Leveraging information from genetic risk scores of coronary atherosclerosis

Abstract

Purpose of Review:

Genome wide association studies (GWAS) have identified ~60 loci for coronary artery disease (CAD). Through genetic risk scores (GRSs), investigators are leveraging this genomic information to gain insights on both the fundamental mechanisms driving these associations as well as their utility in improving risk prediction.

Recent Findings:

GRSs of CAD track with the earliest atherosclerosis lesions in the coronary including fatty streaks and uncomplicated raised lesions. In multiple cohort studies, they predict incident CAD events independent of all traditional and lifestyle risk factors. The incorporation of SNPs with suggestive but not genome-wide association in GWAS into GRSs often increases the strength of these associations. GRS may also predict recurrent events and identify subjects most likely to respond to statins. The effect of the GRS on discrimination metrics remains modest but the minimal degree of improvement needed for clinical utility is unknown.

Summary:

Most novel loci for CAD identified through GWAS facilitate the formation of coronary atherosclerosis and stratify individuals based on their underlying burden of coronary atherosclerosis. GRSs may one day be routinely used in clinical practice to not only assess the risk of incident events but also to predict who will respond best to established prevention strategies.

Introduction

Genome wide association studies (GWAS) have identified a plethora of new loci associated with a broad range of complex traits over the last decade including >60 loci directly associated with clinically significant CAD[1-5]. Each of these loci undoubtedly provides molecular biologists with a golden opportunity to better understand the pathophysiologic processes related to coronary atherosclerosis through mechanistic studies. However, the translation of this knowledge to novel and effective therapeutic agents for most loci will likely take several more years. In the meantime, scientists are leveraging this knowledge through genetic risk scores (GRS) to not only gain a better appreciation of the fundamental mechanisms behind these genetic associations but also to explore their utility in improving risk prediction.

What is a genetic risk score?

A genetic risk score (GRS) provides a means to aggregate the health-related risk of a collection of genetic alleles into a single number [6-8]. A GRS for an individual is calculated by summing the products between the number of high-risk variants inherited at a susceptibility polymorphism identified through GWAS and the log odds ratio for the same variant [9]. GRSs currently represent the most practical way of incorporating genetic risk into clinic practice for common polygenic disorders.

Studies examining GRSs of CAD loci have most frequently incorporated only signal nucleotide polymorphisms (SNP) that have reached genome wide significance (p<5 x 10⁻⁸) although more recent studies have included additional SNPs associated with CAD at more higher p value thresholds of significance[10-18]. Approximately one third of established CAD loci appear to influence risk through effects on traditional risk factors (TRFs)[5]. About three fourths of these SNPs track with plasma lipid levels with the remaining demonstrating evidence of association with blood pressure traits[5].

What are the primary mechanisms of action associated with genetic risk scores of CAD?

The pathophysiology of clinical CAD is complex but can be grossly divided into three major processes including the formation of coronary atherosclerotic plaque, plaque rupture or erosion, and the intraluminal thrombotic response to plaque rupture or erosion[19]. Determining which of these three processes is affected by genetic variation is challenging as we lack the ability to in-vivo non-invasively accurately quantify the burden of coronary atherosclerosis as well as the number of prior symptomatic and silent plaque ruptures[20-22].

One way to circumvent this technical challenge is to examine the relationship between susceptibility loci of clinical CAD and very early lesions of coronary atherosclerosis. Two recently published studies have taken advantage of the NIH’s Pathobiological Determinants of Atherosclerosis in Youth (PDAY) resource to document associations between a GRS involving up to 57 SNPs for CAD and the degree of subclinical coronary and aortic atherosclerotic lesions visualized during the autopsy of young adults dying between 15 and 35 years of age from non-cardiovascular causes [23, 24]. These studies found that the magnitude of association between the GRS and the degree of early raised but uncomplicated lesions not yet prone to plaque rupture or erosion in the right coronary was comparable to the association observed between the same GRS and the degree of fatty streaks in the right coronary as well as the degree of coronary artery calcification in older subjects[23, 24]. The GRS was less strongly associated with raised lesions and not at all associated with fatty streaks in the aorta[24]. Importantly, all associations were robust to the removal of risk factor SNPs[23, 24].

