1. Introduction
While the U.S. Food and Drug Administration has added pharmacogenetic information to over 150 drug labels (
https://www.fda.gov/drugs/science-and-research-drugs/table-pharmacogenomic-biomarkers-drug-labeling, accessed on 10 May 2021), the clinical implementation of this guidance is limited at best. As a result, the Clinical Pharmacogenetics Implementation Consortium (CPIC
®) was created to establish peer-reviewed, evidence-based guidelines for utilizing the results from pharmacogenetic (PGx) tests. CPIC guidelines are conservative and aimed at providing a clinical decision framework for the utilization of PGx to guide pharmacotherapy for certain “priority” drugs. Among the prioritization of CPIC recommendations, drug/gene pairs with Level A notation indicate that “genetic information should be used to change prescribing of affected drug”. Hence, drugs with Level A evidence possess CPIC’s strongest prescribing recommendations.
There exists large disparities between various racial/ethnic/ancestral groups for both cerebrovascular/cardiovascular disease risk as well as response to pharmacotherapeutic interventions for these conditions [
1,
2,
3,
4,
5]. Recent work in the field of pharmacogenomics has increasingly focused on the identification of ethnicity/population-specific variation in drug metabolizing and drug target genes that can help account for these differences. For example, individuals identifying as having African ancestry have a higher risk of developing cardiovascular disease (CVD) than individuals of European-American ancestry (EA) [
1]. African-Americans (AAs) have a higher on-treatment platelet reactivity with antiplatelet drugs such as clopidogrel, suggesting a pharmacogenetic variation in drug metabolism [
6]. Observations of differential racial or ethnic risk of disease/treatment failure is not only limited to thrombotic disease. For instance, Asian ancestry confers a higher risk for intracerebral hemorrhage (ICH) amongst all ethnicities worldwide. In the US, AAs also carry a greater risk of ICH than individuals of European ancestry [
7].
There are three cardiovascular drugs that carry Level A prescribing guidance by CPIC: warfarin, clopidogrel and simvastatin. Warfarin is an oral anticoagulant that targets Vitamin K oxidoreductase complex 1 (VKORC1) thereby preventing the carboxylation/activation of clotting factors. A large amount of research supports variation in
CYP2C9 and
VKORC1 significantly contributing to both the warfarin dose and incidence of severe toxicity (i.e., intracerebral hemorrhage) [
8]. Clopidogrel, a prodrug, is an anti-platelet agent that targets the P2Y12 adenosine receptor. CPIC guidance on clopidogrel focuses on the impact of
CYP2C19 variation and treatment response in patients with acute coronary syndrome (ACS) undergoing percutaneous coronary intervention (PCI). Simvastatin is an HMG-CoA reductase inhibitor used for the treatment of hypercholesterolemia. While generally well-tolerated, simvastatin use has been associated with myalgias, myopathy and, in rare cases, rhabdomyolysis [
9,
10]. While the frequency of reported statin-associated myalgias is no different than placebo [
11,
12], severe muscle toxicity has been associated with allelic variation in the
SLCO1B1 (OATP1B1) gene [
10] which facilitates hepatic uptake of the drug.
The use of race and ethnicity in medicine is controversial due to the large variability amongst different racial/ethnic groups, lack of detailed ethnic information, the co-existence of social and environmental determinants of health, and the broad geographic categorization of “racial” groups (i.e., Asian, African). Yet, race and ethnicity are still employed as surrogate biomarkers in clinical practice [
13,
14] including in many US FDA prescribing guidelines [
15]. For each of the drugs described above, considerable variability amongst different ethnic/ancestral/racial populations [
16,
17,
18,
19,
20] exists within their respective pharmacogenes covered in the CPIC guidelines. Given this variability, and its potential impact on drug response and toxicity, we sought to determine the degree to which race/ethnicity contributed to the frequency of adverse drug reactions for these priority drugs. We performed a retrospective analysis of electronic health records (EHR) for these drugs from a large, diverse medical center in Washington, DC.
