Is it important to measure or reduce C-reactive protein in people at risk of cardiovascular disease?

Systemic and local vascular inflammation is implicated in atherogenesis. High-sensitivity (hs) assays detecting low concentrations of C-reactive protein (CRP) in healthy individuals have delineated associations of this inflammation marker with cardiovascular events years later. A 160 309-person participant level meta-analysis of 54 prospective studies found the relationship between log-CRP concentration and cardiovascular disease (CVD) events to be linear, with a 2.5-fold risk difference in individuals at opposite extremes of the CRP distribution.1In vitro studies and animal experiments point to potentially atherogenic actions of CRP,2 and statins lower both CRP and low-density lipoprotein cholesterol (LDL-C).3 The Justification for the Use of statins in Primary prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER),4 designed to ‘assess the effect of rosuvastatin on first ever cardiovascular events in apparently healthy men and women who do not qualify for statin therapy due to low levels of LDL-C, but who are at increased cardiovascular risk due to elevated levels of hs-CRP’, was halted because of ‘unequivocal evidence of a reduction in cardiovascular morbidity and mortality among patients who received CRESTOR (rosuvastatin) when compared to placebo’ (http://clinicaltrials.gov/ct2/show/NCT00239681). Consequently, the US Food and Drug Administration approved an amended license for rosuvastatin for the primary prevention of CVD events in men and women over 50 and 60 years, respectively, with one other risk factor and a CRP concentration of >2 mg/L. Even before the publication of this trial, the estimated 5 million ‘hs-CRP’ tests were being ordered annually in the USA,5 and updated Canadian guidance on CVD prevention6 now recommends the consideration of ‘hs-CRP’ measurement in those at intermediate risk. The 2010 American College of Cardiology Foundation/American Heart Association Guidelines stated that in ‘men 50 years of age or older and women 60 years of age or older with an LDL-C < 130 mg/dL (3.38 mmol/L) … CRP can be useful in selection of patients for statin therapy’.7 However, the European view has been more cautious. European Society of Cardiology Guidelines considered the measurement of CRP for CVD risk prediction ‘was premature’.8 The European Medicines Evaluation Agency (EMEA) opted to license rosuvastatin in individuals at a high risk of CVD based on the ‘established’ risk factors but was criticized for its failure to incorporate recommending CRP measurement in the licensing decision.9 Here, we critically appraise evidence on the measurement of CRP for CVD risk assessment and targeting of statins and the role, if any, of CRP lowering in the prevention of CVD events. Despite the known causal association of blood pressure (BP) and LDL-C with CVD, and their inclusion in multivariable risk prediction models, it is known that such models are poor at discriminating between individuals who do, or who do not, eventually suffer events.10,11 About 50% of all coronary events occur among individuals with an average level of cholesterol (or at intermediate absolute risk).12,13 Moreover, many CVD events occur even in statin-treated populations. Therefore, interest in reducing the burden of events among those at intermediate risk and those already treated with statins provoked interest in new biomarkers like CRP as predictive tests, therapeutic targets, or both. Three distinct proposals can be distilled from the literature on CRP: first, that CRP measurement aids risk prediction; second, that CRP plays a causal role in CVD, like LDL-C; and third, that CRP measurement identifies an otherwise concealed subgroup of individuals who gain more from statin treatment. But how well does the available evidence support these assertions? The association of a marker with coronary disease fails to guarantee performance as a predictor, even if causal.14 The association of CRP and coronary heart disease (CHD) events is log-linear in shape, and the distribution of CRP in populations is log-normal.1 Thus, as for LDL-C, two systematic reviews found that the majority of CHD events occur among the many individuals with intermediate values of CRP, with a wide overlap in CRP concentration among later cases of CVD and those remaining disease free11,15 (Figure 1). The overall sensitivity (or the disease detection rate) was estimated to be 11% for a 5% false-positive rate. The summary area under the receiver operating characteristic (ROC) curve (an index of the ability of a marker to discriminate cases) for a range of CRP cut-points was 0.57, close to the 0.5 value that indicates no discrimination.11,15 Relative frequency distributions of C-reactive protein values in individuals who subsequently had a coronary heart disease event (affected) and individuals who remained healthy (unaffected), based on data from 22 prospective studies (figure adapted from Wald et al., J Med Screen 2009;16:212–214, © Massachussetts Medical Society15). Detection rates (or sensitivity, i.e. the proportion of those who developed events who tested positive) for three C-reactive protein cut-points (9.74, 6.65, and 3.8 mg/L) are shown corresponding to false-positive rates of 5, 10, and 20%, respectively. The dotted line corresponds to the JUPITER cut-point of 2 mg/L. Two further limitations to the wider implementation of risk prediction based on a