Collectively, these findings support the hypothesis that most newly identified GWAS loci for CAD predispose to plaque development rather than to plaque rupture or thrombosis [23]. The findings also suggest the presence of pathogenetic differences between fatty streaks in the coronary and aorta that may contribute to previously well-documented differences in their predisposition to progress to raised atherosclerosis lesions[25-27]. The less robust associations of the GRS with raised lesions in the aorta compared to the coronary suggests that at least some of the susceptibility loci for CAD operate exclusively in the coronary vascular bed. This hypothesis is already partially supported by population and mechanistic studies for the 9p21 and TCF21 CAD loci with the former locus clearly predisposing to atherosclerosis in all vascular beds but the latter locus demonstrating an effect only in the coronaries[28, 29]. The PDAY resource is too small to allow for definitive conclusions of mechanisms of action on a locus by locus basis. Thus, more studies using this approach are necessary before definitive conclusions can be made for individual CAD susceptibility loci.

A genetic risk score predicts CAD independent of traditional risk factors including family history

The strength of the association between a GRS and CAD does not appear to be influenced to any substantial degree by the inclusion of traditional risk factors into multivariate models[9, 12-16, 30, 31]. No clear explanations for these observations exist but regression dilution has emerged as a leading hypothesis to explain this phenomenon [32, 33]. Future studies may test this hypothesis by incorporating repeated measures of cholesterol and blood pressure into clinical risk scores.

The above-mentioned trends have been reinforced by two recently published studies where the relationship between a GRS with a family history (FHx) of CAD was also carefully examined [12, 13]. FHx would intuitively serve as a substitute for genetic risk but the magnitude of the association between FHx and CAD is not obviously reduced when a GRS is included into a multivariate prediction model [12, 13]. No clear explanations for these observations exists but they do suggest that family history may reflect exposure to common familial environmental risk factors of CAD to a larger degree than previously suspected. Alternatively, current GRSs may not include enough SNPs to noticeably erode the predictive power of family history derived from inherited genetic variation in the context of the power to detect such a difference provided by studies to date. These findings imply that hundreds of SNPs contribute to the heritability of CAD and the erosion of predictive power of FHx may not become evident until a good fraction of these SNPs have been reliably identified[34].

The effect of including variants with suggestive but not genome wide significant association with CAD into genetic risk scores

The utility of genetic risk prediction that incorporates SNPs with suggestive but not genome wide significant associations was first demonstrated in 2009 for schizophrenia, a highly heritable and polygenic disorder [7]. As the polygenic nature of CAD also became apparent, investigators began to also explore the value of expanding a GRS to incorporate a substantially larger number of SNPs that have not yet reached genome-wide significance. The first option in this respect involves adding approximately ~100 SNPs identified by the CARDIoGRAM+C4D with a calculated False Discovery Rate (FDR) < 5%[3]. A second option involves the inclusion of an even larger number of SNPs based strictly on more liberal p value thresholds achieved in GWAS (e.g. 5x10⁻⁶, 5x 10⁻⁵, etc.) combined with some degree of pruning of SNPs due to linkage disequilibrium (LD).

Table 1 summarizes the results of recent studies that have adopted and/or focused on this approach[11-15, 30]. In general, the data support the notion that GRSs incorporating a larger number of SNPs associated with CAD perform better than more restricted GRSs. The strongest association reported to date is in the FINRISK study where an LD pruned set of 49,310 SNPs out of ~79k SNPs included on the Cardio-metabochip resulted in a HR of 1.74 per SD increase in GRS[14]. However, the same GRS demonstrated a substantially lower HR of 1.28 in the Framingham Heart Study which was identical to the peak relative risk identified in the ARIC study using only 7,387 SNPs[11, 14]. Differences in the effect sizes per SD between GRSs tested were least evident in the Rotterdam and GERA studies[3, 4]. Possible reasons for the heterogeneity in findings between studies include the SNPs used to build the GRS and the case mix of the cohorts tested. For example, the GERA study included SNPs in the ALOX5AP region that are of questionable utility[15]. Cohorts with a larger fraction of endpoints related to revascularization may demonstrate improved performance of the GRS if the GRS tracks best with burden of disease[23, 35]. Lastly, cohorts with a higher proportion of endpoints classified as ‘out of hospital incident CHD deaths’ or ‘sudden cardiac death’ may suffer a degradation of performance of the GRS due to the higher probability of misclassification of the primary outcome.

Table 1.

