An update on genetic risk scores for coronary artery disease: are they useful for predicting disease risk and guiding clinical decisions?

KEYWORDS:

• Cardiovascular disease
• coronary artery disease
• genetics
• genetic risk scores
• risk prediction
• risk
• epidemiology
• return of results

1. Background

The susceptibility to coronary artery disease (CAD) and myocardial infarction (MI) is determined by a complex interplay of genetic and lifestyle factors. More than 150 common genetic variants associated with CAD and MI at genome-wide significance have been reported so far [1]. Importantly, the majority of these genetic variants were not correlated with traditional cardiovascular disease (CVD) risk factors [2], underscoring the potential of genetic variants to convey information independent of established CVD risk factors.

Genetic information in an aggregated form – as a genetic risk score (GRS) – can be used to conduct association analyses (relating GRS to a given disease status) [3] or to predict disease risks. Such prediction models, including GRS, could help to identify individuals who are at high risk of developing CAD later in life, and who might benefit from preventive measures at a time when conventional algorithms underestimate their CAD risk. In this editorial, we discuss the current evidence related to the use of GRS for CAD risk prediction, for supporting clinical decisions and we address potential interactions of the GRS with age, lifestyle factors, and race/ethnicity. We identify unanswered questions worthy of additional study (see also Table 1).

Table 1. Open questions in the context of genetic risk scores (GRS) for coronary artery disease (CAD) and how these issues might be addressed.

Potential unanswered questions	Possible scientific approaches to address the issues
What is the appropriate number of SNPs to incorporate into a GRS and what metrics should we use to decide?	Prediction models with millions of SNPs should be compared to more parsimonious SNP sets with respect to their predictive performance.
How is a high GRS defined?	The relation of a GRS distribution to absolute disease risk for CAD should be addressed in cohort studies of incident disease, stratified by relevant subgroups.
Are there specific subgroups of the population where genetic screening for CAD should be performed?
What drives the CAD risk in individuals with a high GRS?	Prospective studies should explore the trajectories of risk factors according to GRS strata in order to better understand at what age the person develops clinically meaningful CAD risk threshold for the initiation of preventive measures; and to define other mechanisms that link GRS to risk of premature disease.
Are there significant statistical interactions of a GRS for CAD with lifestyle factors?	Potential interactions of GRS with age and with various lifestyle factors and in different ethnicities should be explored in longitudinal settings.
Is there an age-related penetrance of genetic variants associated with CAD?
Does the performance of GRS or do risk alleles differ by race/ethnicity, age or diabetes or HTN status?
What are the clinical consequences of having a high GRS? How to avoid ‘deterministic’ perspectives on the part of health care providers?	It should be addressed in randomized trials (RCT; including post-hoc analyses of RCTs with available genetic data) whether individuals with a high GRS benefit from preventive measures.
Does knowledge of a GRS change the clinical management of the patient (and his/her outcome)?	RCTs should address whether the disclosure of genetic information increases patient compliance or motivation. Psychological effects of knowledge of a high GRS must be assessed.
Does knowledge of a GRS change patient motivation for health-related behavior or compliance or provoke excess anxiety?
Is there a potential for genetic discrimination?	The potential for genetic discrimination should systematically be evaluated and legal options for mitigating such risks examined.
Who is responsible for communicating the results of a GRS and what is the impact of such communication on patients’ psychological state and general health?	Research should develop strategies for the regular evaluation of genetic variants and GRS and for communication with the patient when the categorization of SNPs changes. Assessment of impact on disclosure to patients (behavioral, economic, medical outcomes) warrants careful study.
How do we return genetic results to vulnerable populations?	Research into cultural sensitivity, access to educational tools and to health care resources for additional medical follow-up requires consideration.

SNP indicates single nucleotide polymorphism; HTN, hypertension.

2. How are GRS constructed and do they improve conventional risk prediction approaches?

There are different methodological approaches to generate GRS. A method referred to as ‘clumping/pruning and thresholding’ includes a two-step single nucleotide polymorphism (SNP) selection procedure [4]. First, SNPs that are correlated with the index SNP (a SNP associated with the outcome of interest in a given region) are removed (clumping or pruning). Second, the GRS includes only SNPs below a predefined p-value threshold of statistical significance [4]. Subsequently, the risk alleles of these genome-wide significant SNPs are weighted by their effect size and summed up [4]. Several studies have shown that GRS based on genome-wide statistically significant SNPs provided modest improvements in the predictive ability for CAD events beyond established risk factors [5].

