Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Circ Cardiovasc Qual Outcomes. Author manuscript; available in PMC 2011 Mar 1.
Published in final edited form as:
PMCID: PMC2853913

Prediction of cardiovascular death in racial/ethnic minorities using Framingham risk factors



Risk factors for cardiovascular disease (CVD) derived from the Framingham study are widely used to guide preventive efforts. It remains unclear whether these risk factors predict CVD death in racial/ethnic minorities as well as they do in the predominately white Framingham cohorts.

Methods and Results

Using linked data from the National Health and Nutrition Examination Survey III (1988-1994) and the National Death Index, we developed Cox proportional-hazard models that predicted time to cardiovascular death separately for non-Hispanic white (NHW), non-Hispanic black (NHB) and Mexican American (MA) participants aged 40-80 with no previous CVD. We compared calibration and discrimination for the three racial/ethnic models. We also plotted predicted 10-year CVD mortality by age for the three racial/ethnic groups while holding other risk factors constant. 3437 NHW, 1854 NHB and 1834 MA subjects met inclusion criteria. Goodness of fit chi-square tests demonstrated adequate calibration for the three models (NHW, p=0.49; NHB, p=0.47; MA, p=0.55 and areas under the receiver operating characteristic curves demonstrated similar discrimination (c-statistics-NHW=0.8126; NHB=0.7679; and MA=0.7854). Older age was more strongly associated with CVD mortality in NHWs (HR-3.37, 95%CI-2.80, 4.05) than NHBs (HR-2.29, 95%CI-1.91, 2.75) and was intermediate in MAs (HR-2.46, 95%CI-1.95, 3.11). Predicted 10-year mortality was highest for NHBs across all age ranges and was higher for MAs than NHWs until late in the seventh decade.


Framingham risk factors predict CVD mortality equally well in NHWs, NHBs and MAs, but the strength of the association between individual risk factors and CVD mortality differs by race and ethnicity. When other risk factors are held constant, minority individuals are at higher risk of CVD mortality at younger ages than NHWs.

Keywords: Risk factors, cardiovascular diseases, prevention, epidemiology


Cardiovascular disease (CVD) is the leading cause of premature death in the U.S., both in aggregate and subpopulations of whites, blacks and Latinos.1 Since 1950, mortality from cardiovascular disease has fallen steadily,2 but the rate of decline has been greater in whites than in blacks.3, 4 Trends in cardiovascular mortality for Latinos have yet to be thoroughly assessed. In 2001, the proportion of premature deaths (<65 years-old) due to heart disease was highest in blacks (31.5%) and was higher in Latinos (termed Hispanics) than whites (termed non-Hispanic whites) (23.3% vs. 14.4%).5 In 2001, death rates for diseases of the heart and stroke were higher among blacks than whites.6 Since more than half of the decline in mortality has been attributed to adequate treatment of risk factors,7 one explanation for the racial and ethnic difference in trends in cardiovascular outcomes might be differences in prevalence and treatment of cardiovascular risk factors.

The identification of the major independent risk factors for cardiovascular disease - older age, male gender, smoking, diabetes, high total or low HDL cholesterol, and hypertension - is largely a product of the Framingham Heart Study.8, 9 The usefulness of these risk factors has primarily been documented in whites.8 Their value in other racial/ethnic groups may be affected by differences in prevalence and relative importance of these risk factors, or by unidentified risk factors that are specific to non-whites. CVD risk factors remain poorly defined in racial/ethnic minority populations. Although aggregated Framingham risk factors have been shown to underestimate risk in socioeconomically deprived populations in Britain10 and to overestimate risk in Chinese11 and Danish populations,12 limited data are available regarding the ability of the Framingham risk factors to accurately predict cardiovascular disease in African Americans and Latinos.13

This study was designed to assess the ability of Framingham risk factors to predict cardiovascular death in whites compared to Latinos and African Americans.