Recent studies of genetic risk scores of CAD in Europeans and the risk of incident coronary artery disease

Reference	Date Published	Cohort	# of CAD events/non- events	Subgroup	# of SNPs in GRS	Risk factor SNPs included ?	Adjusted for TRFs not including FHx	Further adjusted for FHx	Subjects with baseline CAD excluded ?	RR (95% CI) per SD increase in GRS	HR (95% CI) quintile 5 GRS to quintile 1 GRS
de Vries et al. [30]	#########	Rotterdam Study	964/4450		49	Yes	Yes	No	Yes	1.12 (1.05-1.19)
					152	Yes	Yes	No	Yes	1.13 (1.06-1.21)
					49	Yes	Yes	Yes	Yes	1.11 (1.05-1.19)
					152	Yes	Yes	Yes	Yes	1.13 (1.06-1.20)
Goldstein et al. [11]	#########	ARIC	620/7871		7,387	Yes	Yes	No	Yes	1.28 (1.19-1.38)
Antiochos et al. [12]	#########	Colaus	133/4,150	All	38	No	Yes	No	Yes	1.24 (1.02-1.53)
					53	Yes	Yes	No	Yes	1.34 (1.10-1.64)
					153	Yes	Yes	No	Yes	1.45 (1.19-1.77)
					38	No	Yes	Yes	Yes	1.24 (1.01-1.52)
					53	Yes	Yes	Yes	Yes	1.34 (1.10-1.64)
					153	Yes	Yes	Yes	Yes	1.46 (1.19-1.78)
Tada et al. [13]	#########	MDC	2,213/21,382	All	27	Yes	Yes	No	Yes	1.20 (1.15-1.25)
			2,213/21,382	All	50	Yes	Yes	No	Yes	1.23 (1.18-1.28)
			998/7,790	FHx+	27	Yes	Yes	No	Yes		1.64 (1.34-2.01)
			1,215/13,592	FHx−	27	Yes	Yes	No	Yes		1.67 (1.39-1.99)
			998/7,790	FHx+	50	Yes	Yes	No	Yes		1.75 (1.43-2.15)
			1,215/13,592	FHx−	50	Yes	Yes	No	Yes		1.96 (1.63-2.35)
			651/11,800	≤Median age*	27	Yes	Yes	Yes	Yes		2.35 (1.81-3.06)
			1,562/11,796	>Median age*	27	Yes	Yes	Yes	Yes		1.48 (1.27-1.73)
			651/11,799	≤Median age*	50	Yes	Yes	Yes	Yes		2.19 (1.69-2.84)
			1,562/11,796	>Median age*	50	Yes	Yes	Yes	Yes		1.71 (1.45-2.01)
Abraham et al. [14]	#########	FHS	587/2819		26	Yes	Yes	No	Yes	1.20 (1.07-1.26)
					153	Yes	Yes	No	Yes	1.21 (1.16-1.32)
					49,310	Yes	Yes	No	Yes	1.28 (1.17-1.41)
		FINRISK	757/11919		27	Yes	Yes	Yes	Yes	1.21 (1.12-1.30)
					153	Yes	Yes	Yes	Yes	1.25 (1.16-1.39)
					49,310	Yes	Yes	Yes	Yes	1.74 (1.61-1.89)	4.51 (3.47-5.85)
Iribarren et al. [15]	#########	GERA	1,864/50,090	All	8	No	Yes	Yes	Yes	1.21 (1.15-1.26)	1.82 (1.57-2.11)
					12	No	Yes	Yes	Yes	1.20 (1.15-1.26)	1.81 (1.56-2.10)
					36	No	Yes	Yes	Yes	1.23 (1.17-1.28)	1.78 (1.53-2.06)
					51	Yes	Yes	Yes	Yes	1.23 (1.17-1.28)	1.97 (1.70-2.28)

CAD: Coronary Artery Disease, GRS: genetic risk score, TRFs: traditional risk factors, FHx: family history of CAD, RR: hazard ratio or relative risk, HR: hazard ratio, SD: standard deviation, CI: confidence interval, ARIC: Atherosclerosis Risk in Communities, MDC: Malmo Diet and Cancer, FHS: Framingham Heart Study, FINRISK: National Finrisk study, GERA: Genetic Epidemiology Resource in Adult Health and Aging

