The Effect of Ethnicity on the Relationship Between Premature Coronary Artery Disease and Traditional Cardiac Risk Factors Among Uninsured Young Adults

Study Design and Setting

We conducted a retrospective cohort study of all patients 40 years or younger who underwent coronary angiography at Cook County Hospital between 1993 and 2001. Our large public hospital serves the uninsured, largely minority populations of Chicago.

Data Collection

We used standardized forms to collect data from medical records and electronic and angiographic databases. For missing angiographic reports, we conducted a blinded review of the angiograms (RFK). We collected data on sex, age, family history of first-degree relatives with definite CAD by age 55 years for men or by 65 years for women, hypertension (defined by current use of antihypertensive medication or history of blood pressure >140/95 mm Hg), diabetes mellitus (defined by diabetes therapy or elevated glycated hemoglobin A1c or elevated blood sugar levels), dyslipidemia (defined by lipid-lowering medication or total cholesterol >199 or low-density lipoprotein cholesterol >99), and history of prior myocardial infarction (MI), current smoking, or cocaine use. We calculated body mass index as weight in kilograms divided by the square of height in meters (kg/m²). Obesity was defined as body mass index ≥30 kg/m². Waist circumference could not be adjudicated because of the retrospective nature of this study.

Patients were classified as having definite CAD if there was evidence of an acute MI or angiographic evidence documenting ≥50% stenosis in at least 1 major epicardial coronary artery by visual inspection in ≥2 orthogonal views. We chose the less stringent criterion of 50% stenosis, rather than 70%, because 50% coronary stenosis for persons younger than 40 represents a clinically significant progression of atherosclerosis and portends a premature risk of adverse cardiac and vascular events. Criteria for MI included 2 of the following: angina, electrocardiographic changes, and appropriately elevated cardiac enzymes.

Data Analysis

We used chi-square, Fisher exact, and Mann–Whitney tests for bivariate associations between individual risk factors and our primary outcome, definite CAD. Ethnic group was a variable of special interest. We categorized patients into 4 groups: black/African American, Hispanic/Latino, white, and Asian Indian. We used logistic regression analysis with premature CAD as the dependent variable and the risk factors and 3 dummy variables for the ethnic groups as the independent variables. We also evaluated all 2-way interactions of ethnicity with each of the risk factor variables.

We next created 4 separate multivariable models of CAD risk representing the 4 ethnic groups using stepwise logistic regression analyses. Finally, we created a multivariable model for the entire study population using all available risk factors and all ethnic subgroups. In all multivariable models (the 4 ethnic group models and 1 overall model), we first evaluated for possible interactions by testing all 2-way interactions among potential predictor variables with a chunkwise test. If this chunkwise test was significant, we retained any interaction terms that met all of the following 3 criteria: (1) significance level of P<.05 when included in the final model, (2) prevalence of the combination of factors in at least 1% of the study patients, and (3) clinical and biological plausibility.^11,12

Model discrimination was assessed using the area under the receiver operating characteristic (ROC) curve. This demonstrates how well the model distinguishes between patients with and without CAD. To account for the common problem of model overfitting, which causes extreme and unreliable predictions, we used the bootstrap resampling method recommended by Harrell and colleagues¹² to shrink model coefficients. All analyses were conducted using Stata version 7 (StataCorp, College Station, TX).

Baseline Patient Characteristics

Among the 416 patients aged 40 years or younger who underwent coronary angiography during the 7-year study period, 136 (33%) had premature CAD. Those with CAD had a median of 1 involved coronary vessel (interquartile range, 1–2) and 42 (31%) had documented left main or multivessel disease. The typical patient was a black/African American man in his mid-30s who smoked and had hypertension (Table I). Obesity was also common; present in 36% of the sample. Because 16 of the patients could not be classified into 1 of the 4 ethnic groups, our analysis sample included 400 young adults.

The most common reason for angiography was evaluation for ischemic heart disease (stable angina and acute coronary syndromes). A surprisingly large number of patients had an acute ST-elevation MI as the reason for angiography (86 patients) and another 82 patients had non–ST-elevation MI or unstable angina as the reason (Table I). As expected, the prevalence of CAD was highest among patients presenting with an acute coronary syndrome; the prevalence of CAD was only 15% for patients presenting with a history of stable angina.