4. Discussion
Heterogeneity in drug response is a common issue in all areas of medicine. Although identification of factors accounting for this variability is far from complete, key pharmacogenetic factors that influence their efficacy and/or toxicity have been identified for many drugs. For those factors related to genetic variation, CPIC was established to provide evidence-based guidance to practitioners for certain drug:gene pairs (
https://cpicpgx.org, accessed on 10 April 2021). Extensive genetic variability exists within the human population for most, if not all, of the pharmacogenes covered in the CPIC guidelines. This variability has in part contributed to the use of race and ethnicity in clinical prescribing guidelines [
14]. Here, we have examined the degree to which self-identified race may contribute to the probability patients will experience adverse events with three cardiovascular drugs carrying the strongest prescribing recommendations by CPIC. The results suggest that apparent associations between race and adverse events is highly drug-specific and is oftentimes superseded by other clinical and environmental factors (i.e., clopidogrel). However, in a few instances self-reported race appears to be a significant predictor of drug toxicity in lieu of other covariates (i.e., warfarin bleeding risk in Asians and lower simvastatin adverse events in AA). In these situations, the associations are consistent with known heterogeneity for the relevant pharmacogenes covered in the CPIC guidelines.
The probability of experiencing any adverse drug reaction with warfarin was not associated with a specific race/ethnic group when corrected for comorbidities except in self-identified Asian patients. Consequently, the widespread use of race for the purpose of predicting adverse reactions to warfarin is not warranted based on our study. Predictably, we found that the known risk factors for warfarin toxicity (i.e., hypertension, CAD, COPD, renal impairment, and NSAID usage) [
23], in addition to female gender, were significant predictors of warfarin adverse outcomes (
Table 2). Diabetes (Type I or II) diagnosis was also associated with an increased probability of experiencing any adverse event while taking warfarin (
Table 2). This is potentially related to either the significant pharmacokinetic interactions between warfarin and some antidiabetes drugs [
26] and/or other comorbidities (i.e., hypertension) commonly present in diabetic patients. When we focused on the probability of experiencing any bleeding events while taking warfarin, we observed that hypertension and renal impairment both increased probability (
Table 3) which is consistent with other findings [
23]. The likelihood of having a bleeding event on warfarin was significantly higher in self-identified Asian individuals in both the race-only and full regression models (
Figure 2,
Table 3). A possible explanation for this finding might be related to the higher frequency of carriers of the
VKORC1 haplotypes which confer a low-dose warfarin phenotype in Asian populations [
17,
27] potentially increasing the risk of warfarin toxicity [
19,
28]. Although there is a large degree of ethnogeographic variation amongst different Asian populations [
27], there is an overall greater number of “low dose”
VKORC1 genotypes in Asians compared to other populations [
29]. For example, the frequency of the −1639 CC/CT (rs9923231) and 1173 GG/GT (rs9934438) alleles, which have lower warfarin dose requirements, are much higher in the Chinese vs. Caucasian populations [
29]. Hence, apparent associations between self-identified Asian race and warfarin bleeding risk are plausible and consistent with the known pharmacogenetics of this drug.
Both the U.S. FDA [
30] and CPIC [
31] warn of therapeutic failure of clopidogrel for preventing thrombotic events in individuals who are
CYP219 poor metabolizers. There is considerable worldwide variability within the
CYP219 gene suggesting that certain ethnic or geographically distinct populations may potentially have high on-treatment platelet reactivity with this drug [
32,
33] or potentially greater risk of adverse events [
33]. Similar to warfarin, we did not find that race was associated with any adverse events (
Figure 2A; Black/African-American) or bleeding events (
Figure 2B; Black/African-American, Other/Multiple race) while taking clopidogrel when corrected for covariates. (
Table 4 and
Table 5). Given that many clinical covariates already implicated in predicting clopidogrel response [
32] were significantly predictive of both clopidogrel adverse events (
Table 4) and bleeding events (
Table 5), any apparent associations between race and drug toxicity are better explained by these factors in this study population. Lastly, the primary clinical endpoints of this study focused on adverse events and not treatment failure (i.e., thrombotic events). It is likely that race and ethnicity are not surrogates for any genetic factors that influence the likelihood of experiencing adverse events while taking clopidogrel.