^*57.6 years

A genetic risk score predicts CAD independent of lifestyle factors

Smoking in the only lifestyle measure included in most clinical scores tested to date in conjunction with GRSs of CAD. A recent study examined whether 50 SNPs GRS of CAD modifies the beneficial effects of a healthy lifestyle in the Atherosclerosis (ARIC), Malmo Diet and Cancer Study (MDCS), and the Women’s Genome Health Study (WGHS) cohorts as well as the cross-sectional BioImage Study[16]. The investigators found that a favorable lifestyle, defined as at least three of the four healthy lifestyles of no smoking, no obesity, regular physical activity, and healthy diet, reduced risk of CAD by almost one half and was associated with a reduced prevalence of coronary artery calcification[16]. This reduction of risk was equivalent across all levels of genetic risk. Similar results using a more limited 14 SNP GRS were reported in a case-control study of Hispanics from Cost-Rica[17]. These studies strongly suggest that a GRS influences the risk of CAD not only independent of TRFs but also independent of all major modifiable lifestyle factors. Thus, subjects at high genetic risk should not be led to believe that they cannot reduce their risk dramatically by adopting a healthy lifestyle. Future research in larger cohorts with more events will allow us to further refine these findings by ruling in or ruling out any modification of effects between a GRS and individual components of a healthy lifestyle.

Genetic risk scores and the prediction of recurrent CAD events

Few studies have reported on the ability of CAD loci to predict recurrent CAD events. Initial reports focused on the 9p21 locus with a meta-analysis of 15 studies involving 4,436 subsequent events among 25,163 subjects providing compelling evidence that this locus is unable to predict recurrent MIs (HR, 95% CI: 1.03, 0.93-1.13) but can predict subsequent revascularization procedures (HR, 95% CI: 1.24, 1.12-1.37)[36]. More recent studies have examined not only additional CAD genetic variants but also GRSs[10, 31, 37-39]. The results of these studies are summarized in Table 2. While the GRS reached statistical significance in only one study, a trend towards significance was observed in all but one of the remaining studies even after adjustment for multiple risk factors All studies examined GRSs built with genome wide significant SNPs but one study also reported on the 153 SNP GRS[39]. The associations for the 153 SNP GRS were less significant than the GRS restricted to genome wide significant SNPs but a trend was still noticeable for the outcome restricted to recurrent ACS. Potential sources of heterogeneity to date include the variable length of follow up, the case-mix of subsequent events, and the potential for misclassification of events. For example, the single study that did not show even a trend of association was also the study with the shortest follow up that examined outcomes that may not be directly relevant to CAD such as ‘all-cause mortality’ and ‘cardiac hospitalizations’[38].

Table 2.

Recent studies of genetic risk scores and the risk of recurrent events among European subjects with established coronary artery disease

Reference	Date Published	Cohort	# of SNPs in GRS	Qualifying Event	Recurrent Event Definition	# events / non- outcomes	Follow Up (years)	Covariates	HR (95% CI)	Comparison for HR calculation
Weijmans et al. [10]	01/04/2015	SMART	30	CAD	MI	218/3232	6.5	none	1.11 (0.98-1.27)	per SD increase
								SMART risk score	1.09 (0.96-1.25)	per SD increase
Mega et al. [31]	06/06/2015	CARE	27		CAD death, MI, UA	320/2558	4.94	TRFs	1.14 (0.95-1.32)	per SD increase
		PROVE-IT TIMI22	27		CAD death, MI, UA	229/1770	2.03	TRFs	1.14 (0.95-1.36)	per SD increase
		Meta-analysis	27						1.14 (1.02-1.28)	per SD increase
Labos et al. [38]	01/09/2015	RISCA, PRAXY, TRIUMP	30	ACS	death, recurrent ACS, cardiac hospitalization	389/3503	1	none	0.98 (0.95-1.01)	per allele
								age, sex	0.98 (0.95-1.01)	per allele
								TRFs	0.98 (0.95-1.01)	per allele
								GRACE score	0.98 (0.95-1.01)	per allele
Vaara et al. [39]	01/04/2016	Corogene	47	ACS	recurrent ACS	217/1551	5	TRFs, CHF on CXR, 3VD	1.17 (1.01-1.36)	per SD increase
			153			217/1551			1.13 (0.97-1.30)	per SD increase
			47		CAD death, recurrent ACS	385/1383	5		1.09 (0.98-1.22)	per SD increase
			153			385/1383			1.05 (0.94-1.18)	per SD increase

GRS: genetic risk score, HR: hazard ratio, SD: standard deviation, CI: confidence interval, MI: myocardial infarction, CAD: coronary artery disease, ACS: acute coronary syndrome, UA: unstable angina, TRF: traditional risk factors, CHF: congestive heart failure, CXR: chest X-ray, 3VD: 3 vessel disease, SMART: Second Manifestations of ARTerial disease, RISCA: Recurrence and Inflammation in the Acute Coronary Syndromes, PRAXY: Gender and Sex determinants of cardiovascular disease: From bench to beyond-Premature Acute Coronary Syndrome in men and women, TRIUMPH: Translational Research Investigating Underlying disparities in acute Myocardial infarction Patients' Health status

