transformed to improve normality. Their corresponding HRs were HRs per 1-unit increment in the log-transformed values.
Figure 2.
Forest Plots of Univariate and Multivariate Analysis Results for Cumulative Exposures to Multiple Risk Factors
HR and 95% CI are shown for whites and blacks of men (A) and women (B). AUC = area under the curve; CHD = coronary heart disease; CHF = congestive heart failure; HDL-C = high-density lipoprotein cholesterol; LDL-C = low-density lipoprotein cholesterol; MAP = mean arterial pressure; PP = pulse pressure; TG = triglyceride; other abbreviation as in Figure 1.
Coronary heart disease
Multivariable analysis for CHD in Blacks and Whites is shown in Supplemental Table 2A for men and Supplemental Table 2B for women. Cumulative LDL-C exposure was independently associated with incident CHD in both Black and White men. Cumulative MAP exposure was associated with incident CHD in Black men (HR: 1.28; 95% CI: 1.06-1.56; P = 0.011) but was not independently associated with CHD in White men. In White women, cumulative LDL-C exposure was independently related to incident CHD and there was a similar trend for Black women. On the other hand, cumulative MAP exposure was associated with incident CHD in Black women but not in White women. Never having attended college (grade 12 or less) was associated with increased risk of CHD in White but not in Black participants of either sex.
Congestive heart failure
Multivariable analysis of CHF in Black and White participants is shown in Supplemental Table 3A for men and in Supplemental Table 3B for women. In both Black and White men, MAP at 40 years was the most important clinical risk factor for CHF in the multivariable analysis. Interestingly, never having attended college (grade 12 or less) was associated with CHF in White men but not in Black men. In Black women, cumulative MAP exposure was associated with CHF risk, but only showed a trend in White women.
Stroke
Stroke risk in Black men (Supplemental Table 4A) was associated with cumulative HDL-C and MAP exposures, while in White men, stroke risk was correlated with cumulative LDL-C and MAP at age 40 years. PP at age 40 years was independently associated with stroke risk in White men but achieved only borderline significance in Black men (P = 0.074). Stroke risk in Black women was independently associated with MAP at age 40 years, fasting glucose level, smoking, and family CHD history, while in White women, only cumulative TG exposure and family history of CHD were independently associated with incident stroke risk (Supplemental Table 4B).
Discussion
This study extends prior studies of the association of self-identified race with the incidence of CVD by using propensity score matching to show that Black race is an independent risk factor for incident CVD, independent of other known risk factors, in Black men compared to White men and in Black women compared to White women. Multivariable analyses presented suggest that these differences result from racial differences in the relative importance of the various CVD risk factors in generating incident disease, especially for stroke and CHF. The underlying mechanism(s) of these differences is not known and would be an important topic for future investigation.
Our findings contrast with those in a recent paper by Shah et al that examined this question using multivariable regression analyses of the CARDIA database.11 Shah et al concluded that the higher risk for premature CVD in Black adults compared to White adults could be explained entirely by adjustment for the clinical, neighborhood, and socioeconomic risk factors. In comparison, our study used different design and statistical methods. Based on propensity score matching CARDIA participants in this study, which adjusted for known CVD risk variables previously determined to be important, we showed that race is an independent CVD risk factor. The advantages of comparing the 2 racial groups using propensity score matching employed in this paper, compared to the use of conventional multivariable analysis modeling, reside in the greater assurance of baseline comparability of the groups by accounting for heterogeneous response to risk factors and differing severity of the risk factors within the population and also by avoiding the limitations inherent in regression modeling.18,19
Multivariable analyses of CVD risk factors
Racial differences seen in the association of individual risk factors with incident CVD and with the incidence of its components, such as CHF and stroke, suggest that the racial differences in incidence result from differential associations of the same risk factors for Black vs White participants, mediated by mechanisms that remain to be understood. Specifically, there was substantial heterogeneity of risk factor importance depending on race and sex. In men, Black race was associated with large and significantly increased risk of CVD overall and a significantly increased risk of CHF and stroke. In women, the risks of CVD and also stroke were greater in Black participants than White participants, and there was a trend toward increased CHF risk in Black women. The risk for CHD was not different between Black vs White participants of either sex.
In general, elevated lipid fractions appeared to be more important for CVD risk in White men than Black men and in White women compared to Black women while the opposite was true for blood pressure. In Black men and women, blood pressure appears to be a more important risk factor than lipid fractions.
