Association of Fibrinogen, C-reactive Protein, Albumin, or Leukocyte Count With Coronary Heart Disease:  Meta-analyses of Prospective Studies FREE

VARIOUS epidemiologic studies have reported associations between coronary heart disease (CHD) and blood levels of fibrinogen, C-reactive protein (CRP), and albumin and leukocyte count. These factors can be influenced in the short term by insults that trigger acute-phase reactions.^1- 2 Prospective epidemiologic studies, in which CHD events have been recorded for some years after "baseline" blood collection, should be less prone to bias than retrospective studies, since they limit the influence of preexisting disease itself on the factors being investigated. This should still be the case even in long-term prospective studies in patients with previous vascular disease since, in such studies, subsequent "cases" are compared with "controls" with similar disease at baseline selected from within the same study. To help determine the nature of the associations between CHD and fibrinogen, CRP, albumin, and leukocyte count, we report a systematic overview (meta-analysis) of the available evidence from published prospective epidemiologic studies of these factors.

Long-term prospective studies published before 1998 that reported on correlations between CHD and blood levels of fibrinogen,^3- 22 CRP,^18,23- 27 albumin,^25,28- 34 or leukocyte count^{3,20- 21,35- 47} were sought by MEDLINE searches, scanning of relevant reference lists, hand searching of cardiology, epidemiology, and other relevant journals, and by correspondence with authors of such reports. Computer searches used combinations of key words relating to the blood factors (eg, fibrinogen, C-reactive protein, CRP, albumin,leukocyte, leucocyte, white cell count, acute phase reactants) and to the disease of interest (eg, coronary heart disease, myocardial infarction, atherosclerosis, vascular disease). All relevant studies identified were included. Articles in languages other than English were translated. The following were abstracted: size and type of cohort (ie, population based or selected on the basis of previous vascular disease); mean age and follow-up duration; assay methods; and degree of adjustment for potential confounders. Adjustment as shown in the figures is denoted as + for age and sex only; ++ for these plus smoking; +++ for these plus some other standard vascular risk factors; ++++ for these plus markers of social class; and +++++ for these plus information on chronic disease at baseline.

The log ratio, b, of the risk of disease among individuals in the top third vs those in the bottom third of baseline measurements of the relevant factor was estimated from the published report, assuming a log-linear association with disease risk over the mid range of baseline values of fibrinogen, log CRP, albumin, or leukocyte count. Where this log risk ratio (standardized for certain potential confounders; see above) was not directly available from the published report, it was estimated as 2.18/2.54 times the log risk ratio for the comparison of the top and bottom quarters (or, equivalently, as 2.18/2.80 times that for the comparison of the top and bottom fifths, etc) or, assuming a normal distribution, as 2.18 times the log risk ratio for a 1-SD difference in the baseline value of the factor. If the SD could not be obtained from the published report, then estimates from some of the larger studies were used: 2.4 µmol/L (0.08 g/dL) for fibrinogen⁴; 0.33 for log CRP⁴⁸; 3.0 g/L for albumin²⁹; and 1.8×10⁹/L for leukocyte count.⁴⁴

Most of the published studies related CHD risk only to measurements of these factors made at baseline, even though levels of each factor fluctuate markedly within individuals over time. If the "self-correlation coefficient" between measurements of the factor in blood samples collected some years apart in the same individuals is r, then the difference between the means of the usual values in the bottom and top baseline thirds is 2.18× r×SD (which is narrower than the estimated difference of 2.18×SD between the means of the baseline measurements in these groups). Values of r for these factors are available from certain studies and are in the range of about 0.5 to 0.7 (Table 1). Consequently, correction for this "regression dilution"⁴⁹ leads to associations of disease risk with long-term usual levels of each factor that are substantially stronger than the corresponding associations with baseline levels.

