Mixed Results for Genotype-guided Warfarin Dosing

Update (12/17/2013):

New perspective piece from the New England Journal of Medicine

The December 12 issue of the New England Journal of Medicine  includes a perspective piece discussing the recent trial results on coumarin dosing. This article, Pharmacogenetics and Coumarin Dosing — Recalibrating Expectations, by Zineh et al from the Food and Drug Administration, discusses the different results from three randomized controlled trials published in the same issue of the journal and the implications for public expectations for pharmacogenetics testing. They highlight the importance of including “uncommon but clinically meaningful outcomes” “in addition to intermediate end points (e.g., percentage of time in the therapeutic range) in a totality-of-evidence approach to assessing the usefulness of pharmacogenetic approaches”. They also add that “Many observers have called for randomized, controlled trials to address the translation lag….Randomization, in and of itself, does not accomplish this end. Rather, the choice of control, the treatment setting, characteristics of the population tested, the analytic approach, and end-point definition are likely to be the key considerations that determine the public health relevance of pharmacogenetic trials in the future. Future trials should use various methods to assess the clinical usefulness of pharmacogenetic interventions; these may include designs focused on assessing efficacy (emphasis on internal validity), effectiveness (emphasis on generalizability), and implementation effectiveness (emphasis on adoption and uptake). These approaches are not mutually exclusive and, if combined, may expedite assessment of the effects of pharmacogenetic interventions on patients, providers, and health systems.”

Read the perspective:

Pharmacogenetics and Coumarin Dosing — Recalibrating Expectations

Issam Zineh, Pharm.D., M.P.H., Michael Pacanowski, Pharm.D., M.P.H., and Janet Woodcock, M.D.

From the Center for Drug Evaluation and Research, Food and Drug Administration, Silver Spring, MD.

N Engl J Med 2013; 369:2273-2275December 12, 2013DOI: 10.1056/NEJMp1314529

---------------------------------------------------------------------------------------------------------

Warfarin is a widely used blood thinning agent to prevent strokes, heart attacks, and dangerous blood clots. Though highly efficacious, warfarin use has been challenging due to its narrow therapeutic window and high degree of inter-individual variability. Overdose and underdose of warfarin are both dangerous. Taking too much warfarin could result in bleeding and taking too little may not be able to stop clotting. Many studies have attempted to explain the factors that influence warfarin response and define the optimal dosing algorithm. Clinical factors (eg. race, age, height, body weight, smoking, interacting drugs, comorbidities etc.) as well as genetic factors are all established determinants of variable warfarin response. In particular, genetic variations in two genes, CYP2C9 and VKORC1, have been repeatedly associated with warfarin dosing in various populations. The US FDA had revised the warfarin drug label twice (2007 and 2010) to indicate that CYP2C9 and VKORC1 genotypes may be useful in determining the optimal initial dose of warfarin and provided the recommended initial dosing ranges for patients with different combinations of CYP2C9 and VKORC1 genotypes. 

Despite the large body of literature documenting the significant association between CYP2C9/VKORC1 genotypes and warfarin dose, there is still debate surrounding the clinical utility of this knowledge. A few large, randomized clinical trials are currently underway to determine if using genetic information in warfarin dosing improves clinical outcomes (both efficacy and safety). The results of three studies have just been published online in the New England Journal of Medicine.  

•    Kimmel et al, 2013, COAG Trial: This trial included 1,015 patients (27% black) who were randomized at 18 centers in the U.S. to compare the efficacy of a warfarin-dosing algorithm based on genotype and clinical data with a dosing algorithm based on clinical data only. The authors showed that using a warfarin dosing algorithm based on both clinical and genetic information did NOT increase the percentage of time spent within the therapeutic range as compared to an algorithm based on clinical factors alone at 4 weeks (45.2% versus 45.4%, P=0.91), suggesting that genotype-guided dosing of warfarin did not improve anticoagulation control during the first 4 weeks of therapy. This study also found no difference in the rate of having an INR of 4 or higher, thromboembolism, or major bleeding between the groups (20% versus 21%, P=0.93).

•    Pirmohamed et al, 2013, EU-PACT Warfarin Trial: This trial included 455 patients (98.5% white) recruited from centers in the U.K. and Sweden to compare the effect of genotype-guided dosing with that of standard dosing on anticoagulation control. The authors found that genotype-guided group had higher mean percentage of time in the therapeutic range for the genotype-guided group as compared with the standard dosing group (67.4% versus 60.3%, P<0.001). There were also significantly fewer incidences of excessive anticoagulation (INR ≥4.0) in the genotype-guided group.

