Prediction of Coronary Heart Disease in a Population With High Prevalence of Diabetes and Albuminuria
The Strong Heart Study
Background— The present article presents equations for the prediction of coronary heart disease (CHD) in a population with high rates of diabetes and albuminuria, derived from data collected in the Strong Heart Study, a longitudinal study of cardiovascular disease in 13 American Indian tribes and communities in Arizona, North and South Dakota, and Oklahoma.
Methods and Results— Participants of the Strong Heart Study were examined initially in 1989–1991 and were monitored with additional examinations and mortality and morbidity surveillance. CHD outcome data through December 2001 showed that age, gender, total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein cholesterol, smoking, diabetes, hypertension, and albuminuria were significant CHD risk factors. Hazard ratios for ages 65 to 75 years, hypertension, LDL cholesterol ≥160 mg/dL, diabetes, and macroalbuminuria were 2.58, 2.01, 2.44, 1.66, and 2.11 in men and 2.03, 1.69, 2.17, 2.26, and 2.69 in women, compared with ages 45 to 54 years, normal blood pressure, LDL cholesterol <100 mg/dL, no diabetes, and no albuminuria. Prediction equations for CHD and a risk calculator were derived by gender with the use of Cox proportional hazards model and the significant risk factors. The equations provided good discrimination ability, as indicated by a c statistic of 0.70 for men and 0.73 for women. Results from bootstrapping methods indicated good internal validation and calibration.
Conclusions— A “risk calculator” has been developed and placed on the Strong Heart Study Web site, which provides predicted risk of CHD in 10 years with input of these risk factors. This may be valuable for diverse populations with high rates of diabetes and albuminuria.
- American Indian
- coronary disease
- coronary heart disease
- prediction equation
- risk factors
Received October 5, 2005; revision received April 17, 2006; accepted May 1, 2006.
The prevalence and incidence of coronary heart disease (CHD) and its risk factors have been studied extensively in various populations. The most well-known study is the Framingham Heart Study (FHS), which was initiated in 1948. The FHS has developed mathematical equations to predict the risk of CHD1 on the basis of several of the clinically available risk factors: age, total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), blood pressure/hypertension, diabetes, and cigarette smoking. The equations have been used widely for CHD risk appraisal. A recent examination of the validity and transportability of the FHS equations to other ethnic populations found that the FHS equations performed well for predicting CHD events for white and black men and women. For Hispanic and Puerto Rican men and Native American women, the equations also perform well when adjustments for the different risk factor levels and different incidence of CHD are made.2
Clinical Perspective p 2905
Diabetes is highly prevalent in American Indians, and heart disease has increased rapidly in recent decades3 to become the leading cause of death in most American Indian populations.4 The Strong Heart Study (SHS),5 a longitudinal study of cardiovascular disease in American Indians, initiated in 1988, has collected data on CHD and potential risk factors since 1989. The data provide an excellent opportunity to develop equations to predict CHD risk in individuals without overt CHD on the basis of the levels of CHD risk factors at baseline examination. Our previous analyses showed that risk factors for CHD were similar to those identified in the FHS. However, in the SHS, previous publications showed that albuminuria, fibrinogen, and left ventricular hypertrophy determined by echocardiogram also had a strong association with CHD in the American Indian population.3
This article presents CHD prediction equations from the SHS by gender, with the use of different sets of significant risk factors identified in the study. These equations were developed on the basis of longitudinal data of the SHS during 1989–2001. The primary purpose is to provide physicians who attend American Indians with a tool to predict CHD risk in their patients. A “risk calculator” (available on the SHS Web site: http://strongheart.ouhsc.edu) has been developed for individuals to input their values of the risk factors and instantly obtain a probability of developing CHD in 10 years. This tool should have potential applicability to other populations with high prevalence of diabetes and albuminuria.
Study Population and Variables Included
The SHS examined a total of 4549 American Indian men and women, aged 45 to 74 years, in 13 Indian tribes/communities in Arizona, North and South Dakota, and Oklahoma at its baseline examination (1989–1991). Of the 4549 participants, 4372 (1722 men and 2650 women) were free of CHD and definite stroke at the time of the baseline examination. Survivors of the cohort were reexamined in 1993–1995 and 1996–1999. Mortality and morbidity surveillance was also conducted annually. Data from the 4372 participants who were free of CHD and definite stroke at baseline examination were used to develop the prediction equations.