More recently, Bayesian approaches such as LDpred have been developed that estimate posterior mean effect sizes for each SNP, based on a-priori assumptions about the effect size and about the linkage disequilibrium (LD) pattern, which is obtained from a reference panel [6]. Such approaches have been used to formulate GRS for CAD that include several thousand or even millions of genetic variants [7–9]. These approaches assume that a relevant part of the genetic architecture of CAD may likely be determined or influenced by genetic variants below the conventionally established statistical threshold for genome-wide significance, and that these ‘subthreshold’ (in a statistical sense) variants may be important for predictive purposes [9]. Several such GRS displayed statistically significant associations with CAD in multivariable-adjusted models that incorporate conventional cardiovascular risk factors and family history [8,9]. However, whether such scores that are based on millions of SNPs are clinically meaningful (in terms of better risk prediction, risk stratification, and guided treatment) is a matter of intense debate [10]. Of note, many of these variants have a negligible individual association with CAD [10].

Analyses of data from the UK Biobank revealed that a GRS consisting of more than 6 million genetic variants improved the C-statistic (a measure of discrimination between future CAD cases and those individuals who remain healthy during the follow-up period) as compared to a GRS consisting of 74 genome-wide significant SNPs [7]. However, the absolute increment in the C-statistic was minimal (from 0.791 to 0.806) [7,10]. In that study [7], it has also been reported that individuals at the very end of the GRS distribution (top 8%) have a relative risk for CAD that is comparable to monogenic disease conditions such as familial hypercholesterolemia [7]. However, this interpretation of the data has been debated [10]. In any case, focusing on absolute disease risk (as opposed to relative risk) is most relevant for making clinical decisions.

On a parallel note, a GRS consisting of 1.7 million genetic variants was significantly associated with CAD and led to an improvement in the C-statistics by 2.6% (from C = 0.670 to C = 0.696) when added to a prediction model including six traditional risk factors [8]. Another recent analysis from the UK Biobank (n = 352 660) reported a modest increment in the C-statistic from 0.76 to 0.78 when a GRS was added to the pooled cohort equations in predicting incident CAD [11]. About 4% of all participants were re-classified upon adding the GRS to the pooled cohort equations suggesting modest improvements in risk stratification by the GRS [11]. In a combined sample from the ARIC (Atherosclerosis Risk in Communities) and MESA (Multi-Ethnic Study of Atherosclerosis) cohorts (total n = 7237), a GRS consisting of >6 Million SNPs, was associated with incident coronary heart disease, but did not improve the predictive accuracy or risk stratification when added to the pooled cohort equations [12]. Thus, GRS are significantly associated with CAD independent of traditional risk factors, but improvements in risk prediction metrics were only modest in size, even though often statistically significant. When including the GRS in risk prediction models, it has to be kept in mind that most prediction algorithms were validated for a specific age range. The pooled cohort equations, for example, were validated for the age range of 40–79 years [13], whereas a GRS is valid throughout the entire lifetime. It is unclear, how a GRS could be incorporated for risk prediction outside the range of the pooled cohort equations or other risk prediction algorithms.

3. Can GRS identify individuals at high risk of CAD at a younger age?

Are GRS able to identify individuals who are particularly vulnerable for future CAD events and especially do so at a younger age? In most studies, GRS were associated with prevalent or incident CAD and moderately improved the predictive ability beyond established risk factors [7–9,11]. It is worth noting that the underlying GWAS, that are being used to identify and weight the genetic variants that compose the GRS, often pooled incident and prevalent CAD cases [14]. In most cases, the risk estimates are quantified per standard deviation or per quantile of the GRS. Thus, individuals at the higher end of the GRS distribution have higher relative and absolute disease risks of CAD as compared to individuals at the lower end of the GRS-distribution [7–9].

More emphasis should be given to the absolute disease risk associated with a given GRS because it is the absolute disease risk (not the relative disease risk) that guides clinical decisions [15]. Thus, it is critical to establish quantitative links between the absolute values of a GRS and the absolute disease risks for CAD over well-defined time horizons (separately in each sex and in different age groups and race/ethnic groups).

Furthermore, steeper trajectories in the lifetime risks for CAD in participants with a high GRS as compared to those with lower GRS have been reported [8,9], so that clinically relevant thresholds for the lifetime CAD risk might be reached at an earlier age in those with a high GRS [8,9]. Thus, potentially useful information can be extracted from a GRS if it were obtained at an earlier age.

However, to provide specific guidance at what age a given individual should start monitoring his/her risk factor levels more closely, we need to understand at what age that person’s CAD risk starts to escalate and through what mechanisms (e.g., mediated by which standard or novel risk factors). Therefore, prospective studies should explore more thoroughly the trajectories of individual risk factors in individuals stratified according to GRS.

