PERSPECTIVES of pharmacogenetic associations and may accelerate the implementation of pharmacogenetic-based drug dosing.1 This work is supported by the NIH/NIGMS (R24 GM61374; TE Klein and RB Altman, PIs). We would like to acknowledge the work of others in this area, particularly the Pharmacogenomics Research Network (PGRN) Translational Pharmacogenetics Project, CDC Nomenclature Group and the ClinGen PGx Working Group.
Meaningful Use and Clinical Utility of Preemptive Pharmacogenetic Testing: (Re)View From a CYP2D6 Poor Metabolizer
CONFLICT OF INTEREST RBA and TEK are consultants to Personalis, Inc.
J. Steven Leeder1
ACKNOWLEDGMENTS
C 2014 ASCPT V
1.
Whirl-Carrillo, M. et al. Pharmacogenomics knowledge for personalized medicine. Clin. Pharmacol. Ther. 92, 414–417 (2012). 2. Gray, K.A. et al. Genenames.org: the HGNC resources in 2013. Nucleic Acids Res. 41, D545–D552 (2013). 3. Human Genome Variation Society (HGVS). . Accessed on June 10, 2014. 4. den Dunnen, J.T. & Antonarakis, S.E. Mutation nomenclature extensions and suggestions to describe complex mutations: a discussion. Hum. Mutat. 15, 7–12 (2000). 5. Links to human pharmacogene variant nomenclature committees or resources for genetic variant names can be found here: . Accessed on 10 July 2014. 6. Bustin, S.A. et al. The need for transparency and good practices in the qPCR literature. Nat. Methods 10, 1063–1067 (2013). 7. NCBI Short Genetic Variations (SNV) database, also known as dbSNP. . Accessed on June 10, 2014. 8. National Center for Biotechnology Information (NCBI), . Accessed on June 10, 2014. 9. Wildeman, M., van Ophuizen, E., den Dunnen, J.T. & Taschner, P.E. Improving sequence variant descriptions in mutation databases and literature using the Mutalyzer sequence variation nomenclature checker. Hum. Mutat. 29, 6–13 (2008). 10. International HapMap, C. The International HapMap Project. Nature 426, 789–796 (2003).
The value of pharmacogenetic information at an individual level seems self-evident; demonstrating the value to the individual across a population appears to be much more difficult to demonstrate. Future studies should avoid single gene–drug pairs and a limited number of population-specific alleles that are insufficient to capture all relevant genetic variation within diverse populations, such as CYP2C9*2 and *3. Robust associations between cytochrome P450 (CYP) genotypes and treatment response require knowledge of drug exposure.
The Point/Counterpoint1,2 and accompanying commentary by Gillis and Innocenti3 on the value of preemptive pharmacogenetic testing raise several additional issues that warrant further discussion. Disclosure: I am a CYP2D6 poor metabolizer, and my genotype is CYP2D6*4/*6. As an individual, I don’t care if preemptive genotyping has been demonstrated to have clinical utility at the population level, because if I am prescribed codeine it likely will not provide me with any pain relief, and if I am prescribed a “normal” dose of fluoxetine there is an increased probability that I will experience concentration-dependent toxicity. I am convinced that knowledge of my pharmacogenome has value to me personally, but I don’t have any data to change the position of anyone who considers such knowledge to be useless until suitable evidence is provided. Occasionally I wonder if opponents of preemptive pharmacogenetic testing
would want to know their CYP2C9 and VKORCI genotypes before they are prescribed warfarin? The limitations in extrapolating the results of genome-wide association studies to individuals are well recognized; extrapolating the results of population-based pharmacogenomic clinical utility studies to individual patients may suffer from similar limitations. There are several issues to be considered before the potential of pharmacogenomics and precision medicine can be realized. Issue #1: Developing the value proposition
One message from Janssens and Deverka1 is that the pharmacogenomic community has not adequately developed the value proposition for preemptive pharmacogenetic testing. As Gillis and Innocenti point out,3 the lack of a consensus definition for clinical utility precludes setting standards
1
Marion Merrell Dow/Missouri Endowed Chair in Pediatric Clinical Pharmacology, Director, Division of Clinical Pharmacology, Toxicology and Therapeutic Innovation, Children’s Mercy Kansas City, Professor of Pediatrics and Pharmacology, Schools of Medicine and Pharmacy, University of Missouri-Kansas City, Missouri, USA. Correspondence: J. Steven Leeder ([email protected]) doi:10.1002/cpt.14
CLINICAL PHARMACOLOGY & THERAPEUTICS | VOLUME 97 NUMBER 2 | FEBRUARY 2015
119
