Cardiovascular Pharmacogenetics: From Discovery of Genetic Association to Clinical Adoption of Derived Test

A. Perspective

Clinicians wield a vast therapeutic arsenal endorsed by clinical practice guidelines to prevent and treat cardiovascular (CV) diseases, yet interindividual variability in drug efficacy and safety complexifies optimal management (Wilkinson, 2005). The way people absorb, distribute, metabolize, and eliminate drugs, and how drugs interact with their target receptors, enzymes, or other molecules to produce specific therapeutic or adverse effects, are partly under genetic control (Pirmohamed, 2023). Following the discovery in the 1980s of genetic polymorphisms affecting important drug-metabolizing enzymes such as the cytochrome-P450 enzymes (CYPs) and N-acetyltransferases (Distlerath et al., 1985; Gough et al., 1990; Blum et al., 1991), dozens of genes (known as pharmacogenes) have been associated with drug response. Based on these discoveries, it was anticipated that adding individual pharmacogenetic (PGx) information to other patient-specific data would help select the optimal therapy by identifying drug responders, optimize drug dosage, and partly avoid adverse drug reactions (ADRs) (Roses, 2000; Goldstein et al., 2003).

An extensive review on CV PGx was published in this journal in 2013 (Johnson and Cavallari, 2013). This article offered a comprehensive view of the clinical applications of PGx, with a focus on clopidogrel, warfarin, and statins, the three drugs for which there was already substantial PGx evidence to support therapeutic management. Based on major breakthroughs in our capacity to analyze the human genome and in information technologies (IT), the authors and the community anticipated an exciting next decade in terms of discoveries and implementation of PGx in clinical practice.

B. Publications in Cardiovascular Genetics and Cardiovascular Pharmacogenetics

Indeed, we have witnessed over the past decades a remarkable growth in the number of published studies on genetics pertaining to CV diseases and CV PGx. As a rough illustration, the cumulative number of hits retrieved when querying the PubMed database in July 2023 with the terms “(genetics) OR (genomics) AND (cardiovascular)” increased from 41,929 in 2002 to 239,618 in 2022, an almost sixfold increase. During that same period, the number of hits retrieved when adding “AND (review)” to the query increased by almost sevenfold from 6,057 to 40,422 (Supplemental Fig. 1, blue lines). The cumulative number of published studies using the terms “(pharmacogenetics) OR (pharmacogenomics) AND (cardiovascular)” followed a trajectory that appeared to be relatively steeper, and delayed by a decade, with a remarkable 30-fold increase from 105 hits in 2002 to 3,111 hits in 2022 (Supplemental Fig. 1, orange lines). When examining the absolute number of PubMed hits by 5-year tranches (Supplemental Fig. 1, middle row), one observes a plateau in the production of publications on CV PGx reviews, with even a decline in the proportion of hits retrieved when relating PGx to genetics over the past decade (Supplemental Fig. 1, bottom row).

To better understand this decline in PGx CV hits, we first performed a comprehensive analysis of the 2023 status of PGx for marketed CV drugs, followed by an in-depth investigation of the steps that lead from the discovery of a new association between a genetic variant and CV drug response to clinical adoption of derived PGx test.

C. Scope of the Present Review

The goal of the present article is to provide an overview of how the field of CV PGx has advanced since the publication of the last review on this topic in this journal in 2013 (Johnson and Cavallari, 2013). To that effect, we begin by presenting a brief, up-to-date description of the PGx evidence regarding clopidogrel, warfarin, and statins, the three drugs extensively reviewed by Johnson and Cavallari (2013) (Section II). We then examine the PGx information, if any, in the labels approved by the Canadian, European, Japanese, Swiss, and US regulatory agencies for drugs we have identified to be approved for CV indication (Section III). We also relate this information to PGx-based drug dosing recommendations from two expert panels [i.e., Clinical Pharmacogenetics Implementation Consortium (CPIC) and Dutch Pharmacogenetics Working Group (DPWG)]. Intentionally, we did not cover CV ADRs caused by non-CV drugs, like drug-induced QT prolongation or anticancer drug-induced cardiomyopathies, as this topic has been covered by others (Aminkeng et al., 2016; Garcia-Pavia et al., 2019; Lopez-Medina et al., 2022).

We observed that, out of 221 drugs approved for CV indication, clopidogrel is the only drug so far with unanimous PGx recommendation. This observation prompted us to analyze the full journey of a CV PGx test and the steps leading from the discovery of a genetic variant associated with CV drug response to the clinical implementation of a derived PGx test (summarized in Section IV). In Section V, we provide an in-depth analysis of genome-wide association studies (GWAS) reported so far on the response to CV drugs. We then describe how genomic studies have also revealed intraindividual loads of actionable PGx variants and highlighted broad between-population differences in the prevalence of these variants (Section VI), emphasizing the need for diversity and inclusiveness in PGx research. Next, we summarize the body of published studies that have investigated the medico-economic benefits of PGx testing for CV drugs, and we provide a list of tests commercially available for PGx testing (Section VII). We subsequently examine the factors that still limit the usage of PGx in the clinic and lessons learned from existing PGx implementation initiatives in CV and beyond. Finally, we outline ways forward to fully benefit from PGx to improve CV care while leveraging the latest technologies in data sciences, with an emphasis on pre-emptive PGx (Section VIII).

A. Clopidogrel

A large body of literature has been published since 2013 on clopidogrel and the use of PGx testing in antiplatelets therapy (recently reviewed in Castrichini et al., 2023). Clopidogrel is a platelet aggregation inhibitor approved by the U.S. Food and Drug Administration (FDA) in 1997 and indicated for the medical management of acute myocardial infarction, percutaneous coronary intervention (PCI), acute ischemic stroke, transient ischemic attack, and peripheral artery disease. Postmarketing epidemiological observations showed that a fraction of patients treated with clopidogrel therapy did not benefit from the drug and experienced subsequent recurrent CV morbidity and/or mortality (Gurbel et al., 2003; Sofi et al., 2010). This observation prompted the search for genetic variants that would account for this lack of response. A candidate-gene approach led in 2006 to the identification of variants within the CYP2C19 gene, whose product is responsible for the conversion of clopidogrel to clopidogrel thiol metabolite H4, its active antiplatelet metabolite that inhibits the P2Y12 receptor (Hulot et al., 2006). These loss-of-function variants are associated with an inability to execute this step, so that the drug is ineffective in carriers of these variants. In 2010, the FDA added a warning box to the label for clopidogrel to inform about reduced effectiveness in patients who are poor metabolizers due to genetic variants in CYP2C19 gene (https://www.fda.gov/drugs/postmarket-drug-safety-information-patients-and-providers/fda-drug-safety-communication-reduced-effectiveness-plavix-clopidogrel-patients-who-are-poor).

Several prospective, randomized clinical trials have been conducted to demonstrate the clinical validity and utility of CYP2C19 genotyping in individuals undergoing PCI (Xie et al., 2013; Notarangelo et al., 2018; Claassens et al., 2019; Pereira et al., 2020). Their design varies in terms of study population, clinical settings and outcomes of interest. A first meta-analysis performed in 2021 including 15,949 patients from seven randomized clinical trials showed that the benefit of prasugrel or ticagrelor, two antiplatelets drugs that do not require activation through CYP2C19, over clopidogrel in reducing major adverse cardiovascular events (MACEs) was primarily driven by CYP2C19 genotype status (Pereira et al., 2021). This observation supported the implementation of CYP2C19 PGx testing to identify and treat loss-of-function carriers with ticagrelor or prasugrel, whereas noncarriers would be treated with clopidogrel. A second meta-analysis including 20,743 patients from 14 studies further supported the use of CYP2C19 PGx testing to guide antiplatelet therapy in coronary artery disease (CAD) patients by demonstrating that this strategy significantly improves the composite of MACEs and reduces individual ischemic outcomes, such as CV death, myocardial infarction, stent thrombosis, and stroke, with a significant reduction in minor bleeding (Galli et al., 2021). The benefit of genotyping CYP2C19 when prescribing clopidogrel has also been recently documented for the treatment of stroke and transient ischemic attack, including with prospective, randomized, interventional clinical trials (Y. Wang et al., 2021; Zhang et al., 2022; Ross et al., 2023).

These studies were complemented by several analyses that have investigated the medico-economic benefits of performing PGx testing before prescribing antiplatelet drugs. As an example, AlMukdad et al. conducted a systematic review to examine whether CYP2C19 PGx-guided antiplatelet therapy for acute coronary syndrome after PCI is a cost-effective strategy (i.e., whether the test delivers outcomes that are sufficiently improved compared with the cost of the test itself, even if it does not save money) (AlMukdad et al., 2020). Out of the 13 articles meeting the search criteria, 11 demonstrated that PGx-guided treatment was cost-effective or dominant over universal clopidogrel or other P2Y12 inhibitors (prasugrel or ticagrelor), while only two studies found that universal ticagrelor was cost-effective over PGx-guided prescription of clopidogrel. These conclusions are in line with those from Morris et al., who used a different approach applied to 23 studies (Morris et al., 2022). Of the 22 articles focused on clopidogrel alone, 17 documented cost-effectiveness. In addition, four studies showed that PGx testing was cost-saving (i.e., the PGx test delivers at least comparable outcomes yet costs less), whereas one study concluded PGx-guided clopidogrel is not cost-effective. Overall, for clopidogrel, the balance of evidence weighs heavily in favor of PGx-guided treatment as being cost-effective, and sometimes cost-saving, especially given the high costs of ADRs or poor outcomes (hospital readmissions, additional MACEs) and the cost of PGx testing continuing to decline.

B. Warfarin

As is the case for clopidogrel, a large body of literature, recently reviewed by Asiimwe and Pirmohamed (2022), has been published on PGx and warfarin. In contrast to clopidogrel, however, whether PGx testing is indicated for warfarin prescription remains unresolved. Warfarin is an old but still widely used oral anticoagulant for the prevention and treatment of venous thrombosis and thromboembolic events. It was approved by the FDA in 1954. Warfarin acts as a competitive inhibitor of vitamin K epoxide reductase complex 1 (VKORC1), a limiting-step enzyme responsible for activating the available vitamin K within the body. Close monitoring of patients' coagulation parameters is necessary to guarantee the efficacy and safety of warfarin, given the drug's narrow therapeutic index and the large interindividual variability in dose response. Functional variants in two key genes influencing the dose-anticoagulant effect of warfarin, CYP2C9 and VKORC1, were discovered in the 1990s and 2000s (Aithal et al., 1999; D’Andrea et al., 2005). The FDA added PGx information in the label for warfarin in 2007 but without providing PGx-specific prescribing guidance (Gage and Lesko, 2008). A first GWAS in 2009 confirmed VKORC1 and CYP2C9 as major determinants of warfarin dose and also detected a strong association signal in the CYP4F2 gene (Takeuchi et al., 2009). In parallel, large initiatives, such as the International Warfarin Pharmacogenetics Consortium (Klein et al., 2009) have been launched to test the clinical validity and utility of algorithms using genetic data in optimizing warfarin prescription.