single CRP cut-point value are: first, that population average CRP levels differ according to ethnicity,16 with the differences not being fully explained by differences in cardiovascular risk; and second, that two individuals with the same CRP value, but with a different constituency of risk factors, can have a very different absolute risk CVD (Figure 2). Although CRP is included in the multivariable Reynolds risk score,17 ‘adding’ CRP to scores that already include the established risk factors only marginally improves the area under the ROC curve, partly because CRP is itself associated with these risk factors. Using calibration as a metric, risk models with and without CRP perform almost equivalently.11 CRP also provides little useful reclassification when added to models based on the established risk factors (see Supplementary Data) if reclassification tables are presented appropriately and interpreted on the basis of effect size rather than P-values. (A) Low-density lipoprotein cholesterol is one of several determinants of the absolute risk of cardiovascular disease. Therefore, two individuals with the same low-density lipoprotein cholesterol of 6 mmol/L can be at substantially different risk of cardiovascular disease. (B) Similarly, two individuals with the same C-reactive protein value of 2 mg/L may also be at a different risk of cardiovascular disease. A 10-year risk of cardiovascular disease data for two individuals with differing levels of risk factors were generated using http://www.mycvrisk.co.uk and http://www.reynoldsriskscore.org/, respectively. Thus, judicious interpretation of the evidence indicates that CRP on its own does not usefully discriminate CHD events and, at best, only marginally improves discrimination, calibration or reclassification when included in the established risk models. Statins reduce both LDL-C and CRP.18 In statin-treatment trials, on-treatment CRP concentration is associated with the CVD event rate.19,20 These observations have been interpreted as indicating: (i) that CRP lowering contributes mechanistically to the prevention of CVD events and (ii) the effect is independent of the lowering of LDL-C. But are these interpretations consistent with the evidence? That CRP lowering contributes mechanistically to the benefit of statins implies that CRP itself is ‘causally’ involved in atherosclerosis and its complications. However, convergent evidence suggests that this is not the case. The reportedly deleterious effects of CRP on vascular cells and tissues, used in evidence of a proatherogenic effect, are likely to have been artefactual and due to sodium azide preservative and endotoxin in commercial CRP preparations.21,22 Experiments using pure CRP, free from azide or endotoxin, did not reproduce these effects. Reports of atherosclerosis in mouse models engineered to overexpress human CRP were also not reproduced.23 The association of CRP with CHD in observational studies in humans could be explained by confounding, because CRP is also associated with a very large number of established or suspected risk factors1 or by reverse causation because preclinical atheroma (known to be present from the second decade of life) could provide an inflammatory stimulus for hepatic CRP synthesis. In the individual participant-level meta-analysis of CRP, progressive adjustment for lipids, smoking, and fibrinogen, progressively attenuated the association of CRP with vascular events.1 Single-nucleotide polymorphisms (SNPs) in the CRP gene associated with differences in the CRP level of ∼15–20% per allele are fixed throughout life and unaffected by disease (overcoming reverse causation). As a consequence of their naturally randomized allocation at conception, such SNPs also have no association with confounders, in contrast to CRP itself, akin to a natural trial of a CRP-modifying treatment. However, CRP SNPs are not associated with carotid atheroma, or CVD events,24,25 suggesting that CRP does not influence these outcomes.26 What of statin-induced CRP reductions? A participant-level meta-regression analysis from the Cholesterol Treatment Trialists (CTT) indicates that the benefits of statins can be explained by the LDL-C reduction.27 Although statin-induced reductions in CRP correlate poorly with LDL-C reductions within trials, the reduction in CRP and LDL-C is strongly associated when examined ‘across’ trials (r = 0.80, P < 0.001); the poor correlation within study being explained by the measurement error of the two analytes.18 The statin effect on CRP could operate through rather than independently of LDL-C or may be an off-target consequence of statin treatment. Non-statin LDL-C-lowering interventions also reduce the CHD risk to an extent consistent with their LDL-lowering effect,28 casting doubt on the clinical relevance of the CRP-lowering action of statins.27 What then accounts for the reported association of the ‘on-treatment’ CRP concentration with clinical outcome in the statin trials? A potential explanation is that subgroups defined post hoc according to the achieved CRP value differ in other ways, including the mean on-treatment values of the other risk factors with which CRP is associated and which may not have been adjusted for.29 Such non-randomized comparisons may be as prone to confounding as non-randomized observational studies. A recent substudy from the Anglo Scandinavian Outcomes Trial, in which the appropriate adjustments were made, failed to find evidence for