PERSPECTIVES for demonstrating the value of an intervention, such as preemptive pharmacogenetic testing. The NIH definition: “whether the test can provide information about diagnosis, treatment, management, or prevention that will be helpful to the consumer” implies that value to the individual is a key element of clinical utility. In the simplest sense, a preemptive pharmacogenetic test that is useless, i.e., has no value, is one for which the benefit to the individual is offset by an equivalent amount of harm, or the number of individuals in a population who benefit equals the number who are harmed such that the net value is zero. It is possible that not all individuals in a population will benefit equally from preemptive genotyping. For example, it could be argued that for a CYP2D6 substrate those who benefit most are those whose genotypes place them in the phenotypic extremes of the population—the PMs and the ultra-rapid metabolizers (UMs)—as they represent the subgroups of patients for whom approved dosage recommendations are least appropriate. Unfortunately, few of the 150 FDA-approved products with pharmacogenetic information in their labels actually provide specific dosing recommendations stratified by genotype or phenotype to provide greater value than “may require a reduction in dose.” Similarly, is there truly no value or even harm associated with knowing that a patient has a CYP2D6*1/*1 genotype if no action is necessary (e.g., no alteration in dose) at the time the test result is returned? The answer to this rhetorical question will depend on the particular clinical context at the time the test is being considered. Determining the value of genotype information for genes involved in drug disposition is made more difficult by the fact that there may still be considerable interindividual variability in drug clearance in individuals with nominally identical genotypes. And perhaps this is where some of the problems arise. Janssens and Deverka argue that “the responsible use of PGx testing requires evidence of a robust association between the genotype and treatment response,” but in reality the further away a gene is from drug response, the weaker the association. For example, the proximal phenotype for CYP2C19 technically is the disappearance of the substrate (clopidogrel) 120
and formation of the active plateletinhibiting metabolite; this genotype– phenotype association is several steps removed from the cardiovascular events to be reduced by preemptive genotyping. Is it necessary to demonstrate a robust association between CYP2C19 genotype and a reduction in intervention-related cardiovascular events at a population level to establish sufficient clinical utility for widespread acceptance? Is it possible for individuals to benefit from pharmacogenetic testing without that benefit being evident on a population basis? The randomized controlled trial (RCT) may be the gold standard for comparing the relative effects of two interventions at a population level, but is the outcome of an RCT necessarily applicable to a given individual? The challenge for the pharmacogenomics community is to reassess and better articulate the value proposition posed by specific preemptive testing applications, and to develop appropriate methods to assess the value in a meaningful way. Issue #2: Preemptive pharmacogenetic testing for gene-drug pairs or comprehensive “pharmacogenome” testing?
Both Janssens and Deverka1 and Ratain and Johnson2 make sound arguments in their respective Point/Counterpoint presentations, but are addressing separate issues. Janssens and Deverka may be quite justified in requiring a high level of evidence to establish the clinical utility of specific gene–drug pairs, such as CYP2C19-clopidogrel, especially if pharmacogenetic testing involves a limited number of common variants. A test that only interrogates the most commonly occurring variants in a population of European descent is unlikely to demonstrate clinical utility in a more diverse population representative of a clinical practice setting if: 1) those variants occur rarely, if at all, in non-European population, and 2) other population-specific alleles are functionally relevant. Testing for the CYP2C9*2 and *3 alleles that occur infrequently in African-derived populations but not including the more commonly occurring CYP2C9*5, *6, *8, and *11 alleles can result in misclassification of African Americans as CYP2C9*1/*1 when, in fact, they carry one or more reduced function alleles; this possibility has been offered as a
possible explanation4 of the overanticoagulation of African American participants observed in the US-COAG trial.5 An important question for future consideration centers around how comprehensive the set of variants included in a panel must be to adequately examine clinical utility in a population representative of the patients who would benefit from the test. Ratain and Johnson2 make a distinction between the value of ordering a genetic test and the value of pharmacogenetic test results if they are already available. Genetic testing needs only to be conducted on one occasion, and use of a comprehensive, preemptive pharmacogenetic panel offers a cost-effective means to generate a valuable set of patientspecific information that exceeds the more limited value of a specific gene– drug test. They argue that embedding the results of panel-based preemptive genotyping in the electronic health record, paired with a decision-support system to guide prescribing, represents an optimal approach for clinical implementation. Measuring the impact of clinical implementation in this manner may not be amenable to a randomized-controlled study design, and the emphasis of future research should be to increase the amount and quality of information available to guide therapy. Issue #3. Are the conflicting results of clinical trials evaluating pharmacogenetic testing for clopidogrel, tamoxifen, and warfarin really failures to demonstrate clinical utility, or are they sending us a signal that we need to think about the problem differently?