Data Sources

The data sources for the study were the Third National Health and Nutrition Examination Survey (NHANES III) database and the NHANES III Linked Mortality File. NHANES III is a cross-sectional survey conducted by the Centers for Disease Control and Prevention. During 1988-1994, a representative sample of the civilian non-institutionalized US population was recruited into NHANES III using a multistage, stratified sampling design.14 After an interview in the home, participants were invited to attend 1 of 3 examination sessions. NHANES III oversampled the elderly, non-Hispanic blacks and Mexican Americans. A detailed description of the NHANES III survey and sampling procedures is available elsewhere.14 The NHANES III Linked Mortality File contains the results of matching NHANES III subject identifiers with data available in the National Death Index as of December 31, 2006. Date of death and cause of death in the National Death Index are derived from death certificates.

Study Population

We studied subjects aged 40 to 80 to closely resemble the age range in the study which validated the Framingham global CVD risk tool.15 We excluded subjects who did not self-report being non-Hispanic white (NHW), non-Hispanic black (NHB), or Mexican American (MA). We excluded subjects with a previous myocardial infarction or stroke to eliminate the possible effect of treatment for prior CVD on the association between CVD risk factors and CVD mortality. Finally, we excluded subjects with missing risk factor data.


We chose as independent variables the risk factors identified in models developed by the Framingham study15, 16: gender, age, smoking, diabetes, elevated total cholesterol, low concentrations of high density lipoprotein cholesterol (HDL), systolic blood pressure (SBP) (treated or not). Smoking was defined as participants reporting that they were current smokers. Diabetes and antihypertensive use were defined by self-report. CVD mortality included death from coronary heart disease or cerebrovascular disease and was defined by the principal cause of death on the death certificate listed as ICD9 codes 390.0-459.9 for deaths before or during 1999 or ICD-10 codes I10-15, I20-25, I50-51, I60-69, and I70-73 for deaths after 1999. We chose to investigate overall cardiovascular mortality rather than limit the study to coronary heart disease mortality. Overall cardiovascular mortality includes death from stroke, a particularly important endpoint in minority groups, especially NHBs.

Statistical analysis

Descriptive statistics (means, standard deviations, proportions) were generated to describe the sample. Chi-square tests and one-way ANOVA were used to compare subjects included in the analysis with those excluded because of missing data, and, to compare racial/ethnic groups on sociodemographic and clinical variables. Age, SBP treated, SBP untreated, total cholesterol and HDL were treated as continuous variables. All continuous units were analyzed per 10 units of change (i.e. 10mmHg for SBP). We also naturally logarithmically transformed continuous variables to see if this improved discrimination and calibration of the models by minimizing the influence of extreme observations as in the Framingham model used as the basis for the current study.15 To maximize predictive ability, we developed Cox proportional hazards models with days to death (up to 10 years) from CVD as the dependent variable and each potential risk factor as independent variables separately for NHWs, NHBs and MAs. We checked the assumptions of the proportional hazards model for each variable and assessed for possible interactions between gender and the other risk factors. Subjects who did not complete a full 10-year period of observation or died of causes other than CVD were censored at the time of follow-up. To confirm differences in coefficients by race/ethnicity observed in separate analyses, we tested race/ethnicity by risk factor interactions in a stratified Cox proportional hazards model.

Within each racial/ethnic group (NHWs, NHBs and MAs), subjects were grouped into the following 5 risk categories: ≤ 25th percentile, 25-50th percentile, 51-75th percentile, 76-90th percentile, >90th percentile. These percentile groupings were selected instead of using deciles to provide sufficient event rates in all categories. To assess calibration, Kaplan-Meier survival curves within each racial/ethnic group were used to obtain 10-year mortality rates for each of the 5 strata. These were compared to predicted mortality for each racial/ethnic group by strata and a chi-square goodness-of-fit statistic was calculated. For this test, larger p-values indicate a better fit. Additionally, predicted mortality was compared to observed mortality for the standard risk categories used in clinical practice; <10%, 10-20% and > 20% risk of 10 year CVD mortality. To assess discrimination, we calculated the c-statistic from the Cox regression models for each racial/ethnic group using methods described previously.17, 18 The c-statistic is equivalent to the probability that the predicted risk is higher for a case than a non-case and has a maximum value of 1.19 Predicted 10-year CVD mortality by age for the three racial/ethnic groups was plotted holding the other risk factors fixed at the mean levels of these risk factors for the entire cohort.