Prevalence of Risk Factors by Ethnic Groups

Our ethnically diverse population permitted the analysis of the distribution of risk factors across the 4 ethnic groups (Figure 1). Black patients had a higher prevalence of hypertension, smoking, and cocaine use. Hispanic patients had the highest prevalence of diabetes, dyslipidemia, and family history of CAD. White patients had the highest prevalence of both obesity and smoking but the lowest prevalence of diabetes. Asian Indians had the highest prevalence of male sex but the lowest prevalence of hypertension, dyslipidemia, smoking, and cocaine use.

Ethnic Groups and CAD

The risk of premature CAD was lower among Hispanic (20%) and black (30%) patients and higher among white (50%) and Asian Indian (50%) patients (Figure 1).

Risk Factors, Ethnicity, and CAD

Table II describes the associations of individual risk factors with CAD by ethnic group. Dyslipidemia, family history, diabetes, and smoking were strong risk factors among blacks and Hispanics; smoking was the strongest risk factor among whites; and male sex and hypertension were the strongest risk factors in Hispanics (Table II). Dyslipidemia, smoking, sex, and family history of CAD had statistically significant interactions with ethnicity (Table II and Figure 2). Dyslipidemia and family history of CAD were stronger risk factors for premature CAD in blacks and Hispanics, while male sex and smoking were stronger risk factors for premature CAD in Hispanics and whites.

Multivariable Models of Premature CAD in All Ethnic Groups

The independent association of risk factors with premature CAD in the 4 ethnic groups is shown in Table III. Among black patients, we identified a statistically significant interaction of smoking and sex. In black women, there was a strong association between smoking and premature CAD (odds ratio [OR], 5.2; P=.004), but among black men, the association was significantly weaker (OR, 1.3; P=.45). Among Hispanics, dyslipidemia, family history of CAD, and male sex were independently associated with premature CAD, while smoking was the only independent risk factor among whites. Among Asian Indian patients, we identified no independent predictor of premature CAD, perhaps due to the small sample size (n=28).

Multivariable Model of Premature CAD in the Overall Sample

In the multivariable model for the overall sample (Table IV), 4 of the risk factors—dyslipidemia, diabetes, male sex, and smoking—were all strong independent predictors of premature CAD, with ORs from 2 to 5. On the other hand, hypertension, obesity, and cocaine use were not associated with coronary disease in this cohort. Dyslipidemia was the strongest independent risk factor, increasing the odds of premature CAD by a factor of 4. Diabetes (OR, 2.8) and smoking (OR, 2.4) were also important independent predictors (Table IV). Ethnic group remained an important predictor after adjusting for all other traditional risk factors in the model. The pattern of risk among the 4 ethnic groups observed in bivariate analysis was preserved in the multivariable model: Hispanic and black patients were at lowest risk of premature CAD, compared with whites and Asian Indians, controlling for all other measured risk factors. The multivariable model was able to accurately discriminate between patients with and without premature CAD; the area under the ROC curve, a measure of discrimination, was 0.76. The model had good calibration, which was tested with the Hosmer–Lemeshow goodness-of-fit test (chi-square, 4.09; P=.25).

We identified 3 important interactions among the risk factors that satisfied the criterion of statistical significance. For 2 of them—the interaction between smoking and sex and the interaction between smoking and dyslipidemia—the patterns were similar: smoking had an important effect in the absence of other risk factors, but in the presence of dyslipidemia or male sex, smoking had less effect. Despite their statistical significance, we felt that these 2 interactions were not plausible and therefore were dropped from the final model. Another interaction—between dyslipidemia and ethnic group—did meet criteria for statistical significance and biologic plausibility. In the presence of dyslipidemia, blacks and Hispanics had a much greater risk of premature CAD than whites or Asian Indians. However, there were only 4 patients in the Indian group who had dyslipidemia and, therefore, this interaction was also not included in the model because it did not meet our 1% prevalence threshold. Figure 3 displays predicted risk of premature CAD by all combinations of the 4 important risk factors, illustrating the relative importance of dyslipidemia, diabetes, smoking, and ethnicity.

Modifiable Risk Factors

The prevalence of modifiable risk factors among our study patients was twice as high as the prevalence reported in the other major study on premature CAD¹: diabetes, 28% vs 10%; smoking, 65% vs 35%; and hypertension, 57% vs 29%. In contrast, the prevalence of a positive family history of early CAD—a nonmodifiable risk factor—was less common in our cohort, 35% vs 63%. Thus, the most important risk factors for premature CAD among our uninsured patients are amenable to medical and lifestyle interventions. However, we cannot say whether the differences reported above are due to differences in demographic or socioeconomic characteristics of the 2 patient populations, because Cole and colleagues¹ do not describe their patient population in enough detail.

