Skip to main content
Premium Trial:

Request an Annual Quote

Two Conflicting Prospective, RCTs on Warfarin PGx Provide No Definitive Guidance to Physicians


Originally published Nov. 22.

Doctors have been in a "should we or shouldn't we" conundrum with warfarin pharmacogenetic testing for years.

The US Food and Drug Administration has updated the label for the widely used anticoagulant twice, in 2007 and 2010, informing doctors that patients with certain variants of CYP2C9 and VKORC1 genes may require non-standard warfarin doses when they start therapy. Meanwhile, physician practice guidelines don't recommend genotyping patients ahead of prescribing them warfarin.

As such, the majority of healthcare providers in primary healthcare settings have held off on adopting genotyping strategies to inform warfarin dosing, since there had been no data from gold-standard prospective randomized-controlled trials showing that this additional intervention would improve patient outcomes — such as lowering bleeding risk and reducing hospital stays — compared to standard dosing techniques using the patient's medical data.

This week, independent research groups published findings in the New England Journal of Medicine from two randomized controlled trials comparing genotype-guided warfarin dosing against dosing without genetic testing, but the two trials came to different conclusions. One study, conducted by the EU-PACT group and involving 455 patients, compared a cohort given warfarin with the help of a point-of-care genetic test and a PGx-based algorithm with a cohort given standard dosing. In this study, researchers led by University of Liverpool's Munir Pirmohamed found that in the first 12 weeks of warfarin therapy, patients in the PGx-guided group stayed longer within therapeutic range — the dose range required for desired treatment effect — compared to those in the control arm (67.4 percent versus 60.3 percent).

Another study, called Clarification of Optimal Anticoagulation through Genetics, or COAG, randomized approximately 1,000 patients to receive either a wafarin dose as determined by an algorithm that factored in clinical variables and patients' genotypes or dosing only by a clinical algorithm. Led by University of Pennsylvania's Stephen Kimmel, COAG investigators found that in the first four weeks of treatment, genotype-guided warfarin administration was no better than the clinical algorithm in terms of the mean time that the two groups remained in therapeutic range.

Neither study followed patient outcomes. Still, their divergent conclusions will likely continue to leave healthcare providers in the lurch about whether or not to implement PGx-guided dosing for warfarin. The studies have also revived longstanding questions about whether randomized-controlled trials are truly the best method by which to test the validity of the association between genetic markers and drug response.

"We're in a really tough situation because we've done what we've always wanted to do, which is a randomized, prospective trial," said Howard McLeod, who was not involved in either study and recently left his post as a pharmacogenomics professor at University of North Carolina, Chapel Hill, to join the Moffitt Cancer Center as medical director of its Personalized Medicine Institute. "But, now, we are where we were before."

Although differences in the design of the studies may have impacted their findings, at best these papers will have no effect on the adoption of PGx-strategies for warfarin — at least not until other randomized-controlled trials provide further insights on the clinical effect, cost, and patient outcomes of genetically-guided dosing strategies compared to the standard of care.

"Whatever your belief system is, you can pick a study," McLeod said. "If you're a believer, you've got the EU-PACT study. If you're a non-believer, you've got the COAG study. And we're almost none the wiser."

Different designs, different conclusions

Warfarin keeps the blood from forming clots by inhibiting vitamin K, which clotting factors and other proteins need for coagulation. When patients are given warfarin, it reduces the amount of vitamin K-dependent clotting factor their livers make. But when patients have mutations in the VKORC1 gene involved in this process, their blood doesn't clot as it should, placing them at risk for experiencing adverse reactions due to too much or too little warfarin.

Additionally, warfarin is metabolized in the body by CYP450 enzymes, mainly CYP2C9. Certain variations in the CYP2C9 gene — particularly CYP2C9*2 or *3 alleles — impact the body's ability to metabolize warfarin. According to the published literature, VKORC1 polymorphisms account for 30 percent of the warfarin dose variation in patients, while CYP2C9 variations contributed to 10 percent of the differences in therapeutic dose. People with these gene variations have a heightened risk of life-threatening bleeding events or blood clots if they receive an inadequate dose.

In the labeling for warfarin, the FDA recommends maintenance dosing ranges based on patients' CYP2C9 and VKORC1 genotypes. However, genotyping strategies haven't caught on in clinical practice because the data suggesting a link between these genotypes and patient outcomes mostly come from retrospective or observational trials.