Other studies show racial differences in CVD disease incidence or treatment effect
A number of studies have made clear the presence of racial differences in CVD incidence and treatment response.2, 3, 4, 5, 6, 7, 8, 9, 10 Using multivariable analysis of CARDIA data to study the impact of cumulative risk factor exposure, Domanski et al found that Black race was an independent risk factor for incident CVD.2,3 Bibbins-Domingo et al reported that 26 of the 27 CARDIA subjects who developed early (prior to age 50 years) heart failure were Black.7,10 Feinstein et al reported that data from 3 large epidemiological trials showed that Blacks develop CVD at an earlier age than Whites, possibly attributable to elevated CVD risk factor levels.8
Racial differences in treatment response of Blacks vs Whites have also been demonstrated in studies of CVD treatment. Using the SOLVD (Studies of Left Ventricular Dysfunction) database, Exner et al showed that enalapril therapy reduced heart failure hospital admissions in White but not Black participants with left ventricular dysfunction.4,26 The BEST (Beta-Blocker Evaluation of Survival Trial) was a large, multicenter, randomized controlled trial of beta-blocker heart failure treatment with bucindolol.27 The study was stopped early at the recommendation of the Data and Safety Monitoring Board in the context of other beta-blocker trials showing benefit of beta-blocker treatment of heart failure.28, 29, 30 At the time of trial termination, the results showed a trend toward benefit of treatment with bucindolol that did not achieve statistical significance. However, Domanski et al showed that bucindolol showed a significant treatment benefit in the White participants (similar to other large beta-blocker trials, CIBIS-II (The Cardiac Insufficiency Bioprolol Study II), COPERNICUS (The Carvedilol Prospective Randomized Cumulative Survival trial), and MERIT-H (The Metoprolol CR/XL Randomized Intervention Trial in Congestive Heart Failure), that enrolled predominantly white patients28, 29, 30) whereas there was no benefit for Black subjects and, indeed, a point estimate for harm.5
Study Limitations
The most important limitation of this paper is that the participants studied were Black or White residents of the United States. Given the racial differences in risk factor impact between these 2 races, other racial groups, such as Asians, may have responses to risk factors different than Black and White participants in the CARDIA study.31 A general limitation of propensity score matching is the reduced sample size due to exclusion of individuals who cannot be matched, especially after accounting for the social-economic and neighborhood variables, representing ∼55% of CARDIA participants in this analysis.
Conclusions
Propensity score matching analyses presented in this paper demonstrate that self-identified race remains an independent risk factor for incident CVD and for the risk of incidence of its components, particularly CHF and stroke. Multivariable analyses presented suggest that the racial differences relate to a differing impact of risk factors in the different racial groups, driven by mechanisms that remain to be explained.
Perspectives.
COMPETENCY IN MEDICAL KNOWLEDGE: Recognition of race as an independent risk factor for incident CVD, potentially based on racial differences in the relative importance of CVD risk factors, importantly informs the design of primary prevention programs, primary prevention clinical trials, and personalized treatment plans for individual patients.
TRANSLATIONAL OUTLOOK: Further research should investigate the mechanisms underlying the differences found and also explore risk factor impact heterogeneity across other racial groups in the pursuit of refining population-based primary prevention programs and individualized primary prevention plans for patients.
Disclaimer
The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the NHLBI, the National Institutes of Health, or the U.S. Department of Health and Human Services.
Funding support and author disclosures
CARDIA is conducted and supported by National Heart, Lung, and Blood Institute (NHLBI) in collaboration with University of Alabama at Birmingham (75N92023D00002 & 75N92023D00005), Northwestern University (75N92023D00004), University of Minnesota (75N92023D00006), and Kaiser Foundation Research Institute (75N92023D00003). This paper has been reviewed by CARDIA for scientific content. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Acknowledgments
The authors thank the investigators, staff, and participants of the Coronary Artery Risk Development in Young Adults Study (CARDIA) study for their valuable contributions. The authors would also like to thank the reviewers and editors for many insightful comments and suggestions which greatly improved the presentation of this manuscript.
Footnotes
The authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’ institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information, visit the Author Center.
Appendix
For supplemental tables and material section, please see the online version of this paper.