The results of studies focusing on specific gene–drug pairs demonstrate the potential limitations associated with testing for a limited number of (population-specific) genetic variants. Furthermore, the conflicting results of the COAG5 and EU-PACT7 warfarin studies also illustrate how sensitive the assessment of clinical utility may be to factors such as differences in the design of the studies (e.g., specific dosing algorithms and use of a loading dose8), and the hypotheses they are designed to test.6 However, a more insidious limitation of study designs may be the conceptual gap between
VOLUME 97 NUMBER 2 | FEBRUARY 2015 | www.wileyonlinelibrary/cpt
PERSPECTIVES what is tested and the intended benefit of that testing. To date, pharmacogenetic investigations have targeted genes involved in drug disposition (drug metabolizing enzymes and transporters) with the intent of individualizing doses to achieve a desired therapeutic response. This approach ignores the intermediary role of drug exposure as a more proximal and therefore more important determinant of drug response than is dose. This fact is acknowledged by the EU-PACT investigators,8 but the reality is that drug concentrations generally are not measured in pharmacogenetic studies. Given that multiple pathways are involved in the biotransformation of almost every drug subject to hepatic metabolism, the concentration of therapeutically active drug or bioactivated metabolite can demonstrate considerable interindividual variability among patients with apparently identical genotypes. Even when doses are adjusted based on genotype, observed differences in response may be clarified through assessment of drug exposure, potentially identifying nonadherence or other confounding factors as additional sources of variability. Assessing drug exposure in a pharmacogenetic study represents
an opportunity to assess participants as individuals, rather than as a population. This is the essence of the arguments concerning the value of preemptive genetic testing, as genetic information is one factor that helps to define individuals. Knowing the relationship between the dose of drug, the resulting drug exposure, and finally the response to that exposure for a given individual aids in differentiating patients who do not respond to a drug simply because insufficient drug is present from those who are unlikely to respond to the medication due to genetic variation in the drug target, regardless of how much drug is present. Therefore, components of a pharmacogenetic test should include genes contributing to variability in response (e.g., drug target genes) as well as genes contributing to variability in the dose–exposure relationship (e.g., CYPs and transporters). Studies conducted by pharmaceutical companies determine the dose that, on average, produces the desired response in a treated population; that dose is not necessarily the best dose for every individual patient. In developing pharmacogenetic tests to identify the best dose to provide the desired response in an individual
CLINICAL PHARMACOLOGY & THERAPEUTICS | VOLUME 97 NUMBER 2 | FEBRUARY 2015
patient, the challenge is to develop strategies that are different from the one that created the problem in the first place. CONFLICT OF INTEREST No conflict of interest. C 2014 ASCPT V
1. Janssens, A.C.J.W. & Deverka, P.A. Useless until proven effective — the clinical utility of preemptive pharmacogenetic testing. Clin. Pharmacol. Ther. (in review). 2. Ratain, M.J. & Johnson, J.A. Meaningful use of pharmacogenetics. Clin. Pharmacol. Ther. (in review). 3. Gillis, N.K. & Innocenti, F. Evidence required to demonstrate clinical utility of pharmacogenetic testing: the debate continues. Clin. Pharmacol. Ther. (in review). 4. Cavallari, L.H., Kittles, R.A. & Perera, M.A. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1763 (2014). 5. Kimmel, S.E. et al. A pharmacogenetic versus a clinical algorithm for warfarin dosing. N. Engl. J. Med. 369, 2283–2293 (2013). 6. Kimmel, S.E., French, B. & Geller, N.L. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1763–1764 (2014). 7. Pirmohamed, M. et al. A randomized trial of genotype-guided dosing of warfarin. N. Engl. J. Med. 369, 2294–2303 (2013). 8. Pirmohamed, M., Waldelius, M. & Kamali, F. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1764– 1765 (2014).