Collectively, the data suggest that a GRS of CAD is likely to emerge as an independent predictor of recurrent events in meta-analyses and/or larger studies with prolonged follow up although the magnitude of association may ultimately not be as high as that observed for incident events. The association is likely to be driven by subsequent events leading to revascularization procedures[36]. This pattern of association may also ultimately reflect that a GRS is most effective in stratifying individuals by overall burden of coronary atherosclerosis. When one compares subjects having had their first event to subjects with no prior history of clinical CAD, the mean difference in burden between any single type of case (e.g. MI, angina, revascularization) and controls is likely to be substantial and easily identified through GRS. However, the difference in burden among a cohort of individuals with established disease may be considerably less making it more difficult for a GRS to distinguish individuals. Even if a GRS does emerge as an independent predictor of recurrent events, the clinical utility of this finding will be challenging to establish given the current standard of care which calls for the application of aggressive pharmacotherapy to all subjects with established CAD.

Performance of genetic risk scores in non-European racial/ethnic groups

GRSs of CAD have been tested almost exclusively in European populations because GRSs are expected to perform best in the racial/ethnic groups in which they were discovered. Although the genes contributing to the development of CAD are likely largely the same across all major ancestral groups, the genetic variants within the susceptibility loci associating most strongly with CAD will vary across groups due to differences in linkage disequilibrium[40-42].

Table 3 summarizes recent studies testing the predictive ability of a GRS in non-Europeans populations including Chinese, Blacks, and Hispanics[16-18]. The studies indicate a performance in non-European populations that is arguably better than expected for GRSs constructed with lead SNPs identified largely from white populations. For Chinese, the 156 SNP GRS tested included both East Asian specific CAD loci as well as the low FDR (<5%) SNPS reported by CARDIoGRAM+C4D and produced effects that were even larger than those observed in whites. Overall, these findings support the notion that the same genetic loci contribute to the pathogenesis of CAD across all major ancestral groups and provide optimism that a GRS developed in one race/ethnic group may improve risk prediction in other race/ethnic groups[40-42]. Ultimately, GRS developed from SNPs having reached genome wide significance within a specific race/ethnic groups are expected to deliver the best performance for the same race/ethnic group.

Table 3.

Recent studies of genetic risk scores of CAD and the risk of coronary artery disease in non-European populations

Reference	Date Published	Cohort	Race/Ethnic Group	# of CAD events/non- events	Subgroup	# of SNPs in GRS	Risk factor SNPs included?	Covariates	HR (95% CI) per SD increase in GRS	Other HR (95% CI)
Khera et al. [16]	15/12/2016	ARIC	Black	350/1919		50	Yes			Q5 to Q1: 1.65 (1.16-1.34)
Sotos-Prieto et al. [17]	20/12/2016	CRMIS	Hispanic	1534/1534		14	Yes	age,sex,residence	1.13 (1.06-1.21)	Q3 to Q1: 1.30 (1.07,1.59)
										Q2 to Q1: 1.31 (1.09-1.57)
Chang et al. [18]	01/01/2017	SGHS	Chinese	267/569	Men	156	Yes	TRFs		Q5 to Q1: 3.91 (2.42-6.34)
				128/342	Women	156	Yes			Q5 to Q1: 3.03 (1.43-6.40)

CAD: coronary artery disease, GRS: genetic risk score, SNP: Single nucleotide polymorphism, HR: hazard ratio, SD: standard deviation, CI: confidence interval, ARIC: Atherosclerosis Risk in Communities, CRMIS: Costa Rica Myocardial Infarction Study, SGHS: Singapore Chinese Health Study, TRFs: traditional risk factors, Q1: lowest quintile or lowest tertile of GRS, Q5: highest quintile, Q2: 2nd tertile, Q3: highest tertile

Identifying subjects most likely to respond to pharmacotherapy

In combination, statin and aspirin can reduce the risk of CAD by up to 50%. Whether a GRS of CAD can identify subjects with an optimal rate of response to either one of these medications remains an open question. For aspirin, no data exists but for statins a recent report provides provocative evidence for a GRS-environment interaction[31]. In this report, investigators evaluated the presence of a modification of effect of statins on the risk of incident or recurrent CAD by the level of genetic risk in 4 randomized trials of statin therapy using a 27-SNP GRS. A significant gradient (p=0.03) of increasing relative risk reductions from statin use was observed across the low (13%), intermediate (29%), and high (48%) genetic risk categories. Similarly, a greater absolute risk reduction among individuals at higher genetic risk categories (p=0.01) was found, resulting in a roughly threefold decrease in the number needed to treat to prevent one incident event. The investigators hypothesized that increased burden of coronary atherosclerosis among subjects with higher genetic risk drove the increased benefit of statins.