Ethnic Groups and Their Unique Risk Factor Profiles

Black patients had a very high prevalence of hypertension, smoking, and cocaine use, but only dyslipidemia, diabetes, and smoking were independently associated with premature CAD among blacks. This emphasizes that among blacks, all risk factors associated with premature CAD are modifiable with lifestyle changes and medications (eg, statins and metformin). In contrast, among Hispanic patients, only male sex, family history, and dyslipidemia were independent risk predictors, suggesting that since there are fewer avenues for modifying risk, an aggressive approach to identifying and treating dyslipidemia might be justified. For uninsured white patients, our analyses identified smoking as the most important target for preventing premature CAD. Thus, our study suggests that there might be “risk-factor profiles” unique to each ethnic group, which might have implications for more targeted ethnic-specific interventions.

Race/Ethnicity and Risk of CAD

Although black and Hispanic patients in our study had less premature CAD compared with white and Indian patients, the reasons for these differences are uncertain. In most other studies, black patients have a disproportionately higher cardiovascular mortality rate and higher rates of hypertension, diabetes, carotid disease, and left ventricular hypertrophy.¹³ But other studies found that among patients undergoing coronary angiography, anatomic disease is less extensive among blacks.^14–17 These differences among ethnic groups may disappear at older ages.¹³ Unfortunately, documented ethnic disparities in access to coronary angiography complicate interpretation.

Dyslipidemia in Blacks and Hispanics

In our exploratory analyses, dyslipidemia appeared to be a more powerful risk factor for premature CAD among black and Hispanic patients compared with whites or Indians. Although intriguing, there were too few patients in some race subgroups for the results to be convincing. This interaction has also been identified in other studies.^18,19 We can only speculate that lipid-lowering therapy might have greater absolute benefit among high-risk black and Hispanic patients.¹⁹

Cocaine Use as a Risk Factor

Cocaine use in this study was not associated with a greater risk of premature CAD in adjusted or unadjusted analyses (Table II and Table III). While cocaine use may lead to advanced atherosclerosis, its predominant adverse cardiovascular effects are coronary vasospasm, acute coronary thrombosis, and sympathetic overdrive, which can result in acute coronary syndromes.^20,21 In our study, among 82 patients with unstable angina or non–ST-elevation MI, patients who used cocaine were less likely to have coronary stenosis compared with those who did not use cocaine (37% vs 56%), implying that cocaine use was not a risk factor for atherosclerosis at this age. Measurement error is another potential reason why cocaine use was not a significant predictor. Measuring cocaine use from chart reviews is likely to be very specific but not sensitive. Such misclassification would bias results towards the null.

Limitations

This study has several limitations. First, this was a retrospective study. Therefore the quality and completeness of data collection may not be optimal. Second, the study was conducted at a single urban public hospital, which raises the question of the generalizability of findings. However, two aspects of the sample are noteworthy: the patients were a consecutive series and they represented an important segment of the population that is not often included in large studies of CAD. Thus, this study addresses a gap in our knowledge about the relationships between risk factors and premature CAD among ethnically diverse, uninsured, and minority populations.

A third limitation is that the risk factor profile of patients might have influenced the decision to recommend angiography. If so, then the prevalence of risk factors will be falsely elevated (verification or workup bias). Nevertheless, the ORs assessing the relationship between risk factors and disease should not be biased because the same selection factors would be operating with equal force in patients with and without disease.

Fourth, our measures of risk factors are crude dichotomies based on history, examination, and simple laboratory tests. Because the data on risk factors were recorded in the medical records before angiography, misclassification is likely to be nondifferential and thus only underestimate the true magnitude of risk. However, when imperfect measures are used to adjust for confounding, the adjustments are also imperfect, causing “residual confounding,” which can bias results in either direction—toward or away from the null.²²

Finally, overoptimistic estimates of predictive performance are a common problem in logistic regression models.^11,12 This problem is even more pronounced in small data sets. Since premature CAD is a rare disease, the problem of a small sample size may be hard to overcome. In this study, we attempted to mitigate this problem by using a rigorous modeling strategy that shrank β coefficients (Table III) in order to prevent overoptimistic and unreliable estimates of the strength of associations.^11,12