C-reactive protein, established risk factors and social inequalities in cardiovascular disease – the significance of absolute versus relative measures of disease

The associations between socioeconomic position (SEP) and cardiovascular disease (CVD) are well documented and relatively undisputed [1‐3]. In the search for causal explanations of socioeconomic differences in cardiovascular diseases, it has been shown repeatedly that established risk factors such as smoking, physical inactivity, blood pressure, total-cholesterol, body mass index, and blood glucose, account for less than 50% of the socioeconomic differences in coronary heart disease (CHD) [4, 5]. Much less is known about the reasons for the differences in cerebrovascular disease [6]. Research into causal explanations of socioeconomic differences in CVD has most often used a relative scale, and it has been argued that relative measures are more appropriate for etiological investigations than absolute measures [7]. The increasing awareness of the limitations of established risk factors in explaining socioeconomic differences in CVD has stimulated research in new directions and has revitalized old ideas. One example of this is the increasing recognition of an inflammatory component of atherogenesis [8, 9], where major acute-phase-proteins such as high sensitive (hs) C-reactive protein (CRP) and fibrinogen have been found to predict acute cardiovascular events [8]. Thus, inflammation could in theory be another factor that might explain the social gradient in CVD. However, the notion that established risk factors have only modest importance in explaining socioeconomic differences in CVD, followed by a search for undiscovered risk factors, has been criticized [10, 11]. In a recent study by Lynch et al. [11], it is well illustrated that the conclusions on the importance of various risk factors heavily depend on the scale on which social inequalities are measured. The authors criticize the one-sided use of the relative scale over the past two decades, and they want to make visible that an absolute risk approach focuses attention on those risk factors that cause most cases of disease. The consistent use of the relative scale has important implications for the choice of preventive strategies, in that the quest for new factors may be at the expense of trying to reduce the levels of established risk factors that explain most cases of CVD.

The aim of the present study was to replicate the analyses by Lynch and co-authors [11] on the importance of various risk factors in explaining socioeconomic differences in CVD, this time in a Swedish context using similar methodology. We wanted to examine the importance of established risk factors, as opposed to low-grade inflammation as measured by CRP levels, in explaining socioeconomic differences in the incidence of CVD. However, instead of restricting the analyses to coronary heart disease as in the study by Lynch et al., we wanted to use a broader measure of CVD, including both CE and stroke. In the analyses, social inequalities were measured in relative terms, i.e. as ratios between incidence rates in lower and higher SEP groups, and also in absolute terms, i.e. the absolute difference in incidence rates in lower and higher SEP groups.

As has been shown in earlier studies [1‐3, 24], adjustment for cardiovascular risk factors resulted in a rather small reduction in the relative educational gradient in risk of future CVD, i.e. 8% for CRP and 21% for established risk factors taken together (current smoking, hypertension and hyperlipidemia). When considering absolute educational differences, however, there was a reduction in the absolute gradient of 18% for CRP levels below 3 mg/L and of 69% for the established risk factors taken together. The reduction in the latter group was even more pronounced if also excluding subjects with diabetes from the low-risk population. Thus, conclusions on the importance of "modern" and established risk factors depend on which scale one measures socioeconomic differences in morbidity and mortality. This has important implications for public health strategies and the preventive potential of acting on reducing established risk factors rather than "modern" risk factors. Our results are similar to the results from a recent population-based study of 2,682 men in eastern Finland [11]. For highly prevalent risk factors with strong associations with CVD, as in the case of established risk factors, there were strong reductions in absolute risk in each educational category after stratification (i.e., looking at a low-risk population without smoking, hypertension or hyperlipidemia). There were, however, only rather small changes in the relative gradient, while for the "modern" risk factor (here CRP) there were rather small changes in the relative gradient and only small reductions in absolute risk after stratification (i.e., looking at a population with CRP-levels below 3 mg/L). It is well known that relative measures are cumbersome when it comes to making comparisons, e.g. between age groups, sexes, countries, or populations [25]. Furthermore, when evaluating the importance of mediating factors or confounders, relative comparisons can be misleading as the effect of a given association depends heavily on the distribution of the exposure variable in the different categories of the mediating/confounding factor [11].

Primary and secondary prevention of CVD has mainly included prevention of established risk factors, and the risk of disease has decreased by 40% since the mid-1970s [26]. Regarding social inequalities in cardiovascular health, however, there have been similar or even widening relative social inequalities over time [27, 28]. Those who have benefited most from these interventions appear to have been individuals with a higher level of education, and the relative inequalities are thus widening [27]. If one considers the absolute levels, however, there has been a decrease of CVD in all socioeconomic groups over time. Thus, the conclusion about the importance of established risk factors in explaining the socioeconomic differences depends on the scale on which one measures social inequalities. These are important issues that affect opinions about where to place the effort in preventive programs – on established risk factors or on modern risk factors. It has recently been suggested that even though the use of relative measures is helpful in the search for mechanistic links between SEP and CVD, it is important to interpret the findings against a background of absolute levels of disease, and on what causes disease [11], in order to better interpret the effect of a certain factor on a given association.

