Prediction of coronary artery disease in patients with diabetes and albuminuria

Mathematical equations for the prediction of coronary artery disease (CAD) have been developed, the most well-known of which are from the Framingham Heart Study.¹ The Framingham Heart Study equations are based on several significant risk factors: low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, total cholesterol (TC), blood pressure or hypertension status, cigarette smoking, diabetes, and age. The equations are applicable to American Indian women and Hispanic and Puerto Rican men after proper adjustment for risk factor levels and incidence of CAD.²

Although historically rare among American Indians in the United States, heart disease has become the major cause of mortality in this population in recent years. The Strong Heart Study was begun in 1988 to study the prevalence and incidence of cardiovascular disease and its risk factors in American Indians, aged 45 to 74 years, among 13 tribes and communities in Arizona, Oklahoma, South Dakota, and North Dakota.³ From 1989 to 1992, a total of 4549 participants were interviewed and underwent a baseline examination. Data on cardiovascular disease, diabetes status, and potential risk factors were collected. Surviving participants were reexamined in 1993-1995 and again in 1998-1999. Documentation of mortality and morbidity, including medical chart review and abstraction, was ongoing, and cardiovascular disease events were ascertained for all participants. We used the longitudinal data collected to identify significant CAD risk factors, and these risk factors were used to develop CAD prediction equations.

Subjects and methods

The Strong Heart Study participants who were free of CAD (defined as definite CAD, definite myocardial infarction [MI], and definite electrocardiogram [ECG]-evident MI) and definite stroke at baseline examination were included in the study. The baseline characteristics of these participants and their CAD status at the end of 2001, after an average of 10 years of follow-up, were used to develop the CAD prediction equations. After the baseline examination, an annual mortality and morbidity evaluation by medical record review was conducted, and survivors of this cohort were reexamined in 1993-1995 and 1996-1999. In the morbidity and mortality surveillance, fatal and nonfatal CAD events (nonfatal definite MI, definite CAD, ECG-evident definite MI, fatal definite MI, fatal definite CAD, possible CAD [87% fatal], and sudden death due to CAD) were abstracted from medical records and confirmed by the Strong Heart Study Mortality and Morbidity Review Committees.^4,5 Time to event was defined as the time from the date of the baseline examination to the date of the first CAD event or the last follow-up.

Baseline data collected included age, height, weight, systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting TC level, HDL cholesterol level, LDL cholesterol level, plasma glucose level, and ECG results. We ascertained cigarette smoking status from a personal interview. A urine sample was also collected for albumin and creatinine measurements. Body mass index was computed, and hypertension status was determined based on the Seventh Report of the Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure guidelines⁶: normal (< 120/80 mm Hg and not taking antihypertensive medication); prehypertension (SBP between 120 and 139 mm Hg or DBP between 80 and 89 mm Hg and not taking antihypertensive medication); and hypertension (> 140/90 mm Hg and/or taking antihypertensive medication). Diabetes was defined as having a fasting plasma glucose > 126 mg/dL (6.99 mmol/dL) or receiving insulin or oral hyperglycemic treatment.⁷ The ratio of urinary albumin and creatinine was used to determine the level of albuminuria. A participant was considered to have microalbuminuria if the ratio was at least 30 but less than 300 and macroalbuminuria if the ratio was greater than or equal to 300.

The potential baseline risk factors considered were age, body mass index, SBP, DBP, use of antihypertensive medication, TC level, LDL cholesterol level, HDL cholesterol level, fibrinogen level, diabetes status, current smoking status, log (urinary albumin/creatinine), microalbuminuria, macroalbuminuria, and ECG-evident left ventricular hypertrophy (ECG-LVH). Sex-specific prediction equations were developed using the baseline risk factors that were identified by the Cox proportional hazards model⁸ as having significant predictive value for CAD events: use of antihypertensive medication, SBP, TC level, LDL cholesterol level, HDL cholesterol level, diabetes status, current smoking status, microalbuminuria, and macroalbuminuria. The general equation is:

(1) P(t, x₁, ... , x_p) = 1 - [S₀(t)]^{exp(Σb_i x_i)}

where P(t, x₁, ... , x_p) is the estimated probability of developing CAD during the period between baseline and t, S₀(t) is the estimated baseline "time-to-CAD" function for the time to CAD from the proportional hazards model, x_i is the measurement of the baseline risk factors, and b_i is the estimated coefficient from the proportional hazards model.⁹ The baseline function for the time to CAD was estimated using Breslow's method.¹⁰

The developed equations were internally validated using the bootstrapping method,^10,11 with samples of the same size as the original cohort (4372 participants) taken with replacement 1000 times from the original cohort. A version of the c statistic, similar to the area under a receiver operating characteristic curve, was used to evaluate the equations' ability to discriminate participants who develop CAD from those who do not. A c value of > 0.7 indicates good discrimination ability. In addition, calibration was assessed using a version of the Hosmer-Lemeshow chi-square statistic.¹²

Results

A total of 4549 participants were interviewed and examined at the baseline visit (1989-1991). Among them, 4372 (1722 men) had no history of CAD; at the end of 2001, however, 724 (349 men) had developed CAD. Univariate analyses showed that those who had developed CAD were significantly older and had a significantly higher average SBP, DBP (only in men), TC level, LDL cholesterol level, fibrinogen level, and log (urinary albumin/creatinine). They also were more likely to have diabetes, microalbuminuria, macroalbuminuria, and ECG-LVH (only in women), without a significant difference in current smoking status.