The researchers of the two latest NEJM studies ventured to provide data from randomized-controlled trials that would increase the strength of evidence about the impact of genotype-guided warfarin therapy in clinical care. However, since the studies came to different conclusions, it's up to the experts to dissect the two trials and figure out which study best reflects the experiences of patients and healthcare providers in the real world.

"Randomized controlled trials have both strengths and weaknesses," said Larry Lesko, former FDA official and director for the Center for Pharmacometrics & Systems Pharmacology at the University of Florida. "The most important question for physicians and their patients is the external validity of the randomized-controlled trials and whether or not the results of these trials apply to [their] particular practice setting."

Lesko headed up the Office of Clinical Pharmacology within FDA's drug division during the years the agency updated warfarin's label. At the time, he had indicated that randomized-controlled trials may be an unrealistically high bar for establishing the validity of warfarin pharmacogenetics. As such, in 2010, when the FDA updated warfarin's label with genotype-guided dosing ranges based on studies with various other designs, the agency didn't require doctors to genotype patients ahead of dosing the drug, but gave them the flexibility to decide the best strategy.

In reviewing the NEJM studies, however, he felt that neither answers the question that physicians would most want to know: Does genotype-guided dosing improve on standard of care for high-risk patients? Both studies enrolled patients with genotypes that place them at varying levels of risk for experiencing adverse events with wafarin. The literature suggests that patients who are VKORC1 AA and AG and homozygous CYP2C9*2 and *3 are in the highest risk category and Lesko believes that the studies should have been enriched with this subset of patients.

"We already know that a majority of patients (e.g., wild-type VKORC1 and CYP2C9) are not at risk and will not benefit from genotype-guided dosing," he told PGx Reporter in an email. "So, to draw an analogy, it’s like enrolling 'non-responders' in an efficacy randomized-controlled trial, which dilutes the signal size because of a small percentage of 'responders,' and the population net effect is no benefit."

Beyond genotype, the studies also enrolled patients with other characteristics that influenced their risk of experiencing an adverse event while treated with wafarin, such as age, concomitant diseases, and other drugs they're on. "Distribution of such patients between studies affects results," Lesko said.

Other factors that may impact the studies' findings include the different dosing schedules and strategies researchers employed when administering warfarin to patients in the control arm. In the COAG trial, investigators used a clinical algorithm to initiate patients on warfarin in the control arm, whereas in the EU-PACT trial, researchers used a fixed dose of wafarin for patients in the control arm.

Also, in Lesko's view, the studies may not have followed patients long enough to truly see an impact of the dosing strategies on the time patients were in therapeutic range with warfarin. Patients were followed for 12 weeks in EU-PACT and for four weeks in COAG. "The risks of over- and under-anticoagulation do not stop at four weeks. So, outcome studies need to continue out to six months, at least," he observed, noting that studies comparing warfarin to newer anticoagulants suggest that it could take up to a year to increase the percent of time patients are in therapeutic range from 45 percent to 60 percent.

The COAG investigators acknowledge that their study doesn't address whether a longer duration of genotype-guided dosing would have improved therapeutic control, but they are looking into this in another trial.

In comparing COAG and EU-PACT, Lesko felt that the use of clinical dosing algorithms in COAG moved the study further away from the standard of care. "Neither study reflects current clinical practice exactly but the study by Kimmel et al. is further away from current clinical practice than the study by Pirmohamed et al.," he said. "This fact introduces bias into the study results."

In contrast, Yves Rosenberg, an expert from in the National Institutes of Health, highlighted the COAG study design as superior to EU-PACT. The COAG study was funded by an NIH grant.

"The major difference was that EU-PACT compared genotype-based dosing to a fixed dose of warfarin during the intervention period (first four days of treatment), while COAG used the same dosing formula in both the treatment and comparison groups," said Rosenberg, who is chief of the National Heart, Lung, and Blood Institute's Atherothrombosis and Coronary Artery Disease Branch in the Division of Cardiovascular Sciences. "The only difference between the two groups in COAG was the use of the PGx information allowing for the true testing of the pharmacogenetic hypothesis."