The baseline examination included a personal interview and a physical examination. Age and smoking status (yes or no) were obtained from the interview. The physical examination included measurements of height, weight, and sitting systolic (SBP) and diastolic blood pressure (DBP). Body mass index was calculated from height and weight (kg/m2). Hypertension was classified into 3 categories according to the Seventh Report of the Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure criteria6: normal (SBP <120 mm Hg and DBP <80 mm Hg, not on antihypertensive medication), prehypertension (SBP 120 to 139 mm Hg or DBP 80 to 89 mm Hg, not on antihypertensive medication), and hypertension (SBP ≥140 mm Hg or DBP ≥90 mm Hg and/or on antihypertensive medication). Blood was drawn after a 12-hour fast to measure TC, HDL-C, LDL-C, and fasting plasma glucose. A urine sample was taken to measure albumin and creatinine. Laboratory methods were published previously.5
Diabetes was defined with the use of the 1997 American Diabetes Association criteria,7 ie, fasting plasma glucose ≥126 mg/dL (6.99 mmol/L) or receiving insulin or oral hyperglycemic treatment. Albuminuria was determined with the use of the ratio of urinary albumin and creatinine: microalbuminuria if the ratio was ≥30 and <300 and macroalbuminuria if the ratio was ≥300. Fibrinogen was measured by the von Clauss method.8 An electrocardiogram (ECG) was obtained from every participant. The ECGs were read centrally by 3 cardiologists at the Fitzsimmons Medical Center in Denver.
In some analyses, age was classified into 3 groups: 45 to 54, 55 to 64, and 65 to 74 years; TC into 3 categories: <200, 200 to 239, ≥240 mg/dL (<5.18, 5.18 to 6.21, ≥6.22 mmol/L); HDL-C into 3 categories: <40, 40 to 59, and ≥60 mg/dL (<1.04, 1.04 to 1.54, and ≥1.55 mmol/L); and LDL-C into 4 categories: <100, 100 to 129, 130 to 159, and ≥160 mg/dL (<2.59, 2.59 to 3.35, 3.36 to 4.13, and ≥4.14 mmol/L).9 These analyses were intended to provide information on relative hazards among subcategories. In deriving the prediction equations, these risk factors were used as continuous rather than categorical variables.
Participants who had previous CHD (definite CHD, definite myocardial infarction [MI], definite ECG-evident MI) or definite stroke before or at the baseline examination were excluded. Those who were free of CHD and definite stroke at baseline examination were followed up. CHD events that occurred during the follow-up period were ascertained by annual mortality and morbidity surveillance or at the second and third examinations. In the annual surveillance, participants were contacted to determine their vital status and, if living, whether any cardiovascular events of interest had occurred since last contact. Medical records were abstracted and CHD death and events were ascertained and confirmed by mortality and morbidity review committees using specific criteria.10 Data used in this report included fatal and nonfatal CHD events occurring in the 12-year follow-up period from baseline examination through December 2001. The completion rates for the follow-up of mortality and morbidity events were 99.8% and 99.2%, respectively. These events were nonfatal definite MI, definite CHD, ECG-evident definite MI, fatal definite MI, definite CHD, possible CHD (87% fatal), and sudden death due to CHD. Detailed definitions of the fatal and nonfatal events were described previously.4,11 Time to event was calculated from the date of baseline examination to the date of CHD event or last follow-up. If >1 CHD event occurred in the same individual, the earliest date was used as the date of event in calculating the observed time to event.
Development of Prediction Equations
First, we examined the baseline characteristics of the 4372 participants by gender and CHD status at the end of 2001. Descriptive statistics were obtained and compared by χ2 test or independent t test between participants who developed CHD during the average 10-year follow-up and those who did not. Ten-year incidence rates of CHD in subgroups of the variables by gender were also examined along with the distributions of CHD event–free time, which provided preliminary information on the significant risk factors. The log-rank test12 was used to compare the distributions of time to CHD. Consequently, Cox proportional hazards models13 were fitted to the data with categories of all the significant risk factors to provide information of the hazard ratios between subcategories.
