Author + information
- aInova Heart and Vascular Institute, Falls Church, Virginia
- bDivision of Nephrology/Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland
- ↵∗Address for correspondence:
Dr. Christopher deFilippi, Inova Heart and Vascular Institute, 3300 Gallows Road, Suite 1225, Falls Church, Virginia 22042.
For the inaugural issue of JACC: Heart Failure, Dr. Eugene Braunwald presented a seminal review in which he identified the following 7 domains contributing to the pathophysiology of heart failure (HF): myocardial stretch, myocyte injury, matrix remodeling, inflammation, renal dysfunction, neurohormonal activation, and oxidative stress (Figure 1). Each pathophysiological domain can potentially be represented by soluble biomarkers, which he defined as “substances measured in the blood other than genetic markers, electrolytes, and commonly used markers of hepatic or renal function” (1). Such an approach is supported by the observations that—for atherosclerotic cardiovascular (CV) diseases—combinations of biomarkers representing different pathophysiological domains may be able to both prognosticate and guide medical therapy in patients with stable disease (2).
Procedures for validating a prognostic biomarker for HF have been previously proposed in this journal, and recommendations of an optimal study design to establish a prognostic model have been standardized (3,4). Much of this has been in response to what is perceived as overly optimistic initial publication bias toward biomarkers, which has often been dampened by more modest findings from larger and more heterogeneous populations (Online Appendix). Prior studies in patients with chronic systolic HF that have carefully examined biomarkers from multiple domains have shown positive results, but have experienced from small sample size (Online Appendix) or from incorporating a well-established biomarker, such as a natriuretic peptide, into the model, leaving uncertainty as to the additional prognostic information contained by novel biomarkers representing different pathophysiological domains (Online Appendix). More recently, the investigators of the PROTECT trial took a fairly agnostic approach by looking at both multiple time points for optimal measurement of 48 circulating biomarkers and then multiple follow-up time points for optimal prognosis of outcomes in patients with acute HF (Online Appendix). A statistically rigorous approach required by the “roadmap” for validating the prognostic importance of a biomarker in patients with HF relies on additional testing of multiple statistical measures of accuracy, including discrimination (such as the c-statistic, a measure of how well a marker distinguishes those who have vs. do not have the adverse event), calibration (how well a biomarker predicts the actual probability of an outcome), and reclassification (such as the net reclassification improvement) (3). Even with single biomarkers, there are different statistical methods available to estimate each property, and results may also depend on the “base” model of established risk factors being used. For a multibiomarker panel, there are additional challenges in the selection of individual markers, each of which may have differential correlation with other novel biomarkers and with established risk factors such as age, sex, or renal function. Although approaches to evaluating 48 biomarkers in the PROTECT study (Online Appendix) or the 20 biomarkers considered by Nymo et al. (5) in the CORONA (Controlled Rosuvastatin Multinational Trial in Heart Failure) study in this issue of JACC: Heart Failure require additional rigor, these are just a warm-up of what to expect in the near future. With advances in soluble protein multiplex technology, one might soon anticipate a need to consider hundreds of biomarkers that all represent putative processes that can be referred to 1 of the 7 domains proposed by Braunwald (1). New proximity extension assay technology using oligonucleotide-labeled antibody probe pairs may offer significant advantages over conventional multiplex immunoassays with respect to specificity. Only correctly matched antibody pairs will provide a signal and give rise to new deoxyribonucleic acid amplicons. The deoxyribonucleic acid amplicons are then quantified on a polymerase chain reaction platform. Such an approach has already been applied to atherosclerosis and atrial fibrillation (Online Appendix). One would expect this technology to soon be applied to a plethora of biomarkers potentially associated with prognosis in chronic HF. Without a thoughtful, systematic approach to evaluating multibiomarker predictive models, it could rapidly become challenging for reviewers and readers alike to interpret the findings from such an onslaught of data.