The standard error, s, of the log risk ratio, b, was obtained either from the number of standard errors by which the reported relationship differed from zero (where this z value equals b/s) or, when this was not available, from the approximation s²=2.18×(1/n₁+1/n₂), where n₁ is the number of individuals who had developed the disease and n₂ is the number who had not done so. Summary estimates of the risk ratios from all studies for each factor were obtained by combining the separate estimates of inverse variance–weighted log risk ratios from each study. This was done even when different studies used different assay methods, since cases were compared directly only with controls within the same studies. Confidence intervals (CIs) were obtained by a normal approximation, with 99% CIs (estimated as b±2.576s) used for the several dozen individual study results to take account of the increased scope for the play of chance in multiple comparisons, and 95% CIs used for the overall results. Heterogeneity was assessed by standard χ² tests.

Since the publication of a previous meta-analysis of 6 prospective studies of baseline fibrinogen levels that included 845 cases of CHD,⁵⁰ several of those studies have published additional follow-up, and several others have been published for the first time. Eighteen prospective studies of fibrinogen have now been identified (including 6 studies in patients with either angina or myocardial infarction at baseline), involving a total of 4018 CHD cases with a weighted mean age at baseline of 56 years and a weighted mean follow-up of 8 years^3- 22 (Figure 1). Different studies used different methods to measure fibrinogen levels, including clotting assays (usually gravimetry or the method of Clauss), nephelometry, and immunologic assays. Most studies made adjustments for smoking, blood lipid levels, and other standard vascular risk factors, such as blood pressure.

There was no significant heterogeneity between the published findings of the separate studies (χ²₁₇=14.4; P>.10) or between the 12 population-based cohorts and the 6 studies of people with previous vascular disease (χ²₁=0.4; P>0.10). Overall, comparison of individuals with fibrinogen levels in the top third with those in the bottom third at baseline yielded a combined risk ratio for CHD of 1.8 (95% CI, 1.6-2.0), and the estimated mean usual fibrinogen levels in these 2 groups were 10.3 and 7.4 µmol/L (0.35 and 0.25 g/dL). This risk ratio of 1.8 is similar to that in the previous meta-analysis,⁵⁰ but since it involves 5 times as many cases, it is more reliable. Moreover, since allowance has now been made for fluctuations in fibrinogen levels within individuals over time (ie, "regression dilution"; see "Methods"), the risk ratio relates to a difference in mean usual fibrinogen level of only about 2.9 µmol/L (0.1 g/dL) (rather than one of about 5.0 µmol/L [0.17 g/dL]), indicating a substantially stronger relationship than was previously reported.

The possibility that the association of fibrinogen and CHD is chiefly one of cause and effect is supported by the strength and consistency of the associations reported in prospective (and in other) studies, and by the apparent plausibility of the mechanisms through which such an effect might be produced. For example, fibrinogen is the main coagulation protein in the plasma, it is an important determinant of blood viscosity,⁵¹ and it can act as a cofactor for platelet aggregation.⁵² Blood levels of fibrinogen are, however, moderately correlated with other standard vascular risk factors (Table 1), but this does not necessarily mean that the association of disease with fibrinogen is noncausal. Indeed, fibrinogen could mediate a small part of the effect of smoking on CHD as its levels are about 10% higher in smokers than in nonsmokers, increase with increasing cigarette consumption, and decline to nonsmoker levels after stopping.⁵³ Although it has also been suggested that the associations of various persistent infections with CHD may be importantly mediated through increased fibrinogen (as well as increased CRP or leukocyte count; see below⁵⁴), the evidence for such claims is weak.^55- 56

Seven prospective studies of CRP and CHD were identified (including 2 in patients with either angina or myocardial infarction at baseline), involving a total of 1053 cases with a weighted mean age at baseline of 61 years and a weighted mean follow-up of 6 years^18,23- 27 (Figure 2). All but one of these studies used ultrasensitive CRP assay methods, and all made adjustments for smoking and lipids, but only 3 adjusted for fibrinogen. There was no significant heterogeneity between the published findings of the separate studies (χ²₆=5.3; P>.10). Overall, comparison of individuals with CRP values in the top third with those in the bottom third at baseline yielded a combined risk ratio for CHD of 1.7 (95% CI, 1.4-2.1). The estimated mean usual log CRP values in these 2 groups were 0.38 and 0.02 log mg/L, which correspond to usual CRP values of 2.4 and 1.0 mg/L and a difference of 1.4 mg/L.