121
Meaningful Use and Clinical Utility of Preemptive Pharmacogenetic Testing: (Re)View From a CYP2D6 Poor Metabolizer
CONFLICT OF INTEREST RBA and TEK are consultants to Personalis, Inc.
J. Steven Leeder1
ACKNOWLEDGMENTS
C 2014 ASCPT V
1.
Whirl-Carrillo, M. et al. Pharmacogenomics knowledge for personalized medicine. Clin. Pharmacol. Ther. 92, 414–417 (2012). 2. Gray, K.A. et al. Genenames.org: the HGNC resources in 2013. Nucleic Acids Res. 41, D545–D552 (2013). 3. Human Genome Variation Society (HGVS). . Accessed on June 10, 2014. 4. den Dunnen, J.T. & Antonarakis, S.E. Mutation nomenclature extensions and suggestions to describe complex mutations: a discussion. Hum. Mutat. 15, 7–12 (2000). 5. Links to human pharmacogene variant nomenclature committees or resources for genetic variant names can be found here: . Accessed on 10 July 2014. 6. Bustin, S.A. et al. The need for transparency and good practices in the qPCR literature. Nat. Methods 10, 1063–1067 (2013). 7. NCBI Short Genetic Variations (SNV) database, also known as dbSNP. . Accessed on June 10, 2014. 8. National Center for Biotechnology Information (NCBI), . Accessed on June 10, 2014. 9. Wildeman, M., van Ophuizen, E., den Dunnen, J.T. & Taschner, P.E. Improving sequence variant descriptions in mutation databases and literature using the Mutalyzer sequence variation nomenclature checker. Hum. Mutat. 29, 6–13 (2008). 10. International HapMap, C. The International HapMap Project. Nature 426, 789–796 (2003).
The value of pharmacogenetic information at an individual level seems self-evident; demonstrating the value to the individual across a population appears to be much more difficult to demonstrate. Future studies should avoid single gene–drug pairs and a limited number of population-specific alleles that are insufficient to capture all relevant genetic variation within diverse populations, such as CYP2C9*2 and *3. Robust associations between cytochrome P450 (CYP) genotypes and treatment response require knowledge of drug exposure.
The Point/Counterpoint1,2 and accompanying commentary by Gillis and Innocenti3 on the value of preemptive pharmacogenetic testing raise several additional issues that warrant further discussion. Disclosure: I am a CYP2D6 poor metabolizer, and my genotype is CYP2D6*4/*6. As an individual, I don’t care if preemptive genotyping has been demonstrated to have clinical utility at the population level, because if I am prescribed codeine it likely will not provide me with any pain relief, and if I am prescribed a “normal” dose of fluoxetine there is an increased probability that I will experience concentration-dependent toxicity. I am convinced that knowledge of my pharmacogenome has value to me personally, but I don’t have any data to change the position of anyone who considers such knowledge to be useless until suitable evidence is provided. Occasionally I wonder if opponents of preemptive pharmacogenetic testing
would want to know their CYP2C9 and VKORCI genotypes before they are prescribed warfarin? The limitations in extrapolating the results of genome-wide association studies to individuals are well recognized; extrapolating the results of population-based pharmacogenomic clinical utility studies to individual patients may suffer from similar limitations. There are several issues to be considered before the potential of pharmacogenomics and precision medicine can be realized. Issue #1: Developing the value proposition
One message from Janssens and Deverka1 is that the pharmacogenomic community has not adequately developed the value proposition for preemptive pharmacogenetic testing. As Gillis and Innocenti point out,3 the lack of a consensus definition for clinical utility precludes setting standards
1
Marion Merrell Dow/Missouri Endowed Chair in Pediatric Clinical Pharmacology, Director, Division of Clinical Pharmacology, Toxicology and Therapeutic Innovation, Children’s Mercy Kansas City, Professor of Pediatrics and Pharmacology, Schools of Medicine and Pharmacy, University of Missouri-Kansas City, Missouri, USA. Correspondence: J. Steven Leeder ([email protected]) doi:10.1002/cpt.14
CLINICAL PHARMACOLOGY & THERAPEUTICS | VOLUME 97 NUMBER 2 | FEBRUARY 2015
119
PERSPECTIVES for demonstrating the value of an intervention, such as preemptive pharmacogenetic testing. The NIH definition: “whether the test can provide information about diagnosis, treatment, management, or prevention that will be helpful to the consumer” implies that value to the individual is a key element of clinical utility. In the simplest sense, a preemptive pharmacogenetic test that is useless, i.e., has no value, is one for which the benefit to the individual is offset by an equivalent amount of harm, or the number of individuals in a population who benefit equals the number who are harmed such that the net value is zero. It is possible that not all individuals in a population will benefit equally from preemptive genotyping. For example, it could be argued that for a CYP2D6 substrate those who benefit most are those whose genotypes place them in the phenotypic extremes of the population—the PMs and the ultra-rapid