Despite these efforts, the clinical validity and utility of PGx-guided warfarin therapy is still a matter of debate. A meta-analysis including 5,230 patients from 18 studies concluded that genotype-guided therapy, compared with standard dosing strategies, shortened the time to first therapeutic and stable international normalized ratio, reduced the number of patients with excessive anticoagulation (international normalized ratio ≥ 4) with a relative risk of 0.87 [95% confidence interval (CI) 0.78–0.98], improved (yet minimally) the time in therapeutic range by 3.1% and reduced bleeding events (risk ratio of 0.82, 95% CI 0.69–0.98) but did not significantly reduce the occurrence of clinically important endpoints such as thromboembolism or mortality (Tse et al., 2018).

The ambiguous clinical utility of PGx testing for warfarin is equally reflected in studies that have evaluated the medico-economic benefits of this test. In 2022, Kamil et al. conducted a systematic review of PGx-guided therapy for atrial fibrillation (Kamil et al., 2022). The authors reviewed 18 economic evaluations that comprised 46 comparisons of PGx-guided (CYP2C9 and VKORC1) treatment versus a non-PGx-guided treatment. Of the 41 comparisons that specifically evaluated PGx-guided warfarin dosing, most studies (24 of 41) compared PGx-guided dosing to standard warfarin dosing (without consideration of patient information) or clinical warfarin dosing (with consideration of patient demographic and clinical information, excluding genetics). Out of these 24 studies, 2 concluded that PGx testing was cost-saving, 12 studies concluded that PGx testing was cost-effective, while 7 studies concluded that it was not cost-effective and 3 concluded that PGx testing was dominated (more expensive and less effective). Fourteen comparisons examined PGx testing versus direct oral anticoagulants (DOACs). Half (7 of 14) of the comparisons concluded that PGx testing was cost-effective, while the other studies concluded that it is not cost-effective. This is in line with conclusions made by Morris et al. (2022) who reviewed 16 studies on warfarin, with 7 studies showing cost-effectiveness, 5 studies concluding that PGx-guided warfarin is not cost-effective, and 4 arriving at uncertain conclusions.

Overall, the conflicting evidence regarding clinical endpoints and medico-economic benefits, the important interindividual variability in drug response due to nongenetic factors, the practicalities of the PGx test, as well as the emergence of DOACs collectively explain why the adoption of genotype-guided warfarin therapy by clinicians has been limited so far (Shah, 2020).

C. Statins

As is the case with warfarin, the added value of PGx testing in statin prescription remains uncertain. Statins lower lipids by inhibiting HMG-CoA reductase, a key enzyme involved in cholesterol synthesis. Statins are widely prescribed worldwide for the prevention and treatment of atherosclerotic CV diseases. Despite their proven efficacy, statins are often underutilized or prematurely discontinued because of statin-associated muscle symptoms (SAMS). SAMS encompass a wide spectrum of clinical manifestations associated but not necessarily caused by statins, ranging from mild myalgia to severe myopathy, such as life-threatening rhabdomyolysis with muscle damage and acute renal injury (Cholesterol Treatment Trialists’ Collaboration, 2022). The discovery in 2008 of a common variant strongly associated with the risk of simvastatin-induced myopathy represents one of the first successes of a GWAS approach in PGx (SEARCH Collaborative Group, 2008). The SEARCH trial was a randomized control study designed to test the efficacy and safety of more intensive statin treatment in patients at high CV risk. The study included 12,064 patients allocated 1:1 to either 80 mg simvastatin daily or 20 mg simvastatin daily (SEARCH Collaborative Group, 2008). A higher incidence of statin-induced myopathy was observed in the simvastatin 80 mg group (0.9% vs. 0.03%). This observation prompted a post hoc GWAS analysis including 85 cases who had experienced severe myopathy and 90 controls, all of whom were taking 80 mg of simvastatin daily. The GWAS identified one single nucleotide polymorphism (SNP) (rs4149056 C allele, c.521T>C) located in the SLCO1B1 gene that was strongly associated with the risk of simvastatin-induced severe myopathy. SLCO1B1 encodes OATP1B1, an organic anion-transporting polypeptide responsible for the hepatic uptake of statins, among other drugs. Additional candidate genes or GWAS have confirmed the association of this SNP with SAMS (Voora et al., 2009; Brunham et al., 2012; Carr et al., 2019). Other variants have been associated with SAMS and were reviewed in Kee et al. (2020).

To illustrate the clinical validity of the SLCO1B1c.521T>C variant genotyping, Tonk et al. used two sets of case-control data (Tonk et al., 2017). In these datasets, the variant was present in 25% of the European population. Severe myopathy occurred in 0.8% of patients taking simvastatin 80 mg/day whereas any adverse reaction to various dosages of the drug was reported in 28% of patients. Based on these data, the authors showed that the negative predictive value (i.e., the proportion of noncarriers who will not develop the ADR) was 99.7% for severe myopathy and 83% for any ADR. When considering that 99.2% of patients taking simvastatin 80 mg did not experience severe myopathy, the benefit of systematic testing appears relatively small. In addition, the authors derived the positive predictive value of this PGx test (i.e., the proportion of carriers who will develop the ADR) to be 2% for severe myopathy and 37% for any adverse reaction, respectively. These limited performances partly explain why, despite the existence of actionable guidelines (Cooper-DeHoff et al., 2022) and recent studies showing that the intervention was well received by prescribers and did not result in poorer prevention of atherosclerotic CV diseases (Vassy et al., 2020; Obeng et al., 2023), there is still no clear guidance if genotyping should be performed in every person taking simvastatin (and to a larger extent to all individuals taking statins) or only in people taking statins at high dose or only in people experiencing SAMS. Still, a recent study shows that PGx testing may be of value to reinitiate statin therapy in people who had previously experienced SAMS. In this study, 159 patients with previous SAMS were randomized 1:1 to receiving genotyping-guided statin therapy versus usual care (Peyser et al., 2018). The study showed improved statin reinitiation rates and lipid profiles in the intervention group but no improvement in self-reported adherence.

A. Source of Pharmacogenetics Information

For this analysis, we used the Pharmacogenomics Knowledge Base (PharmGKB). PharmaGKB is an open-access, up-to-date online knowledge resource funded by the National Institutes of Health (NIH) for understanding how genetic variation affects drug response (Whirl-Carrillo et al., 2021). PharmGKB is managed at Stanford University and comprises a panel of field experts dedicated to producing evidence-based annotations of genetic variants and gene-drug-disease relationships via comprehensive literature reviews.

PharmGKB annotates PGx information contained in drug labels approved by the European Medicines Agency (EMA), FDA, Health Canada/Santé Canada, Pharmaceuticals and Medical Devices Agency Japan (PMDA), and the Swiss Agency of Therapeutic Products (Swissmedic). Across disease areas, the total number of drugs whose label contains approved PGx information varies widely between agencies, ranging from 52 drugs for the PMDA (https://www.pharmgkb.org/labelAnnotations) to 363 drugs for the FDA (https://www.fda.gov/drugs/science-and-research-drugs/table-pharmacogenomic-biomarkers-drug-labeling). The FDA also maintains a nonregulatory, information-only list of PGx associations, for which the FDA believes there is sufficient scientific evidence that genetic variants affect drug efficacy or safety (https://www.fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations).

PharmGKB also annotates PGx-based drug dosing guidelines published by professional societies such as CPIC, DPWG, the Canadian Pharmacogenomics Network for Drug Safety (CPNDS), and the French National Network of Pharmacogenetics. The CPIC and DPWG are scientific expert panels formed in 2005 and 2009, respectively, to facilitate PGx implementation and help clinicians interpret available PGx results through clinical practice guidelines (Relling and Klein, 2011; Swen et al., 2011). The guidelines are not designed to inform whether a PGx test should be ordered or whether it is cost-effective. CPIC’s guidelines are endorsed by the Association for Molecular Pathology, the American Society for Clinical Pharmacology and Therapeutics, and the American Society of Health-System Pharmacists. As of November 15, 2023, the CPIC has evaluated 519 drug-gene pairs across multiple therapy areas. Of the evaluated pairs, the CPIC has designated 98 with a final level of evidence A, indicating that PGx information should be considered when prescribing related drugs (https://cpicpgx.org/guidelines/).

B. Pharmacogenetics Information in the Label of Marketed Cardiovascular Drugs

We first generated, in June 2023, a list of CV drugs using the PharmGKB drug database and selected the groups “cardiovascular system” (ID: PA164712600) and “blood and blood forming organs” (ID: PA164712566). We removed duplicates, corticosteroids, or anti-inflammatory drugs; antimicrobial agents; coagulation factors; and antiseptic and disinfectants, which led us to a list of 221 CV drugs. This list includes 26 anticoagulants, 8 antiplatelets, 14 angiotensin-converting enzyme (ACE) inhibitors, 9 angiotensin II-receptor blockers, 22 beta-blockers, 13 calcium channel blockers, 26 lipid-lowering drugs, 21 diuretics, and 82 other drugs (Supplemental Table 1).

PharmGKB indicated that, out of these 221 CV drugs, 36 (16%) drugs had PGx information in their label approved by at least one of the five regulatory agencies (Fig. 1). The PMDA includes two CV drugs (atorvastatin and clopidogrel), whereas the EMA, FDA, Health Canada/Santé Canada, and Swissmedic include 14, 26, 13, and 15 CV drugs, respectively. Only one drug (clopidogrel) has PGx information in its label approved by all five agencies (Fig. 2). Reciprocally, half (17) of the 36 CV drugs have PGx information approved on their label by only one agency, i.e., EMA (two drugs), FDA (nine drugs), and Swissmedic (six drugs).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1

Distribution of CV drugs (N = 221) according to the presence of PGx information and the level of annotation in their label approved by the EMA, FDA, Health Canada/Santé Canada, PMDA, or Swissmedic. In the case of discrepant levels of annotation between agencies, drugs were classified according to the highest level of annotation.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2

Type of PGx information provided in the labels of CV drugs by the EMA, FDA, Health Canada/Santé Canada, PMDA, or Swissmedic. Information was retrieved from the PharmGKB drug label annotations database (https://www.pharmgkb.org/labelAnnotations; accessed April 10, 2024). Clopidogrel is the only CV drug with a concordant level of PGx annotation in its label provided by the five agencies, i.e., “Actionable PGx.” This designation implies that while the PGx information is recognized as valuable for clinical decision-making, it is not positioned as a mandatory directive for prescribing the drug and is not binding to physicians.

PharmGKB curates five levels of PGx annotation for drug labeling (https://www.pharmgkb.org/page/drugLabelLegend): “Testing required,” “Testing recommended,” “Actionable PGx,” “Informative PGx,” and, added in 2024, “Criteria not met.” “Testing required” and “Testing recommended” state or imply that some sort of gene, protein, or chromosomal testing, including genetic testing, functional protein assays, cytogenetic studies, etc., should be conducted, or is recommended, on the gene or gene product mentioned in the annotation before using this drug, respectively. With “Actionable PGx,” the label informs about changes in efficacy, dosage, metabolism, or toxicity due to gene, protein, or chromosomal variants or phenotypes or may mention contraindication of the drug in a specific subset of patients with particular variants/genotypes/phenotypes. “Informative PGx” states that particular gene, protein or chromosomal variants or metabolizer phenotypes do not affect a drug’s efficacy, dosage, metabolism, or toxicity or that particular variants or phenotypes affect a drug’s efficacy, dosage, metabolism, or toxicity, but this effect is not “clinically” significant. Finally, “Criteria not met” means that the drug’s label has been evaluated for PGx relevance but does not currently meet the specific criteria set by PharmGKB for classification as having either “Testing required,” “Testing recommended,” “Actionable PGx,” or “Informative PGx” implications.