It is a continuous challenge for healthcare providers to extrapolate what the findings of a tightly controlled clinical trial mean for their patients, because care in the real world is rarely given in such an ideal setting. "If you're at Washington University [in St. Louis] with Brian Gage, you're already getting excellent care," Moffitt's McLeod said. Gage, a WashU professor and an author on the COAG trial, is one of the leading experts on wafarin pharmacogenetics. He developed a widely used nomogram for dosing warfarin that factors in patients' clinical variables and genotypes.

"Brian does nothing but manage these patients, and he has a team of people to do that," McLeod said. "So, you're getting good care. Most patients aren't even managed by a coagulation service, much less an algorithm."

Rapid testing

Finally, a key difference between COAG and EU-PACT was the specific genotyping technology employed. Industry experts have long believed that without the availability of rapid turnaround tests, PGx-guided dosing for warfarin would never take off. EU-PACT used a point-of-care genotyping instrument developed by the firm LGC that yielded results in approximately two hours.

COAG researchers used a test developed by GenMark Diagnostics and AutoGenomics with longer turnaround times. As such, in COAG, in the genotype-guided arm, test results weren't available for 55 percent of patients and so their first warfarin doses weren't genotype guided. "Whether this influenced the results is unknown," Kimmel et al. wrote in the NEJM paper. "However, the effect of missing genetics data on day one on the dose administered during the first three days of therapy was trivial."

Pirmohamed, the lead investigator from EU-PACT, noted that the point-of-care test was critical to the design of the study. Without it, researchers would not have been able to employ a loading dose algorithm that factored in the effect of CYP2C9 allelic variants in the first three days after warfarin initiation. "COAG ignored the effect of CYP2C9 on the first day," he said. "They also did not use a loading dose strategy, with the initial dose determined by a maintenance dose algorithm."

In EU-PACT, however, the point-of-care test yielded incorrect VKORC1 results for six patients. The study authors attributed this to operator error or problems with the stability of the reagents, but noted that the sensitivity and specificity of the assay remained high regardless.

Impact on African Americans

While COAG found no meaningful difference in mean percent of time in therapeutic range between the genotyped group and the control arm (45.2 percent versus 45.4 percent), a surprising finding was that African American patients fared worse on a genotype-guided warfarin dosing strategy than when given the drug with a clinical algorithm alone.

When stratified by race, African American patients in the genotype-guided group had a mean percent of time in therapeutic range of 35.2 percent compared to 43.5 percent in the comparator cohort. Non-African Americans did somewhat better on the genotype-guided strategy, with a mean percentage of time in therapeutic range of 48.8 percent compared to 46.1 percent. This finding was not statistically significant, however.

"Although the interaction between race and dosing strategy with respect to the primary outcome could be due to chance, the analysis was pre-specified and was consistent with our a priori hypothesis that there would be race-based differences," Kimmel and colleagues explained in their paper.

According to Lesko, this finding reflects the fact that dosing algorithms are unique to specific patient populations. "If a study with 98 percent Caucasians was used for developing an algorithm, it would perform poorly in African Americans because of the different frequencies of CYP2C9 and VKORC1 alleles among the two populations," he said. "For example, in African Americans, the frequency of the 'at risk for bleeding' AA genotype of VKORC1 is 0 percent to 2 percent and in Caucasians it is near 20 percent."

Pirmohamed similarly noted that the fact that the African American population in COAG did worse on genetically guided warfarin dosing is "very difficult to explain" since only a minority have the at risk variants. In the EU-PACT paper, the majority of patients were of European ethnicity and Pirmohamed and colleagues noted in their paper that "although the same genes determine warfarin dose requirements in different ethnic groups, the frequency of the individual gene variants differs, and algorithms that are specific to ethnic groups will need to be developed."

Gage explained that in COAG the same PGx dosing algorithm was used for African-American and in European-American patients. The algorithm factored in the prevalence and effect of the CYP2C9 and VKORC1 alleles in African Americans, but did not include rare polymorphisms that "might have improved the accuracy of dosing" in this patient subset, such as CYP2C9*5, CYP2C9*6, CALU rs339097, and CYP2C9 rs12777823.

"Individually, each of these SNPs likely plays only a small role in predicting the warfarin dose, but when considered together, they might improve the accuracy of pharmacogenetic dosing considerably in African Americans," Gage said.

