C-Reactive Protein and Prognosis in Diabetes: Getting to the Heart of the Matter

C-reactive protein (CRP) is a liver-derived protein whose concentration increases manyfold during inflammation or infection, making it a useful diagnostic and treatment measure in diseases such as rheumatoid arthritis. However, in the last decade, CRP has garnered interest as a biomarker of vascular risk that has stimulated more studies than perhaps any other “novel” biomarker. In these studies, the much lower CRP values seen in healthy individuals free from overt infection or inflammation, detected by a new generation of sensitive assays, have been associated in study after study with the risk of a first cardiovascular event (rev. in (1). Yet, despite the consistency of this association, whether CRP is causally involved in atherogenesis or whether it helps in prediction of cardiovascular disease (CVD) events remains intensely controversial (2,3).

In light of this controversy, studies examining the association of CRP with vascular outcome and mortality in patient populations, such as those with diabetes, where prospective data remain relatively sparse are to be welcomed. In the current issue of Diabetes, Bruno et al. (4) report that although CRP was related to risk of all-cause and CVD mortality in their 5-year follow-up of the Casale Monferrato Study, the improvement in individual risk prediction beyond established risk markers, evaluated by a range of statistical tests, was marginal at best. The results from this helpful study, therefore, currently caution against routine measurement of CRP for risk prediction in diabetes patients.

Why should CRP not better predict risk in individuals, given its consistent association with CVD and mortality? Several issues contribute. The association of CRP with incident events is linear when both CRP values and risk are plotted on a log scale, with no threshold value of CRP above which risk is substantial and below which it is negligible. This means that individuals in a population with intermediate values of CRP are at moderate risk of events. Moreover, the distribution of CRP in a population is log (normal), such that most individuals have intermediate values of CRP. Thus, a large proportion of events would be expected among people with nearly average CRP values, which explains the wide overlap in the distribution of CRP values among those who remain disease free and those who later suffer events (whether fatal or nonfatal) (respectively 2.5 mg/l [95% CI 1.2–5.3] and 3.7 mg/l [1.6–9.1] in the study by Bruno et al. [4]). With the broad overlap, it is difficult to set CRP cut point values that adequately discriminate later cases from those remaining disease free, accounting for the modest C-statistic. Additionally, the modest incremental risk prediction of CRP also stems from the correlation of CRP with a range of established risk factors (e.g., lipids, BMI, blood pressure, and glycemia indexes), an observation confirmed by Bruno et al. (4).

In a recent systematic review of 31 prospective cohorts (1), CRP did not perform better than the established Framingham risk equation for discrimination. Moreover, improvement in risk stratification or reclassification from the addition of CRP was small and inconsistent (1). As such, the conclusion by Bruno et al. (4) regarding the marginal value of CRP in predicting mortality risk in diabetic patients should not be too surprising. The efforts of the Emerging Risk Factors Collaboration (5), which has established a central database of over 1.1 million participants from 104 prospective population-based studies, in which subsets have information on inflammatory markers as well as major cardiovascular morbidity and cause-specific mortality, will help define whether there are subgroups of risk in which CRP measurement will be more helpful.

Of course, CRP is not the only “new” risk factor being considered for a role in CVD risk prediction (6). Indeed, measures of obesity, socioeconomic status, and family history of premature CVD all in part capture inflammation signals, and all may be more easily obtained than CRP. A significant correlation to inflammation is also evident for the ankle-brachial index (7), a measure that appears to significantly enhance risk prediction beyond the Framingham risk score (8).

Irrespective of its use as a risk predictor (which does not require that CRP be causally involved in vascular disease), there is equal interest in whether CRP is causal and thus whether lowering it will reduce CVD risk. Established and proven causal risk factors such as blood pressure and cholesterol are individually poorly predictive of incident disease (9). This is again because of the log-linear association with risk and the normal distribution of these risk factors in populations, which again means that a major proportion of events occur among those with average blood pressure or cholesterol. Thus, a risk factor that is causally involved in atherosclerosis need not be a good predictive tool. Although some in vitro studies have suggested a causal role for CRP in atherogenesis, some of the effects may have been due to the contamination of commercial CRP preparations (10). Observational studies are also not able to assess unequivocally whether CRP is causally related to vascular disease because higher CRP values in people at higher risk of disease could simply mark other risk factors (confounding) or subclinical atherosclerosis (reverse causation). In the absence of a clinical trial of a specific CRP-lowering agent, genetic studies utilizing common variants in the CRP gene that affect its level can be used as a type of natural randomized trial (11). Genetic variants are allocated at random according to Mendel's Laws (Mendelian randomization), which means that factors that could confound the association of CRP itself with disease risk should distribute evenly among the genotypic groups (12). If CRP was causally involved in vascular disease, individuals with a high CRP genotype should have a greater degree of atherosclerosis and a higher risk of incident events. Such studies to date provide no strong evidence for a causal role for CRP in atherosclerosis or the risk of a first vascular event (13,–15). A very-large-scale analysis using this approach is now in progress (16). Moreover, multiple observations demonstrating that CRP is as strongly linked to risk of noncardiovascular mortality (2) further argue against a causal role for CRP in CVD but suggest that the association might, in part, reflect reverse causation. Of interest, interleukin (IL)-6 may be linked more strongly to risk for CHD events in the general population than CRP (17), and there are more reasons to consider IL-6 as a potentially causal factor. However, such work is in its infancy, and no firm conclusions on IL-6 can currently be made.

Some commentators have argued that the results of the recent JUPITER (Justification for the Use of Statins in Primary Prevention: an Intervention Trial Evaluating Rosuvastatin) trial provide confirmation of the value of CRP in determining CVD risk (18). However, JUPITER was a randomized controlled trial of a statin in people with low LDL cholesterol but high CRP. Statins lower CRP but also lower LDL cholesterol, and so it can be difficult to distinguish whether reductions in risk observed were due to LDL lowering, CRP lowering, or some combination. This is the case even among people with low starting LDL cholesterol concentration because of the log-linear association of LDL cholesterol with vascular risk over the whole range of usual values (19), which suggests that a salutary effect of LDL cholesterol lowering might be expected irrespective of the starting LDL cholesterol value.

In summary, therefore, although CRP commands much interest as a CVD risk factor, higher-resolution observational epidemiology (such as that in the Emerging Risk Factors Collaboration [5]), continued genetic epidemiology, and eventually specific CRP inhibitors (20) are needed to help the vascular risk community make a more balanced judgment on any causal role and utility of CRP as a predictor of risk. Such efforts also need to be cognizant of a range of competing and overlapping risk factors, since some may have distinct benefits over CRP. Presently, therefore, the current focus in clinical practice should remain on established risk factors (e.g., smoking, lipids, and blood pressure), both in determining coronary heart disease risk and as targets for prevention.