As is the case for the cumulative number of drugs with PGx information approved in their label, the level of annotation varies widely between agencies for CV drugs (Fig. 2). Clopidogrel is the only CV drug with a concordant level of annotation (i.e., “Actionable PGx”) endorsed by the five agencies. Testing is required for three drugs. Two of them are lipid-lowering drugs: inclisiran (by the FDA only) and lomitapide (by the EMA only) as the indication for these drugs is limited to the treatment of molecularly documented familial hypercholesterolemia. The third one is mavacamten (by the EMA only), a first-in-class cardiac myosin ATPase inhibitor, as CYP2C19 intermediate or poor metabolizer status may result in higher systemic drug concentrations, potentially leading to an increased risk of heart failure. Swissmedic, but none of the other four agencies, recommends PGx testing for simvastatin. In most cases, the PGx information on the labels is not binding for prescribers. The higher level of annotation is “Actionable PGx” for 16 drugs and “Informative PGx” for five drugs. For the remaining 11 CV drugs, the PGx information does not meet the criteria for the aforementioned levels of annotation (“Criteria not met”).

C. Pharmacogenetic Tests Indicated for Marketed Cardiovascular Drugs and Corresponding Drug Dosing Guidelines

The pharmacogenes indicated for analysis for the 36 CV drugs for which there is PGx information in their label, and the phenotype associated with functional variants within these genes, are listed in Table 1. This table also includes five additional CV drugs for which genotype-specific prescribing information is available in current clinical guidelines from the CPIC or DPWG, yet no PGx information is provided in their label. Twelve lipid-lowering drugs, 13 antihypertensives, 8 anticoagulants or antiplatelets, and 8 drugs from other groups are listed. Also described is information collected from the FDA Table of PGx Associations (https://www.fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations). Finally, Table 1 summarizes, when available, the prescribing information by the CPIC (for 10 drugs) and DPWG (for 15 drugs), with the year these guidelines were first published. PGx information is provided by the three organizations (CPIC, DPWG, and FDA) for 4 drugs; the remaining 37 drugs have PGx information provided by only one or two of these organizations.

Overall, Table 1 contains 41 CV drugs with 55 drug-gene pairs from 19 genes. Most genes (12/19, 63%) are involved in the metabolism of drugs [eight CYP genes, four transporters (ABCB1, ABCG2, SLCO1B1, and NAT2)], three genes are implicated in the clearance of low-density lipoprotein (LDL) particles (LDLR, PCSK9, and APOB), whereas the remaining four genes belong to other categories. CYP2D6 is the pharmacogene that is paired with the largest number of CV drugs (n = 10), followed by SLCO1B1, which is paired with seven drugs, i.e., statins.

There is a limited match between the information in the approved labels and the recommendations by the CPIC and DPWG. In our list of 221 CV drugs, 15 (7%) drugs had both PGx information in one or more of the approved labels and a prescribing guidance from the CPIC or DPWG. PGx recommendations may differ from the PGx information on the label. As an example, six drug-gene pairs (acenocoumarol and CYP2C9, aspirin and G6PD, carvedilol and CYP2D6, hydralazine and NAT2, procainamide and NAT2, and ticagrelor and CYP2C19) have “Actionable PGx” in their drug label, yet neither the CPIC nor the DPWG recommend specific actions for these pairs due to limited PGx evidence. Conversely, five drugs (bisoprolol, flecainide, phenprocoumon, lovastatin, and fluvastatin) do not have PGx information in their label yet have clinical guidelines from the CPIC or DPWG. Table 1 also shows that 16 CV drugs have “Actionable PGx'' in their label. All but one (warfarin and VKORC1) of these drug-gene pairs are related to safety, i.e., ADRs or pharmacokinetics (PK). Lastly, clinical guidelines are available for one out of five drugs for which the PGx label is “Informative PGx” and for three out of 11 drugs for which the annotation level is “Criteria not met.”

The observation that only one CV drug, i.e., clopidogrel, has been granted so far unanimous recommendation for PGx actionability, and that PGx testing is still a matter of debate or insufficiently explored for the remaining 220 CV drugs is striking, considering the thousands of PGx studies that have been performed so far, including in CV (Supplemental Fig. 1) and the millions of people who are prescribed this type of drugs. In this regard, one may keep in mind that atorvastatin and simvastatin were the first and 13th most prescribed drugs in the United States in 2020 with an estimated 26,640,141 and 8,557,525 people receiving these drugs, respectively. Clopidogrel came 29th with 4,340,688 individuals and warfarin 58th with 2,424,821 patients (https://clincalc.com/DrugStats/Top200Drugs.aspx). This situation forces us to acknowledge a certain frustration. It also prompts the need to better understand the various steps that lead from the discovery of an association between a genetic variant and response to a specific CV drug to the clinical translation into a new, approved, derived PGx test. This is the topic of the next section.

A. The Steps Toward a New Pharmacogenetics Test

Irrespective of the therapy area, the discovery and development of a PGx test follow the same steps as for any biomolecular marker used in the clinic like NT-pro-BNP or troponin. These steps are illustrated in Fig. 3. Remarkably enough, the processes of bringing innovative PGx tests (Fig. 3, upper panel) to the market share a number of commonalities with the one needed to bring new drugs to the clinic (Fig. 3, lower panel) in terms of duration (many years), costs (millions of U.S. dollars), and risk (low probability of success). For the sake of completeness, we will summarize the contribution that human genetics has had to the discovery of new CV drugs and the repositioning of existing drugs for CV indications.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3

Chevron diagrams illustrating the steps that lead from the discovery to development and commercialization of novel PGx tests or new drugs. Top and bottom diagrams illustrate the steps leading to new marketed PGx tests or drugs, respectively. Arrows at the bottom point to steps where human molecular genetics has been shown to be instrumental in supporting go/no-go decision-making to progress to next steps in the drug discovery-development process. The red arrow corresponds to our operational definition of PGx.

B. Contribution of Human Genetics to Target Identification and Validation for Cardiovascular Drugs

Today, dozens of marketed drugs have been developed after their target had been discovered through seminal human molecular genetic studies (Trajanoska et al., 2023). Moreover, robust analysis of historical data has unambiguously demonstrated that the availability of human genetic associations between variants in the target gene and the disease indication more than doubles the probability of success for investigational drugs to reach the market (Nelson et al., 2015; King et al., 2019). As such, having human genetic data to substantiate new therapeutic targets has now become, for a large proportion of drug discovery campaigns, an essential component of their validation.

In the CV field, PCSK9 inhibitors represent an emblematic illustration of the impact of human genetics in drug discovery. The target was originally identified through genetic analyses of individuals with extreme plasma levels of LDL-cholesterol (Abifadel et al., 2003). These studies revealed gain- and loss-of-function variants within the PCSK9 gene responsible for very high, or very low, LDL cholesterol levels, respectively, and a corresponding risk of CAD (Abifadel et al., 2003; Cohen et al., 2006). The observation that reduced activity of this enzyme was also associated with a reduced risk of CAD pointed to PCSK9 as a potential new target for the prevention and treatment of this condition. Moreover, the demonstration that carriers of null-alleles (i.e., “human knockout”) in the PCSK9 gene did not have any other phenotype but very low blood LDL cholesterol levels (Zhao et al., 2006) provided a strong safety signal, indicating that limited, if any, adverse on-target effects would be expected from specific PCSK9 inhibitors. The first PCSK9 inhibitor was approved by the FDA in 2015, 12 years after the initial discovery of the enzyme (Raedler, 2016). For this class of drugs, genetics also had an impact beyond target discovery and validation. Recent post hoc analyses of two large phase III trials with two distinct PCSK9 inhibitors (Damask et al., 2020; Marston et al., 2020) consistently showed that patients who benefited the most from PCSK9 inhibition are those with a high polygenic predisposition to CAD. This polygenic predisposition is reflected in polygenic risk score (PRS), which captures an individual's genetic susceptibility to a specific trait or disease by considering the additive risk conferred by numerous genetic variants across the genome (Torkamani et al., 2018).

C. Contribution of Human Genetics to Repositioning Existing Drugs for Cardiovascular Indication

In parallel, human genetic studies have already pointed to a series of repositioning opportunities that have been materialized in approved new indications for existing drugs (Trajanoska et al., 2023). In the CV area, tocilizumab, a monoclonal antibody to the IL-6 receptor originally developed to treat rheumatoid arthritis, is currently being tested to treat CAD based on genetic studies. In this case, repurposing was based on results from large-scale Mendelian randomization analyses (Georgakis et al., 2020) an approach that assesses causal relationships between an exposure (in this particular case genetic variants within the IL-6R gene) and an outcome (CAD).

A. Detection of Adverse Drug Reactions and Variations in Drug Efficacy

ADRs and interindividual variability in efficacy are detected during the development phase of the investigational drug or through postmarketing surveillance, pharmacovigilance reporting systems and epidemiological studies, or case reports for marketed drugs (Spear et al., 2001; Pirmohamed et al., 2004). Another strategy, adopted for instance by the CPNDS, is to build a nationwide network infrastructure to actively collect data on drug safety and efficacy and maintain a longitudinal database to perform PGx research (Tanoshima et al., 2019). Between 2005 and 2017, the CPNDS has collected Canada-wide 93,974 reports of medication use, 10,475 of which are ADR cases and 83,499 of which are non-ADR reports. The majority (72.6%) of the cases are pediatric cases, as the CPNDS, previously known as Genotype-Specific Approaches to Therapy in Childhood, was initially set up to study the genetics of ADR in children (Carleton et al., 2009). Patients who experienced ADRs in each active surveillance site are referred to CPNDS surveillors who enroll them, as well as drug-matched controls. In addition to clinical data collection, genomic DNA is extracted from blood or saliva and stored in a central laboratory. Genomic analyses rapidly expanded from candidate SNPs testing to genome-wide genotyping and whole exome sequencing (WES). Using this approach, the CPNDS has successfully identified genetic variants that are involved, among others, in cisplatin-induced hearing loss, anthracycline-induced cardiotoxicity, and opioid-induced toxicities (Ross et al., 2009; Madadi et al., 2013; Visscher et al., 2013, 2015). More recently, the Medicines and Healthcare products Regulatory Agency in the UK, in collaboration with Genomics England, has established the Yellow Card Biobank to facilitate understanding of the role of genetics in drug safety by collecting genetic samples from patients experiencing ADRs, with an initial focus on direct oral anticoagulants and allopurinol (https://yellowcard.mhra.gov.uk/biobank). This is the first genetic biobank in the world to be operated by a drug regulatory agency.

B. Genome-Wide Technologies to Detect Pharmacogenetics Associations

Completion of the Human Genome Project in 2001 was a landmark in genetics by providing the first comprehensive map of the human genome (Lander et al., 2001; Venter et al., 2001). Mapping the human genome enabled PGx research to move from candidate-gene to hypotheses-free GWAS. In this type of analysis, variants along the entire human genome are tested for associations with a variety of traits, including response to drugs (Giacomini et al., 2017; Uffelmann et al., 2021).