Nymo et al. (5), using the well-characterized CORONA trial and 20 previously measured biomarkers representing the 7 spokes of chronic HF pathophysiology, test a methodology for developing and validating biomarker selection for long-term prognosis of a composite clinical outcome and its individual components. Next, they examine the influence of the study drug rosuvastatin on the biomarker levels over time to provide insight into the potential physiological mechanisms of HF that may be influenced by this drug. Last, they examine whether multiple biomarker levels can identify specific groups of participants who may derive benefit from rosuvastatin in an overall neutral study (Online Appendix). Multiple biomarkers have been well studied in the CORONA study and have been the subject of prior publications, as outlined by the authors of the present paper. Based on the proposed roadmap for validation of new HF prognosis biomarkers, they initially establish the association of a multibiomarker panel with the primary composite outcome (CV mortality, nonfatal myocardial infarction, and stroke), in addition to individual outcomes of CV-related and total mortality separately. They then provide estimates of the 3 statistical properties including calibration, incremental discrimination, and incremental (or net) reclassification when compared to the Seattle Heart Failure Score (SHFS) and a CORONA-specific prognostic score. The authors likewise present a reasonably robust process for selection of individual markers from among the 20 potential markers for their panel, using 3 different statistically-driven methods and also a complementary knowledge-based approach using scientific judgment and review of the published data. Furthermore, they use separate nonoverlapping random subgroups of participants for biomarker selection and validation to minimize overly optimistic results that can occur when a model is derived and validated among the same individuals. Finally, they use two clinical “base” models, one derived from the CORONA cohort, and the second from the well-established SHFS. To both they add natriuretic peptide levels (N-terminal pro–B-type natriuretic peptide [NT-proBNP]), already a well-established and ubiquitously available prognostic biomarker.
Using the statistically-driven selection methods, the authors identify 6 biomarkers representing unique pathophysiological domains; a partially overlapping panel of 5 biomarkers is identified using the knowledge-based approach. Although both panels are associated with risk of the primary composite and secondary outcomes and are reasonably calibrated, there is only a modest improvement in the measures of discrimination and reclassification when added to the CORONA-based risk score of standard risk factors. Furthermore, even the small improvement in net reclassification improvement observed was predominately due to more correct (i.e., lower risk) reclassification of those without the adverse event. It is uncertain whether such improved reclassification of lower-risk individuals has the same clinical importance and influence as correct reclassification of higher-risk patients. When added to the SHFS incorporating NT-proBNP, there was likewise minimal improvement in discrimination, especially for CV-related and total mortality, and the modest improvement in reclassification was again driven predominately by correct reclassification of lower-risk patients. As the authors note, even these estimates of prognostic accuracy may be optimistic in that they were derived in an internal rather than true external cohort of subjects.
Beyond prognosis, biomarkers could be ideally suited for a precision-medicine approach to target specific interventions, and provide biological plausibility for patient selection. Interestingly, unlike C-reactive protein in the CORONA study, the influence of rosuvastatin on this panel of inflammatory biomarkers was modest even if the changes were statistically significant, consistent with earlier suggestions that the anti-inflammatory effects of statins on factors influencing HF may be modest compared with factors influencing atherosclerotic disease. This is supported by the finding that a positive effect of rosuvastatin on adverse outcomes was only seen in subjects with the lowest tertile of biomarker scores, a finding first noted with NT-proBNP in this study (Online Appendix). Overall, the CORONA investigators in this study have done much to move the field toward a consistent methodology to evaluate multiple biomarkers for prognosis in chronic HF.
For supplemental references, please see the online version of this article.
↵∗ Editorials published in JACC: Heart Failure reflect the views of the authors and do not necessarily represent the views of JACC: Heart Failure or the American College of Cardiology.
Drs. deFilippi and Seliger receive grant support to their institutions from Roche Diagnostics. Dr. deFilippi is a consultant for Alere, Ortho Diagnostics, Roche Diagnostics, and Siemens Healthcare Diagnostics; and is on an endpoint committee for Radiometer.
- 2017 American College of Cardiology Foundation
- Braunwald E.
- Sabatine M.S.,
- Morrow D.A.,
- de Lemos J.A.,
- et al.
- Ahmad T.,
- Fiuzat M.,
- Pencina M.J.,
- et al.
- Nymo S.H.,
- Aukrust P.,
- Kjekshus J.,
- et al.