C-Reactive Protein, Fibrin D-Dimer, and Risk of Ischemic Heart Disease
The Caerphilly and Speedwell Studies
Background— There is increasing interest in the predictive value of C-reactive protein (CRP) and fibrin D-dimer in the prediction of ischemic heart disease (IHD). We assessed their joint and independent associations with IHD in a large combined analysis of 2 population cohorts.
Methods and Results— Men aged 49 to 66 years from the general populations of Caerphilly and Speedwell were studied between 1982 and 1988 and re-examined for new IHD events at fixed intervals of ≈105 months (Caerphilly) and 75 months (Speedwell). 3213 men had CRP and D-dimer measured at baseline and 351 (11%) had a new IHD event. Mean levels of CRP and D-dimer were significantly higher among men in whom IHD developed. The relative odds of IHD in men in the top 20% of the distribution of CRP was 2.97 (95% CI, 2.04, 4.32) and for D-dimer was 2.40 (95% CI, 1.69, 3.40); CRP and D-dimer had additive effects on risk of IHD. Multivariate analysis reduced the size of the relative odds, which remained significant for D-dimer.
Conclusions— Both inflammatory and thrombogenic markers are important (and potentially additive) predictors of coronary risk.
C-reactive protein (CRP), a marker of the reactant plasma protein component of the inflammatory response, has been associated with the risk of future ischemic heart disease (IHD) in several prospective studies.1 A recent meta-analysis found a combined risk ratio for IHD by tertiles (after adjustment for other risk factors) of 1.45 (95% CI, 1.25 to 1.68).1 The determinants of CRP in samples of the general population have not been clearly defined. Chronic infections (eg, Helicobacter pylori, Chlamydia pneumoniae) or plasma homocysteine concentrations are not strongly associated with circulating CRP levels.1 Cigarette smoking and obesity1–6 are associated with increased CRP levels,1 whereas long-term treatment with pravastatin reduces CRP levels.7
Fibrin D-dimer, a marker of cross-linked fibrin turnover, has also been associated with the risk of future IHD; a recent meta-analysis found a combined risk ratio for IHD by tertiles (after adjustment for other risk factors) of 1.8 (95% CI, 1.4 to 2.4).8 Local fibrin formation and lysis are part of the inflammatory response, and fibrin degradation products (including D-dimer) have been shown to have diverse effects on inflammatory processes and acute phase responses, including hepatic synthesis of acute-phase proteins, such as CRP and fibrinogen.9–11
We recently hypothesized that plasma levels of CRP and D-dimer and their associations with incident IHD in the general population might be linked, possibly caused by vessel wall-related fibrin turnover producing D-dimer and other fibrin degradation products, which stimulate an inflammatory response including CRP.6 In an analysis of stored plasma samples from 1690 men in the Speedwell Study of IHD risk, we observed that CRP was associated with D-dimer; both were associated with incident IHD after follow-up for 75±4 months; and that each blood measure reduced the strength of the association of the other blood measure with incident IHD.6 We suggested that further larger studies were required to assess their relative predictive value. We therefore assayed CRP in stored plasma samples from the Caerphilly Study of IHD12 (which shares a common protocol with the neighboring Speedwell Study) and report a combined analysis of CRP, D-dimer, and incident IHD in an extended follow-up (averaging 90 months) of the combined Caerphilly–Speedwell cohort, comprising 4453 men, of whom 473 had major IHD events.
The Caerphilly and Speedwell population-based studies began in 1979 and have been previously described.6,12 The Caerphilly subjects of this report are those men who were seen at the first re-examination between 1984 and 1988, when they were aged 49 to 66 years. A fasting blood sample was obtained from 2225 (93%).12 The Speedwell subjects of this report are those men who were seen at the first re-examination between 1982 and 1985 when the men were also aged 49 to 66 years.6 A nonfasting blood sample was obtained from 2016 (98%) of the 2055 men.
Blood Collection, Storage, and Analysis
In the Caerphilly study, a fasting blood sample was taken. Fibrinogen was measured using heat precipitation nephelometry.12 D-dimer and CRP were measured in a citrated sample that had been centrifuged within 1 hour and stored at −70°C. D-dimer was assayed with an enzyme-linked immunoassay (Agen Gold; Agen, Parsippany, NJ).13 One batch of samples was unavailable for the present analyses, so D-dimer was measured on 1947 samples. During 2000, CRP was measured on the residue of these samples (samples available 1523) using a high-sensitivity nephelometric assay (Dade-Behring, Germany).
