Coagulation Factors in Ischemic Heart Disease
Answers From a Mendelian Randomization Study Inspire Further Questions
See Article by Zhao and Schooling
In recent years, Mendelian randomization (MR) became the causal inference tool de rigueur in observational epidemiology. Originally proposed more than 3 decades ago,1 MR has risen in popularity because the evidence of spurious associations in observational studies emerged as a critical challenge to the validity of most epidemiological findings. Canonically, this problem has been illustrated by examples such as the Women’s Health Initiative, in which observational associations failed to reproduce in a randomized trial because of residual confounding and other forms of bias.2 MR promises to estimate unconfounded associations without requiring an actual randomized experiment, thus obviating many ethical and logistical hurdles—but relying on its own set of assumptions. In this issue of Circulation: Genomic and Precision Medicine, Zhao and Schooling3 used MR to explore potential causal associations between several coagulation factors and the risk of ischemic heart disease (IHD) using data from the CARDIoGRAM consortium (Coronary Artery Disease Genome Wide Replication and Meta-Analysis).
Understanding the biology of thrombotic risk in patients with IHD has significant implications for clinical practice. Recently, the large-scale COMPASS trial (Cardiovascular Outcomes for People Using Anticoagulation Strategies) (NCT01776424) demonstrated that an enhanced antithrombotic regimen that included addition of the factor Xa inhibitor rivaroxaban in patients with stable atherosclerotic vascular disease (90% of whom had a history of IHD) reduced the composite of cardiovascular death, stroke, or myocardial infarction by 24% relative to aspirin alone.4 However, the regimen also increased the risk of major bleeding by 70%, compellingly illustrating the need to identify novel therapeutic targets that may offer a favorable risk–benefit balance. Further etiologic insights could also lead to the development of individualized strategies to identify IHD patients at relatively higher thrombotic risk, perhaps through the use of genetic markers. Hence, the ambitions and potential extensions of the questions asked by Zhao and Schooling3 are highly relevant to clinical practice.
The basic paradigm of MR is as follows. During meiosis, genotypes at specific loci are established randomly and independently (Mendel’s law of independent assortment). This property can be exploited to mimic a randomized trial by using this random allocation of genotype (which serves as a proxy, or instrumental variable, for the exposure of interest) to explore effect on outcomes of interest. Within that framework, the measured and unmeasured confounders are (at least in theory) on average evenly distributed across genotype groups, and the resulting estimate of association between the genetic instrumental variable and the outcome is more resistant to bias because of such confounding. Instrumental variables may be either individual single-nucleotide polymorphisms (SNPs) or summary risk scores based on multiple variants. For most complex traits, summary scores make stronger instrumental variables than single variants because they explain a larger proportion of variance.5
To strengthen their inferences, Zhao and Schooling3 created summary genetic risk scores for each major coagulation factor based on top SNPs from published genome-wide association studies. Although genome-wide association studies for factor V, VII, Xa, or thromboxane were not available, the search for genetic instruments for vWF (von Willebrand factor), endogenous thrombin potential, d-dimer, factor VIII, tPA (tissue-type plasminogen activator), and its inhibitor (plasminogen activator inhibitor)-1 proved fruitful, enabling estimation of causal effects of these coagulation factors on the outcomes of interest: acute myocardial infarction and its composite with stable coronary artery disease (myocardial infarction), as well as plasma lipids.
Three key assumptions must be fulfilled to ensure validity of MR estimates (Figure). First, the instrumental variable must be robustly associated with its proxy, as Zhao and Schooling3 report in their Table I in the Data Supplement. Second, none of the SNPs included in the summary score can be directly associated with any potential confounders of the relationship between the coagulation factors and IHD. This assumption is challenging to test empirically because some confounders can be unknown or unmeasured. Zhao and Schooling,3 however, were able to partially address this assumption by conducting sensitivity analyses that excluded SNPs located in the ABO gene, a plausible confounder6,7 of the association of interest. Finally, the third assumption precludes any pleiotropic effects of the genetic predictors on the outcome, that is, each SNP included in the risk score must only affect the risk of IHD via its association with the coagulation factor for which it serves as the proxy. To satisfy that assumption, Zhao and Schooling3 checked for known association between the proxy SNPs and other phenotypes and tested for unknown pleiotropic effects using the weighted median and the MR-Egger method.8 In addition, the presented MR analysis was extended to test for the associations between genetically predicted coagulation factors and blood lipids (an intermediate phenotype on the pathway to IHD) using data from the Global Lipids Genetic Consortium.
Most robust finding by Zhao and Schooling3 implicated endogenous thrombin potential as a modest risk factor for both IHD outcomes, but paradoxically, also as a predictor of increased HDL (high-density lipoprotein) cholesterol. The other examined coagulation factors were inconsistently associated with the outcomes, with notable evidence of pleiotropic effects and ABO polymorphism involvement. Because of the difficulties inherent to validating some of the principal MR assumptions, these findings should not be construed as definitive evidence of causal relationships or lack thereof. Indeed, the investigators found that variation in F10—the gene encoding factor Xa— did not increase the risk of IHD. This is in contrast to aforementioned findings from the COMPASS trial, which reported considerable reductions in composite cardiovascular risk achieved with factor Xa inhibition. The trial was not, however, powered to look at the effects on myocardial infarction alone, and indeed this individual end point was nonsignificantly different between the 2 groups (P=0.14). The differences in composite cardiovascular risk could instead be driven by other clinical events (eg, stroke), which were not addressed in the study by Zhao and Schooling.3 Furthermore, the importance of these coagulation factors may vary between primary versus secondary prevention or between specific clinical populations, potentially limiting the generalizability of the etiologic inferences drawn from the CARDIoGRAM consortium data.
At the very least, the findings published in this issue of Circulation: Genomic and Precision Medicine3 represent a useful starting point for designing further investigations. Of note, Vanderweele et al9 have previously shown that by interpreting MR findings as the presence/absence of an effect rather than actual estimates, the impact of potential biases is reduced. Therefore, future studies can benefit from this evidence of a causal relationship between thrombin generation, plasma lipids, and IHD without unnecessary preoccupation with the exact magnitude of effect.
MR results by Zhao and Schooling3 open several opportunities for further inquiry. First, it is unclear whether the positive relationship between endogenous thrombin potential and IHD is partially mediated by plasma lipids, as the reported association with HDL cholesterol would suggest; furthermore, the underlying mechanisms remain to be elucidated. Second, ongoing clinical trials of anticoagulant therapy represent another domain of evidence that will need to be interpreted in the context of this study. Third, the lack of robust causal effects reported for coagulation parameters other than endogenous thrombin potential can help focus the search for the most promising therapeutic approach to IHD, once again highlighting the value of null findings to the scientific community.
Sources of Funding
NIH grant K01 HL136700 (Dr Aslibekyan).
Circ Genom Precis Med is available at http://circgenetics.ahajournals.org.
- © 2018 American Heart Association, Inc.
- Anderson GL,
- Limacher M,
- Assaf AR,
- Bassford T,
- Beresford SA,
- Black H,
- et al
- Zhao JV,
- Schooling CM
- Eikelboom JW,
- Connolly SJ,
- Bosch J,
- Dagenais GR,
- Hart RG,
- Shestakovska O,
- et al
- Song J,
- Chen F,
- Campos M,
- Bolgiano D,
- Houck K,
- Chambless LE,
- et al
- Chen Z,
- Yang SH,
- Xu H,
- Li JJ