•    Verhoef et al, 2013, EU-PACT Acenocoumarol and Phenprocoumon Trial: This trial included 548 patients (>96% white) with atrial fibrillation or venous thromboembolism treated with acenocoumarol or phenprocoumon to compare the effect of a genotype-guided dosing algorithm with the effect of a dosing algorithm based solely on clinical factors (control). Similar to the study by Kimmel et al, this study also found that genotype-guided dosing did not increase the time spent within the therapeutic range through 12 weeks as compared to the control (61.6% versus 60.2%, P=0.52). 

All three studies are large, multi-center randomized trials and they all measure the same primary endpoint, the percentage of time that a patient is within the therapeutic range during the initial phase of treatment. Two studies (Kimmel et al, Verhoef et al) suggested no significant difference between the genotype-guided group vs. control in terms of the primary outcome, while one study (Pirmohamed et al) suggested  positive improvement (though modest) with added genetic information. The different findings from the three studies might partly be due to factors besides genetics that determine warfarin dose, eg. race, age, weight, smoking, concomitant drugs or comorbidities. Additionally, these studies examined if genotyping improves the initial time in therapeutic range, and yet, they were not powered to examine the effect on the secondary clinical outcome (eg. the rate of bleeding and thrombotic complications) and neither were they designed to address whether a longer duration of genotype-guided dosing would have improved INR control. We eagerly await publications of trials focusing on these aspects. The genetics-informatics trial of warfarin (GIFT Trial) is one of such trials that evaluates whether the addition of genotyping will reduce the risk of venous thromboembolism (VTE) and severe bleeding associated with warfarin management (PMID: 21606949). Even though the current clinical utility studies showed mixed results, the results of these trials are highly valuable. Given that preemptive genotyping is occurring on a more frequent basis, when genotypes are already available, it is in the best interest of the patient to use that information along with their clinical information to achieve a more optimum starting dose of therapy (PMID:19228618).

Read the articles:

A Pharmacogenetic versus a Clinical Algorithm for Warfarin Dosing

Stephen E. Kimmel, M.D., Benjamin French, Ph.D., Scott E. Kasner, M.D., Julie A. Johnson, Pharm.D., Jeffrey L. Anderson, M.D., Brian F. Gage, M.D., Yves D. Rosenberg, M.D., Charles S. Eby, M.D., Ph.D., Rosemary A. Madigan, R.N., M.P.H., Robert B. McBane, M.D., Sherif Z. Abdel-Rahman, Ph.D., Scott M. Stevens, M.D., Steven Yale, M.D., Emile R. Mohler, III, M.D., Margaret C. Fang, M.D., Vinay Shah, M.D., Richard B. Horenstein, M.D., Nita A. Limdi, Pharm.D., Ph.D., James A.S. Muldowney, III, M.D., Jaspal Gujral, M.B., B.S., Patrice Delafontaine, M.D., Robert J. Desnick, M.D., Ph.D., Thomas L. Ortel, M.D., Ph.D., Henny H. Billett, M.D., Robert C. Pendleton, M.D., Nancy L. Geller, Ph.D., Jonathan L. Halperin, M.D., Samuel Z. Goldhaber, M.D., Michael D. Caldwell, M.D., Ph.D., Robert M. Califf, M.D., and Jonas H. Ellenberg, Ph.D. for the COAG Investigators

New England Journal of Medicine November 19, 2013, DOI: 10.1056/NEJMoa1310669

A Randomized Trial of Genotype-Guided Dosing of Warfarin.

Munir Pirmohamed, Ph.D., F.R.C.P., Girvan Burnside, Ph.D., Niclas Eriksson, Ph.D., Andrea L. Jorgensen, Ph.D., Cheng Hock Toh, M.D., Toby Nicholson, F.R.C.Path., Patrick Kesteven, M.D., Christina Christersson, M.D., Ph.D., Bengt Wahlström, M.D., Christina Stafberg, M.D., J. Eunice Zhang, Ph.D., Julian B. Leathart, M.Phil., Hugo Kohnke, M.Sc., Anke H. Maitland-van der Zee, Pharm.D., Ph.D., Paula R. Williamson, Ph.D., Ann K. Daly, Ph.D., Peter Avery, Ph.D., Farhad Kamali, Ph.D., and Mia Wadelius, M.D., Ph.D. for the EU-PACT Group

New England Journal of Medicine November 19, 2013 DOI: 10.1056/NEJMoa1311386 .