Most PGx candidate-gene and GWAS reported so far have been performed using genotyping arrays. These latter technologies interrogate hundreds of thousands of specific variants in the genome. Variants that are not directly tested on the array can be inferred using imputation, a statistical tool that uses haplotype or genotype reference panels to estimate missing genotypes (Li et al., 2009). Newer genome-wide arrays have been designed to increase their PGx content (Verlouw et al., 2021). This expansion makes these arrays suitable and affordable for genome-wide screening for markers associated with drug efficacy and ADRs in large cohorts (Arbitrio et al., 2021). Still, one major limitation of array-based genotyping platforms is that they are best suited to detect relatively common variants, i.e., variants with a minor allele frequency >0.1%. In addition, structural variants (SVs) and variants located in highly polymorphic regions of the genome, such as the CYP genes encoding cytochromes P450 enzymes and the human leukocyte antigen region, which are involved in the response to many drugs, tend to be less well captured or have low imputation quality (Karnes et al., 2017; Nofziger and Paulmichl, 2018).

To address these limitations, while taking advantage of the dramatic reduction in the costs of DNA sequencing (https://www.genome.gov/about-genomics/fact-sheets/Sequencing-Human-Genome-cost) and the availability of WES and whole genome sequencing (WGS) data from large cohorts (Denny et al., 2019; Locke et al., 2019; Backman et al., 2021; Halldorsson et al., 2022), sequencing data are more and more used for PGx applications (Tafazoli et al., 2021). Sequencing technologies have been demonstrated to outperform array-based technologies for capturing rare and copy number variants and may facilitate the discovery of new loci (Mu et al., 2019; Halldorsson et al., 2022). One factor that determines which technologies are optimal to capture PGx variants is their location in the coding or noncoding regions of the pharmacogenes. Both imputed array and WES represent suitable approaches for screening pharmacogenes for which variants of interest are mapped in the exons; these genes include CFTR, CYP2C9, TPMT, CYP4F2, and DPYD (McInnes et al., 2021). On the other hand, WES appears to outperform imputed array for certain genes like CYP2B6, CYP3A5, and NUDT15. In contrast, both imputed array and WES have a low performance to analyze genes that carry both exonic and nonexonic PGx variants like UGT1A1, CYP2D6,or SLCO1B1 (McInnes et al., 2021). Finally, as expected, several known nonexonic PGx variants in CYP2C19*17, CYP3A5*3, UGT1A1*28/*37, and VKORC1*2 are missed by WES (Reisberg et al., 2019; Caspar et al., 2020; Lanillos et al., 2022). Taken together, these considerations highlight the importance of performing WGS, or a fusion of WES and WGS (Bhérer et al., 2023), if one wants to capture the full spectrum of PGx variants that a given individual carries. This is particularly relevant for pre-emptive PGx discussed in Section VIII.

CYP2D6 illustrates well the complex architecture of certain pharmacogenes. Its product is involved in the metabolism of 20% of drugs including drugs prescribed in CV diseases, psychiatry, and cancer (Taylor et al., 2020). Sequencing analyses have documented more than 100 haplotypes, i.e., combinations of variants from multiple regions of the gene on a single chromosome. Certain haplotypes are associated with specific protein activity levels, defined by the Pharmacogene Variation consortium (Gaedigk et al., 2021). The different haplotype patterns are called star alleles (*). Star alleles are annotated as CYP2D6*1, *2, *3, etc. and are determined by SNPs, insertion-deletion variations, and/or SVs. Knowing the combination of variants within a given haplotype is crucial for proper interpretation of PGx results. CYP2D6*1 indicates a normal enzyme activity; CYP2D6*10 is associated with a decreased enzyme activity, whereas CYP2D6*4 results in undetectable (null) enzyme activity. Given the complexity of PGx regions, traditional tools for calling SNPs and insertion-deletion variations have a limited capacity to derive PGx variant star (*) alleles or define highly polymorphic pharmacogenes. Specific bioinformatic tools such as Stargazer, Aldy, Astrolabe, and Cypiripi (Supplemental Table 2) have been developed to address this need. These tools use Pharmacogene Variation and PharmGKB to automate the detection of haplotype combinations of pharmacogenes and facilitate clinical interpretation (Gaedigk et al., 2021; Whirl-Carrillo et al., 2021). Currently, Stargazer (version 2.0) appears to be the most comprehensive PGx tool for detecting SNPs, insertion-deletion variations, and SVs mapping to 58 known pharmacogenes.

Most GWAS studies exploit large cohorts and biobanks, which are becoming invaluable resources for establishing relationships between genetic and phenotypic data extracted from electronic medical records (EMRs), questionnaires, and other sources (Abdellaoui et al., 2023). Some studies enable the longitudinal screening of wide-range of outcomes relevant to PGx (McInnes and Altman, 2021). An additional advantage of large biobanks is the possibility to detect rare ADRs and to include phenotypes that may not have been recorded in clinical trials. A proof-of-concept was recently published to illustrate the power of this approach. A GWAS performed on EMR data from 81,739 participants in the Vanderbilt University Medical Center (VUMC) BioVU DNA Biobank identified seven genetic variants associated with ADRs and confirmed known genetic associations such as CYP2D6 and OPRM1 for CYP2D6-metabolized opioid-related ADRs (Zheng et al., 2021). This type of analysis can be complemented by phenome-wide association studies, which allow hypothesis-free testing of associations between variants in a given gene (here pharmacogenes, drug target genes, or specific PGx variants) and a broad range of phenotypes (Diogo et al., 2018; Carss et al., 2022).

C. Genome-Wide Association Studies on the Response to Cardiovascular Drugs

The vast majority of published GWAS are included in the GWAS Catalog, an open-access resource funded by the U.S. National Human Genome Research Institute (Sollis et al., 2023). The catalog is a comprehensive and updated database of known SNP-trait associations and GWAS. The data release on June 22, 2023, contained 6,422 publications yielding 529,713 top SNP-trait associations (P value ≤ 10-6). Close to 10% of these studies (598 studies, 9.3%) are PGx-related (i.e., the trait under investigation contains either “response to” or “drug-induced"). The largest proportion was observed in 2015 with 17% of reported GWAS being PGx-related (McInnes et al., 2021). Cancer drugs represent the most common class of drugs examined using GWAS approaches, with methotrexate and paclitaxel being the most studied drugs. Oncology drugs are followed by psychiatry drugs (mostly antidepressants) and lipid-modifying agents. Remarkably, the vast majority of genes (>94%) discovered in PGx GWAS had not been previously incorporated in earlier candidate-gene association studies (Linskey et al., 2021).

To obtain a full picture of GWAS reported to date on the response to CV drugs, we interrogated the GWAS Catalog (v1.0.2) using the name of each individual drug from our list of 221 CV drugs (Supplemental Table 1). No GWAS were retrieved for 147 (67%) drugs like prasugrel, labetalol, or lovastatin. Out of the 74 drugs with reported GWAS studies, the GWAS Catalog provided specific results for 35 drugs, whereas the name of the remaining 38 drugs like perindopril or carvedilol was mentioned in the GWAS Catalog yet no relevant data were reported on the response to these drugs. For three drugs with reported results, GWAS had examined blood levels of the endogenous substance (i.e., adenosine, dopamine, and urokinase), as opposed to PK parameters following administration of a xenobiotic. We also excluded eight GWAS listed for lisinopril, metoprolol, theobromine, or valsartan because these studies investigated the metabolomic profile of patients as opposed to the specific response to the drugs (Shin et al., 2014; Al-Khelaifi et al., 2019; Z Wang et al., 2021; Hysi et al., 2022; Rhee et al., 2022; Tahir et al., 2022; Yin et al., 2022; Schlosser et al., 2023). Dalcetrapib was listed in the GWAS Catalog yet is not approved to date for CV indication. These filters led to 71 GWAS, which had investigated the specific response to 27 CV drugs: nine independent studies were reported on the response to warfarin, eight to fenofibrate, seven to aspirin, six to hydrochlorothiazide, five to clopidogrel and to simvastatin, four to heparin and to atorvastatin, three to chlorthalidone and to rosuvastatin, and two to phenylephrine and to pravastatin, whereas a single study was reported for the remaining 15 drugs. An additional 30 independent GWAS performed on response to CV drug classes were retrieved from the GWAS Catalog. Eleven of these studies investigated the collective response to statins, eight to ACE inhibitors and/or angiotensin II receptor blockers, five to beta-blockers, four to drugs grouped as antihypertensives, and three to diuretics, and one study was reported for calcium-channel inhibitors. Of the total 101 GWAS, 41 were published before 2015 (the first one in 2008). The characteristics of these 101 studies are described in Table 2. Of note, we have chosen not to report index SNPs because their identification may vary due to differences in genotyping platforms, haplotype structures, and population-specific linkage disequilibrium patterns. In addition, we have omitted the names of causal SNPs or genes because these have not been definitively established across all signals in subsequent post-GWAS analyses.

GWAS had been performed to investigate the genetic basis of CV drugs’ pharmacodynamics (PD; i.e., drug efficacy like blood pressure response to antihypertensive agents), PK characteristics (using maintenance dose or blood levels of the drug or its metabolites as phenotype under investigation), and ADRs (such as myopathy associated with statins). Taken together, we identified 69 traits that have been investigated by CV PGx GWAS (Table 2, fourth column). PD traits were the most studied phenotypes (51%), followed by ADR (32%) and PK (17%) traits. We also examined the number of loci that had been identified with a Pvalue for association ≤ 5 × 10⁻⁸. These loci mapped to 306 distinct genes according to the GWAS Catalog (Table 2, fifth column). No genome-wide significant locus was reported for one-third (20/69) of the studied PGx traits.

We then compared the results of GWAS with PGx recommendations, as discussed in Section III. A total of 11 out of the 27 CV drugs for which the response had been investigated using GWAS and results reported in the GWAS Catalog had some PGx recommendations. These recommendations included the analysis of eight different genes (Table 2, right column). Six of these eight genes recommended for PGx testing were shown to be significantly associated with the corresponding trait in GWAS. These include CYP2C19 for clopidogrel; VKORC1, CYP2C9, CYP4F2 for warfarin; SLCO1B1 for rosuvastatin and simvastatin; and ABCG2 for rosuvastatin. Reciprocally, only a very small fraction (6/306, 2%) of the genes significantly associated with a PGx trait in GWAS had a corresponding PGx recommendation.

D. Learnings from Genome-Wide Association Studies on the Response to Cardiovascular Drugs

Our search was limited to the GWAS Catalog, so it is possible that we missed PGx GWAS that have not yet been reported (including very recently published GWAS that have not been curated by the GWAS Catalog) or published (it is reasonable to assume that most unpublished studies did not find genome-wide statistically significant variants). Still, several learnings can be gathered and lessons learned from our analysis. First, only a small fraction of the 221 drugs (27, 12%) we had identified to be indicated for CV disease prevention or treatment had their response investigated using GWAS approaches. We can postulate that the response to at least a fraction of the remaining 194 drugs also exhibit a certain degree of interindividual variability, which may equally warrant future detailed genetic analyses.

Next, GWAS revealed significant genome-wide associations for two-thirds of the PGx traits under investigation (48/69), with no significant locus reported for the remaining 21 traits (12 PD traits, 7 ADR traits, 1 PK trait, and 1 PD/PK trait). The absence of detected significant loci could be due to biological factors and the complexity of deriving the correct metabolic phenotype from PGx data, as it depends on the number of alleles interrogated, population-specific differences in allele frequencies, allele-specific substrate selectivity, and phenoconversion, especially for CYPs (Shah et al., 2016). Still, one major limitation of most reported PGx GWAS is their restricted sample size, hence limited power to detect associations, and the lack of replication in an independent dataset. Such a low yield should prompt for careful power calculation prior to performing this type of study and for planning independent replication.