an association between on-treatment CRP and cardiovascular outcomes.30 In a post hoc, subgroup analysis of the AFCAPS/TexCAPs primary prevention trial of lovastatin, the relative risk reduction (RRR) was statistically significant in subgroups with a high starting value of CRP, even if the LDL-C value was low, but was non-significant in a group with a low value of LDL-C as well as low CRP.31 The interpretation was that individuals with a high CRP might be targeted as a distinct high-risk group particularly responsive to statins. However, tests of significance within trial subgroups can be misleading. The preferred analysis, a statistical test for a treatment–subgroup interaction (Figure 3), did not corroborate the original interpretation. When the same question was addressed in another statin trial (PROSPER),32 the correct analysis provided no evidence for a CRP subgroup by treatment interaction (Table 1). Among 20 536 participants from the Heart Protection Study, the proportional reduction in the incidence of a first major cardiovascular event was also consistent in patients from six different categories of baseline CRP values from <1.25 to ≥8 mg/L.29 A prior analysis also showed that age, gender, BP, diabetes, smoking, or starting value of high-density lipoprotein cholesterol (all associated with CRP concentration) do not modify the treatment effect of statins.27 Effect of pravastatin on cardiovascular events by tertile of baseline C-reactive protein in the PROSPER trial32 HR, hazard ratio. Negative numbers represent a lower event rate in pravastatin- vs. placebo-treated subjects. Data are presented for all subjects and are subdivided into those with and without a history of vascular disease. The HR was adjusted for treatment allocation, age, sex, country, LDL-cholesterol, HDL-cholesterol, systolic BP, smoking status, history of diabetes, and history of hypertension. Probability values are the significance of interaction term for tertile-by-treatment effect. aCHD death, nonfatal myocardial infarction, or fatal or nonfatal stroke. Effect of pravastatin on cardiovascular events by tertile of baseline C-reactive protein in the PROSPER trial32 HR, hazard ratio. Negative numbers represent a lower event rate in pravastatin- vs. placebo-treated subjects. Data are presented for all subjects and are subdivided into those with and without a history of vascular disease. The HR was adjusted for treatment allocation, age, sex, country, LDL-cholesterol, HDL-cholesterol, systolic BP, smoking status, history of diabetes, and history of hypertension. Probability values are the significance of interaction term for tertile-by-treatment effect. aCHD death, nonfatal myocardial infarction, or fatal or nonfatal stroke. Re-analysis for a treatment–subgroup interaction in the AFCAPS/TexCAPs study comparing the RRR for individuals with a high starting C-reactive protein value and individuals with low levels of low-density lipoprotein and C-reactive protein.31 In summary, critical scrutiny of the evidence prior to the JUPITER trial reduces confidence in the role of CRP as a predictive test, therapeutic target, or to guide statin therapy. What then could explain the apparently dramatic findings of the JUPITER trial in people with a high CRP but low LDL-C? In the JUPITER trial, screening of 90 000 individuals was required to identify 17 802 with the low LDL/high CRP profile necessary for inclusion.33 This placebo-controlled trial lacked a comparator arm of individuals with a similar CVD risk but with an average or low CRP. Scheduled to last 4 years, the trial was halted early by the independent Data and Safety Monitoring Board because of ‘unequivocal evidence of a reduction in cardiovascular morbidity and mortality in patients treated with rosuvastatin compared to placebo’. Prior statin trials had supported the prediction of observational epidemiology that LDL-lowering should produce a similar ‘proportional’ reduction in the CVD risk at all levels of the absolute risk, even among those with below-average LDL values. So did the findings of the JUPITER trial actually deviate from expectation? The perception of a larger than expected RRR was supported by a graph in the Supplementary Data of the trial report, reproduced here as Figure 4A, in which the JUPITER trial appears as an apparent outlier. However, in a formal meta-regression analysis, in which each trial contributes a separate data point, the JUPITER trial looks less like an outlier (Figure 4B). The meta-regression analysis also concurs with the updated CTT Collaboration meta-analysis that added a further seven trials (including JUPITER) to the previous overview.34 In the updated analysis, a larger LDL-C reduction was associated with a greater RRR in major cardiovascular events but there was no significant residual variation between trials after adjustment for LDL-C differences. Indeed, in JUPITER itself, the LDL reduction and relative hazard of a CVD event during rosuvastatin treatment did not differ by the stratum of CRP, or by stratum of Framingham risk score, with which CRP is highly correlated.35 Relationship of the proportional reduction in the cardiovascular event rate and mean low-density lipoprotein cholesterol difference between treatment groups in published statin trials. (A) Figure originally reported in Ridker et al. with permission from Elsevier.4 The grey square represents summary data from randomized trials of statin therapy vs. placebo as summarized