A Randomized Trial of Genotype-Guided Dosing of Acenocoumarol and Phenprocoumon

Talitha I. Verhoef, M.Sc., Georgia Ragia, Ph.D., Anthonius de Boer, M.D., Ph.D., Rita Barallon, Ph.D., Genovefa Kolovou, M.D., Ph.D., Vana Kolovou, M.Sc., Stavros Konstantinides, M.D., Ph.D., Saskia Le Cessie, Ph.D., Efstratios Maltezos, M.D., Ph.D., Felix J.M. van der Meer, M.D., Ph.D., William K. Redekop, Ph.D., Mary Remkes, M.D., Frits R. Rosendaal, M.D., Ph.D., Rianne M.F. van Schie, Ph.D., Anna Tavridou, Ph.D., Dimitrios Tziakas, M.D., Ph.D., Mia Wadelius, M.D., Ph.D., Vangelis G. Manolopoulos, Ph.D., and Anke H. Maitland-van der Zee, Pharm.D., Ph.D. for the EU-PACT Group

New England Journal of Medicine November 19, 2013DOI: 10.1056/NEJMoa1311388

Also read the editorial:

Do Pharmacogenetics Have a Role in the Dosing of Vitamin K Antagonists?

Bruce Furie, M.D.

New England Journal of Medicine November 19, 2013DOI: 10.1056/NEJMe1313682

Controversy over the PGx of Cisplatin-Induced Hearing Loss in children

Hearing loss is a common side effect in cancer patients treated with cisplatin, with significant effects on quality of life. Age, dosage, cranial irradiation and concomitant vincristine use are all established risk factors for cisplatin ototoxicity. However, there is debate surrounding genetic variants associated with risk and the clinical utility of this knowledge. This controversy is highlighted in a new commentary available online in Clinical Pharmacology & Therapeutics.

Openly published history:
•    Ross et al, 2009: TPMT rs12201199 allele T (+ strand) and COMT rs9332377 allele T (+ strand) are significantly associated with an increased risk of ototoxicity (discovery and replication cohorts combined = 162 Canadian children in total, statistically significant after Bonferroni multiple testing correction). 
•    Pussegoda et al, 2013 (members of the same group): TPMT variants; rs12201199, rs1142345, rs1800460, COMT variants; rs4646316, rs9332377, and ABCC3 rs1051640 are significantly associated with risk of ototoxicity (previous cohort + replication cohort combined, total of 317 Canadian children, adjusted for age, vincristine treatment, germ-cell tumor and cranial irradiation). Of note, the COMT variants were not significantly associated in the replication cohort alone.
•    Yang et al, 2013 (same issue of CP&T): no association between the 3 TPMT variants and 2 COMT variants with risk of cisplatin ototoxicity (an independent cohort, total of 213 children enrolled in clinical trials at St Jude’s Hospital). No association with cisplatin cytotoxicity in vitro with human cell lines carrying the variants,  and no differences in hearing loss or cochlea hair cell damage between TPMT knockout and wild-type mice treated with cisplatin.

The controversy
Members of the group who published the positive genetic association findings submitted provisional patents for the analytical approach of looking for genetic variants highly predictive of cisplatin-induced hearing loss as early as 2006, but these have never been disclosed in their publications. After the Ross et al, 2009 publication, they were invited to speak at the FDA’s Drug Safety Oversight Board in 2010. In April 2011 members of the same group published an article regarding the economic benefits of testing for these genetic variants in children before cisplatin treatment. Later the same year, the FDA added information to the cisplatin drug label regarding an increased risk of hearing loss in children with certain genetic variants of TPMT.

Issues regarding data discrepancies and statistical methodology are also raised in the commentary, including significant differences in concomitant vincristine use in cases compared to controls not disclosed in the original analysis. High variation in allele frequencies between different populations exists for some of the SNPs examined, which authors of the commentary argue were not correctly addressed. In addition, p-values given in Pussegoda et al, 2013 for this original cohort are those before Bonferroni correction, whereas Ross et al, 2009 report only two statistically significant variants after correcting for multiple testing.

Transparency is key
The authors of this commentary appeal for openness and availability of data from these cohorts so they can be re-analyzed by others. They also argue for publishers to require disclosure of patents filed by authors that could be regarded as a conflict of interest. They request that the FDA revoke its amendments to the cisplatin drug label regarding TPMT genetic variants and ototoxicity. This tale also raises the issue of the difficulty in publishing negative results but the importance of this information in assessing evidence for associations.