Supplementary data
References
- 1.Vaduganathan M., Mensah G.A., Turco J.V., Fuster V., Roth G.A. The global burden of cardiovascular diseases and risk: a compass for future Health. J Am Coll Cardiol. 2022;80:2361–2371. doi: 10.1016/j.jacc.2022.11.005. [DOI] [PubMed] [Google Scholar]
- 2.Domanski M.J., Tian X., Wu C.O., et al. Time course of LDL cholesterol exposure and cardiovascular disease event risk. J Am Coll Cardiol. 2020;76:1507–1516. doi: 10.1016/j.jacc.2020.07.059. [DOI] [PubMed] [Google Scholar]
- 3.Domanski M.J., Wu C.O., Tian X., et al. Association of incident cardiovascular disease with time course and cumulative exposure to multiple risk factors. J Am Coll Cardiol. 2023;81:1151–1161. doi: 10.1016/j.jacc.2023.01.024. [DOI] [PubMed] [Google Scholar]
- 4.Exner D.V., Dries D.L., Domanski M.J., Cohn J.N. Lesser response to angiotensin-converting-enzyme inhibitor therapy in black as compared with white patients with left ventricular dysfunction. N Engl J Med. 2001;344:1351–1357. doi: 10.1056/NEJM200105033441802. [DOI] [PubMed] [Google Scholar]
- 5.Domanski M.J., Krause-Steinrauf H., Massie B.M., et al. A comparative analysis of the results from 4 trials of beta-blocker therapy for heart failure: BEST, CIBIS-II, MERIT-HF, and COPERNICUS. J Card Fail. 2003;9:354–363. doi: 10.1054/s1071-9164(03)00133-7. [DOI] [PubMed] [Google Scholar]
- 6.Cooper R., Cutler J., Desvigne-Nickens P., et al. Trends and disparities in coronary heart disease, stroke, and other cardiovascular diseases in the United States: findings of the national conference on cardiovascular disease prevention. Circulation. 2000;102:3137–3147. doi: 10.1161/01.cir.102.25.3137. [DOI] [PubMed] [Google Scholar]
- 7.Bibbins-Domingo K., Pletcher M.J., Lin F., et al. Racial differences in incident heart failure among young adults. N Engl J Med. 2009;360:1179–1190. doi: 10.1056/NEJMoa0807265. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Feinstein M., Ning H., Kang J., et al. Racial differences in risks for first cardiovascular events and noncardiovascular death: the atherosclerosis risk in communities study, the cardiovascular Health study, and the multi-ethnic study of atherosclerosis. Circulation. 2012;126:50–59. doi: 10.1161/CIRCULATIONAHA.111.057232. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Howard G., Safford M.M., Moy C.S., et al. Racial differences in the incidence of cardiovascular risk factors in older black and white adults. J Am Geriatr Soc. 2017;65:83–90. doi: 10.1111/jgs.14472. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lloyd-Jones D.M., Lewis C.E., Schreiner P.J., et al. The coronary artery risk development in young adults (CARDIA) study: JACC focus seminar 8/8. J Am Coll Cardiol. 2021;78:260–277. doi: 10.1016/j.jacc.2021.05.022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Shah N.S., Ning H., Petito L.C., et al. Associations of clinical and social risk factors with racial differences in premature cardiovascular disease. Circulation. 2022;146:201–210. doi: 10.1161/CIRCULATIONAHA.121.058311. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Inoue K., Seeman T.E., Horwich T., Budoff M.J., Watson K.E. Heterogeneity in the association between the presence of coronary artery calcium and cardiovascular events: a machine-learning approach in the MESA study. Circulation. 2023;147:132–141. doi: 10.1161/CIRCULATIONAHA.122.062626. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Friedman G.D., Cutter G.R., Donahue R.P., et al. CARDIA: study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol. 1988;41:1105–1116. doi: 10.1016/0895-4356(88)90080-7. [DOI] [PubMed] [Google Scholar]
- 14.Cutter G.R., Burke G.L., Dyer A.R., et al. Cardiovascular risk factors in young adults. The CARDIA baseline monograph. Control Clin Trials. 1991;12:1S–77S. doi: 10.1016/0197-2456(91)90002-4. [DOI] [PubMed] [Google Scholar]