In the Speedwell study, a nonfasting blood sample was taken. Fibrinogen was measured by the same laboratory as for Caerphilly using the same heat precipitation nephelometry method. Lipids were not measured on this nonfasting sample; hence, the fasting levels of total cholesterol from the first recruitment examination are used in this report. EDTA plasma samples from the nonfasting re-examination were stored at −20°C. CRP and D-dimer were measured on these samples in 1997 by the same laboratory performing these measurements in Caerphilly. CRP was measured as described for Caerphilly. Because of discontinuation of the assay used in Caerphilly, D-dimer was measured using a different enzyme-linked immunoassay (Biopool, Umea, Sweden);6 however, there was a high correlation (r=0.78) between these 2 assays when we compared them directly in 600 samples from a previous separate population study (unpublished data). One batch of samples was unavailable for the present analyses so that CRP was measured for 1690 and D-dimer for 1719 men.6
There were no significant differences in age or incident IHD between subjects with and without information on CRP and D-dimer. Hence, data were “missing at random” and unlikely to bias the results.
Follow-Up Procedure and Definition of Incident IHD
The incidence of IHD was measured up to the third re-examination of the men. In Caerphilly, this took place between 1993 and 1997 with a nearly constant interval of 105 (SD=6) months after the baseline first re-examination. In Speedwell, the third re-examination took place between 1988 and 1991, again at a nearly constant interval of 75 (SD=4) months. Incident IHD comprised fatal coronary events, myocardial infarction validated from hospital records using World Health Organization criteria, and the new appearance of major or moderate Q waves on electrocardiograms at a subsequent follow-up examination, as described previously.12
The distributions of CRP and D-dimer both had a marked positive skew. When used as continuous variables, they were thus first transformed to logarithms. This transformation produced distributions that were close to Gaussian, with transformed-back geometric mean values for CRP of 1.73 mg/L (Caerphilly) and 1.57 mg/L (Speedwell) and for D-dimer of 11.7 and 42.0 ng/mL for the 2 areas, respectively. The large difference in D-dimer levels between the areas arises principally because of the different antibodies used in the different enzyme-linked immunoassays used for the 2 areas.6,12 Other factors that may contribute to the difference are the different types of samples (fasting citrated blood in Caerphilly, nonfasting EDTA in Speedwell) and, possibly, genuine differences between the 2 cohorts. These latter factors may also explain the small area difference in levels of CRP.
Adjusted mean differences in CRP and D-dimer between men in whom IHD developed and those in whom it did not were obtained by analyses of covariance, using standard multiple regression techniques. The adjusted mean differences on the logarithmic scale were transformed-back to give age-adjusted ratios of geometric means together with 95% CI. The remainder of the analysis was performed using multiple logistic regression analysis with the occurrence, or not, of any of the 3 types of incident IHD as the dependent variable. Logistic regression takes no account of the timing of events or of the duration of follow-up, but this is likely to be immaterial with the nearly constant lengths of follow-up in each area.
The difference in incidence in the 2 areas arising from their different lengths of follow-up has been accommodated by including area as a 2-level factor in all regression analyses. The assumption that the relations between incidence of IHD and both CRP and D-dimer in the 2 areas are parallel on the logit scale was tested, and no statistically significant interactions were found. This absence of interactions is an essential prerequisite for the combination of the 2 areas.
In the logistic regression analyses, both CRP and D-dimer were first divided into 5 equally sized groups using the 4 quintile cut-points. Results are presented as the odds of IHD in each group, relative to the odds in the 20% of men with the lowest levels. Tests for trend were then obtained by entering CRP or D-dimer as logarithmically transformed continuous variables in standard deviation units. The trends were summarized by the standardized relative odds (SRO): the odds associated with a 1-SD change in the transformed CRP or D-dimer levels.