The clinical relevance of PGx GWAS analyses needs to be examined critically as well, especially for studies investigating PD parameters. Discovering loci associated with blood pressure or lipid response to antihypertensive or lipid-lowering drugs, respectively, may contribute to the generation of PRS for drug efficacy and may point to interesting genes that could eventually represent new therapeutic targets. Still, from a clinical perspective, it is unlikely that a genetic profile will ever replace the measurement of intermediate phenotypes themselves, i.e., blood pressure or blood lipid levels. In contrast, finding genetic variants associated with end-points, i.e., CV events or death, would be highly desirable to target populations who will mostly benefit from pharmacological intervention, as appears to be the case for PCSK9 inhibitors described earlier (Damask et al., 2020; Marston et al., 2020).

Most cataloged PGx GWAS have been performed on PD traits. We identified a limited number of GWAS (24/101) that have revealed loci significantly associated with ADRs to CV drugs. These include heparin-induced thrombocytopenia, warfarin-associated bleeding, ACE inhibitors/angiotensin II-receptor blockers intolerance or induced angioedema, myopathy or liver injuries induced by statins, metabolic disturbances induced by thiazides or thiazide-like drugs, and new-onset diabetes after exposure to beta-blockers. This low number of ADR-directed GWAS sounds somehow paradoxical considering that risk alleles for ADR are usually frequent in the population as, unlike disease-causing variants, they have not been filtered during humankind evolution, and the associated odds ratios are usually high, as illustrated by the simvastatin example given earlier (SEARCH Collaborative Group, 2008). This small number of ADR CV PGx studies is partly accounted for by the difficulties in recruiting large enough groups of patients who have experienced ADRs and controls exposed to the same drug (Giacomini et al., 2017). Taken together, these points highlight the necessity to perform additional, well-designed, large-scale PGx GWAS investigations on more drugs, including broader and more diverse patient populations, and to collect high-quality phenotypic data. A protocol has recently been proposed that may be of great help in the design and execution of future CV PGx studies (McDonough, 2021). Moreover, consortia, such as the International Clopidogrel Pharmacogenomics Consortium (Bergmeijer et al., 2018) and the International Consortium for Antihypertensive Pharmacogenomic Studies (McDonough et al., 2021) provide frameworks for the organization and execution of large PGx studies where multiple investigators share relevant genetic, phenotypic and outcomes data to identify additional genetic markers. As discussed in Section VIII, pre-emptive PGx and access to EMRs linked to genomic profiling may help address these critical issues.

Finally, the fact that a very small fraction of genes detected by GWAS to be associated with CV drug response have presently a corresponding PGx drug-gene recommendation (6/306, 2%) is consistent with a particularly high attrition rate and may indicate that only a small proportion of these variants are deemed sufficiently informative to derive a PGx test. This observation is also consistent with the decades-long delay between the discovery of a PGx marker and the clinical translation into a PGx test. Considering that 90% of the GWAS we identified were reported after 2010, one may expect that the PGx discoveries made more recently should translate to new PGx testing in the coming years.

A. Intraindividual Load of Actionable Pharmacogenetics Variants

Large genome-wide profiling studies have documented and quantified the presence and the cumulative number of PGx variants within a given individual. Studies conducted in different biogeographical populations showed that more than 95% of individuals carry at least one clinically relevant PGx variant (Bush et al., 2016; Dong et al., 2018; Mostafa et al., 2019; McInnes et al., 2021; Jithesh et al., 2022). In a prospective PGx implementation study in Europe involving 6,944 participants, 93% and 30% of participants carried at least one or two actionable PGx variants, respectively, and 25% carried an actionable variant for the drug they were taking, with CYP2D6 being the most frequently implicated gene (Swen et al., 2023).

B. Between-Populations Differences in Pharmacogenetics Variants Frequency

Drugs prescribed for CV diseases and other medical conditions are used worldwide. Yet, the vast majority of GWAS-based PGx studies published so far involve primarily populations of European descent (McInnes et al., 2021). Building a comprehensive and inclusive map of allele frequencies for clinically relevant PGx variants is essential for the development of reliable and equitable PGx tests and clinical guidelines. Part of this knowledge gap has been filled over the past decade (Zhou and Lauschke, 2022; Li et al., 2023). For instance, a recent study analyzed the ancestry-specific allele frequency of 282 important PGx variants mapping to 152 genes, the variation in the human leukocyte antigen region, and the full deletion/duplication of CYP2D6 (Markianos et al., 2023). This study included 658,585 individuals representative of the diverse US population [Europeans (70%), Africans, Hispanics, and Asians] and documented substantial variation in the prevalence of a subset of PGx variants.

The allelic frequencies for key genes involved in the response to clopidogrel (CYP2C19), propafenone (CYP2D6), warfarin (CYP2C9), and statins (SLCO1B1) have been examined in several populations and have been curated by PharmGKB (https://www.pharmgkb.org/page/pgxGeneRef). The derived functional, phenotypic impact of genetic variation on the metabolism of these drugs is presented in Fig. 4. For instance, the contrast in PGx phenotype frequencies between Oceanians and Europeans is striking, with more than half of Oceanians, compared with a very small fraction of Europeans, being CYP2C19 poor metabolizers. Conversely, the proportion of Oceanians who are at genetic risk of statin-induced myopathy due to poor or intermediate function of OATP1B1, encoded by SLCO1B1, is almost null compared with close to 25% of Europeans carrying this risk. These observations fully illustrate the clinical importance of diversity in CV PGx; the challenges that it poses are discussed in Section VIII.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4

Distribution of individuals from various ethnic groups according to the rate they metabolize CV drugs. CYP2C19, CYP2D6, and CYP2C9 are key enzymes in the metabolism of clopidogrel, propafenone, and warfarin, respectively, whereas OATP1B1 is a transporter encoded by SLCO1B1 involved in the metabolism of statins. Data were extracted from the PharmaGKB PGx gene-specific information tables (https://www.pharmgkb.org/page/pgxGeneRef; accessed June 26, 2023). CPIC standardized terms were used for allele function assignment (https://cpicpgx.org/resources/term-standardization/). Where applicable, phenotypes were derived from diplotype activity scores (a diplotype is a combination of CYP alleles a particular individual harbors; based on function, each allele is assigned a value to calculate the activity score of the diplotype). For CYP2C19, ultrarapid and rapid, poor and likely poor, intermediate and likely intermediate phenotypes were grouped together. For CYP2D6 (activity scores ranging from 0.0 to 6.0), individuals were considered as ultrarapid/rapid, normal, intermediate, or poor metabolizers if their score was 2.5 or more, ranged between 1.25 and 2.25, 0.25 and 1.0, or was 0.0, respectively. For CYP2C9 (activity score ranging from 0.0 to 2.0), individuals were categorized as normal, intermediate, or poor metabolizers if their score was 2.0, 1.0 or 1.5, or 0.5 or 0.0, respectively.

A. Documentation of the Clinical Validity and the Clinical Utility of Cardiovascular Pharmacogenetics Tests

Irrespective of the therapy area, documenting the performance of a PGx test and its clinical validity (i.e., its ability to predict the occurrence of an ADR or the efficacy of a particular drug) and demonstrating its clinical utility (i.e., its ability to prevent ADRs or improve outcome through differentiation of treatment based on the test results) (Tonk et al., 2017) are critical for its eventual progression toward commercialization. At the population level, the clinical validity of a PGx test designed to prevent ADRs is characterized by its sensitivity (i.e., the proportion of people who are positive for the test and experience ADR), its specificity (i.e., the proportion of people who are negative for the test and do not experience ADR), and the population-attributable risk (i.e., the proportion of ADR cases attributed to carriage of the risk allele, which mirrors the proportion of cases that could be avoided if the test was applied and appropriate changes were made to prescription). At the individual level, clinical validity is characterized by the positive predictive value and the negative predictive value, as discussed earlier for simvastatin. Prospective randomized trials are optimally positioned to document the clinical validity and utility of PGx tests. These studies need to be complemented by medico-economic, cost-effectiveness, and cost-saving studies and backed by robust and reliable analytical capabilities to obtain regulatory approval, access the market, and be successfully reimbursed.

There is a limited number of prospective, randomized, interventional clinical trials run in CV that provide empirical data to demonstrate the clinical validity and utility of PGx testing. Most have been performed on clopidogrel and warfarin, as discussed earlier. In absence of empirical data from interventional trials, the clinical validity of a PGx test designed to prevent ADRs needs to be derived from hypothetical data of observational studies. The odds of experiencing ADR, the frequency of the risk allele, and the incidence rate of ADR are used to determine the performance of the corresponding PGx test. This has been elegantly and comprehensively documented by Tonk et al. for simvastatin, mentioned earlier (Tonk et al., 2017).

B. Medico-Economic Studies to Show Cost Effectiveness and Cost Savings for Cardiovascular Pharmacogenetics Testing

Once the clinical validity and the clinical utility of a biochemical test, including a PGx assay, have been documented, medico-economic studies are usually required to demonstrate cost-effectiveness and cost savings, as presented earlier for clopidogrel and warfarin. Describing the magnitude of cost-effectiveness is rarely as simple as providing an average dollar figure of savings per patient but rather invokes metrics such as incremental cost-effectiveness ratio, incremental cost utility ratio, quality-adjusted life-years, and willingness to pay. None of these concepts are immediately intuitive to physicians and pharmacists tasked with ordering PGx tests and prescribing based on the results and certainly not to patients either (Neumann et al., 2010). Yet these are the metrics used by many regulatory authorities and health agencies when determining reimbursement.

In the CV area, much has been written about the medico-economics of PGx testing. Here again, most studies relate to clopidogrel and warfarin, discussed earlier (Section II). Perhaps unsurprisingly, while there have been studies demonstrating lower costs and more effectiveness of PGx-guided therapy over usual care (Moloney et al., 2019), there are also earlier studies that found PGx-guided therapy to provide limited medico-economic benefits (Meckley et al., 2010). These studies have been incorporated into several systematic reviews. There are considerable issues when weighing the evidence presented in this type of analysis. For example, rather than standardize the threshold, some systematic reviews use the thresholds for cost-effectiveness established in the original papers when drawing conclusions, which may or may not reflect thresholds commonly used in the country the studies were conducted. Previously, the World Health Organization recommended that national cost-effectiveness thresholds fall within a range of one to three times gross domestic product per capita (The World Health Report: 2002: Reducing Risks, Promoting Healthy Life, 2002) However, in 2021 the World Health Organization removed gross domestic product-based cost-effectiveness thresholds from their new generalized cost-effectiveness analysis methodology (Bertram et al., 2021). Whether the obsolescent thresholds were used or not, it is rarely clear how national thresholds are arrived at (Cameron et al., 2018). Thus, the authors of such studies could possibly draw inaccurate conclusions from methodologies not based on quality evidence. Furthermore, many of the studies incorporated into these systematic reviews were conducted in populations that lack genetic diversity.