On a parallel note, it should be explored whether there are specific subgroups of the population (e.g., categories defined on the basis of race or ethnicity, select ages or those with a family history of premature disease, etc.) that could benefit more from genetic screening for CAD [16]. Defining answers to these questions requires studies of large diverse groups and also use of additional endpoints such as age at onset of premature incident CAD [14].

4. Are there interactions of GRS with age, lifestyle factors, or race/ethnicity?

Most associated genetic variants by themselves increase the risk of CAD only very modestly. Nevertheless, even genetic risk variants with small effect sizes could potentially increase the risk for CAD substantially if these risk alleles influenced disease risk over the entire life course. However, it is unknown currently whether there is an age-dependent penetrance of genetic variants associated with CAD and whether the penetrance or expressivity of such risk alleles is influenced by lifestyle or other environmental factors (effect modification). Khera and colleagues explored the interrelation of a CAD-GRS consisting of 50 SNPs with an index combining different lifestyle factors (smoking, obesity, physical activity, diet) [17]. As expected, individuals with a high GRS (compared to those with a low GRS) and individuals with an unhealthy lifestyle (compared to those with a healthy lifestyle) had higher absolute risks of CAD. However, the relative benefits (relative risk reductions) associated with a healthy lifestyle (compared to an unhealthy lifestyle) were similar across individuals with a low, intermediate, or high genetic risk [17]. Thus, a healthier lifestyle was beneficial in all genetic risk groups. Yet, the greatest differences in absolute CAD risk associated with a healthy compared to an unhealthy lifestyle were observed in those with high genetic risk [17].

In another report, investigators observed that Chinese children aged between 7 and 11 years experienced substantially reduced relative risks for obesity with adoption of a healthy lifestyle as compared to other children with an unhealthy lifestyle. Of note, the relative risk reductions in this report were comparable (ranging between 77% and 91%) in those with low, intermediate, and high genetic risk, based on a GRS that included 11 body mass index-associated SNPs [18]. These observations are consistent with the concept that the relative risk reductions associated with lifestyle modifications might be comparable across different groups of genetic risk for CAD. The greatest reductions in absolute CAD risk, however, might be expected in those with the highest genetic risk. Both premises warrant further investigations.

In addition, there is growing evidence that the performance of GRS differs between races/ethnicities, e.g., due to differences in allele frequencies and in the underlying LD structures [19]. Furthermore, several races/ethnicities are underrepresented in genomic research [19,20]. Therefore, more efforts to study GRS in such underrepresented populations are necessary.

5. What would be the clinical consequences of a high GRS over the life course and can GRS modify therapeutic approaches to lowering CVD risk?

There are some initial studies evaluating whether and how GRS might affect clinical decision making, but overall, the clinical consequences of a high GRS for CAD are not well studied. Does knowledge of a GRS has the potential to change patient management by the clinicians? Does it change the compliance of the patients with the use of select medications or does it increase motivation of patients for adopting a healthier lifestyle? These are important heretofore incompletely answered questions because if a GRS does not bear the potential of changing clinical decision-making or health-related behavior of the patients, it is unlikely to change patient outcomes and it may not be cost-effective for large-scale clinical use. Some initial evidence is summarized below.

In a retrospective analysis of four randomized controlled trials (RCTs) involving statins, those individuals with higher genetic risk burden displayed larger absolute and relative risk reductions associated with statin therapy in primary and secondary prevention settings [21]. Similar observations were reported from the WOSCOPS trial [22]. On a parallel note, two recently published studies suggest that the efficacy of PCSK9 inhibition might be modified by an individual’s genetic susceptibility to CAD [14,23,24]. Within the ODYSSEY OUTCOMES (Evaluation of Cardiovascular Outcomes After an Acute Coronary Syndrome During Treatment With Alirocumab) trial and the FOURIER (Further Cardiovascular Outcomes Research With PCSK9 Inhibition in Subjects With Elevated Risk) trial, post-hoc analyses suggested that those individuals with the highest genetic risk burden had the highest absolute and relative risk reductions upon treatment with PCSK9 inhibitors [23,24]. The two studies used GRS consisting of more than 6 Million SNPs and 27 SNPs, respectively [23,24]. However, in a case-control study nested within the ACCELERATE secondary prevention trial, there was no evidence that the efficacy of evacetrapib (a cholesteryl ester transfer protein inhibitor) treatment in reducing major CVD was modified by a GRS consisting of 6.6 Million SNPs [25]; there was no association between evacetrapib treatment and CVD events across GRS quintiles [25].