All statistical analyses were performed using SAS version 9.2 (SAS Institute Inc., Cary, N.C.).


Exclusions from the study population are presented in Figure 1. The final sample included 3437 NHW, 1854 NHB, and 1834 MAs participants. Subjects included in the analysis were similar to subjects excluded in terms of HDL and SBP (p>0.05), but those excluded due to missing values were less likely to be male (43.4% vs. 47.2%, p=0.01), older (61.1 years vs. 57.7 years, p<.0001), more likely to smoke (27.1% vs. 24.2%, p=0.02), more likely to have diabetes (12.5% vs. 10.5%, p=0.03), and had higher total cholesterol (230.1 vs. 217.7, p<0.01).

Figure 1
Flow diagram of deriving study population.

Racial/ethnic group risk factors and CVD mortality are presented in Table 1.

Table 1
Cardiovascular risk factors and mortality by gender and race/ethnicity, NHANES III

Parameter estimates and hazard ratios from the Cox models for the three racial/ethnic groups are presented in Table 2--4.4. Models using continuous variables per 10 units fit as well as models using log-transformed continuous variables and therefore are presented here for ease of interpretation. Models using log-transformed continuous variables are available as supplemental material (Supplemental Tables 1-3). No significant gender interactions were found. Hazard ratios were generally of the expected magnitude and directions with two exceptions. First, the association between age and CVD mortality was stronger for NHWs (HR-3.37, 95%CI-2.80, 4.05) than for NHBs (HR-2.29, 95%CI-1.91, 2.75) and MAs HR (HR-2.46, 95%CI-1.95, 3.11). This was confirmed in the stratified Cox model that included a racial/ethnic group by age interaction effect that was significant for NHBs compared to NHWs (p=<.001) and MAs compared to NHWs (p=0.03)

Table 2
Hazard Ratios for 10 year CVD mortality for individual CVD risk factors for Non-Hispanic Whites.
Table 4
Hazard Ratios for 10 year CVD mortality for individual CVD risk factors for Mexican Americans.

Second, the associations between cholesterol (total and HDL) and CVD were notably weaker than expected. Stratified analysis revealed a significant racial/ethnic group by HDL interaction effect for MAs compared to NHWs (p<.01).

The calibration goodness-of-fit X2 statistics for CVD mortality for NHWs (X2 (4df) =3.44, p=0.49), NHBs X2 (4df) =3.56 (p=0.47), and MAs (X2 (4df) =3.04, p=0.55), indicate adequate fit for all racial/ethnic groups when models are developed separately for each group (See Figure 2a--c).c). Observed CVD mortality by commonly used risk categories is presented in Table 5. As observed in this table, stratification of risk using conventional AHA/ACC risk categories, were generally well calibrated; however, confidence intervals sometimes crossed the defined stratum boundaries.

Figure 2a
Predicted vs. observed 10 year CVD mortality in non-Hispanic whites
Figure 2c
Predicted vs. observed 10 year CVD mortality in Mexican-Americans
Table 5
Estimated vs. Observed 10 year CVD mortality using clinically relevant Framingham risk categories (using 10 unit increments)

C-statistics summarizing the ability of racial/ethnic prediction functions to discriminate between subjects who experienced CVD mortality from those who did not were 0.8126 for NHWs, 0.7854 for MAs and 0.7679 for NHBs.

Predicted 10-year CVD mortality across age ranges are presented in Figure 3. Predicted mortality for NHBs is highest across all age ranges. Predicted mortality is higher for MAs than NHWs until late in the seventh decade.

