Poor predictive value of high-sensitivity C-reactive protein indicates need for reassessment.

Inflammation appears to be an integral part of the process of atherosclerosis that leads to coronary artery disease (CAD) and acute ischemic syndromes (1, 2). Data indicate that high-sensitivity C-reactive protein (hs-CRP) and other inflammatory markers are associated with atherosclerosis and that hs-CRP decreases with statin treatment (3, 4). These are important findings that support the inflammatory disease hypothesis. Moreover, it was suggested that hs-CRP could be used to assess risk of CAD for clinical purposes (5 ), and several hs-CRP assays are commercially available (6 ). In this Journal, Rifai and Ridker (7 ) suggested an algorithm for assessing risk of CAD based on hs-CRP in conjunction with the ratio of HDL-cholesterol to total cholesterol. The algorithm was later modified to use hs-CRP cutoffs of 1, 1–3, and 3 mg/L in conjunction with the LDL-cholesterol (LDLC) concentration or the Framingham 10-year risk assessment (8 ). Several editorials that have accompanied reports on hs-CRP cautioned that use of this test for clinical purposes was premature (9–12). In spite of these warnings, an American Heart Association/CDC Scientific Statement (13 ) recommended the use of this test to enhance risk evaluation in certain population groups, although they noted that the benefits of this strategy or any treatment based on it remain uncertain. The clinical recommendations of others argued for even wider use of the test, suggesting that it should be an adjunct for initial screening for global risk assessment in conjunction with conventional lipid testing (8 ). We contend that because of its poor predictive value, until its use is better demonstrated, hs-CRP should not be recommended for defining risk. As discussed below, our argument is based on recently published reports and reassessment of previously published reports. We have shown that hs-CRP exhibits a very low Bayesian positive predictive value when used alone as a marker for predicting CAD (14 ). The positive predictive value in the highest quartile for hs-CRP was calculated to be 0.86%. hs-CRP varies with many conditions, e.g., acute infections, chronic diseases, smoking, hormone replacement therapy, obesity, age, diabetes, and atrial arrhythmias, and it is associated with the metabolic syndrome that is strongly linked to a proinflammatory state (14, 15). A recent report in this Journal indicated that 40% of women between ages 30 and 39 years have hs-CRP concentrations 3.5 mg/L and that this high percentage was similar in healthy women and those treated with hormone replacement therapy (16 ). However, the risk of a coronary event in these women is very low, which is consistent with the poor predictive value of hs-CRP. In another recent article, Danesh et al. (17 ), who studied 2459 persons who had suffered a coronary event, concluded that recommendations regarding the use of hsCRP in predicting the likelihood of disease should be reviewed because the adjusted risk ratio ( 1.45) was much lower than indicated in some previous reports. This conclusion was reinforced by a metaanalysis comparing earlier studies with those conducted more recently (17 ). On the basis of our previous calculations (14 ), after adjustment for conventional risk factors, this would translate into a much lower positive predictive value than the 0.86% that we calculated previously. An editorial (18 ) accompanying the publication by Danesh et al. (17 ) noted that further research is needed to clarify the usefulness of hs-CRP in the clinical setting. In yet another recent report, risk ratios were calculated after binning of data for 27 939 women into deciles (19 ). For the entire cohort, after adjustment for Framingham risk factors, the risk ratios for hs-CRP values between 0.64 mg/L (third decile) and 7.73 mg/L (ninth decile) ranged between 1.7 and 2.1, with substantial overlap between confidence intervals (CIs; 1.0–2.8 for the third decile and 1.3–3.4 for the ninth decile). This indicates no significant discrimination over this wide range of hs-CRP concentrations and is in agreement with the conclusions suggesting very poor positive predictive values (14 ). The discrimination was even poorer when adjusted for diabetes. After results for women on hormone replacement therapy were removed from the analysis, there was some improvement in discrimination, but the fourth through eighth deciles varied between risk ratios of 1.7 and 1.8 and still indicated no significant discrimination based on CIs. In the editorial (18 ) accompanying the report by Danesh et al. (17 ), Professor Tall pointed out that it remains unclear whether hs-CRP is a marker for a lowgrade inflammatory systemic state, a marker for the metabolic syndrome, or a marker for both (18 ). Regarding this syndrome, which is rapidly becoming the largest risk factor for CAD, after analysis of hs-CRP in 14 719 women, it was concluded by Ridker et al. (8, 20) that hs-CRP adds independent prognostic information on risk at all degrees of severity of the metabolic syndrome. After an appraisal of the data used to come to this conclusion (20 ), we came to a different conclusion, namely, that hs-CRP concentrations add little clinically useful prognostic information to the factors defined by the National Cholesterol Education Program (NCEP) ATP III for assessing risk (15 ). Table 3 from the article testing 14 719 women is partially reproduced below as Table 1. In the third and fourth rows of Table 1, all groups show an increased risk. In addition, there is no overlap of the CI with 1.0, which indicates that this is a significant difference. We conclude from the data in these rows that all persons with metabolic syndrome should be treated to reduce its potential effects based on NCEP guidelines (15, 21) regardless of the concentration of hs-CRP. There is no evidence that treatment should be held back on the basis of this test. Opinion