Incidence of IHD
Among the 2398 men from Caerphilly, IHD developed in 282 (11.8%) over the follow-up period of nearly 9 years, giving an average annual incidence of 1.3%. In Speedwell, IHD developed in 191 (9.3%) of the 2055 men over the follow-up period of just >6 years, giving an average annual incidence of 1.5%. Of the combined total of 473 major IHD events, 204 (43%) were fatal. CRP was measured in 3213 men; major IHD developed in 351 (10.9%) of these 3213. D-dimer was measured in 3666 men; major IHD developed in 398 (10.9%) of these 3666.
The baseline characteristics of the men in whom major IHD developed are compared with the men who did not (Table 1). Men in whom IHD developed were slightly older than those in whom it did not, were more likely to be current smokers and much less likely to be lifetime nonsmokers; and had higher blood pressures, cholesterol, and body mass index. They were also more likely to have diabetes and to have a family history of myocardial infarction. There was no evidence that incidence of IHD was related to social class (manual/nonmanual). The pattern, and magnitude, of differences between men in whom IHD developed and those in whom it did not is very similar in the 2 areas.
Univariate Analyses for CRP, D-dimer, and Incident IHD
Table 2 shows that the geometric mean levels of both CRP and D-dimer were significantly higher among the men in whom IHD developed in both Caerphilly and Speedwell. The age-adjusted ratios of geometric means were similar in the 2 areas for CRP, but were higher in Caerphilly for D-dimer.
The incidence of IHD increased steadily from 6.2% in the 20% of men with the lowest levels of CRP to 16.5% among the men with the highest levels, in the combined areas. Table 3 shows that, unadjusted, the corresponding relative odds increased to 2.97 (95% CI, 2.04 to 4.32) in the top 20%. The trend was highly statistically significant (P<0.00001) and the standardized relative odds (SRO), the odds associated with a 1-SD change in the logarithm of CRP, was 1.41. Adjustment for age and area slightly reduced the odds in the top 20% to 2.71 (95% CI, 1.85 to 3.97) and the SRO to 1.36, but the trend remained highly significant (P<0.00001). Table 3 also shows that the pattern of relative odds by fifths of CRP was very similar in the 2 areas. Age-adjusted relative odds increased to 2.67 in the top 20% of men in Speedwell and to 2.73 in Caerphilly. The corresponding SROs were 1.39 (P=0.0001) and 1.33 (P=0.0003).
For D-dimer, the incidence of IHD increased steadily from 7.0% in the 20% of men with the lowest levels to 15.3% in the top 20% in the combined areas. The corresponding age-adjusted relative odds increased steadily to 2.17 (95% CI, 1.52 to 3.09) in the top 20% with an SRO of 1.36 (P<0.00001). Table 3 shows that the pattern of age-adjusted relative odds by fifths of D-dimer was fairly similar in the 2 areas for the lower four-fifths, but in the top 20% of the distribution the relative odds decreased to 1.40 in Speedwell, whereas it increased to 3.02 in Caerphilly. This is reflected in the corresponding SRO of 1.21 (P=0.017) and 1.50 (P<0.00001), as it was in the ratio of geometric means in Table 2. In a logistic model including D-dimer, age, area, and a term for an interaction between D-dimer and area, the interaction term was not statistically significant (t=1.37; P=0.17). This implies that the difference in the pattern of IHD incidence by fifth of D-dimer between the 2 areas is no more than might have been expected by chance and that it is therefore legitimate to combine the data from the 2 areas, as has been performed in the remainder of the analyses.
Associations With Cardiovascular Risk Factors
Table I (available online at http://atvb.ahajournals.org) shows that both CRP and D-dimer were strongly associated with both smoking habit and evidence of ischemia at baseline. Current smokers had higher levels of CRP compared with lifetime nonsmokers. There was no clear dose response among the current smokers. There was, however, among the ex-smokers, a clear trend with the length of time since quitting. The association between smoking and D-dimer was not as strong, but the pattern was similar. Current smokers also had higher levels of D-dimer, with no clear dose response among the current smokers. Again, there were clear trends among the ex-smokers for levels to decrease with increasing length of time since quitting.
All manifestations of ischemia at baseline were associated with higher levels of both CRP and D-dimer in both areas. Men with ischemia at baseline had higher levels of CRP than those without, with the larger differences being found in Speedwell. As with smoking, the associations with D-dimer were less strong.