These findings require further validation but have the potential to change the decision of prescribing a statin in a considerable number of patients. They suggest that a GRS has the potential to serve not only as a prognostic marker, but also as a marker that can predict response to the single most important therapy available for CAD [43]. Such a combination for a single biomarker is uncommon, has generally been observed only for certain markers of cancer [44].

Are the effect sizes observed with GRS clinical useful for primary prevention?

Incorporating GRS of CAD into clinical practice has been hampered by several factors including the high cost of genotyping, the more modest effects of genetic variants on the risk of CAD than originally anticipated, and the challenge of improving clinical risk scores for CAD that already perform quite well [8, 9, 45]. While the cost of genotyping has dropped dramatically making large scale genotyping in clinical practice feasible in the near term, it remains difficult to demonstrate substantial improvements to standard model performance metrics, such as the C-statistic, with the addition of a GRS, even though many studies have highlighted its ability to independently predict incident CAD events. Not surprisingly, many studies reporting such discrimination metrics express reservations on the clinical utility of the GRS including many of the new studies reviewed here.

The minimal degree of improvement in the metrics needed to make any biomarker clinical useful remains unclear as it depends on several factors including the characteristics of the population being tested, which model performance metric is being embraced, and the risk-benefit tradeoff for the therapeutic intervention being considered. For atherosclerotic related outcomes, the degree of improvement in prediction (e.g. ΔAUC) may not have to be substantial given the size of the population at risk, the prevalence of disease, and the availability of relatively safe and effective pharmacologic and non-pharmacologic interventions [46, 47]. We note that any single traditional risk factor currently included in widely used clinical risk scores also contributes little to AUC when added last to a multivariate model[48].

The challenge of demonstrating the clinical utility of a novel biomarker through observational studies has been further amplified by concerns of the power and reliability of some of the most commonly used model performance metrics for novel biomarkers [49-51]. For example, Pepe et al. show that testing the null hypothesis that a biomarker is not independently associated with an outcome through likelihood tests is theoretically equivalent to testing the null hypothesis of 9 other commonly used discrimination metrics including delta Area-under-the-curve (ΔAUC) = 0, net reclassification index (NRI) = 0, and net benefit (NB) = 0[49], but testing the regression coefficient of a biomarker has substantially higher power than other tests including ΔAUC = 0. Furthermore, procedures for fitting risk regression models and for testing coefficients in regression models are highly developed while this is not the case for ΔAUC where standard tests may not be valid[50, 51]. These deficiencies may be the primary reason the ΔAUC of a biomarker is frequently not significant even when the biomarker is clearly independently associated with the outcome in a multivariate regression model. Lastly, substantial concerns have also arisen on the validity of the NRI and related indexes [50, 51]. Ultimately, large clinical trials may be required to prove clinical utility of a GRS for CAD. In the setting of a mega-cohort accruing many events annually, the incremental benefit of incorporating a GRS for CAD into a standard clinical risk score may become clear within just a couple of years of follow up.

Conclusion

Novel GWAS loci of CAD appear to predominantly facilitate the formation of coronary atherosclerosis. GRSs constructed from lead SNPs of these loci are independently associated with incident CAD using standard multivariate regression models. These associations are evident in multiple race/ethnic groups. GRSs may also predict recurrent events as well as those individuals most likely to respond to statins. The clinical utility of GRSs of CAD remains unproven but current discrimination metrics used in observational studies may be underestimating their performance. Additionally, the performance of GRSs is expected to improve over time as larger scale GWAS and sequencing studies are conducted and more susceptibility loci are discovered. Nevertheless, large scale clinical trials may ultimately be required to prove their clinical utility.

Key Points.

• Studies of GRSs of CAD suggest that most novel susceptibility loci of CAD identified through GWAS to date promote the development of coronary atherosclerosis.

• GRSs of CAD predict incident CAD events in multiple prospective cohorts independent of all traditional and lifestyle risk factors and in multiple race/ethnic groups.

• GRSs of CAD may predict recurrent events and may identify subjects most likely to benefit from statins.

• The performance of GRSs of CAD is expected to improve over time as more loci are discovered and they are likely to be incorporated into routine clinical risk assessment in the future.