Improving prediction of anticoagulant‐related major bleeding in atrial fibrillation: The search for new biomarkers

Abstract

While highly efficacious for stroke prophylaxis in patients with atrial fibrillation (AF), oral anticoagulants (OAC) carry a risk of major bleeding, including intracranial hemorrhage (ICH). Clinicians concerns regarding this risk have contributed to a substantial underuse of OACs in eligible patients with AF. Compared with warfarin, the direct oral anticoagulants (DOAC) have a more favorable bleeding profile,1 which has been an important factor in their widespread adoption over the last decade. Still, in the pivotal DOAC trials, the annual incidence of major bleeding ranged from 1.6% to 3.6%.2– 5 In view of the tradeoff between less stroke and more bleeding with anticoagulation, clinicians must carefully weigh these risks when making decisions about initiating OAC for AF. Unfortunately, clinical risk tools for assessing OACrelated bleeding risk, such as the HASBLED score, are notoriously limited in their prognostic performance.6 As such, professional society guidelines continue to discourage their use for withholding OAC from patients with AF.7 In this issue of JTH, Siegbahn and colleagues report intriguing new data that reveal possible novel biomarkerbased approaches to enhance this critical stratification of risk.8 Accumulating evidence suggests that circulating biomarkers may improve the prediction of OACrelated bleeding risk in this patient population.9 The ABC (Age, Biomarkers, Clinical history)bleeding risk score, which was developed by many of the coauthors of the present study, incorporates three biomarkers— growth differentiation factor15 [GDF15], highsensitivity cardiac troponin [hsTn], and hemoglobin— to estimate OACrelated major bleeding risk in patients with AF.10 This score has now been validated both in patients receiving and not receiving anticoagulation11,12; however, it remains controversial whether the ABCbleeding risk score should supplant bleeding risk scores based on clinical variables only.7,13 One of the enduring challenges of bleeding risk stratification in patients with AF is that many of the clinical risk factors for bleeding are also risk factors for stroke, including advanced age and renal dysfunction.14 Similarly, higher concentrations of both GDF15 and hsTn are not only associated with higher bleeding risk but also with higher stroke risk.11 Given this limitation of the circulating biomarkers that have attracted the most attention to date, there is room for additional, ideally more specific, biomarkers for optimal prediction of OACrelated bleeding. Siegbahn and colleagues used multiplex protein screening to identify novel biomarkers of OACrelated major bleeding risk in 4200 patients with AF enrolled in the ARISTOTLE (Apixaban for Reduction in Stroke and Other Thromboembolic Events) trial (identification cohort) and 1368 patients with AF enrolled in the RELY (Randomized Evaluation of LongTerm Anticoagulation Therapy) trial (replication cohort).8 Traditional candidate biomarker studies evaluate outcome associations of a limited set of biomarkers, which may be selected based on specific mechanistic hypotheses. By contrast, targeted proteomic approaches, like the one used in this analysis, cast a wider net, with simultaneous testing of a broader range of biomarkers. Although this strategy creates a “multiple comparisons” problem (and thus requires careful attention to control of type I error), it also affords the opportunity to discover pathobiological axes of risk that may not have been previously anticipated. Indeed, targeted proteomic discovery efforts have recently been employed to identify potential novel biomarkers of stroke risk in patients with AF.15,16 In this analysis, Siegbahn and colleagues measured 268 unique protein biomarkers in plasma samples collected at the time of randomization using conventional immunoassays and the Olink Cardiovascular Disease (CVD) II, CVD III, and Inflammation proteomic panels, which leverage oligonucleotidelabeled antibodies for protein detection. The study designs varied between the