CRP increased with age (r=0.14 in Caerphilly and r=0.15 in Speedwell) and showed positive associations with other conventional cardiovascular risk factors such as total cholesterol (r=0.06 and 0.08), diastolic blood pressure (r=0.08 and 0.07), and body mass index (r=0.15 and 0.14). All these are statistically significant (P<0.01) but are modest in size. There was a much stronger association with fibrinogen (r=0.52 and 0.42). D-dimer also increased with age (r=0.24 and 0.15) and with fibrinogen (r=0.27 and r=0.16) but showed no association in either area with total cholesterol, diastolic blood pressure, or body mass index.
There was a positive association between CRP and D-dimer (r=0.29 in Caerphilly, r=0.21 in Speedwell; P<0.00001 in both areas). This association did not arise purely because both were positively associated with age, smoking habit, evidence of ischemia at baseline, and fibrinogen. On adjusting for all these factors, the partial correlation between CRP and D-dimer declined only to ≈0.15 (P<0.00001) in both areas.
Incidence of IHD Jointly With CRP and D-dimer
The Figure shows that the incidence of IHD increased with CRP at each of 3 levels of D-dimer, and also that it increased with D-dimer at each of 3 levels of CRP. The trend for incidence of IHD to increase with CRP, independent of the level of D-dimer, was highly statistically significant (P<0.00001), as was the trend for incidence to increase with D-dimer, independent of the level of CRP (P=0.00012). There was no evidence that the association between incident IHD and CRP was different at different levels of D-dimer or that the association between incident IHD and D-dimer varied with level of CRP (test for interaction, χ2 [4 degrees of freedom]=3.71; P=0.45).
Multivariate Analysis for CRP
Table 4 gives the results from a series of multiple logistic regression analyses that show how the association between incident IHD and CRP changes as groups of cardiovascular risk factors are added to the regression model. The results are presented as the relative odds of IHD in the 20% of men with the highest levels of CRP together with a 95% CI, plus a standardized relative odds ratio that summarizes the trend, together with a probability value. When the model contains only age and area, the relative odds in the top 20% are 2.61 (95% CI, 1.78 to 3.83); the standardized relative odds are 1.35 (P<0.0001). These figures differ from those in Table 3 only because those in Table 3 were based on all 3213 men with a measurement of CRP, whereas those in Table 4 are based on the 3065 men with a complete set of data for all the variables included in the models of Table I.
The addition of a set of conventional cardiovascular risk factors (smoking habit, body mass index, diastolic blood pressure, total cholesterol, and evidence of ischemia at baseline) reduced the relative odds in the top 20% to 1.72 (95% CI, 1.14 to 2.58) and the standardized relative odds to 1.19 (P=0.005). Both the relative odds and the standardized relative odds remained statistically significant on the further addition of fibrinogen to the model. However, both became nonsignificant when D-dimer was then added to the model; the relative odds in the top 20% was reduced to 1.51 (95% CI, 0.98 to 2.33) and the standardized relative odds to 1.13 (P=0.093).
A stepwise multiple logistic regression was then performed and is summarized in Table II (available online at http://atvb.ahajournals.org).
The variables considered for inclusion in the stepwise regression were all those appearing in the models of Table 4. The baseline model contained CRP, age, and area. At each subsequent step, the variable that reduced the standardized relative odds by the greatest amount was added to the model. The first variable to be added was D-dimer, which reduced the SRO from 1.35 to 1.28, with the test for trend remaining highly significant (P=0.00005). The next factor to be added was evidence of ischemia at baseline, which reduced the SRO to 1.22 (P=0.0013). The further addition of smoking habit and the diastolic blood pressure reduced the SRO to 1.17 (P=0.012) and 1.15 (P=0.032), respectively. It was only with the addition of a fifth factor, fibrinogen, that the SRO was reduced sufficiently (to 1.13) for the test for trend to become nonsignificant (P=0.086). The further addition of body mass index and total cholesterol produced no material change, as can be seen by comparing the last lines from Table 4 and Table II.