- 15.Levey A.S., Stevens L.A., Schmid C.H., et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604–612. doi: 10.7326/0003-4819-150-9-200905050-00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Jacobs D.R., Jr., Yatsuya H., Hearst M.O., et al. Rate of decline of forced vital capacity predicts future arterial hypertension: the Coronary Artery Risk Development in Young Adults Study. Hypertension. 2012;59:219–225. doi: 10.1161/HYPERTENSIONAHA.111.184101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Park K., Gross M., Lee D.H., et al. Oxidative stress and insulin resistance: the coronary artery risk development in young adults study. Diabetes Care. 2009;32:1302–1307. doi: 10.2337/dc09-0259. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Dehejia R.H., Wahba S. Causal effects in nonexperimental studies: reevaluating the evaluation of training programs. J Am Stat Assoc. 1999;94:1053–1062. [Google Scholar]
- 19.Dehejia R.H., Wahba S. Propensity score-matching methods for nonexperimental causal studies. Rev Economics Statistics. 2002;84:151–161. [Google Scholar]
- 20.Austin P.C. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat. 2011;10:150–161. doi: 10.1002/pst.433. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Navas-Nacher E.L., Colangelo L., Beam C., Greenland P. Risk factors for coronary heart disease in men 18 to 39 years of age. Ann Intern Med. 2001;134:433–439. doi: 10.7326/0003-4819-134-6-200103200-00007. [DOI] [PubMed] [Google Scholar]
- 22.Gidding S.S., Robinson J. It is now time to focus on risk before age 40. J Am Coll Cardiol. 2019;74:342–345. doi: 10.1016/j.jacc.2019.04.064. [DOI] [PubMed] [Google Scholar]
- 23.Shapiro M.D., Bhatt D.L. “Cholesterol-Years” for ASCVD risk prediction and treatment. J Am Coll Cardiol. 2020;76:1517–1520. doi: 10.1016/j.jacc.2020.08.004. [DOI] [PubMed] [Google Scholar]
- 24.Lau B., Cole S.R., Gange S.J. Competing risk regression models for epidemiologic data. Am J Epidemiol. 2009;170:244–256. doi: 10.1093/aje/kwp107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Austin P.C., Putter H., Lee D.S., Steyerberg E.W. Estimation of the absolute risk of cardiovascular disease and other events: issues with the use of multiple fine-gray subdistribution hazard models. Circ Cardiovasc Qual Outcomes. 2022;15 doi: 10.1161/CIRCOUTCOMES.121.008368. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Exner D.V., Dries D.L., Waclawiw M.A., Shelton B., Domanski M.J. Beta-adrenergic blocking agent use and mortality in patients with asymptomatic and symptomatic left ventricular systolic dysfunction: a post hoc analysis of the Studies of Left Ventricular Dysfunction. J Am Coll Cardiol. 1999;33:916–923. doi: 10.1016/s0735-1097(98)00675-5. [DOI] [PubMed] [Google Scholar]
- 27.Eichhorn E.J., Domanski M.J., Krause-Steinrauf H., Bristow M.R., Lavori P.W. A trial of the beta-blocker bucindolol in patients with advanced chronic heart failure. N Engl J Med. 2001;344:1659–1667. doi: 10.1056/NEJM200105313442202. [DOI] [PubMed] [Google Scholar]
- 28.CIBIS-II Investigators and Committees The cardiac insufficiency bisoprolol study II (CIBIS-II): a randomised trial. Lancet. 1999;353:9–13. [PubMed] [Google Scholar]
- 29.Packer M., Coats A.J., Fowler M.B., et al. Effect of carvedilol on survival in severe chronic heart failure. N Engl J Med. 2001;344:1651–1658. doi: 10.1056/NEJM200105313442201. [DOI] [PubMed] [Google Scholar]
- 30.Hjalmarson A., Goldstein S., Fagerberg B., et al. Effects of controlled-release metoprolol on total mortality, hospitalizations, and well-being in patients with heart failure: the Metoprolol CR/XL Randomized Intervention Trial in congestive heart failure (MERIT-HF). MERIT-HF Study Group. JAMA. 2000;283:1295–1302. doi: 10.1001/jama.283.10.1295. [DOI] [PubMed] [Google Scholar]
- 31.Gasevic D., Ross E.S., Lear S.A. Ethnic differences in cardiovascular disease risk factors: a systematic Review of north American evidence. Can J Cardiol. 2015;31:1169–1179. doi: 10.1016/j.cjca.2015.06.017. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.