Because CHD risk differs between men and women, prediction equations were developed by gender. We first considered the significant variables that are commonly measured in the clinic, ie, age, blood pressure, LDL-C, HDL-C, diabetes status, albuminuria, and cigarette smoking status. We also considered fibrinogen and ECG-evident left ventricular hypertrophy (ECG-LVH). Because fibrinogen is not assessed routinely in most clinics and the ECG is not a routine screening procedure, these variables are of limited use in clinical prediction of CHD. We therefore will only report briefly the findings and will not present any prediction models that include ECG-LVH or fibrinogen. The second set of models was similar to the first except that TC and HDL-C were used instead of LDL-C and HDL-C.
The Cox proportional hazards model with the significant risk factors (or covariates) was used to develop the prediction equations. Age, SBP, TC, LDL-C, and HDL-C were treated as continuous variables in the equations. With the use of the estimated coefficients, an equation that estimates the probability of developing CHD in a certain number of years was derived. The mathematical details are given in the online-only Data Supplement. The estimated probabilities of developing CHD in 10 years based on this equation are provided on the SHS Web site (http://strongheart.ouhsc.edu).
Discrimination, Calibration, and Validation of the Prediction Equations
We used methods that account for the CHD-free time to assess calibration and discrimination. To assess the ability of the prediction equations to discriminate patients who develop CHD from those who do not, we used a version of the c statistic, which was calculated on the basis of all usable pairs of participants.14 Analogous to the area under the receiver operating characteristic curve, c represents an estimate of the probability that the equation assigns a higher risk to participants who develop CHD early in the follow-up period (10 years) than to those who develop CHD late or never develop the disease in the follow-up period. A c value of ≥0.7 indicates good discrimination ability, and the closer the c value is to 1.0, the better is the discrimination ability. The performance of the developed equations was also assessed for calibration by the use of a version of the Hosmer-Lemeshow statistic.14 Participants were divided into deciles according to their predicted risk of CHD in 10 years with the use of the proposed equations, and the Hosmer-Lemeshow statistic was calculated to compare the differences between the predicted and actual proportions of CHD events. Values of such statistics <20 are considered good calibration. We used D(orig) to denote the c statistic or the probability value for the Hosmer-Lemeshow statistic obtained by applying the developed prediction equation to the original cohort that was used to derive the equation.
In addition, the generated equations were validated internally with the use of bootstrapping methods.14,15 Samples of the same size as the original cohort (n=4273) were taken with replacement 1000 times from the original cohort. The procedure was to select 4273 individuals from the original cohort 1 at a time with replacement. This procedure was then repeated 1000 times. For each of these 1000 samples, the following analyses were performed: (1) the proportional hazards model was fitted, coefficients were estimated, and a prediction equation was generated; (2) the c statistic and the probability value for the Hosmer-Lemeshow statistic were calculated [denoted by D(boot)]; and (3) the prediction equation generated in analysis 1 was then applied to the original cohort, and a c statistic and the probability value for the Hosmer-Lemeshow statistic were obtained. Let D(apply) denote the c statistic or the probability value obtained in analysis 3. The bootstrap-corrected performance of the original CHD prediction equation was assessed by the difference between D(orig) and the average of the differences, D(boot)−D(apply), from the 1000 bootstrap samples. The difference [D(boot)−D(apply)] is called the “optimism” in the fit from the bootstrap samples.14 Let O denote the average of the 1000 optimism values, then D(orig)−O is an estimate of internal validation. The smaller the value of O is, the better is the validity of the developed prediction equation.
The authors had full access to the data and take full responsibility for its integrity. All authors have read and agree to the manuscript as written.
At the end of 2001, among the 4372 participants who were CHD free at the baseline examination, 724 (375 women and 349 men) had developed CHD. Table 1 gives summary statistics of several covariates at baseline by gender and CHD status at the end of 2001. In men, the participants who developed CHD were significantly older; had significantly higher average SBP, DBP, TC, LDL-C, fibrinogen, and log(urinary albumin/creatinine ratio); had higher prevalence of diabetes, macroalbuminuria, and microalbuminuria; and had significantly lower HDL-C. Results from women were similar except that the participants who developed CHD did not have significantly higher DBP but had higher prevalence of ECG-LVH.