metabolizers (UMs)—as they represent the subgroups of patients for whom approved dosage recommendations are least appropriate. Unfortunately, few of the 150 FDA-approved products with pharmacogenetic information in their labels actually provide specific dosing recommendations stratified by genotype or phenotype to provide greater value than “may require a reduction in dose.” Similarly, is there truly no value or even harm associated with knowing that a patient has a CYP2D6*1/*1 genotype if no action is necessary (e.g., no alteration in dose) at the time the test result is returned? The answer to this rhetorical question will depend on the particular clinical context at the time the test is being considered. Determining the value of genotype information for genes involved in drug disposition is made more difficult by the fact that there may still be considerable interindividual variability in drug clearance in individuals with nominally identical genotypes. And perhaps this is where some of the problems arise. Janssens and Deverka argue that “the responsible use of PGx testing requires evidence of a robust association between the genotype and treatment response,” but in reality the further away a gene is from drug response, the weaker the association. For example, the proximal phenotype for CYP2C19 technically is the disappearance of the substrate (clopidogrel) 120
and formation of the active plateletinhibiting metabolite; this genotype– phenotype association is several steps removed from the cardiovascular events to be reduced by preemptive genotyping. Is it necessary to demonstrate a robust association between CYP2C19 genotype and a reduction in intervention-related cardiovascular events at a population level to establish sufficient clinical utility for widespread acceptance? Is it possible for individuals to benefit from pharmacogenetic testing without that benefit being evident on a population basis? The randomized controlled trial (RCT) may be the gold standard for comparing the relative effects of two interventions at a population level, but is the outcome of an RCT necessarily applicable to a given individual? The challenge for the pharmacogenomics community is to reassess and better articulate the value proposition posed by specific preemptive testing applications, and to develop appropriate methods to assess the value in a meaningful way. Issue #2: Preemptive pharmacogenetic testing for gene-drug pairs or comprehensive “pharmacogenome” testing?
Both Janssens and Deverka1 and Ratain and Johnson2 make sound arguments in their respective Point/Counterpoint presentations, but are addressing separate issues. Janssens and Deverka may be quite justified in requiring a high level of evidence to establish the clinical utility of specific gene–drug pairs, such as CYP2C19-clopidogrel, especially if pharmacogenetic testing involves a limited number of common variants. A test that only interrogates the most commonly occurring variants in a population of European descent is unlikely to demonstrate clinical utility in a more diverse population representative of a clinical practice setting if: 1) those variants occur rarely, if at all, in non-European population, and 2) other population-specific alleles are functionally relevant. Testing for the CYP2C9*2 and *3 alleles that occur infrequently in African-derived populations but not including the more commonly occurring CYP2C9*5, *6, *8, and *11 alleles can result in misclassification of African Americans as CYP2C9*1/*1 when, in fact, they carry one or more reduced function alleles; this possibility has been offered as a
possible explanation4 of the overanticoagulation of African American participants observed in the US-COAG trial.5 An important question for future consideration centers around how comprehensive the set of variants included in a panel must be to adequately examine clinical utility in a population representative of the patients who would benefit from the test. Ratain and Johnson2 make a distinction between the value of ordering a genetic test and the value of pharmacogenetic test results if they are already available. Genetic testing needs only to be conducted on one occasion, and use of a comprehensive, preemptive pharmacogenetic panel offers a cost-effective means to generate a valuable set of patientspecific information that exceeds the more limited value of a specific gene– drug test. They argue that embedding the results of panel-based preemptive genotyping in the electronic health record, paired with a decision-support system to guide prescribing, represents an optimal approach for clinical implementation. Measuring the impact of clinical implementation in this manner may not be amenable to a randomized-controlled study design, and the emphasis of future research should be to increase the amount and quality of information available to guide therapy. Issue #3. Are the conflicting results of clinical trials evaluating pharmacogenetic testing for clopidogrel, tamoxifen, and warfarin really failures to demonstrate clinical utility, or are they sending us a signal that we need to think about the problem differently?
