C-Reactive Protein Modulates Risk Prediction Based on the Framingham Score
Implications for Future Risk Assessment: Results From a Large Cohort Study in Southern Germany
Background— The Framingham Coronary Heart Disease (CHD) prediction score is recommended for global risk assessment in subjects prone to CHD. Recently, C-reactive protein (CRP) has emerged as an independent predictor of CHD. We sought to assess the potential of CRP measurements to enhance risk prediction based on the Framingham Risk Score (FRS) in a large cohort of middle-aged men from the general population.
Methods and Results— We measured CRP and traditional cardiovascular risk factors at baseline in 3435 white men of German nationality, 45 to 74 years of age. All men were drawn from 3 random samples of the general population in the Augsburg area located in Southern Germany in 1984 to 1985, 1989 to 1990, and 1994 to 1995 (response rate, 80%), and the FRS was calculated in all of them. Outcome was defined as nonfatal and fatal coronary events, including sudden cardiac death. During an average follow-up of 6.6 years, a total of 191 coronary events occurred. Cox regression showed a significant contribution of CRP to coronary event risk prediction independent of the FRS (P=0.0002). In stratified analysis for 5 categories of FRS, CRP significantly added prognostic information to the FRS in subjects in 2 intermediate risk categories (P=0.03 and P=0.02).
Conclusions— Our results suggest that CRP enhances global coronary risk as assessed by the FRS, especially in intermediate risk groups. This might have implications for future risk assessment.
Received April 25, 2003; de novo received October 28, 2003; revision received December 17, 2003; accepted December 29, 2003.
The Framingham Risk Score (FRS) is widely recommended to assess global risk for coronary heart disease (CHD) events.1 However, classic risk factors do not account for all incident coronary events, and there may be a substantial number of subjects with normal lipoprotein concentrations who have the disease.2 This has led to the search for additional diagnostic tools,3 and in more than 15 large prospective studies, C-reactive protein (CRP), through the use of new high-sensitivity (hs) assays, has emerged as a strong and consistent predictor of an incident cardiovascular event in initially healthy subjects.4 A recent prospective study in women has suggested that CRP may even better predict future cardiovascular events than LDL cholesterol2 and that it may add to the prediction of the estimated 10-year risk according to the FRS.
We sought to investigate the potential of CRP measurements to modify risk prediction based on the FRS in a large CHD-event free cohort of middle-aged white men of German nationality sampled from the general population.
Study Design, Population, and Follow-Up
The population-based MONICA (MONItoring of trends and determinants in CArdiovascular disease) Augsburg studies (Southern Germany) conducted between 1984 and 1998 were used as the database.5
Three independent cross-sectional surveys covering the city of Augsburg and two adjacent counties were conducted in 1984 to 1985 (S1), 1989 to 1990 (S2), and 1994 to 1995 (S3) to estimate the prevalence of cardiovascular risk factors among men and women. Altogether, 13,428 white subjects of German nationality (6725 men, 6703 women; response rate, 77%), 25 to 74 years of age, randomly drawn from the general population, participated in at least one of the three studies. In a follow-up study, vital status was assessed for all persons of the three surveys in 1998. During the observation period, 772 participants (531 men, 241 women) had died, and vital status could not be assessed for 56 persons (31 men, 25 women) who had moved to an unknown location.
The outcome variable was a combination of incident fatal or nonfatal acute myocardial infarction (MI) and sudden cardiac death. They were identified through the MONICA/KORA (KOoperative Gesundheitsforschung in der Region Augsburg) coronary event register of the 25- to 74-year-old study population and censored at the 75th year of age.6 According to the MONICA manual,5 the diagnosis of a major nonfatal MI event was based on symptoms, cardiac enzymes, and typical ECG changes. Deaths from cardiovascular causes were validated by autopsy reports, death certificates, chart review, and information from the last treating physician.
The present analysis was restricted to men 45 to 74 years of age (response rate, 80%) at the baseline examination (n=3667). Of those, 990 (27%) subjects were from survey S1, 1324 (36%) from S2, and 1353 (37%) from S3. A total of 48 subjects had missing values on CRP or other variables. Participants with prevalent MI (n=184) were excluded. Thus, 3435 subjects were available for the present analysis.
All participants completed a standardized questionnaire, including medical history, lifestyle, and drug history. Blood pressure, body height (meters) and body weight (kilograms), body mass index (BMI, kg/m2), smoking behavior, and alcohol consumption (g/d) were determined as described elsewhere.7 The number of education years was calculated on the basis of the highest level of formal education completed.
A nonfasting venous blood sample was collected from all participants in a supine resting position. Samples for measurement of CRP were stored at −70°C until analysis. Serum CRP concentrations were measured with the use of an hs-immunoradiometric assay (range, 0.05 to 10 mg/L), as previously described.8 The coefficient of variation (CV) for repeated measurements was 12% over all ranges. Total serum cholesterol (TC) and HDL cholesterol (HDL-C) were measured in multiple batches by routine enzymatic methods. Corresponding CVs were between 1% and 3% for TC and between 3% and 4% for HDL-C.