Risk Factors and Incident CHD
Table 2 gives the number of CHD events, person-years of follow-up, and incidence per 1000 person-years by gender and subcategories of the covariates. All of the variables were significantly related to the CHD-free time except current cigarette smoking and ECG-LVH (in men only). CHD incidence increased with increasing age, blood pressure, TC, LDL-C, fibrinogen, severity of albuminuria, and decreasing HDL-C in both genders. Women and men with diabetes had 3 times and >2 times higher CHD incidence than those without, respectively.
To identify important predictive variables, age, body mass index, SBP, DBP, HDL-C, LDL-C, and fibrinogen were first treated as continuous variables in a Cox proportional hazards model with categorical variables: diabetes status, current smoking status, microalbuminuria and macroalbuminuria, and ECG-LVH. The stepwise variable selection procedure identified age, SBP, DBP, HDL-C, LDL-C, diabetes status, current smoking status, and albuminuria as significant risk factors for CHD. We also combined SBP, DBP, and antihypertensive treatment into a categorical variable “blood pressure” with 3 categories: normal, prehypertension, and hypertension. A proportional hazards model was fitted with all of the aforementioned significant risk factors as categorical variables. Table 3 gives the estimated coefficients, standard errors, probability values, and hazard ratios (relative risks) with their 95% confidence intervals (CIs) of these 7 risk factors. Women who were 65 to 74 years old, who had LDL-C ≥160 mg/dL, who had diabetes, and who had macroalbuminuria had >2 times higher risk of developing CHD in 10 years than their respective reference groups. Men who were between 65 and 74 years of age, who had hypertension, whose LDL-C was >160 mg/dL, or who had macroalbuminuria had >2-fold higher risk of developing CHD in 10 years.
Development of Risk Prediction Equations
A number of models were considered in the process of developing the most appropriate predictive equations, including models with various forms of the significant risk factors, for example, treating age as a continuous variable; replacing categorical hypertension with continuous SBP and DBP measurements; considering age, SBP, DBP, LDL-C, and HDL-C as continuous variables; and use of hypertension medication, diabetes status, current smoking status, microalbuminuria, and macroalbuminuria as categorical variables with and without interaction terms such as age×LDL-C, age×HDL-C, age×SBP, age×DBP, and ln(albumin/creatinine)×diabetes. None of the interaction terms were found to be significant in either men or women after having age, SBP, DBP, LDL-C, HDL-C (all as continuous variables), use of hypertension medication, diabetes status, current smoking status, microalbuminuria, and macroalbuminuria in the model.
The model that included age, SBP, LDL-C, HDL-C (all as continuous variables), use of hypertension medication, diabetes, smoking status, microalbuminuria, and macroalbuminuria was found to be most appropriate on the basis of the c statistic. Categorizing some or all of the continuous variables or including DBP in the model did not increase the c statistic by any meaningful amount. We therefore propose this model for practical use. The equation is given in the online Data Supplement.
Table 4 gives the estimated coefficients underlying the CHD prediction equations we propose for men and women. For example, the exponent term in Equation 1 in the online Data Supplement for men was as follows:
Σbixi=0.0492×age+0.1776 (if hypertension medication is used)+0.0094×SBP+0.0106×LDL-C−0.0166×HDL-C+0.5105 (if diabetes present)+0.3117 (if current smoker)+0.3087 (if microalbuminuria present)+0.6965 (if macroalbuminuria present)
For a continuous variable (or risk factor), exp(bi) represents the increase in risk of developing CHD corresponding to a 1-unit increase (if the sign is positive) or decrease (if the sign if negative) in the variable. For example, for 1-year increase in age, the risk of CHD for men increases 5.0% [exp(0.0492)=1.050], and for every 10-unit increase in HDL-C, the risk of CHD decreases 15.3% [exp(−0.166)=0.847]. The prediction equation for women was derived in a similar manner with the use of the corresponding regression coefficients in Table 4. As an example, for a 65-year-old male smoker, who has diabetes but no albuminuria, who is on hypertension medication, and whose SBP is 140 mm Hg, LDL-C is 140 mg/dL, and HDL-C is 35 mg/dL (3.63 and 0.91 mmol/L), the estimated probability of developing CHD in 10 years would be 57% (95% CI, 43% to 68%). On the other hand, a nonsmoking man without diabetes or albuminuria at the same age, whose SBP is 110 mm Hg, LDL-C is 110 mg/dL, and HDL-C is 60 mg/dL (2.85 and 1.55 mmol/L), would have a risk of only 11% (95% CI, 7% to 14%) to develop CHD in 10 years. Similarly, for women, a prediction equation was developed with the use of the estimated coefficients from the proportional hazards model given in Table 4.