Taken together, these concepts highlight the monumental effort, time, and costs required to demonstrate the clinical validity, clinical utility, cost-effectiveness, and cost savings, if any, of any PGx test. Moreover, since most medico-economic studies regarding PGx and CV diseases are based on models or observational studies, it is not clear whether the costs of IT, particularly integration of PGx testing results into EMR systems, are factored into economic evaluations. Jiang et al. specifically addressed the cost of a clinical decision support (CDS) tool as part of an economic model of PGx testing for acute coronary syndrome and atrial fibrillation (Jiang et al., 2022). The use of a CDS was compared with no CDS in a modeled health system with 500,000 members aged 18 to 100 years, wherein every year 20% of members aged 55 to 65 received PGx tests for CYP2C19, CYP2C9, CYP4F2, and VKORC1, assuming annual turnover in membership. The model then compared outcomes and cost-effectiveness if a CDS tool was used or not, even though in both scenarios PGx testing was employed. If the CDS tool was used, an estimated 3,169 alerts would be issued over a 20-year period, and an additional 16 major clinical events and 6 deaths would be avoided for acute coronary syndrome and 2 major clinical events and 0.9 deaths would be avoided for atrial fibrillation, over the estimated event and death avoidance due to PGx testing alone. Despite the low estimates for incremental events and deaths, the CDS tool was still found to be cost-effective compared with no CDS alert tool, even if patients received PGx testing in both scenarios.

C. Regulatory Approval of Cardiovascular Pharmacogenetics Assay

Most PGx tests, in CV and other therapy areas, have not been reviewed by the FDA or any other health authority. Furthermore, neither the Centers for Medicare and Medicaid Services nor the Centers for Disease Control and Prevention evaluate PGx tests or examine marketing or scientific claims before these PGx tests are marketed (Rubinstein and Pacanowski, 2021). In 2018, the FDA notified the public regarding its concerns with PGx tests, since most of the marketed tests were not approved by FDA. The FDA had specific concerns about direct-to-consumer (DTC) PGx tests (Shuren and Woodcock, 2018). This came one day after the FDA had approved its first DTC PGx test (https://www.fda.gov/news-events/press-announcements/fda-authorizes-first-direct-consumer-test-detecting-genetic-variants-may-be-associated-medication). In 2020, the FDA published the first iteration of a table of PGx associations, including drug–gene pairs wherein there was, in the FDA’s opinion, sufficient scientific evidence of altered drug metabolism or differential therapeutic effects. This list is a nonregulatory table which contains, as of July 2023, 149 drug–gene interactions, including 13 (9%) in the CV therapeutic area. The FDA stopped short of recommending PGx testing for any of the drug-gene pairs unless the drug labels already required PGx testing (Shuren, 2020).

Currently, many PGx tests are available as laboratory developed tests (LDTs). For decades, the FDA has exercised enforcement discretion, rarely requiring premarket review for LDTs. However, the regulatory review of PGx tests is becoming standardized. On June 14, 2023, the FDA posted notice of a proposed rule to “amend the Food and Drug Administration’s regulations to make explicit that LDTs are devices under the Federal Food, Drug, and Cosmetic Act” (https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=202304&RIN=0910-AI85). This change in policy by the agency comes after years of enforcement discretion and a very deliberative process reconsidering this policy, which followed a 2010 workshop to obtain input from stakeholders, draft guidance published in 2014, a public workshop in 2015 and analysis of hundreds of comments, and a discussion paper issued in 2017 in lieu of issuing final guidance (https://www.fda.gov/medical-devices/in-vitro-diagnostics/laboratory-developed-tests).

Another way that the FDA might eventually regulate LDTs is through passage of the Verifying Accurate Leading-edge IVCT Development (VALID Act). The VALID Act is a bipartisan and bicameral bill that has been under development for years (https://www.congress.gov/bill/117th-congress/house-bill/4128; https://www.congress.gov/bill/117th-congress/senate-bill/2209). It would provide a regulatory framework to set quality and performance standards and require premarket review for LDTs but would do so in a flexible manner, utilizing a risk-based system of oversight and exempt existing LDTs. This would result in a more comprehensive change to regulation of LDTs and allow the FDA to effectuate oversight.

Another bill that could grant the FDA greater oversight of PGx testing for ADRs is the Right Drug Dose Now Act (https://www.congress.gov/bill/117th-congress/house-bill/6875). The proposed legislation mandates measures to tackle ADRs, including the use of PGx testing. The bill obliges the Department of Health and Human Services to formulate a national prevention plan, coordinate its report with related federal agencies, and assemble a steering committee for plan updates. The bill also enforces health information technologies to signal the necessity of PGx testing prior to completing a medication order and mandates Department of Health and Human Services to enhance EMRs for improved data collection in PGx. Collaboration with the FDA is required to upgrade reporting processes for ADRs. Lastly, the bill instructs the US National Human Genome Research Institute to promote awareness about ADRs and to educate healthcare professionals on PGx testing.

D. Marketed Pharmacogenetics Tests for Cardiovascular and Other Drugs

While PGx testing can be performed using LDTs, there are several commercially available PGx tests. Some, like 23andMe’s test, are DTCs. Most, however, are prescribed by physicians, particularly since understanding and applying the results of the test to medical practice should be done by physicians and only one DTC PGx test is authorized by the FDA (https://www.fda.gov/medical-devices/in-vitro-diagnostics/direct-consumer-tests). Like many LDTs for PGx, most commercially available PGx tests look for specific variants employing either microarrays, polymerase chain reactions or mass spectrometry. Those that we have identified with easily discoverable lists of genes tested, costs and technology used, are summarized in Table 3. This list does not include Invitae, which had performed PGx testing in the past but announced on June 29, 2023, that they would put this activity on hold (https://www.invitae.com/en/providers/mental-health/personalize-medication-prescribing?tab=for-you).

All the tests in Table 3 cover variants that are within recommendations by the CPIC, DPWG, and/or FDA. Fulgent Genetics' PGx Focus is the only test that reports variants in the CYP4F2, G6PD, and NAT2 genes. While the NAT2 gene is present in the FDA Table of PGx Associations (level 2 for procainamide and level 3 for hydralazine), G6PD PGx actionability for CV drugs is not supported by either the CPIC, DPWG, or FDA. The 23andMe test is limited to two genes relevant to CV PGx, which have an impact on response to clopidogrel and statins. As for the method of detecting variants, only the PGx Focus test is conducted entirely by sequencing. The 23andMe test uses microarray technology (Bead Array) and the Sonic Genetics test uses both mass spectrometry (for SNPs) and real-time polymerase chain reaction (for copy number variants). The other tests rely on Taqman genotyping (with polymerase chain reaction amplification). As for cost, the range is fairly broad with the most expensive test costing nearly twice as much as the least expensive one, although a direct comparison would require consideration of all the non-CV variants included in the tests.

Every test manufacturer provides reports that include clinically actionable information to inform medical decisions. While analysis of the manufacturer-issued reports and their understandability by physicians and patients is beyond the scope of this review, it is important to highlight the necessity for medical professionals and patients alike to comprehend the information provided and put it into the appropriate context of medical care. These reports rely on guidelines from scientific societies, such as the CPIC and DPWG, which were discussed in Section III.

A. Barriers to Adoption of Cardiovascular Pharmacogenetics

Despite the wealth of data that has been generated over the past 30 years, the abundance of literature in the field of CV PGx (Supplemental Fig. 1), medico-economic studies demonstrating cost-effectiveness and cost savings of incorporating PGx in patient care, at least for clopidogrel (Sections II and VII), the multitude of tests available commercially (Section VII), the incorporation of PGx in the label of 36 CV drugs, and the recommendations made by expert groups such as the CPIC and DWPG (Section III) and by cardiology societies like the European Society of Cardiology discussed earlier, it is fair to say that adoption of PGx testing has been limited so far for CV care.

Several factors appear to contribute to this situation, and understanding the root for limited adoption and implementation of PGx in CV diseases prevention and treatment is important to reap the full benefits from scientific developments in the field. One key factor is the need for unambiguous demonstration of the clinical validity and the clinical utility of PGx testing. Moreover, to be broadly adopted by clinicians and patients requires PGx testing to be reimbursed. Robust data demonstrating the medico-economic benefits of such testing, above and beyond existing standards of care, are usually required to convince payors to reimburse new tests. Beyond the studies mentioned earlier showing cost-effectiveness and cost savings for clopidogrel (AlMukdad et al., 2020; Galli et al., 2021; Pereira et al., 2021; Morris et al., 2022) and, to a certain extent, for warfarin (Tse et al., 2018; Kamil et al., 2022; Morris et al., 2022), more medico-economic studies are needed for other CV drugs. Analyses of large cohorts linked to EMR may partly address this point.

Another factor playing a role is the complexity in interpreting PGx testing. Deriving the degree to which individuals metabolize a particular drug (i.e., metabolic phenotype) based on array- or sequencing-detected variants in genes involved in its metabolism requires specific knowledge. In addition, the way allelic variation correlates with metabolic phenotype is population-specific and may be incompletely captured by array-based genotyping. Moreover, the genotype-phenotype correlation for a given enzyme, especially CYPs, may be drug-specific due to the enzyme's substrate selectivity. Finally, this correlation may be affected by phenoconversion, a situation in which factors such as drug*drug interactions, changes in metabolic state due for instance to pregnancy or inflammatory conditions, or other genetic variations influence the metabolic phenotype (Shah et al., 2016). On the other hand, PGx resources are scattered and siloed, leading to a lack of consistency and standardization when reporting PGx findings. To address this challenge, a group of experts affiliated with PharmGKB has announced the development of a unified PGx resource called ClinPGx (Klein et al., 2023) based on the model of ClinGen (a standardized infrastructure for sharing relevant data on genes, genetic conditions, and causal variants, characterized by a worldwide expert curation ecosystem) (Rehm et al., 2015).

As for any other molecular biomarker, time is the essence. It takes years, or even decades, for companies to build up the necessary knowledge and develop a test that has the satisfactory characteristics and performance to obtain regulatory approval and be adopted by clinicians. This delay explains in part why only a few PGx tests have reached the market. As discussed in Section V, one may reasonably expect that, for at least a certain fraction of the PGx markers recently discovered through GWAS, corresponding PGx tests to assay them will one day reach the market.

As mentioned, most PGx studies so far have been performed on populations of European descent. Populations may vary widely in the prevalence of PGx-relevant markers (Fig. 4), emphasizing the need to perform these types of studies in diverse populations, which should equally benefit from PGx. This is a key priority for initiatives like the Human Heredity and Health in Africa Consortium (H3Africa Consortium et al., 2014) which has facilitated research to better capture the genetic variation in the African populations, including for pharmacogenes (Choudhury et al., 2021; Asiimwe et al., 2022; Twesigomwe and Drögemöller, 2023). The African Pharmacogenomics Consortium was created in 2018 under the umbrella of H3Africa and the African Academy of Sciences, with the overarching goal of setting up in the coming years a platform for regulated data sharing and collaborative research in the field of PGx (Dandara et al., 2019). Regarding CV PGx, one can cite the example of the African American Cardiovascular Pharmacogenetic Consortium. The project aimed to build a large cohort of African Americans for both discovering and validating new genetic associations in CV drug responses that have been well studied in Europeans, such as clopidogrel response and warfarin-associated bleeding risk (Friedman et al., 2019). Additionally, translational projects were planned, such as making pre-emptive PGx results available to clinicians through a CDS. Yet, only a small proportion of the anticipated 1,000 participants (N = 135) have been included in the published findings so far (Saulsberry et al., 2021), and the researchers did not provide clear information on the barriers that affected the enrollment process. Additional efforts in assessing the allele frequencies and functional impacts of important PGx variants in ancestry-diverse populations are needed to facilitate the development and validation of adequate PGx tests and clinical guidelines in CV care and beyond.