Means or proportions (adjusted for age and survey) for baseline demographic and clinical characteristics were computed by ANCOVA for men with and without incident coronary events. Cox proportional hazards analysis was used to assess the independent risk for the incidence of a first fatal or nonfatal acute coronary event separately in quartiles of TC/HDL-C and in quartiles of CRP. Relative risks for both variables were adjusted for age and survey (S1, S2, or S3) and were further adjusted for BMI (according to Bray, 7), current smoking (yes/no), hypertension (blood pressure less than versus ≥140/90 mm Hg), education years (less than versus ≥12 years), alcohol consumption (0, 0.1 to 39.9, versus ≥40 g/d), physical activity (inactive versus active, that is, ≥1 hour in at least 1 season), and history of diabetes (yes versus no) in all other models. Results are presented as hazard ratios (HR), together with their 95% confidence intervals. Probability values are based on the Wald statistic.
Finally, using Cox proportional hazards analysis, we assessed whether CRP contributed information on risk beyond that conveyed by the 10-year risk calculated with the FRS according to the formula of Wilson et al9 and beyond the risk associated with the ratio of TC/HDL-C. In these analyses, CRP concentrations were categorized according to recently proposed cut-points (<1.0 mg/L; 1.0 to 3.0 mg/L; >3.0 mg/L).10 Moreover, to estimate the effects of the additional value of CRP on the prediction of coronary events in different FRS categories, separate Cox proportional hazards analyses stratified for FRS categories (3 or 5, respectively) were performed. For each of the 3, respective 5 FRS categories, a Cox proportional hazards model was calculated with the inclusion of CRP (3 categoreis) as exposition variable and FRS (continuous) and survey as confounding covariates. The percentages of a first incident coronary event within 10 years estimated by these Cox models were compared according to percentages estimated by Cox models with FRS categories (3 or 5, respectively) adjusted only for survey.
To assess the goodness of fit of the different prediction models, we calculated Akaike’s Information Criterion (AIC) regarding an AIC difference between two models of >10 as essentially different.11,12 Although ROC analysis can be only a rough estimate for the predictive value of a Cox proportional hazards model in our study design with censored data, we performed such analysis for reasons of comparison with other studies. We estimated the area under the curve (AUC) as a measure for the predictive ability of a Cox proportional hazards model. Different AUCs were tested by a nonparametric approach of DeLong et al.13 The “change-in-estimate” method (CIE) was used to evaluate the impact of the addition of CRP on the HRs of the FRS with a 10% criterion.14 Significance tests are 2 tailed, and probability values <0.05 were considered statistically significant. All analyses were performed with the use of the Statistical Analysis System (Version 8.2 for Unix, SAS Institute Inc).
During an average follow-up of 6.6 years, a total of 191 coronary events occurred. Men with an incident event were significantly older; had a higher TC/HDL-C ratio, higher systolic blood pressures, and elevated CRP concentrations; were more frequently smokers; and had a higher prevalence of obesity and diabetes (Table 1).
Relative Risks Associated With CRP and the Ratio of TC/HDL-C
In multivariable analyses, the relative risks associated with increasing quartiles as compared with the bottom quartiles were 1.21, 1.43, and 2.03 for CRP (P=0.0079) and 1.72, 1.76, and 2.62 for the ratio of TC/HDL-C (P=0.0006).
Additive Effect of CRP to the Risk Associated With the Ratio of TC/HDL-C and the FRS
We performed Cox proportional hazards analyses to assess the influence of CRP measurements on the risk of a first coronary event, independent of the TC/HDL-C ratio and the calculated FRS. For this purpose, TC/HDL-C was categorized into tertiles and the FRS was divided into low (<6%), intermediate (6% to 19%), and high (≥20%) risk over 10 years (FRS 1), according to Greenland et al,15 and into 5 risk categories <6%, 6% to 10%, 11% to 14%, 15% to 19%, and ≥20% (FRS 2) (Table 2). For each of these factors, two Cox regression models were calculated: The first model included only the factor under concern as covariate (model without CRP) and the second model additionally included CRP as covariate (model with CRP). As shown in Table 2, CRP significantly contributed to the prediction of incident coronary events through the use of the TC/HDL-C ratio, FRS 1, and FRS 2 (probability values <0.001). As suggested by AIC, with differences >10 between models without CRP and those including CRP, the fit of each model containing CRP was superior to that of the TC/HDL-C ratio and FRS alone.