The results of studies focusing on specific gene–drug pairs demonstrate the potential limitations associated with testing for a limited number of (population-specific) genetic variants. Furthermore, the conflicting results of the COAG5 and EU-PACT7 warfarin studies also illustrate how sensitive the assessment of clinical utility may be to factors such as differences in the design of the studies (e.g., specific dosing algorithms and use of a loading dose8), and the hypotheses they are designed to test.6 However, a more insidious limitation of study designs may be the conceptual gap between
VOLUME 97 NUMBER 2 | FEBRUARY 2015 | www.wileyonlinelibrary/cpt
PERSPECTIVES what is tested and the intended benefit of that testing. To date, pharmacogenetic investigations have targeted genes involved in drug disposition (drug metabolizing enzymes and transporters) with the intent of individualizing doses to achieve a desired therapeutic response. This approach ignores the intermediary role of drug exposure as a more proximal and therefore more important determinant of drug response than is dose. This fact is acknowledged by the EU-PACT investigators,8 but the reality is that drug concentrations generally are not measured in pharmacogenetic studies. Given that multiple pathways are involved in the biotransformation of almost every drug subject to hepatic metabolism, the concentration of therapeutically active drug or bioactivated metabolite can demonstrate considerable interindividual variability among patients with apparently identical genotypes. Even when doses are adjusted based on genotype, observed differences in response may be clarified through assessment of drug exposure, potentially identifying nonadherence or other confounding factors as additional sources of variability. Assessing drug exposure in a pharmacogenetic study represents
an opportunity to assess participants as individuals, rather than as a population. This is the essence of the arguments concerning the value of preemptive genetic testing, as genetic information is one factor that helps to define individuals. Knowing the relationship between the dose of drug, the resulting drug exposure, and finally the response to that exposure for a given individual aids in differentiating patients who do not respond to a drug simply because insufficient drug is present from those who are unlikely to respond to the medication due to genetic variation in the drug target, regardless of how much drug is present. Therefore, components of a pharmacogenetic test should include genes contributing to variability in response (e.g., drug target genes) as well as genes contributing to variability in the dose–exposure relationship (e.g., CYPs and transporters). Studies conducted by pharmaceutical companies determine the dose that, on average, produces the desired response in a treated population; that dose is not necessarily the best dose for every individual patient. In developing pharmacogenetic tests to identify the best dose to provide the desired response in an individual
CLINICAL PHARMACOLOGY & THERAPEUTICS | VOLUME 97 NUMBER 2 | FEBRUARY 2015
patient, the challenge is to develop strategies that are different from the one that created the problem in the first place. CONFLICT OF INTEREST No conflict of interest. C 2014 ASCPT V
1. Janssens, A.C.J.W. & Deverka, P.A. Useless until proven effective — the clinical utility of preemptive pharmacogenetic testing. Clin. Pharmacol. Ther. (in review). 2. Ratain, M.J. & Johnson, J.A. Meaningful use of pharmacogenetics. Clin. Pharmacol. Ther. (in review). 3. Gillis, N.K. & Innocenti, F. Evidence required to demonstrate clinical utility of pharmacogenetic testing: the debate continues. Clin. Pharmacol. Ther. (in review). 4. Cavallari, L.H., Kittles, R.A. & Perera, M.A. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1763 (2014). 5. Kimmel, S.E. et al. A pharmacogenetic versus a clinical algorithm for warfarin dosing. N. Engl. J. Med. 369, 2283–2293 (2013). 6. Kimmel, S.E., French, B. & Geller, N.L. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1763–1764 (2014). 7. Pirmohamed, M. et al. A randomized trial of genotype-guided dosing of warfarin. N. Engl. J. Med. 369, 2294–2303 (2013). 8. Pirmohamed, M., Waldelius, M. & Kamali, F. Genotype-guided dosing of vitamin K antagonists [letter]. N. Engl. J. Med. 370, 1764– 1765 (2014).
121