In light of this controversy, the PharmGKB clinical annotations for TPMT and COMT variants & cisplatin in children are assigned a level of evidence of 3, as they lack clear evidence:
 

TPMT	rs1142345
TPMT	rs1800460
TPMT	rs12201199
COMT	rs4646316
COMT	rs9332377

Read the commentary:
Challenges in Interpreting the Evidence for Genetic Predictors of Ototoxicity.
Mark J. Ratain, Nancy J. Cox, Tara O. Henderson.
Clinical Pharmacology & Therapeutics. 2013 Dec;94(6):631-5. doi: 10.1038/clpt.2013.178.

Also read:
Genetics of cisplatin ototoxicity: confirming the unexplained?
Boddy, A.V.
Clinical Pharmacology & Therapeutics. 2013 Aug;94(2):198-200.

Other genetic variants that have been reported with an association with risk of cisplatin-induced toxicity:
•    GSTP1       rs1695
•    LRP2         rs2075252
•    XPC           rs2228001
•    SLC31A1  rs10981694

PharmGKB Drug Labels: Description, Update and New Features

PharmGKB has recently added new information to drug label entries, standardized the annotations, and updated the website display.  The overview page for drug labels contains high level information about each label at a glance, including:

• Whether the drug label is currently on the FDA's Table of Pharmacogenomic Biomarkers in Drug Labels.  Drugs are occasionally added to and removed from this list, so it changes over time.
• The level of pharmacogenomic information in the label.  Does the label require or recommend genetic testing?  Or does the label merely mention genes involved in the metabolism of the drug without discussing gene variation?
• The date of the last major update to the label annotation on PharmGKB.

The user has the ability to sort by level of pharmacogenomic information or restrict the list of labels to those on the FDA's Biomarker list.

Drug label curation can be subjective, and some of the issues are discussed in a previous blog.  Below, we describe how the PharmGKB drug labels are sourced and curated, the updates we have just released, and new features of these labels.

FDA-approved drug labels

• Source:
•  The most recent FDA-approved drug labels for drugs found in the FDA's Table of Pharmacogenomic Biomarkers in Drug Labels are retrieved through the link to Drugs@FDA or alternatively from DailyMed. 
• Some labels are of FDA-approved drugs that are not within the PGx Biomarker table, however we have found PGx related information in them. These are sourced from DailyMed.
• Curation process:
• PharmGKB Curators download the PDF for the most recently approved label and highlight PGx-related information within this label. The PDF, along with a summary explaining the PGx and excerpts from the label, are added to the drug page.
• Labels are tagged with the Biomarker icon if they are found in the FDA's PGx Biomarker table.     
• Each label is now assigned a PGx Level (see definitions): 

• Phenotypes/diseases that have been found in the label by the PHONT (PGRN Ontology Network Resource) group are provided at the bottom section of PharmGKB's drug label page.  The phenotype terms from the drug label have been standardized to the MeSH ontology, and link to PharmGKB phenotype web pages.  The section(s) of the label in which each phenotype term was found is provided.
• Genes and/or phenotypes found in the label by manual curation may also be added by PharmGKB Curators.  This may include many genes not found in the Biomarker table that are discussed in the label.  The section(s) of the label in which each gene/phenotype term was found is provided.  In most cases, PharmGKB curators specify the relationship of the gene to the drug using these terms:  toxicity, dosage, efficacy, metabolism/PK.
• The date a label was last updated is provided on the drug label page. All FDA drug labels were updated on 25th October 2013.  

 EMA-approved drug labels

• Source:
• There is no PGx Biomarker table for European drug labels. We are working with the EMA to establish a list of European Public Assessment Reports (EPARs) that contain PGx information. 
• We are constructing this list by initially searching for drugs in the FDA PGx Biomarker table - of these 44 EMA EPARs were identified.  
• Curation process:
• PharmGKB Curators download the EPAR Product Information PDF and highlight PGx-related information within this label. The PDF, along with a summary explaining the PGx and excerpts from the label, are added to the drug page. 
• A PGx level is added to each EMA label. 

You can view all PharmGKB drug labels and Filter by PGx level or if they are found in the FDA's Biomarker table.  

PG&G publication: Mycophenolic acid Pathway

Mycophenolic acid (MPA) is an immunosuppressive agent used to prevent rejection in organ transplantation.

We recently have published this pathway summary in the Pharmacogenetics and Genomics  Journal.