The AUC as a rough estimate (in our study design) for the predictive value of the different Cox models revealed a statistically significant increase from 0.735 (model with FRS 2) to 0.750 (model with FRS 2 and CRP) (P=0.0163). The inclusion of CRP was associated with a remarkable decrease in the HR of the FRS. The HR of the highest FRS 2 category (risk >20%/10 years) compared with the lowest FRS 2 category (risk <6%/10 years) decreased from 6.2 to 5.0, indicating a CIE of 19.4%. All CIEs are greater than the 10% criterion, indicating a strong impact of CRP on the HRs of FRS 2. Similar results were observed for the FRS 1 model.
Moreover, we compared the proportions of incident coronary events within 10 years estimated by the Cox model for the five categories of FRS 2 alone (Figure, left panel) and for different CRP categories in each category of FRS 2, adjusted for survey and FRS (Figure, right panel). Probability values of the stratified analyses are given in the Figure (right panel, above each FRS category). Cox regression revealed a considerable modification in coronary event incidence based on CRP concentrations and, more importantly, in categories of FRS 2 associated with a 10% to 20% risk per 10 years, elevated concentrations of CRP were consistently and statistically significantly associated with a further increased risk (P=0.03 and P=0.02). In contrast, in men with a risk <6% and 6% to 10% per 10 years, CRP had no statistically significant additional effect on the prediction of a first coronary event. Regarding the different AUCs, a remarkable increase was found for the intermediate FRS categories of 11% to 14% and 15% to 19% (increase in the AUC from 0.725 to 0.776 and from 0.695 to 0.751).
In this prospective population-based study, increased CRP concentrations and an elevated TC/HDL-C ratio were both independently related to incident coronary events. However, even if the strongest lipid/lipoprotein variable was chosen for risk assessment, our data clearly show that the measurement of CRP contributes significantly to the prediction of a first coronary event and adds clinically relevant information to the TC/HDL-C ratio. Finally, and most importantly, these prospective data from a large European cohort of middle-aged men clearly suggest that CRP modulates the risk conveyed by the FRS as the HRs from the top to the bottom category decreased remarkably after inclusion of CRP in the various models. This was observed in particular in those with an FRS between 10% and 20% over a period of 10 years. These men may benefit from additional noninvasive tests such as determination of CRP by a hs assay.
Such conclusions are in agreement with recent findings by Albert et al.16 They found significant correlations between CRP concentrations and the FRS but only minimal correlations with its individual components. However, our data are in contrast to a report from the Rotterdam Study,17 which found no additional value of CRP in elderly people. Such discrepancy may be explained by differences in age, the additional inclusion of left ventricular hypertrophy, another important risk marker, and the fact that CRP concentrations in incident cases were considerably lower than in our study despite the older average age in the Rotterdam population.
In summary, our data confirm and extend results from the Women’s Health Study2 in a large sample of men from the general population. If these findings can be replicated in other populations, this may represent a strong argument for the inclusion of CRP as an additional variable to further improve risk prediction in asymptomatic subjects at intermediate risk of CHD. This would be in line with recent American Heart Association/Centers for Disease Control guidelines.10
The authors thank Gerlinde Trischler for excellent technical assistance and Andrea Schneider for expert data handling. We also thank Lloyd Chambless for critical comments on the revised manuscript.
Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Summary of the 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). JAMA. 2001; 285: 2486–2497.
Pearson TA. New tools for coronary risk assessment: what are their advantages and limitations? Circulation. 2002; 105: 886–892.
Ridker PM. Clinical application of C-reactive protein for cardiovascular disease detection and prevention. Circulation. 2003; 107: 363–369.
Hense HW, Filipiak B, Döring A, et al. Ten-year trends of cardiovascular risk factors in the MONICA Augsburg region in southern Germany. Results from 1984/1985, 1989/1990, and 1994/1995 surveys. CVD Prevention. 1998; 1: 318–327.
Hutchinson WL, Koenig W, Fröhlich M, et al. Immunoradiometric assay of circulating C-reactive protein: age-related values in the adult general population. Clin Chem. 2000; 46: 934–938.
Wilson PWF, D‘Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation. 1998; 97: 1837–1847.
Pearson TA, Mensah GA, Alexander WR, et al. Markers of inflammation and cardiovascular disease: application to clinical and public health practice: a statement for healthcare professionals from the Centers for Disease Control and Prevention and the Am Heart Association. Circulation. 2003; 107: 499–511.
Akaike H. Information theory as a extension of the maximum likelihood principle. In: Pterov BN, Csaki F, eds. Second International Symposium on Information Theory. Budapest, Hungary: Akademiai Kiado; 1973: 267–281.
Burnham KP, Anderson DR. Model selection and inference: a practical information-theoretic approach. New York, NY: Springer; 1998: 43–48.
Greenland P, Smith SC, Grundy SM. Improving coronary heart disease risk assessment in asymptomatic people: role of traditional risk factors and noninvasive cardiovascular tests. Circulation. 2001; 104: 1863–1867.
Albert MA, Glynn RJ, Ridker PM. Plasma concentration of C-reactive protein and the calculated Framingham coronary heart disease risk score. Circulation. 2003; 108: 161–165.