In recent years, there has been increasing recognition of the inflammatory component of atherogenesis. Levels of major acute-phase-proteins such as CRP and fibrinogen have been found to predict acute cardiovascular events in prospective studies [8]. The inflammatory marker CRP has been found to be associated with the presence of atherosclerosis and to double the risk of future cardiovascular events when exceeding 3 mg/L [8]. However, there have also been studies that have shown that the usefulness of CRP in prediction beyond that of established risk factors is small [29, 30]. Furthermore, previous studies have shown an inverse association between SEP and levels of CRP, serum amyloid A (SAA), and fibrinogen [31‐37]. In a recent study on the MDCS cohort, we found that low SEP was strongly associated with CRP levels, independently of potential mediating factors e.g., smoking and factors involved in the metabolic syndrome. Furthermore, CRP levels were found to be associated with the extent of carotid atherosclerosis. However, there was only a minor attenuation of the relative SEP gradient in carotid atherosclerosis after adjustment for CRP [9]. Similarly, in the present study there was only a small reduction in the relative educational gradient in the risk of cardiovascular events, and also a small reduction in the absolute educational gradient after stratification for CRP below 3 mg/L. It should be noted that this was for the unadjusted measure of CRP, i.e. some of these reductions in the gradients were due to established risk factors known to be associated with CRP such as smoking, hypertension, and BMI [8]. It is difficult to directly compare the effects of CRP and established risk factors for several reasons. Firstly, stratification of established risk factors also results in a reduction in mean CRP levels and stratification for CRP results in lower levels of the established risk factors. The absolute risk reductions are thus influenced by other risk factors that correlate with the risk factors for which the analysis was stratified. Secondly, some of the effects of CRP may be mediated through the development of other risk factors. Longitudinal studies have shown that low-grade inflammation is associated with the development of hypertension, diabetes and large weight gain [38‐40]. Similarly, the effects of the established risk factors could be mediated through their proinflammatory effects. Thirdly, stratification of CRP resulted in a larger group (n = 3,296) than the group without major risk factors (n = 1,092), and the low-risk group was therefore a more selected group with lower cardiovascular risk.

Certain methodological issues must to be addressed. Firstly, misclassification of endpoint is a possible cause of bias. However, about 95% of the cases of stroke were confirmed by CT and/or autopsy [23]. Vital status of all individuals at the end of follow-up was updated by data linkage with the regional stroke register in Malmö, and with the regional and national myocardial infarction register [19, 21]. The STROMA is a population-based register that is known to be better than hospital-based registers with regard to coverage and potential selection bias [41]. The completeness and validity of the national myocardial infarction register and STROMA has been documented in several other studies [22, 42]. The proportion of non-hospitalized cases is very small in Sweden. There is no reason to believe that incomplete retrieval of cases biased the results.

Misclassification of exposure is a potential cause of bias. Educational level, usually attained in early adulthood, should not to be influenced by clinical atherosclerotic disease, which occurs later in life. The use of education as a marker of SEP has been shown to be reliable, to have a low non-response rate and, as it is usually attained in early adulthood, not to be subject to reverse causation [43, 44]. However, education might be a problematic indicator in a study covering a wide range of age cohorts, i.e. social and economical values might differ between various birth cohorts [45]. Other studies have shown a weaker association between SEP and various disease outcomes in groups close to retirement age, which might be partly attributable to survivor bias [46]. However, an earlier study on the MDCS showed a similar association between educational level and risk of future coronary events in subjects below 60 years of age and in the whole population [47].

Furthermore, misclassification of mediating factors is another potential source of bias. CRP has been suggested to be a good indicator of low-grade inflammation since the levels appear to be reasonably stable over time, with little seasonal variation [48, 49]. While some studies have shown signs of diurnal variation [49], others have not [50]. CRP has been shown to be related to future cardiovascular events [8], and this was also true in our study, even after adjustment for potential confounders such as smoking, hypertension, hyperlipidemia, BMI, and diabetes mellitus. The definition of a low-risk population in the present study was based on the three major cardiovascular risk factors, i.e., current smoking, hypertension and hyperlipidemia. The definition of hypertension using a systolic blood pressure exceeding 160 mm Hg is in accordance with the national guidelines at the time of baseline investigation in the early 1990s.

This study is based on a community-based sample of the general population, which makes it less sensitive to selection bias than samples based on workplace or populations in clinical settings. However, excluding subjects with known CVD together with the fact that people who participate in public health surveys are generally healthier than the non-participants might lead to an underestimation of the true associations between the measures of education and incident cardiovascular events in our study.

As shown in earlier studies, adjustment for cardiovascular risk factors resulted in a rather small reduction in the relative educational gradient in incidence of CVD, i.e. 8% for CRP and 21% for established risk factors taken together. However, in terms of absolute educational differences in CVD incidence, the risk reductions were 18% and 69%, respectively. Thus, the conclusion on the importance of a risk factor depends on which scale one measures social inequalities. In the case of highly prevalent risk factors with strong associations with CVD, as in the case with established risk factors, there was a strong reduction in absolute risk in each educational category after stratification (i.e. looking at the low-risk population without smoking, hypertension or hyperlipidemia), resulting in rather small changes in the relative gradient. With regard to the "modern" risk factor (here CRP), however, there were rather small changes in the absolute gradient after stratification (i.e., looking at subjects with CRP-levels below 3 mg/L), also resulting in rather small changes in the relative gradient. Even though, for several reasons, it is difficult to directly compare the effects of CRP and the established risk factors, it is essential to be aware of the fact that the interpretation of the importance of various risk factors in the association between SEP and CVD differs depending on which scale is used to measure social inequalities. These are important issues that affect opinions of where to invest effort in population-wide preventive programs and on the preventive potential of acting on reducing established risk factors as opposed to "modern" risk factors. The one-sided use of the relative scale can thus inhibit efforts to reduce established risk factors in that it gives priority to research and preventive programs looking in new directions.