Minimal and null predictive effects for the most popular blood biomarkers of cardiovascular disease.

Hundreds of thousands of papers have addressed cardiovascular biomarkers in the last few decades. Blood biomarkers in particular have been a very attractive topic of research efforts for myriads of investigators, given that they require only a blood draw. Emerging biomarkers hold diverse promises.1,2 In theory, they may inform about cardiovascular risk; reclassify patients into more appropriate risk categories versus what can be achieved on the basis of traditional risk factors; improve outcomes when used as appropriate guidance for treatment initiation, management, and monitoring; or point to other modifiable factors on which one can act. In this Perspective, we have reassessed the evidence on the performance of the most popular emerging blood biomarkers for cardiovascular disease. Popularity was defined on the basis of the number of PubMed articles retrieved on each of these biomarkers with a cardiovascular focus using the search string [cardiovascular OR heart OR cardiac OR myocardial OR coronary OR angina OR hypertension OR “blood pressure”] (search performed on January 14, 2012). We screened 36 emerging cardiovascular biomarkers measured in blood, serum, or plasma that are listed in 2 relevant reviews.2,3 This does not include traditional risk factors for cardiovascular disease that are already included in the Framingham Risk Score.4 Of those 36 candidates, we focus here on the 10 biomarkers that had at least 6000 articles each identified with the cardiovascular search string (Table 1). Search terms appear in the Appendix. Of note, not all of these articles directly address with primary data the main promises of biomarkers that are mentioned in the previous paragraph. As in any biomedical research field, approximately half of the articles are nonsystematic reviews, editorials, or news items, and many of the remaining half pertain to other, often tangential questions. Nevertheless, with a literature of anywhere between 6538 and 35 700 papers for each, clearly all of these biomarkers have been highly popular in the scientific literature. In fact, with 1 exception (brain-type natriuretic peptide), the overall literature for these biomarkers is far broader than the cardiovascular-related papers, which constitute only 10% to 40% of the total scientific output on them. Most of these biomarkers have had a very nonspecific trajectory, with interest attracted for their potential role in very diverse processes, conditions, and diseases. This is most prominent for “transdisciplinary” markers such as myeloperoxidase, serum albumin, and interleukin 6. How much can these popular markers inform cardiovascular risk prediction? Table 2 summarizes effect sizes from the most recent and largest meta-analyses for associations of these biomarkers with cardiovascular disease, focusing on studies in general populations. Preference was given to meta-analyses of individual-level data whenever available. We standardized the relative risks to correspond to risks per 1 standard deviation (SD) of each biomarker; otherwise, some effects may have appeared to be large simply because extreme exposure contrasts were selected.5 Given the potential prevalence of publication and other selective reporting biases in this literature, many of the relative risks derived from these meta-analyses are likely to be inflated.6 Nevertheless, despite the potential inflation, none of the listed biomarkers had a reported summary relative risk exceeding 1.45 per 1 SD in analyses that adjusted for other known cardiovascular risk factors (Table 2). One should acknowledge that the relative risk may not be an optimal metric to assess the discriminative ability of a biomarker; however, such relative risks are expected to translate to improvements in the range of 0.01 or less in the area under the curve when added to traditional risk factors.7,8 In fact, some of these markers have nonstatistically significant predictive effects that cannot be differentiated from the null, as in the case of triglycerides and intercellular adhesion molecule 1. In other cases (interleukin 6, serum albumin, homocysteine, and uric acid), the effects are very small, and the 95% confidence intervals exclude the possibility of even The opinions expressed in this article are not necessarily those of the editors or of the American Heart Association. Original received January 20, 2012; revision received January 26, 2012; accepted January 30, 2012. From the Stanford Prevention Research Center, Department of Medicine and Department of Health Research and Policy, Stanford University School of Medicine, and Department of Statistics, Stanford University School of Humanities and Sciences, Stanford, CA (J.P.A.I.), Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece and Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, UK (I.T.). Correspondence to John P.A. Ioannidis, MD, DSc, Stanford Prevention Research Center, 1265 Welch Rd, MSOB X306, Stanford, CA 94305. E-mail [email protected] (Circ Res. 2012;110:658-662.) © 2012 American Heart Association, Inc.