Multivariate Analysis for D-dimer
Table 4 shows that the addition of the same set of conventional cardiovascular risk factors reduced the relative odds in the top 20% of the distribution of D-dimer from 2.15 to 1.76 (95% CI, 1.18 to 2.64). Correspondingly, the SRO was reduced from 1.33 to 1.26, but the test for trend remained highly significant (P=0.0001). The further addition of fibrinogen and CRP produced further small falls in the relative odds in the top 20% to 1.69 and then 1.61 (95% CI, 1.06 to 2.44); the corresponding falls in the SRO were to 1.24 and then 1.22. However, even after the final addition of CRP, the test for trend for incidence of IHD to increase with D-dimer remained highly statistically significant (P=0.0012).
The stepwise multiple logistic regression analysis (Table II) shows that the inclusion of first CRP and then evidence of ischemia at baseline reduced the SRO to 1.22 (P=0.0011)—very close to the final figure in Table 4. The inclusion of fibrinogen then reduced the SRO very slightly to 1.21 (P=0.0017), but the further addition of smoking habit, diastolic pressure, total cholesterol, and body mass index actually caused the SRO to increase marginally back to 1.22, the final value in Table 4.
By assaying CRP in stored plasma samples from the Caerphilly study, we have been able to compare CRP and D-dimer in the prediction of major IHD events in the combined Caerphilly–Speedwell studies (which shared a common protocol), comprising 3213 men, of whom 351 had events over an average follow-up period of almost 8 years. Our principal findings from this combined study were: (1) that both CRP and D-dimer were strong predictors of IHD events (P<0.00001), of similar magnitude to total cholesterol and blood pressure; (2) that the combination of CRP and D-dimer appeared a potentially useful predictor of IHD (incidence of IHD in men in the highest third of CRP and D-dimer 19.3%, compared with 7.0% in men in the lowest third of both variables, Figure); and (3) that on multivariate analyses including each other, as well as classic IHD risk factors and fibrinogen, D-dimer remained a significant predictor of IHD, whereas CRP did not.
As expected, men in the Caerphilly–Speedwell studies in whom IHD events developed had higher baseline levels of classic IHD risk factors compared with men who did not. The associations of CRP with baseline risk factors (and baseline evidence of IHD) in the Caerphilly study were similar to those in the Speedwell study,6 and in previous epidemiological studies.1 Adding the Caerphilly CRP data, we confirmed our preliminary findings from the Speedwell study6 that the combination of CRP and D-dimer increased their potential predictive value for IHD compared with either variable alone (Figure).
Combining the Caerphilly and Speedwell data also allowed us to confirm our preliminary findings from the Speedwell study6 that CRP and D-dimer were weakly, but significantly (r=0.15, P<0.00001), correlated after adjustment for age, smoking habit, baseline evidence of IHD, and fibrinogen. We have hypothesized that this association may arise because of the positive effects of D-dimer (and other fibrin degradation products) on secretion of interleukin 6, which stimulates hepatic synthesis of acute-phase proteins including CRP.6,9–12 Combining the Caerphilly and Speedwell data, we have shown that on multivariate analyses including each other, as well as classic IHD risk factors, D-dimer remained a significant predictor of IHD (SRO 1.22, P=0.0012), whereas CRP did not (SRO 1.13, P=0.093) (Table I).
The most likely explanation for this finding is that CRP levels in the general population are correlated with several other classic risk factors for IHD, including age, smoking habit, and obesity;1–5 hence, the association of CRP with risk of IHD is reduced when these classic risk factors are included in multivariate analyses.1 In contrast, D-dimer is only weakly associated with age and smoking habit; hence, adjustment for classic factors has little effect on its association with risk of IHD.8 Another possible explanation is that CRP may have greater biological variability (interindividual variation) than D-dimer.14,15 Although we have no data at present on this from Caerphilly or Speedwell, existing comparative data suggest that biological variability over several years is similar for both variables.1,8 However, further studies are required to investigate this possibility.
Conversely, the addition of CRP to classical IHD risk factors in multivariate analysis of the relationship of D-dimer to incident IHD might be expected to reduce the strength of the D-dimer–incident IHD relationship. However, D-dimer retained a significant association with incident IHD, suggesting that hypercoagulability (increased fibrin turnover) may play a causal role in the pathogenesis of IHD, independently of markers of low-grade inflammation such as CRP.