P

Cumulative incidence rates of CAD increased significantly ( < .005) with age, blood pressure, TC level, LDL cholesterol level, albuminuria level, and fibrinogen level in both men and women. As expected, incidence rates of CAD decreased with HDL cholesterol level in both sexes. Women with ECG-LVH at baseline examination had a significantly higher incidence of CAD. Diabetes was associated with a 2-fold higher CAD incidence in men and a 3-fold higher incidence in women compared with those without diabetes. Figures 1 and 2 show the cumulative incidence rates by diabetes status and albuminuria status.

Figure 1. Incidence of coronary artery disease (CAD) by diabetes status.

Figure 2. Incidence of coronary artery disease by albuminuria status.

All of the potential risk factors were included in the Cox proportional hazards model. Several models were considered using these risk factors in different forms, for example, treating age as a continuous variable or as a categorical variable and using categorical hypertension or continuous SBP and DBP measurements. We also included various interaction terms, for example, age × LDL cholesterol level and age × HDL cholesterol level. None of the interaction terms were significant in either men or women after age, SBP, DBP, LDL cholesterol, HDL cholesterol, use of antihypertensive medication, diabetes status, current smoking status, and albuminuria were included in the model. We identified the model that included age, SBP, LDL cholesterol, HDL cholesterol (all as continuous variables), antihypertensive medication, diabetes, current smoking status, microalbuminuria, and macroalbuminuria as the most appropriate model. Table 1 shows the estimated coefficients (b_i in Equation 1) and S₀(10), the estimated baseline time-to-CAD function evaluated at 10 years, of the prediction equations that we proposed for men and women. Using these values and the observed values of the risk factors (x_i), the estimated probability of developing CAD in 10 years can be obtained from Equation 1. For example, the exponent term in Equation 1 for women was:

(2)

Σb_ix_i = 0.0382 (age) + 0.3632 (if taking hypertension medication) + 0.0049 (SBP) + 0.0064 (LDL cholesterol) — 0.0122 (HDL cholesterol) + 0.7803 (if diabetes present) + 0.3624 (if current smoker) + 0.2693 (if microalbuminuria present) + 0.9645 (if macroalbuminuria present)

Substituting the above value of Σb_ix_i and S₀(10) = 0.9971 into Equation 1, the probability of developing CAD for a woman with the risk factors can be obtained. For example, from Equations 1 and 2, a 60-year-old woman who is a current smoker, has no diabetes or albuminuria, who is taking antihypertensive medication, and whose SBP, LDL cholesterol level, and HDL cholesterol level are 130 mm Hg, 120 mg/dL, and 55 mg/dL, respectively, has a 12% probability of developing CAD in 10 years. Similarly, using the estimated coefficients for men, the probability of developing CAD for a man with a given set of values for these risk factors can be obtained.

P

c

c

The Hosmer-Lemeshow chi-square statistics obtained for assessing the calibration of these equations were 7.18 and 7.25 (both nonsignificant = .45) for men and women, respectively, indicating good agreement between the observed and predicted number of CAD events over 10 years, or good calibration. The statistics were 0.71 for men and 0.73 for women, indicating good ability to discriminate participants who developed CAD from those who did not. The bootstrap-corrected statistics were 0.70 for men and 0.72 for women, indicating good internal validation.

c

Another set of prediction equations were developed using TC and HDL cholesterol instead of LDL cholesterol and HDL cholesterol, employing the estimated coefficients given in Table 1 (To see Table 1, please consult the print version of this paper). Probabilities of developing CAD in 10 years can be calculated following the same procedure described earlier. The statistics obtained for men and women were similar to the equations including LDL cholesterol and HDL cholesterol; calibration and internal validation, however, were moderate to good.

Discussion

Although rare in the past, cardiovascular disease has become the leading cause of death in the American Indian population in recent years.¹³ The CAD prediction equations presented in this article were derived using longitudinal data over an average of 10 years from men and women aged 45 to 74 years from 13 American Indian tribes and communities in Arizona, Oklahoma, South Dakota, and North Dakota. We believe that the equations are applicable to other American Indians of similar ages in the United States. Several methods were used to assess the discrimination ability, calibration, and internal validation of these equations, and the results were good.

Diabetes has been considered a significant risk factor for CAD; in the American Indian population, however, albuminuria was found to be another significant independent risk factor for CAD after adjusting for conventional risk factors, including diabetes, hypertension, and high cholesterol levels. Men and women with macroalbuminuria had almost a 3-fold and more than a 4-fold higher 10-year cumulative CAD incidence, respectively, than those without albuminuria. Strong Heart Study participants had a high prevalence and incidence of diabetes and its vascular complications, including albuminuria. It is likely that the proposed prediction equations will be applicable to other ethnic groups in which diabetes and albuminuria are highly prevalent.

Conclusions

Using data from the Strong Heart Study, we developed CAD prediction equations. The proposed equations can be used by physicians in patient education regarding CAD risk factors and probability of developing CAD. They can also be used to evaluate the efficacy of CAD prevention and intervention programs.