PGx testing is an interrogation of the most intimate layer of individuals and, as such, carries a strong legal, ethical, and emotional component. These concerns are legitimate when one considers that genetic variants are shared in families and transmitted through generations and may have important predictive implications for disease prediction, prevention, and treatment. One particular advantage of PGx testing, compared with other genetic tests, is that most PGx tests are predictive of the risk of experiencing ADRs when exposed to an environmental trigger, i.e., a drug, but do not predict per se any risk of disease, so that genetic counseling is not required. In addition, most actionable PGx variants, which have not been selected through evolution of human humankind, are much more prevalent than disease-causing variants. Still, strong governance and policies are required to avoid any misuse of PGx data, especially in large biobanking initiatives (Joly and Knoppers, 2006; Sillon et al., 2008).

Successful implementation of PGx requires a careful understanding of patients’ perspectives and needs. A group of investigators explored barriers and facilitators for PGx testing acceptance among 31 participants of the Pharmacogenomics Testing for Veterans Program in the United States (Melendez et al., 2023). Acceptors of PGx testing prioritized the potential health benefits of PGx testing, while decliners prioritized the possibility of a data breach or negative impact on health insurance. Another group of respondents who were “contemplating” PGx testing wanted more information before making a decision. Comprehensive patient education, clear communication channels, and ongoing research to understand patient acceptance and concerns are needed. A recent scoping review on patient experience with PGx listed five themes to focus on during patient counseling on PGx. These include 1) reasons for testing/perceived benefit, 2) understanding of results, 3) psychological response, 4) impact of testing on patient-provider relationship, and 5) concerns about testing/perceived harm (Allen et al., 2022). Importantly, educational barriers also exist for a variety of stakeholders in addition to patients (Luzum et al., 2021). Perception of lack of evidence is often cited by healthcare providers, including noninterventional cardiologists (Deininger et al., 2019). An illustrative example of an initiative to overcome this barrier is the Introductory Tutorial on CV PGx for Healthcare Providers published by the Pharmacogenomics Global Research Network (Oni-Orisan et al., 2023). Unsurprisingly, this tutorial focuses on clopidogrel, statins, and warfarin.

Finally, one critical challenge in the clinical implementation of PGx is turn-around time and delays in returning results from PGx testing to clinicians. Solutions to address this issue include the development of reactive and pre-emptive PGx testing. These solutions are discussed next.

B. Reactive Testing to Facilitate Pharmacogenetics Adoption in the Clinic

Reactive PGx testing refers to performing a PGx analysis in response to a specific trigger, such as ordering a drug for which there is PGx recommendation, specific procedures like PCI, or unexpected events including the occurrence of ADRs or lack of drug efficacy (Haidar et al., 2022). This approach implies that the PGx test is performed in-office or at the bedside and that results are returned within a short turn-around time, typically less than one week. Several point-of-care testing solutions have been developed to meet these requirements. A selection of seven examples of successful reactive PGx initiatives is presented in Table 4. The first one was launched in 2004 at the Cincinnati Children’s Hospital and the remaining six after 2011. Four initiatives are US-based and monocentric; the remaining three are collaborations between five and seven centers, including the Implementing Genomics in Practice (IGNITE) program and the Translational Pharmacogenetics Program in the United States and Ubiquitous Pharmacogenomics (U-PGx) in Europe. These initiatives interrogated a limited number of candidate genes (ranging from 5 to 120), using genotyping arrays from various manufacturers.

The IGNITE network was established in 2013 to support and coordinate six research projects aiming at developing, investigating, and improving the use of genomics in clinical practice (Weitzel et al., 2016). Three projects were PGx-oriented and were led by the University of Florida, Indiana University, and VUMC. CYP2C19-clopidogrel testing was reactively ordered in a pilot implementation project at the University of Florida if the patient underwent PCI (Johnson et al., 2013). At Indiana University, the INGENIOUS trial was performed to test the performance of reactive PGx testing for 28 index drugs (Eadon et al., 2016). At VUMC, patients who underwent left heart catheterization were reactively tested for CYP2C19-clopidogrel. In this institution, investigators also pre-emptively tested adult outpatients in primary care, cardiology, and endocrinology clinics (Van Driest et al., 2014; Luzum et al., 2017) (the pre-emptive approach is discussed in Section VIII.C). In 2015, the Pharmacogenetics Working Group was formed to share and disseminate PGx implementation metrics from the IGNITE network (Cavallari et al., 2017). This consortium led the largest real-world investigation on the clinical value of CYP2C19 testing to guide antiplatelet therapy in patients undergoing PCI (Beitelshees et al., 2022) The next phase of IGNITE began in 2018 with the goal of conducting pragmatic clinical trials of genomic medicine interventions, including in PGx (Cavallari et al., 2022).

The Translational Pharmacogenetics Program (TPP), part of the National Institutes of Health Pharmacogenomics Research Network, was created in 2011 to study the PGx implementation programs in eight US healthcare systems and to identify logistic barriers to implementation and propose solutions (Shuldiner et al., 2013). By 2016, nearly 100,000 PGx test results had been entered in the respective EMR of the participating centers, and the TPP collected and normalized numerous metrics of PGx implementation across these centers (Luzum et al., 2017). The median turnaround times were 2.6 days (range 0.3 to 16 days) and 14 days (range 1 to 249 days) for reactive and pre-emptive testing, respectively. Importantly, one out of four PGx tests had potentially actionable results. Even though CPIC guidelines were widely used for PGx test interpretation and clinical recommendations, a key barrier to broad PGx clinical adoption identified by the TPP was the heterogeneity in PGx data representation in the EMR (genomic data can be stored and reported in different formats), which complicated the portability and interoperability of the results.

Given their central role in prescription management and patient advice, community pharmacists are at the front line to implement PGx (Padgett et al., 2011). The Implementation of Pharmacogenomics into Primary Care Project tested the adoption of PGx among pharmacists and used a reactive strategy (panel-based for 10 drugs) (Bank et al., 2019; van der Wouden et al., 2019). In this study, PGx-based recommendations were made in 31% of incident drug prescriptions and were followed by 89% of the clinicians.

Reactive PGx testing has limitations. First, it is technically demanding and costly over time, as tests are carried out individually rather than in batches. Because most reactive PGx tests interrogate a limited number of variants within a restricted number of genes, they may miss variants that would have been detected through more extensive genotyping or through sequencing. Similarly, PGx testing may have to be repeated several times for the same patient over the course of their life. In that sense, reactive PGx testing may be particularly effective when assaying simultaneously several pharmacogenes (i.e., the panel is ordered reactively but some genes are pre-emptively tested). In particular, the University of Florida Health Precision Medicine Program (Table 4) has adopted this approach following the successful implementation of several single-gene PGx tests (Marrero et al., 2020).

In the muticenter, controlled, crossover randomized implementation PREPARE study run by the U-PGx Consortium, 50 PGx variants in 12 pharmacogenes were tested simultaneously (Swen et al., 2023). The panel was ordered once a patient was initiated on a drug impacted by DPWG guidelines (N = 42, notably clopidogrel, atorvastatin, simvastatin, and warfarin), and the results were available to healthcare providers within seven days. Atorvastatin was the most common medication in the study. Physicians and pharmacists followed 70% of the PGx recommendations they received during the study period. Overall, the incidence of clinically relevant ADRs over a 12-week follow-up period was 628 (21.5%) for 2,923 patients in the study group and 934 (28.6%) for 3,270 patients in the control group [odds ratio 0.70 (95% CI: 0.61-0.79); P value < 0.0001] (Swen et al., 2023). This study suggests a benefit of a pre-emptive approach, as 953 (13.7%) patients were later prescribed a second actionable drug with PGx recommendation, 79 (1.2%) a third drug, 6 (0.1%) a fourth drug, and 1 a fifth drug. A cost-effectiveness analysis is currently underway. Several limitations should be mentioned. First, 96% of participants had European self-reported ancestry, limiting the external applicability of the findings to other ancestry groups. Second, a similar risk reduction in ADR was observed in all patients of the intervention group, including those without actionable variants. This unexpected result may be due to the inclusion of additional recruiting centers with different prescribing patterns during the study period, potentially leading to higher rates of adverse reactions in the control group. Third, data on drug efficacy or specific patient-reported outcomes such as treatment satisfaction were not collected. Future studies should include more diverse ancestry groups, explore additional outcomes, and assess the benefit of performing sequencing rather than genotyping and WES- or WGS-based analysis rather than panel-based.

C. Pre-Emptive Testing as an Efficient Solution for Clinical Pharmacogenetics Implementation

In contrast to reactive PGx testing, pre-emptive PGx testing refers to the analysis of PGx variants prior to, and independently from, any specific trigger and incorporation of the results into EMRs. That way, clinicians are immediately informed, at the time of prescription, whether their patient is genetically at risk for developing ADRs or would minimally benefit from the drug (Dunnenberger et al., 2015; Haidar et al., 2022). By design, pre-emptive PGx addresses the issue of turn-around time, does not require prior selection of genes for testing, and can be coupled with CDS notifications that prompt for action at the time of drug prescription. In addition, as pre-emptive PGx interrogates numerous pharmacogenes simultaneously, testing is performed once and can be used during a lifetime. Moreover, the benefits of pre-emptive PGx testing apply whenever patients are prescribed drugs that are paired with actionable PGx variants, irrespective of any specific disease or therapeutic area. This is particularly relevant considering that, as mentioned earlier (Section VI), we almost all carry at least one actionable PGx variant and close to 30% of the population carries two or more such variants (Pirmohamed, 2023; Swen et al., 2023).

Nine representative initiatives in pre-emptive PGx are listed in Table 5. Five are monocentric, based in the United States (n = 4) or in Singapore (n= 1); the other four are consortia based in the United States or the UK. The former ones were initiated between 2007 and 2014 and the latter one in 2023. Most initiatives rely on genotyping using various platforms, with the number of genes tested ranging from 34 to 230.

A pioneering initiative in pre-emptive PGx is the PREDICT program at VUMC. PREDICT was initiated in 2010 (Van Driest et al., 2014). Within 4 years, 10,000 patients were successfully genotyped using a panel of 34 genes and demonstration was made that a pre-emptive approach significantly reduced the burden of PGx testing compared with a reactive strategy (Van Driest et al., 2014; Liu, Vnencak-Jones, et al., 2021). Moreover, PREDICT provided proof-of-principle for reinterpretation and reprocessing of PGx results. This represents a complex but necessary step to ensure the lifetime value of pre-emptive PGx and the incorporation of new knowledge in the interpretation of PGx data. In 2020, VUMC systematically reinterpreted historic CYP2C19 and CYP2D6 genotypes to update phenotypes in the context of a new CDS for selective serotonin reuptake inhibitors (Liu, Van Driest, et al., 2021). Updating PGx results required an important multidisciplinary team effort but was shown to be feasible and affordable. A total of 289 PREDICT participants (corresponding to 2.4% of the whole cohort still alive) were taking a selective serotonin reuptake inhibitors and had new actionable PGx variants. However, VUMC was not able to outreach clinicians for one-quarter of them (72/289). This experience underlines the importance of ensuring the portability and interoperability of pre-emptive PGx results.