In addition to these long-term prospective studies, some short-term studies have reported CRP to be predictive of cardiac complications, such as the need for coronary revascularization, among patients recently hospitalized for acute coronary syndromes.^57- 67 Those studies were not included in the present analyses because the apparent relevance of CRP to short-term complications may be partly attributable to residual confounding by the severity of the disease that caused hospitalization, despite attempts to adjust for this.

C-reactive protein has been less extensively investigated than fibrinogen, and too little is yet known about CRP to determine whether the association with CHD chiefly reflects causality. Although a variety of mechanisms by which CRP might promote thrombosis and atherosclerosis have been proposed,^68- 73 there is no proven relevance for these or any other putative mechanism. Nor is the suggestion in 1 report statistically convincing that the vascular protective effects of aspirin increase with increasing CRP level.^23,74 The first report from a long-term prospective study of an association between CRP and CHD risk was, however, published only in 1996, and it may be that some of the studies published so far would not have been if they had observed less striking results.⁷⁵ Further measurement of CRP (and of various possible confounders or mediators) in some large studies might, therefore, substantially change the present overall results and their interpretation.

Eight prospective, population-based studies of albumin and CHD were identified, involving a total of 3770 cases with a weighted mean age at baseline of 64 years and a weighted mean follow-up of 11 years (Figure 3).^28- 34 All used standard albumin assays (generally involving bromocresol green) and adjusted for smoking and lipid levels, and several also adjusted for blood pressure, obesity, and socioeconomic status. There was no significant heterogeneity between the published findings of the separate studies (χ²₇=7.5; P>.10). Overall, comparison of individuals with albumin levels in the bottom third with those in the top third at baseline yielded a combined risk ratio of 1.5 (95% CI, 1.3-1.7). The estimated mean usual albumin values in these 2 groups were 38 and 42 g/L (Table 1).

In 2 of these studies of CHD^28- 29 and in 3 other prospective studies,^76- 78 associations of albumin with total vascular mortality (most of which was CHD) have been reported, involving 2626 such deaths. As for the studies of CHD, the weighted mean age at entry in these studies was 64 years, and the weighted mean follow-up was 11 years, and all adjusted for smoking and other vascular risk factors. There was no significant heterogeneity between the published findings of these 5 studies of vascular mortality (χ²₄=2.9; P>.10), and comparison of individuals in the bottom vs top thirds of baseline albumin yielded a combined risk ratio of 2.0 (95% CI, 1.7-2.4). The 3 studies not included in the analysis of CHD cases yielded a similar combined risk ratio for vascular mortality but with wider confidence limits. These and other long-term prospective studies have also reported associations of low albumin with all-cause mortality (combined risk ratio, 1.9; 95% CI, 1.6-2.3)^{28- 30,76- 85} and with total cancer mortality (combined risk ratio, 1.9; 95% CI, 1.5-2.4).^{28- 30,76- 78,86- 88} But as separate results for CHD or for other vascular disease were not reported in several of those studies, it is possible that the available evidence for those outcomes has been somewhat exaggerated by preferential reporting of more extreme associations (ie, publication bias⁷⁵ ). Any bias owing to the absence of CHD results from studies that reported only on total mortality or cancer is, however, not likely to be substantial, since these studies included only about 10% of the deaths in all available studies.

Certain conditions, such as severe renal disease, might reduce albumin and so produce spurious inverse associations.⁸⁹ But attempts to control for the possible effects of preexisting disease on albumin by statistical adjustment for disease at baseline, by the exclusion of those known to have preexisting diseases, or by omission of any vascular events during the first few years of follow-up did not substantially change the strength of the associations with albumin. The apparent relevance of low albumin to such a wide range of causes of death is surprising, and many hypotheses have been proposed to explain the risks associated with low albumin.^90- 98 But the consistency of the associations of low albumin with CHD and with other diseases remain more convincing than any of the explanations thus far proposed for them.