Other studies evaluated whether the disclosure of genetic information to the patient improves clinical decision making or patient outcomes.

Knowles and colleagues reported that disclosure of a GRS in a clinical setting is feasible, but it did not lead to statistically significant improvements in CVD risk factor levels after 3 months [26]. It is unclear if the period of observation was too short to demonstrate clinically significant benefits in the aforementioned report [26]. In contrast, Dr. Kullo and colleagues reported in a modest-sized investigation that in individuals at intermediate CHD (coronary heart disease) risk, providing genetic information in addition to a 10-year risk estimate, based on traditional risk factors (compared to disclosing only the 10-year CHD risk estimate without providing genetic information) led to a higher proportion of individuals on statin medication and, consequently, to significantly lower levels of LDL cholesterol after 6 months of follow-up [27]. Importantly, the disclosure of CHD risk (with or without genetic information) was accompanied by a shared decision making by the patients and their physicians regarding the initiation of statin therapy [27]. However, disclosure of the genetic information in the context of the 10-year CHD risk did not lead to relevant changes in diet or physical activity [27].

While the above-mentioned analyses are intriguing, more RCTs may be needed to assess whether GRS can reliably identify subgroups of people (e.g., those with a high GRS) who would benefit from well-defined preventive measures, such as lifestyle interventions or certain medications.

It has to be kept in mind that CAD risk calculations are traditionally based on the levels of conventional risk factors. Even though GRS might improve the predictive ability of risk prediction models, it is not entirely clear how a high GRS should be interpreted in individuals with a normal cardiovascular risk profile. Should individuals with a high GRS be treated with medications such as aspirin or statins – perhaps for decades – even if their cardiovascular risk profile remains normal over select time intervals? Currently, there is no evidence to support such a strategy. Recent evidence suggests that aspirin should not be offered unselectively to everyone in primary CVD prevention settings given that the benefits may not outweigh potential risks [15]. It is conceivable but unproven that GRS might help identify population subgroups that may potentially benefit more from aspirin in the primary prevention setting.

6. Potential for genetic discrimination and communication of genetic results

Potential downsides of comprehensive genetic testing at an early age should also be kept in mind. The exact clinical pathways from research-based GRS to clinical grade testing followed by genetic counseling and clinical decision support systems (steps implemented for Mendelian disorders) remain to be defined [14].

As genetic predictions for complex disease condition (including CAD) are becoming more and more accurate, the potential for genetic discrimination remains real, despite the Genetic Information Nondiscrimination Act (GINA). Whereas GINA prohibits potential discrimination by employers or by health insurance companies [28], it does not apply to other kinds of insurances, such as life insurance and disability insurances [28,29].

Furthermore, there remain many challenges related to the communication of potential risk variants/GRS. For example, there might be new disease-associated variants discovered over time so that the composition of a GRS might change. Alternatively, the classification of known genetic variants (as benign or as pathogenic) might change over time [30,31].

Finally, a note of caution. The past century has witnessed an unprecedented rise in CAD incidence and prevalence in many parts of the world, which is not well explainable by the genetic underpinning of the disease, but rather by fundamental lifestyle changes. Thus, research around the clinical implications of GRS should be placed in an appropriate perspective and not be deterministic [32]. The communication of genetic results to vulnerable segments poses additional challenges related to the need for cultural sensitivity, access to educational tools, and referral for additional medical follow-up in the context of lack of health-care access or insurance [32]. More research on GRS and its utility for CAD prevention must occur hand-in-hand with targeting known key environmental/lifestyle drivers of the global CAD/CVD epidemic [16].

In conclusion, knowledge about the genetic architecture of CAD and its risk factors is growing. Many common genetic variants and several GRS, some of them including millions of SNPs, are significantly associated with CAD, and modestly improve the predictive ability of risk prediction models beyond established risk factors. However, the clinical utility of GRS for identifying people at risk of premature CAD and the specific sub-populations that might benefit from genetic screening at a younger age for CAD risk variants warrant comprehensive investigations before GRS might be applied more broadly in routine clinical care.

Declaration of interest

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Additional information

Funding

Dr. Vasan acknowledges the support from contracts NO1-HC-25195, HHSN268201500001I and 75N92019D00031 and grant [U01HL146382] from the National Heart, Lung and Blood Institute; NIH grants [RF1AG063507, R01HL143295, and R01HL093328]; and the Evans Medical Foundation and the Jay and Louis Coffman Endowment from the Department of Medicine, Boston University School of Medicine. Dr. Lieb acknowledges funding from the German Research Foundation [EXC 2167] and the German Ministry for Education and Research [01ER1801A].