Another early initiative in pre-emptive PGx is the Coriell Personalized Medicine Collaborative, a prospective study launched in 2007. This study recruited participants from various settings (community, oncology, and chronic diseases) to assess the clinical relevance of reporting genetic risk factors, including PGx, to patients and physicians (Keller et al., 2010). As of 2009, 4,372 participants from the community were enrolled and genotyped. The investigators developed a systematic process, called PhAESIS (Pharmacogenomics Appraisal, Evidence Scoring and Interpretation System) to evaluate which gene-drug associations were ready for implementation into patient care and demonstrated its application for seven commonly prescribed drugs, including clopidogrel, statins, and warfarin (Gharani et al., 2013). The end of the enrollment phase was announced in October 2019, with more than 7,000 participants having been enrolled (https://www.genomeweb.com/genetic-research/coriell-personalized-medicine-collaborative-ending-recruitment-phase).

After initially focusing on single-gene reactive PGx tests, St. Jude Children’s Hospital started preemptive PGx testing for 230 genes in 2011, using an innovative protocol they called PG4KDS (Hoffman et al., 2014). This protocol allows for a stepwise process for pre-emptive PGx testing, organizing and reporting PGx data, and obtaining consent from patients and families to withhold or share results, including incidental findings. Over a 2-year period, more than 1,500 patients consented, and four gene tests were released into the EMR for clinical implementation (TPMT, CYP2D6, SLCO1B1, and CYP2C19).

Capitalizing on the success of these initiatives limited to single or few centers, larger PGx networks have been built to encourage the adoption of pre-emptive PGx, including the NIH-funded Electronic Medical Records and Genomics (eMERGE) program in the United States, the EU-funded U-PGx, and the National Health Service-funded PROGRESS Program in the UK. eMERGE is the largest and probably most advanced pre-emptive PGx consortium. This network was initiated in 2007 with the goal of combining biorepositories with EMR to foster genomic discoveries and facilitate genomic medicine implementation and validation (eMERGE Consortium, 2021). It currently includes 10 sites across the United States. The PGx component of eMERGE, called eMERGE-PGx, is a partnership between the eMERGE group and the Pharmacogenomics Research Network (Rasmussen-Torvik et al., 2014). Major objectives of eMERGE-PGx were the implementation of a sequencing platform specifically dedicated to assessing PGx variants (called PGRNseq) and sequencing of nearly 9,000 individuals over a 3-year period, the integration of a CDS with validated PGx clinical guidelines in the EMR, and the assessment of clinical outcomes data. Additionally, the project aimed at facilitating PGx research through linked repositories of PGx variants of unknown significance and EMR-based phenotypic data (SPHINX tool). PGx variants discovered in 33,966 patients are listed in the SPHINX repository. This tool was also configured to integrate dynamic updates from DrugBank (linking variants to drug-gene pathways), the GWAS Catalog (linking variants to phenotypic associations), dbSNP (standardization of variant naming, genomic position, and population frequency), and PharmGKB (linking variants with existing PGx evidence) resources. eMERGE has clearly demonstrated that real-world PGx data sharing in a structured repository is necessary for the implementation of PGx testing, that ongoing education is critical for provider utilization, and that SNP coverage in targeted panels should be sufficient for interpreting the full range and types of variants for clinical return (eMERGE Consortium, 2021).

Expanding its expertise in integrating genomic data into the EMR, eMERGE is presently focusing on the development of methods to return disease risk via “genome-informed risk assessment” reports, which combine PRS, monogenic risk, family history, and clinical risk assessment (Linder et al., 2023). PRS have gained considerable interest in the scientific community for various clinical applications, such as disease prediction, risk stratification, drug development, and PGx, including in CV clinics (Kullo et al., 2022; O’Sullivan et al., 2022; Chung et al., 2023). PRS can serve to identify individuals who are most likely to benefit from a drug (because of drug efficacy or drug indication based on risk/disease stratification) or to experience ADRs. PRS in PGx have been recently reviewed (Cross et al., 2022; Johnson et al., 2022; Simona et al., 2023), and specific methods have been proposed to generate PGx-PRS (Zhai et al., 2022). There is emerging evidence of the clinical utility of PGx-PRS for CV diseases, notably for statins (Mega et al., 2015; Natarajan et al., 2017), clopidogrel (Lewis et al., 2020), and beta-blockers (Lanfear et al., 2020) efficacy. Regarding the use of statins in primary prevention, for example, PRS have the potential to markedly reduce the number needed to treat to prevent coronary heart disease events (Cross et al., 2022). Results are still conflicting regarding antihypertensive drugs, such as calcium channel blockers, thiazides, and angiotensin-II receptor blockers (McDonough et al., 2013; Sánez Tähtisalo et al., 2020). Along the same line, PGx-PRS might also be successfully applied to predict CV ADRs, such as drug-induced long QT syndrome (Strauss et al., 2017).

Altogether, these initiatives have been shown to be of unique value for retrospective and prospective validation of new PGx testing, anticipating cost-benefits from such interventions, and expanding our knowledge in PGx beyond European populations. Key elements of success include strong institutional support and leadership, a robust IT infrastructure with a versatile EMR and seamless CDS tool, an experienced clinical laboratory, a process to ensure portability (interoperability) and updating of standardized results, a process to manage return of results and incidental findings, and comprehensive clinician education and behavioral interventions. Important areas requiring further exploration include ethical, legal, and social aspects of pre-emptive PGx (notably equity and inclusion, perceptions and needs of the public) and the potential of cascade PGx screening. Ultimately, regulators as well as the pharmaceutical and diagnostics industries should be involved in PGx efforts—both in research and in clinical implementation—toward public-private partnerships (Swift, 2022).

Presently, CDS systems that support PGx vary in comprehensiveness, cost, and interoperability. A study compared three commercial CDS tools that use next-generation sequencing for guiding therapy in oncology patients (Perakis et al., 2020). In this particular use case, the three CDS tools had low concordance, running counter to the ideal of PGx-guided therapy. Wake et al. reviewed the types, functions, and limitations of CDS tools for PGx that are in practice (Wake et al., 2022). Additionally, Smith et al. reviewed the metrics used to quantify the clinical utility of PGx CDS (Smith et al., 2023). While not specific to CV disease, this paper provides a comprehensive review of the current state of PGx CDS and concludes that few data are available to support approaches for optimal implementation. It should be noted that with rapid advances in technology, some of these tools might have evolved significantly, while others might cease to exist. As advocated by the U-PGx initiative, the harmonization of existing PGx information in drug labels and guidelines, as well as the facilitation of CDS adoption by healthcare providers, are major challenges to reaching the full benefits of CDS (Blagec et al., 2022).

Finally, learnings from the pre-emptive PGx initiatives described here have been foundational for a very recent shift in clinical care efforts to overcome inertia to bring PGx testing to the bedside by payers and healthcare systems. Instead of highly targeted testing initiated by a fraction of physicians, these systems are attempting to integrate PGx testing across the board as a means of driving medico-economics. Frederick Health is a US community health system in the state of Maryland that is rolling out pre-emptive PGx testing. The goal is to provide PGx testing for all its employees initially and then for many commercially insured patients meeting certain criteria (e.g., number of medications, multiple hospital admissions). The PGx panel is free for employees and costs the organization less than USD 300 per employee. Currently, physicians at Frederick Health receive genetic test reports as PDF files that might be easy to print or view on screen but are not machine-readable, and the company is planning on integrating a CDS tool into the EMR system (Versel, 2023). Similarly, companies have started offering PGx testing as a service for employers or healthcare systems in their entirety as cost-effective solutions. For example, a partnership between the Teachers' Retirement System of the State of Kentucky, Coriell Life Sciences, and the Know Your RX Coalition documented a reduction in direct medical costs for Medicare Advantage patients (aged ≥ 65 years) of USD ∼7,000 per patient during the first 32 months of a comprehensive medication management program with pre-emptive PGx (5,288 participants compared with 22,357 controls) (Jarvis et al., 2022).

D. Pre-Emptive Pharmacogenetics Opportunities for Conditions Predisposing to Cardiovascular Diseases

Most CV diseases are typically diseases of aging, and the incidence of CV diseases is amplified in patients with other common age-related conditions, including obesity, diabetes, or depression. These diseases have reached noncommunicable epidemic status and affect more than 30%, 20%, and 10% of the population aged 65 years or older, respectively, with a corresponding CV risk increased by approximately 30%, two- to four-fold, and 40%, respectively (https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight; https://www.who.int/news-room/fact-sheets/detail/depression; https://www.who.int/news-room/fact-sheets/detail/diabetes). Survivors of breast cancer, and possibly other conditions like long COVID-19, also appear to be at increased risk for CV diseases (Koric et al., 2022; Xie et al., 2022). Individuals with such pre-existing conditions usually take several drugs and are exposed to drug*drug interactions. As an illustration, the median number of drugs taken daily by patients with diabetes is 4.1 in the United States (Boye et al., 2020). The risk of ADRs may be further increased by the decline with aging of kidney and liver function and cognitive impairments. Moreover, as the recent studies summarized earlier (Section VI) show, almost everybody carries actionable PGx variants (Pirmohamed, 2023). Accordingly, polypharmacy patients may also be at high risk of drug*drug*gene interactions, as response to several non-CV drugs including antidepressants have been shown to have a genetic component; PGx tests are available and, for some of them, have been shown to be cost-effective (Perlis et al., 2018). Finally, interindividual variability in the efficacy and ADR response to novel, transformative, and expensive molecules that show great promise for obesity and diabetes may have a genetic component, and more data are needed to anticipate high responders (Dawed et al., 2023).

Even if the medico-economic benefits of this approach remain to be demonstrated and despite the practical difficulties in its implementation, pre-emptive PGx appears to be ideally suited to optimize therapy in these high-risk populations and to genetically identify, using PRS, those individuals who would optimally benefit from preventive measures. For diabetes, a few pilot studies have already been performed in this direction (Tremblay et al., 2021). These studies have provided early evidence that pre-emptive PGx has the capacity to avoid ADR. In a small randomized control study including 110 patients aged 50+ years referred to home health after hospital discharge, pre-emptive PGx testing of five CYP genes, compared with usual care, led to a reduction of 52% in the mean number of hospital readmissions and a reduction of 42% in the mean number of emergency department visits by day 60 (Elliott et al., 2017). Additionally, there is evidence that this approach is cost-saving, with a USD 4,382 per patient cost savings at 60 days of follow-up. Finally, more complete genetic information may allow detection of monogenic forms of diabetes (called maturity-onset diabetes of the young), which appears to represent up to 5% of the population diagnosed as having type 2 diabetes (Younis et al., 2022). Detection of maturity-onset diabetes of the young is important for selection of specific treatment and may allow cascade screening among relatives (GoodSmith et al., 2019). For reasons described earlier (Sections III and IV), WGS may be the best approach to derive optimal benefits from pre-emptive PGx in these populations.

While these studies might not be considered as purely pre-emptive, which could be conceptualized as conducting PGx at birth or at least while a young adult and relatively healthy, it is tempting to hypothesize, based on these results, that a lifetime of pre-emptive PGx-guided treatment would result in better outcomes and potentially better medico-economics. This is in clear contrast to the mixed results of PGx testing conducted at the time of treatment, reviewed in Section VII. More work and innovative IT solutions are needed to demonstrate the value of this approach and its impact on healthcare.