by the CTT Collaboration, and solid squares represent results of individual trials comparing different intensities of statin therapy. Open circle represents projected benefit of rosuvastatin and closed circle represents observed benefit in the JUPITER trial. Vertical lines are 95% confidence intervals. (B) A new meta-regression analysis based on the following trials: LIPS, AFCAPS/TEXCAPS, LIPID, CARE, PROSPER, WOSCOPS, Post-CABG, ASCOT-LLA, CARDS, HPS, 4S, TNT, Prove-IT, A to Z, IDEAL, and JUPITER. We previously simulated the JUPITER outcome based on established risk assessment methods and the predicted LDL-C-lowering, without consideration of CRP, and submitted the findings as a British Medical Journal rapid response, 24 h before the outcome of the JUPITER trial was first reported (http://www.bmj.com/cgi/eletters/337/aug28_2/a1371). At the time, we indicated that it would be ‘of interest to recalculate the expected treatment effect in the simulation “post-JUPITER”, based on the achieved LDL reduction in the JUPITER trial, … without measuring CRP’. The results of both (Table very the observed event the predicted a highly significant difference in event rates between treatment after 2 The JUPITER trial was after a of of cardiovascular disease events each in two simulated trials of a low-density lipoprotein cholesterol lowering statin of similar to rosuvastatin 20 The values are based on an average baseline CVD risk similar to the JUPITER trial, estimated using the Framingham risk without consideration of CRP, and on the known relationship between LDL-C-lowering and CVD risk estimated the average risk of CVD and CHD in JUPITER participants by the proportion of and using published on baseline at using an Framingham risk The CHD risk was estimated to be risk and the CVD risk risk The risk reduction from 20 rosuvastatin was based on an estimated LDL-C reduction of A meta-analysis of trials of estimated rosuvastatin 20 reduces LDL-C by but LDL-C reductions may be attenuated in the We estimated that LDL would be by with this of The corresponding expected relative reduction in the CVD risk of in in and in was using from et The absolute number of events is based on individuals in each arm and the of and that all participants are at risk throughout the trial. simulation was using a baseline CVD risk to the CVD event rate reported in the JUPITER trial itself and the LDL-C reduction of mmol/L observed in the trial. of cardiovascular disease events each in two simulated trials of a low-density lipoprotein cholesterol lowering statin of similar to rosuvastatin 20 The values are based on an average baseline CVD risk similar to the JUPITER trial, estimated using the Framingham risk without consideration of CRP, and on the known relationship between LDL-C-lowering and CVD risk estimated the average risk of CVD and CHD in JUPITER participants by the proportion of and using published on baseline at using an Framingham risk The CHD risk was estimated to be risk and the CVD risk risk The risk reduction from 20 rosuvastatin was based on an estimated LDL-C reduction of A meta-analysis of trials of estimated rosuvastatin 20 reduces LDL-C by but LDL-C reductions may be attenuated in the We estimated that LDL would be by with this of The corresponding expected relative reduction in the CVD risk of in in and in was using from et The absolute number of events is based on individuals in each arm and the of and that all participants are at risk throughout the trial. simulation was using a baseline CVD risk to the CVD event rate reported in the JUPITER trial itself and the LDL-C reduction of mmol/L observed in the trial. These support the view that the RRR observed in JUPITER was within the expected from the of LDL-C residual might be explained by the of the trial which is known to to of treatment individuals with a similar absolute risk but with a different of risk factors should the same risk reduction from cholesterol lowering with statins as the JUPITER but would be if the high LDL-C were as a of population for statins based on the absolute risk than but with a lower for is likely to be more and of men over 50 and women over 60 would be expected to have a high CRP and a low LDL-C, based on screening and to the JUPITER Therefore, the number to = of would be In based on the for men for over age and women for the age of 60 would be for statins based on a absolute risk estimated using established risk factors, a rate than using the JUPITER Moreover, the CRP measurement with an is in risk assessment is likely to have a larger than risk based on the established risk factors. an age has been as an even for statin treatment. About 95% of all CHD events occur the age of 50 in men and in and age later cases of CHD as well of the available risk factors or risk As well as the of these screening it would be to their to and, of individuals being targeted for Despite two of observational and a large clinical trial, and early in scrutiny of the available evidence does not provide evidence for the CRP measurement for risk prediction or the targeting of that CRP lowering is a therapeutic for the primary prevention of The with CRP should now the of other biomarkers of CVD risk. was supported by a British Heart was supported by a British Heart is by the to the but was not for this of has received for at to risk factor and primary in or to We the who the which was not involved in with this

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.