While all statins carry a small risk of myopathy and elevations in creatinine kinase and transaminase levels, simvastatin still has a relatively favorable safety profile [
9], which is consistent with the low percentage of adverse events (0.9%) observed in this investigation (
Table 6). In contrast to both warfarin and clopidogrel, the race-only model for simvastatin failed to yield any associations with adverse events (
Figure 3). However, patients identifying as Black/African-American or of Other/Multiple race were significantly less likely to report an adverse event (
Table 6) compared to self-reported White individuals when considering all other covariates. To the best of our knowledge, this is the first study to report such an association. Given the relatively small number of adverse events, these results are somewhat surprising and likely benefited from the relatively large sample size (
n = 11,522). The hepatic uptake of simvastatin is mediated by the organic anionic transporter SLCO1B1 and associations between lower activity
SLCO1B1 allelic variants and simvastatin myopathy has been observed [
34]. Consequently, CPIC recommends either dose-reduction or switching to an alternative statin (i.e., pravastatin, rosuvastatin) for patients carrying decreased function
SLCO1B1 alleles [
10]. Interestingly, the African population has a lower frequency of
SLCO1B1 alleles conferring lower transporter activity [
16,
34]. For example, the
SLCO1B1*15 allele occurs in approximately 2.9% of Africans and 13% of Europeans and the
SLCO1B1*5 allele (2.8% of Europeans) has not been detected in African populations [
35]. Moreover, individuals of European ancestry have a greater area under the curve (AUC) and slower clearance for pravastatin than African-Americans [
36]. While we have no information on
SLCO1B1 genotype in our study population, it is possible that individuals identifying as Black/African-American in this study were less likely to carry decreased functioning
SLCO1B1 alleles relative to the “White” patients taking simvastatin. However, it is equally possible that environmental/nongenetic factors can account for this discrepancy in adverse events. For example, Black/African-American and/or Other/Multiple race patients could have been less likely to report adverse reactions or seek medical care while taking simvastatin. Indeed, racial disparities in adverse drug reactions have been observed in other studies [
5,
28]. As a result, these findings require further detailed investigation.
This study has several strengths including its diverse study population (>50% of patients identified as non-White) and the relatively large sample size of each drug group. However, there are several limitations to the study that warrant discussion. First and foremost, we have relied on EHRs for all of our analyses. These are inherently incomplete and oftentimes lack detailed information. We also utilized ICD-10 codes for our endpoints, and many covariates, which are frequently nonspecific and somewhat subjective. In addition, many studies have highlighted the limited usefulness of social constructs such as race and ethnicity data in retrospective studies. Indeed, the self-reported racial and ethnic information present in the EHR lacks granular information on specific racial and ethnic subgroups that would have strengthened this investigation and the validity of its conclusions. Along these same lines, we were unable to obtain genotypic information on the relevant pharmacogenes for each drug which would have allowed analysis to include genetic predisposition in addition to race/ethnicity. Similarly, we lacked DNA/data on ancestry informative markers as a more accurate index of self-reported race and ethnicity which would have more accurately characterized the ethnogeographic composition of our study population. Consequently, the broad categorization of race/ethnicity in the EHR is somewhat reflected in the large variability of probabilities observed in the race-only models. Taken together, any of the observed correlations between self-reported race/ethnicity and adverse treatment outcome may not be applicable to other populations in more rural areas and/or datasets that contain more detailed ethnoracial information.
The use of race and ethnicity in clinical decision-making and in U.S. FDA package inserts are controversial given that they do not adequately reflect subgroup populations and are confounded by social determinants of health and health inequities. This being said, both terms are often used as surrogate genetic markers of disease risk and therapeutic outcome. This work highlights the somewhat limited biological value of utilizing race and ethnicity as predictors of drug toxicity. As we have shown, apparent associations between adverse drug outcomes and race can often be explained by other clinical (i.e., co-morbidities) and environmental (i.e., drug interactions, tobacco usage) factors. In certain situations where the variability of pharmacogenes is geographically diverse, race/ethnicity may be informative variables to include along with other factors commonly used in therapeutic decision-making (i.e., liver/kidney function, BMI) for predicting ADRs. While interesting, the results of this study require further prospective analyses in large, diverse populations accompanied by biological samples that will help yield genetic factors that can potentially account for the findings observed in this investigation.