Figure 3
Predicted mortality by racial/ethnic group across age ranges.*


Principal findings

We assessed the ability of the well-known Framingham CVD risk factors (age, gender, systolic blood pressure, cholesterol, smoking and diabetes) to predict cardiovascular death separately in NHWs, NHBs, and MAs. Distinct racial/ethnic specific risk factor associations with CVD mortality were demonstrated; however, we found that survival models based on Framingham risk factors were well calibrated in all racial/ethnic groups when models were developed separately within each group. The discriminative capacity of our models was also similar for all three groups. Older age was more strongly associated with CVD mortality in NHWs than NHBs and MAs. With all other risk factors held constant minority participants were at higher risk for cardiovascular death at younger ages compared to white participants.

Previous studies of Framingham risk functions in U.S. racial/ethnic minorities

Our findings are similar to those of D'Agostino et al.13 This study compared calibration, recalibration and discrimination of gender specific Framingham risk models and risk models for coronary heart disease (CHD) in specific racial/ethnic cohorts for black men and women in the Atherosclerosis Risk in Communities Study (ARIC 1987-88) and for Hispanic men in the Puerto Rico Heart Health Program (1965-1968). 13 They found that actual CHD event rates were similar to predicted event rates predicted event rates for blacks (male X2(8df) =6.2, p=.62 and females X2(8df) =5.0, p=.76). The calibration for Puerto Rican men was poor (X2(8df) =142, p=<.001); however, recalibration of this model using Puerto Rico Heart Program cohort mean values for risk factors and coronary heart disease (CHD) incidence improved the performance of Framingham prediction functions (X2(8df)=7.2, p=.51). Discrimination of the models in this study (c-statistics=0.67-0.85) were comparable to our findings.

Similarly, Liao et al., 20 pooling data from NHANES I and II (1976-1985), assessed CHD risk functions based on Framingham risk factors in whites and African Americans (termed blacks). Applying risk equations derived from white men to African American men over predicted CHD mortality by 60%. Discrimination of CHD risk functions were similar to those found in our study and similar across racial/gender groups (c-statistic for white men=0.77, African American men=0.76, white women= 0.84 and African American women =0.82).

These previous studies differed from the current study in that they were restricted to an evaluation of CHD; we investigated overall CVD mortality in order to capture the importance of stroke in minority populations. Also, the previous studies used cohorts initiated in years prior to that used in our study. Enrollment in the Puerto Rico Heart Health Program, ARIC and NHANES 1 and II occurred one to two decades prior to NHANES III enrollment (1988-1994), and before the widespread use of therapies such as reperfusion for acute myocardial information and effective cholesterol reduction with statins. Additionally, our study is based on a more recent interpretation of the Framingham risk function15 in which systolic blood pressure, total and HDL cholesterol are treated as continuous variables, blood pressure treatment is considered, and HDL is considered separately from total cholesterol.

Previous studies of associations between cardiovascular risk factors and CVD mortality in racial/ethnic minorities

Few studies have evaluated the associations between risk factors and cardiovascular mortality by different racial/ethnic groups. In the study of D'agostino et al.,13 CHD mortality was associated with hypertension and elevated total cholesterol at varying levels, and diabetes in white, black and Puerto Rican men. CHD mortality was associated with hypertension for white and black women. Additionally, elevated total cholesterol, low HDL, diabetes and current smoking were associated with CHD mortality for black women. There was no association with HDL cholesterol and CHD mortality in the black and Puerto Rican men studied and there was no association with age in any of the women studied.

In the study of Liao et al.20, there were no consistent differences in the associations between risk factors and CHD mortality across black and white groups studied except there were significant differences in coefficients for smoking status among men and for age among women.

Most current understanding of the relationships between Framingham risk factors and CVD mortality in MAs stems from the San Antonio Heart Study. Wei et al21 found that diabetes and hypertension were independently associated with CVD mortality, but current smoking and total cholesterol >240mg/dl were not. None of these 4 variables were associated with CVD mortality in the non-Hispanic white comparison group. In another San Antonio Heart Study,22 age, gender, diabetes, hypertension (yes vs. no), smoking, total cholesterol and HDL were all significantly associated with CVD mortality. However, another study23 of the same population showed only a trend for low HDL to predict CVD mortality in individuals without known CVD (HR 1.46, 95%CI-0.94-2.27).