The Hosmer-Lemeshow statistics yielded by these equations were 7.18 (P=0.45) and 7.25 (P=0.51) for men and women, respectively, indicating good agreement between the observed and predicted number of CHD events in 10 years or good calibration. The c statistics were 0.73 for women and 0.71 for men, indicating good discrimination ability. The internal validation results in which the bootstrapping method was used showed a bootstrap-corrected c statistic of 0.70 (with an O value of 0.0069) for men and 0.72 (O=0.0056) for women. The small O values indicate good internal validation. The probability value for the Hosmer-Lemeshow statistic was 0.60 (O=−0.147) for men and 0.71 (O=−0.198) for women. These Hosmer-Lemeshow probability values and c statistics indicated good calibration and discrimination ability.
To investigate whether fibrinogen and ECG-LVH warranted further consideration, we added these 2 variables to the proportional hazards model with the aforementioned 7 significant clinically available risk factors. It was found that, after adjustment for the clinical variables in the proposed equation, neither fibrinogen nor ECG-LVH was significant in either men or women. Significance levels for the other 7 covariates were similar to those given in Table 4.
Table 4 also gives the estimated coefficients in the Cox model when TC and HDL-C were used instead of LDL-C and HDL-C. Estimated probabilities of developing CHD in 10 years by gender can be calculated by following the same aforementioned procedure. The discrimination ability of these equations was similar to those equations in which LDL-C and HDL-C were used, as indicated by the c statistics (0.70 for men and 0.73 for women). However, calibration was moderate to good, as indicated by the Hosmer-Lemeshow χ2 statistics of 11.72 (P=0.164) for men and 16.11 (P=0.041) for women. The internal validation results in which the bootstrap method was used showed that the bootstrap-corrected c statistic was 0.68 (O=0.0069) for men and 0.73 (O=0.0063) for women, and the probability value for the Hosmer-Lemeshow statistic was 0.382 (O=−0.218) for men and 0.199 (O=−0.158) for women. Again, when fibrinogen and ECG-LVH were included as potential risk factors, neither was significant in either gender.
A major new finding in the SHS was that albuminuria is an independent significant risk factor for CHD in the American Indian population, with macroalbuminuria having a stronger effect than microalbuminuria. The effects of diabetes may be stronger in American Indian populations than in the general population. For example, when age, blood pressure, TC, HDL-C, diabetes, and smoking were used as covariates, the estimated regression coefficients for diabetes as a predictor of CHD in the SHS were 0.6825 for men and 1.0819 for women. These coefficients, adjusted for the other covariates, are >68% for men and 89% for women, larger than those obtained in the FHS (0.4055 and 0.5710, respectively).1 This was also reflected in the higher adjusted hazard ratio (or relative risk) associated with diabetes in the SHS, particularly in women (1.98 for men and 2.95 for women versus 1.50 and 1.77 in the FHS1). In men, high total cholesterol had a stronger effect on CHD in the SHS participants than in the FHS participants. The risk of developing CHD for men whose cholesterol levels were >240 mg/dL (6.22 mmol/L) was 2.7 times that of men whose cholesterol levels were <200 mg/dL (5.18 mmol/L), which was much larger than the 1.9-fold increase in the FHS. Although blood pressures seem to be less affected by obesity and hyperinsulinemia in American Indians compared with other populations,16 the effects of hypertension among the SHS participants were similar to those in the FHS.