Nineteen prospective studies of leukocyte count and CHD were identified (including 4 in patients with previous myocardial infarction and 1 in patients with coronary stenosis at baseline), involving a total of 7229 cases with a weighted mean age at baseline of 55 years and a weighted mean follow-up of 8 years^{3,20- 21,35- 47} (Figure 4). All measured leukocyte counts by standard methods (usually using Coulter counters), most made adjustments for smoking, blood pressure, and obesity, and several also adjusted for fibrinogen. There was some evidence of heterogeneity between the published findings of the separate studies (χ²₁₈=31.0; P<.03), but no significant heterogeneity between the 14 population-based cohorts and the 5 studies of people with previous vascular disease (χ²₁ = 0.80; P>.10). More extreme risk ratios in studies that involved fewer cases and less adjustment for potential confounders may be due both to preferential publication of small studies with striking findings⁷⁵ and to residual confounding. For example, as smokers tend to have higher leukocyte counts than nonsmokers,⁹⁹ and as adjustment for self-reported smoking substantially reduces the strength of the association between leukocyte count and CHD, full adjustment for smoking should reduce it still further (although some studies have reported an association even among nonsmokers³⁶).

Both publication bias and confounding should be limited by restricting attention to just the larger published studies, since the 7 studies that involved more than 400 cases all adjusted for at least smoking and other standard vascular risk factors (Figure 4). Together those 7 studies involved 5337 cases, or about three quarters of the total, and there was no significant heterogeneity between them (χ²₆=6.1; P>.10). Overall, comparison of individuals with a single baseline leukocyte count in the top third with those in the bottom third yielded a combined risk ratio of 1.4 (95% CI, 1.3-1.5), which is similar to the overall risk ratio in all 19 studies of 1.5 (95% CI, 1.4-1.6). The estimated mean usual leukocyte counts in these 2 groups were 8.4 and 5.6×10⁹/L (Table 1), and the difference was 2.8×10⁹/L. These and several other long-term prospective studies have also reported moderately strong associations between leukocyte count and total cancer mortality¹⁰⁰ or all-cause mortality.^{21,35,42,46,101- 103} But as was the case with albumin, those studies of leukocyte count and all-cause mortality or cancer mortality that did not report separate findings for CHD or other vascular disease tended to be small.

For leukocyte count, a causal relationship with CHD is particularly difficult to establish because leukocytes have such a wide range of biological effects, some potentially protective against vascular disease and some potentially damaging.^104- 107 Moreover, even if some particular aspect of leukocyte activity was independently associated with CHD, the extent of this activity would not necessarily be associated with total leukocyte count, or even with the circulating levels of some specific type(s) of leukocyte.⁴⁰

For each of these factors, the published results from the prospective studies (or, for leukocyte count, the larger such studies) are remarkably consistent, indicating moderate but highly significant associations with CHD that deserve further study. More detailed combined analyses, perhaps based on individual participant data from each of the prospective studies, could help to characterize the shapes of the dose-response relationships, reduce any bias related to the selection of particular cutoff levels, allow more complete adjustment for other risk factors, and assess the associations in particular subgroups. Repeated measurements of these factors in larger subsamples would also help correct more reliably for fluctuations over time within individuals,⁴⁹ and additional studies of these and of various other potential correlates of CHD might help elucidate causality. (If, for example, several different factors were found to be closely associated with each other and with CHD, it might suggest that they are all just general indicators of some underlying process more fundamentally relevant to the disease.) By analogy with familial hypercholesterolemia, studies of CHD incidence in people with some rare genetic abnormality or chronic disease state that persistently and grossly alters one of these factors might help to confirm or refute causality. Randomized trials might eventually become relevant, if treatments become available that specifically affect fibrinogen, CRP, albumin, or particular types of leukocytes (or some common underlying process). Even though the mechanisms that account for these 4 associations (Figure 1, Figure 2, Figure 3 and Figure 4) are not yet clear, the existence of such consistent and highly significant relationships with such a common cause of death is of substantial interest.