Inflammatory markers versus lipid levels in the risk of CAD

A recent study reported in Cardiology Review showed that classic risk factors—cholesterol levels, blood pressure, and smoking habit—accounted for at least 75% of major coronary artery disease (CAD) events during 10 years of follow-up in a large British cohort.1 In our studies based on two British cohorts, we evaluated the contribution to subsequent risk of additional lipids (high-density lipoprotein [HDL] cholesterol and triglycerides2) and inflammatory markers (fibrinogen, viscosity, and white blood cell count). Our initial study showed that total cholesterol, HDL cholesterol, and triglycerides act largely independently of each other in predicting subsequent CAD risk.2 Our recent study compared the predictive value of lipid markers with inflammatory markers, both in addition to smoking habit, blood pressure, and body mass index. Those findings are reviewed in this article.

Patients and methods

Two populations from the West of England and from South Wales, United Kingdom, were examined using a common protocol between 1979 and 1983. Men were selected from the general population and were 45 to 63 years of age when first examined. About 90% of patients were eligible for the study, or a total of 4,860 men.

A standard (epidemiological) cardiovascular examination was performed and a venous blood sample taken after an overnight fast. Suitable samples were available from 4,641 men.

The follow-up procedure consisted of a repeated cardiovascular examination at 5-year intervals for one population and 3-year intervals for the other. Hospital admissions in the survey district were used to check major coronary events, and the validity of clinical diagnoses was checked using hospital notes. Death certificates for men who died during the 10-year follow-up period were obtained from the National Health Service Central Register. In addition to fatal and nonfatal coronary events, silent myocardial infarctions were ascertained from electro-cardiograms done at follow-up examinations according to World Health Organization criteria.3

Full details of the laboratory analyses done at the time of the baseline examinations have been reported previously.2,4 Calculations of the laboratory precision and biological variability of these assays were also performed. Regression dilution effects were calculated by the parametric method of Rosner and colleagues5 using samples taken in an identical manner to those taken at baseline examination (early morning after an overnight fast) from subjects rescreened at 5 years in one of the study populations. Details of all statistical methods used to evaluate risks at 10 years of follow-up are reported elsewhere.

Results

In the examined study population of 4,325 men, 525 (12%) had experienced a major CAD event during 10 years of follow-up. Of 1,006 men who had clinical or historical evidence of CAD at baseline screening, 234 (23%) had had a CAD event during 10 years of follow-up. Among the 3,319 men without evidence of CAD at baseline, 291 men (9%) experienced a major CAD event.

The distributions of the clinical risk factors, lipids, and inflammatory markers differed in the expected directions between those who had had a major CAD event and those who had not. All risk factors showed statistically significant differences in the two groups when adjusted for age and area (study population). These data are shown in Table 1.

Relative odds (risk) of a major CAD event by 10 years were calculated from logistic regression analyses comparing a model incorporating total cholester-ol, HDL cholesterol, triglycerides, and standard nonlipid clinical risk factors (age, preexisting CAD, blood pressure, smoking habit, and body mass index) with a model incorporating these same nonlipid risk factors and three hemostatic factors (fibrinogen, viscosity, and white blood cell count). Two methods were then used to compare the risk prediction derived from these models: (1) risk scores were calculated for each individual from regression coefficients and used to construct receiver operating characteristic (ROC) curves, with the area under this curve providing a measure of usefulness for prediction (0.5 indicated no better prediction than expected by chance, and 1.0 indicated perfect prediction); and (2) absolute risk in 10 subgroups of men defined by the deciles of predicted risk according to each model.

Relative odds of a major CAD event at 10 years were statistically significant in both lipid and hemostatic models with the exception of body mass index, which failed to achieve significance. Correction for regression dilution tended to increase the relative odds for the continuously distributed variables (with the exception of body mass index), but decreased the relative odds for some categorical defined variables.

Results for the predictions derived from areas under the ROC curves are shown in Table 2. For the whole cohort, the prediction derived from the lipid model was similar to that derived from the hemostatic model, but the combined model was statistically significantly better for prediction than either of the separate models. When men without evidence of CAD at baseline were considered, a similar pattern was shown in the results, but the predictive ability was reduced.

Absolute risks were calculated for 10 subgroups of men defined from the deciles of predicted risk for each of the models—lipid and hemostatic—when all other nonlipid clinical risk factors were included in the model (as described previously). These analyses are summarized in the Figure and show that men in the highest group based on their lipid values actually had a 34.3% chance of a major CAD event by 10 years (35.5% predicted) compared with 3.2% for those men in the lowest risk group (3.1% predicted). Absolute risks for the hemostatic model were similar (34.5% for men in the highest risk group and 2.1% for men in the lowest risk group).

Discussion

Three main points can be highlighted from these results: (1) in these particular British populations, three inflammatory markers predict CAD as well as three lipid markers, taking into account all other established risk factors; (2) prediction of events from either of these models is relatively weak; in men in the top 10th of the distribution of either of these classes of risk factors, just over one third of the CAD events occurred, although the majority of the events occurred at lower levels of these risk factors. When this was repeated for fifths of the distribution of risk

factors, up to half of CAD events occurred in the top fifth of men for the lipid, hemostatic, or combined model; and (3) adjustment for regression dilution does not result in any improvement in risk prediction, at least when assessed in the same dataset that was used to derive the logistic model. These three points are addressed in the following discussion.

1. In countries such as Japan and China, the prevalence of CAD is low; a large prospective study from Shanghai showed that there was a graded relationship between the incidence of CAD and the level of serum cholesterol in men or women at recruitment, although the population levels of cholesterol were low by Western standards.6 In the present study, cholesterol levels were relatively high at baseline recruitment (≥ 6.0 mmol/L), and it is possible that inflammatory markers may be either less or more predictive with lower population levels of cholesterol. This can only be tested in prospective studies in different populations at different levels of lipid risk factors.

2. These data broadly support the view that risk factor thresholds may not be relevant for reporting risk of CAD7 and also show that risk factor scores fall considerably short of perfect prediction of risk during a 10-year period of follow-up. The claim that 75% of major CAD events could be accounted for by smoking habit, serum cholesterol levels, and blood pressure by the authors of the British Regional Heart Study report1 is dependent on the choice of cut points; if low threshold values for risk factors are used, then a higher percentage of events will be attributed to these risk factors. In particular, the data, although similar

to those reported in the British Regional Heart Study1 and with similar levels of serum cholesterol, show that the general level of risk prediction,

taking into account all known risk factors, is relatively poor. This is likely to indicate the existence of unknown

or poorly measured risk factors, such

as the level of physical activity or

psychosocial stressors. Some indica-tion of the importance of the latter, although in a limited case-control design, has been shown in the INTERHEART Study in 52 countries across the globe8; it seems important to attempt to reproduce these findings in prospective studies.

3. The model that had been corrected for regression dilution bias was no better for prediction than the uncorrected model. Rather than assessing it on the original dataset that was used in its derivation, it may have performed better had it been applied to a new dataset or to a dataset incorporating repeated assessments of risk factors to minimize their measurement error.

Conclusions

These data support the view that inflammatory factors are important in the pathogenesis of CAD. It has been suggested that statins, in addition to lowering cholesterol, also have antiatherothrombotic properties9 and can lower plasma viscosity.10 Laboratory evaluation of inflammatory markers, such as those described in this article, is inexpensive and may be helpful in risk prediction, although this should be assessed in future prospective studies.