Cardiovascular disease was rare among American Indians decades ago. However, in recent years, the rates of CHD in this population have exceeded those in other populations. The SHS participants who are members of 13 American Indian tribes in 3 of the regions with the largest American Indian populations are representative of the American Indian tribes in these regions. Therefore, we believe that the prediction equations developed with the use of the 10-year follow-up data from these SHS participants are reasonably applicable to a large proportion of the American Indian population.
We took advantage of the Internet to provide a risk calculator on the SHS Web site for convenient reference. Although Internet access may not be available to every American Indian, it is available to most, if not all, tribal offices and providers who deliver healthcare to American Indians. Users can log on to the SHS Web site and select the link to “risk calculator.” The risk calculator asks the user to enter the values or status of the risk factors. After these values are entered, an estimated probability of developing CHD in 10 years will be given immediately.
We understand that all risk schemes perform better in the cohort in which they are derived. The best assessments we could perform were with the use of the c statistic for discrimination ability, the Hosmer-Lemeshow statistic for calibration, and the bootstrapping method for internal validation (because there are no other similar longitudinal data of 10 years for us to validate our equations). All of the assessment results were reasonably good. The equations proposed in this article were derived with the use of data over an average of &10 years of follow-up from men and women aged 45 to 74 years, and therefore we cannot assess its utility in younger individuals. These equations may be applicable to other ethnic groups with similar propensity for diabetes. In this regard, several issues must be considered. A large percentage of the participants were either overweight (26% in men and 31% in women) or obese (36% in men and 41% in women), and the average level of physical activities was low (38% of men and 40% of women reported no physical activity during past week17). Obesity and physical activity were not included in the equation because they were not significant. An explanation is that obesity and inactivity may lead to diabetes. Thus, the effects of these proximal risk factors disappeared after adjustment for diabetes. Albuminuria was found to be a significant independent predictor of CHD in this population, after adjustment for conventional risk factors including diabetes. The probability of developing CHD increased considerably in participants with albuminuria, particularly macroalbuminuria. For a man who had macroalbuminuria, the probability of developing CHD is twice that of a man without albuminuria with exactly the same other risk factors, and in women the effect is even greater. Thus, the contribution of macroalbuminuria to CHD is crucial and cannot be ignored in this population. Albuminuria should be considered in other populations with a high rate of diabetes, and the proposed equations should have potential applicability to other populations with high prevalence of diabetes.
Because of the differences in the impact of these risk factors among American Indians, the proposed equations and the resulting individual projections have particular value for healthcare providers of this population. The equations will afford clinicians the opportunity to easily calculate an American Indian–specific CHD risk that can be used in patient and family education as well as in decision making about modifications of their risk intervention goals and therapies. For example, a high estimated risk may prompt the clinician to examine carefully all of the risk factors for the patient and modify or intensify the treatment plan for some risk factors that may have otherwise been considered less important. Indeed, providers within the Indian health system initially requested the development of these equations to better evaluate patient risk and intervene more appropriately. Healthcare providers in the American Indian health system plan to incorporate these equations broadly into the risk calculators within their cardiovascular disease management system, which is an electronic reminder system that automatically populates registries of levels of cardiovascular disease risk for specific prevention interventions as well as a patient information system for patient handouts related to risk modification opportunities. We hope that the equations and particularly the calculated probabilities are useful to healthcare providers of the American Indian population and that they will be considered for use in other similar populations.
The authors acknowledge the support, assistance, and cooperation of the Ak-Chin Tohono O’Odham (Papago)/Pima, Apache, Caddo, Cheyenne River Sioux, Comanche, Delaware, Fort Sill Apache, Gila River Pima/Maricopa, Kiowa, Oglala Sioux, Salt River Pima/Maricopa, Spirit Lake Sioux, and Wichita Indian tribes and communities and the participation of their members in the SHS. The authors also acknowledge the support and assistance of the Indian Health Service hospital and clinic at each center, field directors of the Strong Heart Study clinics and their staff, and the physicians who performed the mortality and morbidity reviews.
Sources of Funding
This study was supported by cooperative agreement grants (U01HL-41642, U01HL-41652, and U01HL-41654) from the National Heart, Lung, and Blood Institute.
Wilson PW, D’Agostino RB, Levy D, Belanger AM, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998; 97: 1837–1847.