The article and included figure describe the metabolism of MPA and its prodrug mycophenolate mofetil. MPA is a noncompetitive, selective and reversible inhibitor of inosine monophosphate dehydrogenase (IMPDH). The pharmacodynamic section shows several mechanisms of actions.

Interindividual variability in the MPA exposure levels within and among populations has been observed. The article summarizes studies reporting the clinical association of genetic variants in PK and PD genes with MPA exposure and risk of rejection or GI toxicity (one of the major toxicities associated with MPA).

Find out more...

View the Mycophenolic acid pathway on PharmGKB.

Read the publication:

PharmGKB summary: mycophenolic acid pathway.

Lamba V, Sangkuhl K, Sanghavi K, Fish A, Altman RB, Klein TE.

Pharmacogenet Genomics. 2013 Nov 11.

PMID: 24220207

View all pathways on PharmGKB.

FDA-Approved Drug Labels, Pharmacogenomics and PharmGKB Evaluation

FDA-approved drug label standardization and pharmacogenomics
People often use the wording "FDA drug labels" instead of "FDA-approved drug labels."  To be clear, the FDA does not write drug labels; it is the responsibility of the drug company to provide them.  The FDA reviews and approves labels, and has implemented rules to standardize them.  Since 2005, submitted labels are required to be in Structured Product Labeling (SPL) format.  Starting in 2006, label highlights are required to have bulleted boxed warnings, and include indications and a Table of Contents.  These requirements greatly improve the organization and readability of labels for humans, but stop short of standardization enabling automated parsing and interpretation of labels.  It remains difficult to use techniques like Natural Language Parsing (NLP) to automatically extract information from drug labels.  And even with the current rules, sometimes vital information remains missing from approved drug labels, such as data regarding toxicity and uncertainty regarding the net benefit,  and comparative effectiveness.

The language used in drug labels can be vague and often stops short of requiring action.  For example, the azathioprine label states "It is recommended that consideration be given to either genotype or phenotype patients for TPMT."  Do they recommend testing?  They don't mention testing anywhere on the label, only that consideration should be given to the genotype or phenotype.  With a new patient, how would you have this information without testing?

The FDA maintains a list of labels containing pharmacogenomic information, called the Table of Pharmacogenomic Biomarkers in Drug Labels.  It is important to note that this table is (1) not a complete list of labels that reference genes, nor does it reflect (2) all genes of importance on labels listed or (3) indicate that the genes that are listed in the table are important for prescribing the drug.  Here is an example of each case (as of 10/20/13):

(1)  Many drug labels that discuss G6PD deficiency are not listed in the table (eg. glibenclamide, pegloticase, primaquine - this also mentions CYPB5R3 deficiency on the label!), but the chloroquine label mentioning G6PD deficiency is listed.  Why chloroquine is listed but not other drugs warning against G6PD deficient patients is unclear.

(2) The valproic acid (Depakene) label is listed in the table with NAGS, CPS1, ASS1, OTC, ASL, ABL2.  These are genes related to urea cycle disorder which is a contraindication of the drug, though the genes are not listed by name on the label.   However the label does specifically discuss POLG gene mutations that predispose patients to increased risk of liver failure and death.  POLG is a mitochondrial DNA polymerase associated with hereditary neurometabolic syndromes such as Alpers Huttenlocher Syndrome, and the valproic acid label states "POLG mutation testing should be performed in accordance with current clinical practice for the diagnostic evaluation of such disorders."  POLG is not listed in the Biomarker table with this or any other drug.

(3) The prasugrel (Effient) label is listed in the table with CYP2C19 (no other genes).   The label states "There is no relevant effect of genetic variation in CYP2B6, CYP2C9, CYP2C19, or CYP3A5 on the pharmacokinetics of prasugrel's active metabolite or its inhibition of platelet aggregation." The drug-gene pair is likely in the table because CYP2C19 poor metabolizers may be poor responders to clopidogrel, another anti-platelet drug.  For these cases, prasugrel is a typical alternative because CYP2C19 variation does not affect prasugrel use in any way.   This is an example where a drug-gene pair on the Biomarker table does not indicated a relevant PGx relationship between the two.

PharmGKB
When PharmGKB curators curate drug labels, they need to interpret them, and that can be subjective.   PharmGKB has recently reviewed the drug labels from the FDA's Biomarker Table and improved the label annotations.  Curators have tried to define clearly how they interpret the level of PGx information on labels, resulting in definitions that are really a rather long list of criteria.  These definitions may change over time as new label wording comes up and definitions need to adjust for these new cases.