As in the Speedwell study,6 assays of CRP and D-dimer in the Caerphilly study were performed after storage for 12 to 15 years. Although both of these proteins appear resistant to proteolysis during storage, and the assays used in these studies appear sufficiently robust to withstand the effects of long-term storage, it is also important to perform further prospective cohort studies in which these assays are performed together in samples stored at low temperatures (−70°C or colder, as in the Caerphilly study). Although the Caerphilly and Speedwell studies shared a common protocol, there were some differences in sample anticoagulant, duration of storage, duration of follow-up, and type of D-dimer assay (see Methods). However, inspection of these variables, the inclusion of an “area adjustment” (Caerphilly versus Speedwell), and the similarity of the results from each area justify the grouping of the Caerphilly and Speedwell data (see Methods).
In conclusion, we have shown that (regardless of causative relevance) single measures of both CRP and D-dimer predict IHD risk, and that their combination is even more powerful (Figure). This may be clinically useful for risk stratification or selecting the intensity of secondary prevention measures. However, there are several different D-dimer assays on the market, which give different absolute values, and standardization or harmonization of the various assays is an important consideration before D-dimer could be considered for risk prediction. Further studies of the biological variability of both CRP and D-dimer are also required. We suggest that further prospective epidemiological studies (and collaborative meta-analyses) of CRP and D-dimer and incident IHD are required to clarify their relative (and additive) predictive values for IHD events.
We thank the Medical Research Council (UK) and the British Heart Foundation for financial support. The Caerphilly study was undertaken by the former MRC Epidemiology Unit (South Wales) and was funded by the Medical Research Council of the United Kingdom. The data archive is maintained by the Department of Social Medicine, University of Bristol.
D.B. is deceased.
- Received January 30, 2004.
- Accepted July 12, 2004.
Tracy RP, Psaty BM, Macy E, Bovill EG, Cushman M, Cornell ES, Kuller LH. Lifetime smoking exposure affects the association of C-reactive protein with cardiovascular disease risk factors and subclinical disease in healthy elderly subjects. Arterioscler Thromb Vasc Biol. 1997; 17: 2167–2176.
Danesh J, Muir J, Wong Y-K, Ward M, Gallimore JR, Pepys MB. Risk factors for coronary heart disease and acute-phase proteins: a population-based study. Eur Heart J. 1999; 20: 954–959.
Yudkin JS, Stehouwer CDA, Emeis JJ, Coppach SW. C-reactive protein in healthy subjects: associations with obesity, insulin resistance, and endothelial dysfunction. A potential role for cytokines originating from adipose tissue? Arterioscler Thromb Vasc Biol. 1999; 19: 972–978.
Hak AE, Stehouwer CDA, Bots ML, Polderman KH, Schalkwijk CG, Westendorp ICD, Hoffman A, Witterman JCM. Associations of C-reactive protein with measures of obesity, insulin resistance, and subclinical atherosclerosis in healthy, middle-aged women. Arterioscler Thromb Vasc Biol. 1999; 19: 1986–1991.
Lowe GDO, Yarnell JWG, Rumley A, Bainton D, Sweetnam PM. C-reactive protein, fibrin D-dimer, and incident ischemic heart disease in the Speedwell Study: Are inflammation and fibrin turnover linked in pathogenesis? Arterioscler Thromb Vasc Biol. 2001; 21: 603–610.
Ridker PM, Rifai N, Pfeffer MA, Sacks F, Braunwald E; for the Cholesterol and Recurrent Events (CARE) Investigators. Long-term effects of pravastatin on plasma concentration of C-reactive protein. Circulation. 1999; 100: 230–235.
Danesh J, Whincup P, Walker M, Lennon L, Thomson A, Appleby P, Rumley A, Lowe GDO. Fibrin D-dimer and coronary heart disease: prospective study and meta-analysis. Circulation. 2001; 103: 2323–2327.
Ritchie DG, Levy BA, Adams MA, Fuller GM. Regulation of fibrinogen and fibrin: an indirect feedback pathway. Proc Natl Acad Sci U S A. 1982; 79: 1530–1534.
Edgington TS, Curtiss LK, Plow EG. A linkage between the haemostatic and immune systems embodied in the fibrinolytic release of lymphocyte suppressive peptides. J Immunol. 1985; 134: 471–477.