Howard BV, Lee ET, Cowan LD, Devereux RB, Galloway JM, Go OT, Howard WJ, Rhoades ER, Robbins DC, Sievers ML, Welty TK. Rising tide of cardiovascular disease in American Indians: the Strong Heart Study. Circulation. 1999; 99: 2389–2395.
Lee ET, Cowan LD, Howard WJ, Sievers M, Welty TK, Wang W, Yeh JL, Rhoades ER, Devereux RB, Fabsitz RR, Go OT, Howard BV. All-cause mortality and cardiovascular disease mortality in three American Indian populations, aged 45–74 years, 1984–1988: the Strong Heart Study. Am J Epidemiol. 1998; 147: 995–1008.
Lee ET, Welty TK, Fabsitz R, Cowan LD, Lee N-A, Oopik AJ, Cucchiara AJ, Savage PJ, Howard BV. The Strong Heart Study: a study of cardiovascular disease in American Indians: design and methods. Am J Epidemiol. 1990; 132: 1141–1155.
Chobanian AV, Bakris GL, Black HR, Cushman WC, Green LA, Izzo JL Jr, Jones DW, Materson BJ, Oparil S, Wright JT Jr, Roccella EJ. The seventh report of the Joint National Committee on Prevention, Evaluation, and Treatment of high blood pressure: the JNC 7 report. JAMA. 2003; 289: 2560–2572.
Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes. 1997; 20: 1183–1197.
National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) final report. Circulation. 2002; 106: 3143–3421.
The National Heart, Lung and Blood Institute, The Strong Heart Study Coordinating Center. Strong Heart Study Operations Manual. Oklahoma City, Okla: University of Oklahoma Health Sciences Center; 1989.
Howard BV, Lee ET, Cowan LD, Fabsitz RR, Howard WJ, Oopik AJ, Robbins DC, Savage PJ, Yeh JL, Welty TK. Coronary heart disease prevalence and its relation to risk factors in American Indians: the Strong Heart Study. Am J Epidemiol. 1995; 142: 254–268.
Peto R, Peto J. Asymptotically efficient rank invariant procedures. J R Stat Soc. 1972; series A (135): 185–207.
Cox D. Regression models and life tables. J R Stat Soc. 1972; 34: 187–220.
Howard BV, Lee ET, Yeh JL, Go OT, Fabsitz RR, Devereux RB, Welty TK. Hypertension in adult American Indians: the Strong Heart Study. Hypertension. 1996; 28: 256–264.
Welty TK, Lee ET, Yeh J, Cowan LD, Go O, Fabsitz RR, Lee N-A, Oopik AJ, Robbins RC, Howard BV. Cardiovascular disease risk factors among American Indians: the Strong Heart Study. Am J Epidemiol. 1995; 142: 269–287.
The proposed coronary heart disease (CHD) prediction equations are derived from a large cohort of American Indians, aged 45 to 74 years, who participated in the Strong Heart Study and therefore are specifically applicable to middle-aged and senior American Indians. The American Indian population is known to have high prevalence of diabetes and albuminuria. Thus, the proposed equations, which include diabetes and albuminuria as major CHD predictive factors, have a high potential of applicability to other ethnic populations in which the prevalence proportions of diabetes and albuminuria are high. Clinicians can use the prediction equations as a patient education tool to advise the patients (and their families) how they can reduce the risk of CHD by modifying their health behaviors or lifestyles and consequently improve their risk factors. They can also teach their patients to monitor their own CHD risk by using the user-friendly online risk calculator. A high risk could prompt them to change their health behaviors, and a reduced risk would encourage them to continue the healthy habits or improve even more. The prediction equations provide an overall prediction on CHD outcomes based on the most important risk factors. They can serve as an aide to the clinicians in determining intervention goals and treatment strategies for their patients. For example, a high estimated risk may prompt the clinician to examine carefully all of the risk factors for the patient and modify or intensify the treatment plan for some risk factors that may have otherwise been considered less important.
The online-only Data Supplement, which contains an appendix, can be found at http://circ.ahajournals.org/cgi/content/full/CIRCULATIONAHA.105.593178/DC1.
The views expressed in this article are those of the authors and do not necessarily reflect those of the Indian Health Service.