Potential explanation for current findings

We propose two potential explanations for the finding that minority individuals are at higher risk for CVD mortality at younger ages than NHWs. First, minority subjects may have received less effective care subsequent to baseline visits. Second, low socioeconomic position is common to US minorities and low socioeconomic position has been suspected to be a risk factor for poor health24 and has been shown to be a risk factor for cardiovascular disease in particular in other populations.10 Although the reasons underlying this association are not agreed upon, environmental factors such as chronic emotional stress and poor diet have been suggested.

With regard to the relationship between CVD mortality and total and HDL cholesterol, one potential explanation for the weakness of association between CVD mortality is that the inception of the NHANES III cohort coincided with the introduction of statin therapy. Widespread use of these agents during the follow-up period may have negated much of the impact of baseline hyperlipidemia on outcome. Individuals with nonfatal events may have been more likely to have their hyperlipidemia treated leading to lower death rates and weakening the relationship between lipids and CVD mortality. It is also possible that an association between cholesterol levels and outcomes would have been statistically significant had our sample size, particularly in minority subjects, been larger, however, often the relationships we found between cholesterol levels and CVD mortality were in the opposite than expected direction suggesting power was not the issue. Still, these findings are surprising and warrant further examination in contemporary cohorts.

Strength and Limitations

To our knowledge, this is the first study to examine the use of the Framingham risk factors in a study that includes large numbers of all three major racial/ethnic groups in the U.S. using a national dataset. Our study had several limitations, however. Our definition of cardiovascular mortality relies on death certificate diagnoses which are subject to error in the certification of the underlying causes of death. Due to the relatively few deaths in the racial/ethnic groups of interest, we were unable to perform gender specific analyses. We did, however, account for gender as a separate risk factor. Also, we were not able to measure nonfatal events because end points were determined solely from the National Death Index. An analysis based on non-fatal CVD events would be an improvement on this study, but such data is currently unavailable. By excluding non-fatal events we may have underestimated associations between risk factors and CVD. However, where a difference in predicted and observed outcomes has been found, it is similar in fatal and non-fatal events.25 Lastly, we were unable to validate our models in different cohorts.


Taken in aggregate, Framingham risk factors predicted cardiovascular mortality in nationally representative samples of NHBs and MAs as well as they did in NHWs. Individually, however, the risk factors carried different strengths of association with CVD mortality and prevalences of all risk factors differed across racial/ethnic groups. Holding other risk factors constant, estimated mortality from cardiovascular disease is higher for NHBs than NHWs across all age ranges and higher for MAs than NHWs until late in the seventh decade. The reasons for these differences deserve further investigation.

Figure 2b
Predicted vs. observed 10 year CVD mortality in non-Hispanic blacks
Table 3
Hazard Ratios for 10 year CVD mortality for individual CVD risk factors for Non-Hispanic Blacks.

Supplementary Material



Funding Sources: Funding for this work was supported by the NHLBI 5 U01 HL079208-05.


Conflict of Interest Disclosures:


This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


1. Heart Disease Facts and Statistics [May 20, 2008]. Available at: www.cdc.gov/heartdisease/facts.htm.
2. Decline in deaths from heart disease and stroke--United States, 1900-1999. MMWR Morb Mortal Wkly Rep. 1999;48:649–656. [PubMed]
3. Trends in ischemic heart disease death rates for blacks and whites--United States, 1981-1995. MMWR Morb Mortal Wkly Rep. 1998;47:945–949. [PubMed]
4. Cooper R, Cutler J, Desvigne-Nickens P, Fortmann SP, Friedman L, Havlik R, Hogelin G, Marler J, McGovern P, Morosco G, Mosca L, Pearson T, Stamler J, Stryer D, Thom T. 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. [PubMed]
5. Disparities in premature deaths from heart disease--50 States and the District of Columbia, 2001. MMWR Morb Mortal Wkly Rep. 2004;53:121–125. [PubMed]
6. Mensah GA, Mokdad AH, Ford ES, Greenlund KJ, Croft JB. State of disparities in cardiovascular health in the United States. Circulation. 2005;111:1233–1241. [PubMed]
7. Hunink MG, Goldman L, Tosteson AN, Mittleman MA, Goldman PA, Williams LW, Tsevat J, Weinstein MC. The recent decline in mortality from coronary heart disease, 1980-1990. The effect of secular trends in risk factors and treatment. JAMA. 1997;277:535–542. [PubMed]
8. Anderson KM, Odell PM, Wilson PW, Kannel WB. Cardiovascular disease risk profiles. Am Heart J. 1991;121:293–298. [PubMed]
9. [June 18, 2008]. Available at: http://www.nhlbi.nih.gov/about/framingham/.
10. Brindle PM, McConnachie A, Upton MN, Hart CL, Davey Smith G, Watt GC. The accuracy of the Framingham risk-score in different socioeconomic groups: a prospective study. Br J Gen Pract. 2005;55:838–845. [PMC free article] [PubMed]
11. Liu J, Hong Y, D'Agostino RB, Sr., Wu Z, Wang W, Sun J, Wilson PW, Kannel WB, Zhao D. Predictive value for the Chinese population of the Framingham CHD risk assessment tool compared with the Chinese Multi-Provincial Cohort Study. JAMA. 2004;291:2591–2599. [PubMed]
12. Thomsen TF, McGee D, Davidsen M, Jorgensen T. A cross-validation of risk-scores for coronary heart disease mortality based on data from the Glostrup Population Studies and Framingham Heart Study. Int J Epidemiol. 2002;31:817–822. [PubMed]
13. D'Agostino RB, Sr., Grundy S, Sullivan LM, Wilson P. Validation of the Framingham coronary heart disease prediction scores: results of a multiple ethnic groups investigation. JAMA. 2001;286:180–187. [PubMed]
15. D'Agostino RB, Sr., Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, Kannel WB. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation. 2008;117:743–753. [PubMed]
16. Anderson KM, Wilson PW, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation. 1991;83:356–362. [PubMed]
17. Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387. [PubMed]
18. Pencina MJ, D'Agostino RB. Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation. Stat Med. 2004;23:2109–2123. [PubMed]
19. Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation. 2007;115:928–935. [PubMed]
20. Liao Y, McGee DL, Cooper RS. Prediction of coronary heart disease mortality in blacks and whites: pooled data from two national cohorts. Am J Cardiol. 1999;84:31–36. [PubMed]
21. Wei M, Mitchell BD, Haffner SM, Stern MP. Effects of cigarette smoking, diabetes, high cholesterol, and hypertension on all-cause mortality and cardiovascular disease mortality in Mexican Americans. The San Antonio Heart Study. Am J Epidemiol. 1996;144:1058–1065. [PubMed]
22. Hunt KJ, Resendez RG, Williams K, Haffner SM, Stern MP, Hazuda HP. All-cause and cardiovascular mortality among Mexican-American and non-Hispanic White older participants in the San Antonio Heart Study- evidence against the “Hispanic paradox”. Am J Epidemiol. 2003;158:1048–1057. [PubMed]
23. Hunt KJ, Resendez RG, Williams K, Haffner SM, Stern MP. National Cholesterol Education Program versus World Health Organization metabolic syndrome in relation to all-cause and cardiovascular mortality in the San Antonio Heart Study. Circulation. 2004;110:1251–1257. [PubMed]
24. Robert S. Socioeconomic position and health: the independent contribution of community socioeconomic context. Annual Review of Sociology. 1999;25:489–516.
25. Brindle P, Emberson J, Lampe F, Walker M, Whincup P, Fahey T, Ebrahim S. Predictive accuracy of the Framingham coronary risk score in British men: prospective cohort study. BMJ. 2003;327:1267. [PMC free article] [PubMed]
PubReader format: click here to try


Save items

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


  • Cited in Books
    Cited in Books
    PubMed Central articles cited in books
  • PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...