Back in the real world …

Warfarin has been available since 1954 as a drug that prevents the formation of clots in blood vessels and keeps clots from migrating to other parts of the body and cutting off blood flow to vital organs like the heart or lungs. Although new anticoagulants have come on the market, such as Boehringer Ingelheim's Pradaxa (dabigatran), warfarin remains the most widely used, costing as low as $48 for a year of treatment. Although doctors have extensive experience dosing the drug through trial-and-error, factoring in patients' clinical data, and through international normalized ratio (INR) monitoring — which gauges how the blood is clotting after administering warfarin — such strategies are imprecise.

Warfarin is consistently at the top of the list for drug-related adverse events. An analysis of medical claims for atrial fibrillation patients who received warfarin between January 2003 and December 2007 found that the cost per patient with warfarin-related intracranial bleeding was nearly $42,000 and more than $40,000 for each patient with a major gastrointestinal bleed. Given the cost to the healthcare system, strategies that enable doctors to more quickly identify the appropriate dose for warfarin could potentially reduce healthcare costs.

Most private insurers don't pay for PGx testing for warfarin. And similarly, in 2009, the Centers for Medicare & Medicaid Services decided not to cover genetically-guided warfarin dosing for its Medicare beneficiaries. Instead, CMS agree to a "coverage with evidence development" scheme in which it would pay for PGx-based warfarin dosing only for Medicare beneficiaries who are part of a prospectively designed randomized-controlled trial showing pharmacogenomics-guided dosing improves health outcomes over standard dosing methods.

WashU's Gage is conducting one such study under CMS's CED framework, called Genetic InFormatics Trial (GIFT) of Warfarin Therapy to Prevent Deep Vein Thrombosis. "Given the conflicting results of these two trials, it's critical that we complete the third multi-center pharmacogenetic trial, GIFT," Gage said. To date, the trial has enrolled more than 700 elderly orthopedic patients. It is evaluating an additional gene, CYP4F2, and is investigating up to 11 days of PGx dosing. "Thus, the putative advantages of pharmacogenetic dosing may be magnified in GIFT," he noted.

While the research community continues to try to gain convincing evidence to back PGx-guided warfarin use, newer anticoagulants like Pradaxa have entered the market that carry some bleeding risks, but not the dosing variability of warfarin. According to data reported by the FDA, from October 2010 to December 2011 intracranial and gastrointestinal hemorrhage events were 1.8 to 2.6 times higher for new users of warfarin than for new users of Pradaxa.

Although the use of Pradaxa and newer anticoagulants such as Janssen's Xarelto (rivaroxaban) and Pfizer/Bristol-Myers Squibb's Eliquis (apixaban) are on the rise, they are priced at a significant premium compared to generic warfarin. For example, one year of treatment with Pradaxa costs approximately $3,000. As such, the relatively lower cost of wafarin therapy makes it more widely used in the US and in the world. In the UK, data from the National Health Service have showed that warfarin prescriptions rose from 9.4 million in 2011 to 10.2 million in 2012. Comparatively, Pradaxa scripts jumped from 3,200 in 2011 to 48,300 in 2012.

"Outside the US, and outside the richer countries, warfarin is still the go-to drug, mainly because of cost," Moffitt's McLeod said. "And warfarin still has the majority [of prescribers] because of familiarity." As the safety profiles of the newer agents become more apparent, there may come a time when warfarin isn't as widely used, he added.

Based on available data, McLeod acknowledged that if he needed to be on an anticoagulant, he would want to be on something other than warfarin. "But I'm in a privileged situation where I can both afford it and have colleagues who can manage it carefully. So, I think where most people are being managed in a family practice or an internal medicine practice, where they don't have all the latest knowledge, warfarin is still the way they're going."

Despite the recently published studies, Moffitt plans to conduct PGx testing for warfarin preemptively, so the genotype information is in the patient record and available for doctors to use should the need arise. The University of Florida, where Lesko works, has similarly instituted a preemptive PGx testing program for clopidogrel, another drug about which there is no consensus within the medical community as to whether to administer with a genotyping strategy. Other universities have also employed PGx programs where test results are stored in patients' electronic medical records for future use.

If genotype data is available in the EMR ahead of time, "I do think genetics is useful," McLeod noted. "After looking at the two studies, I say that having genetic data is never bad but it could be good. In a preemptive setting, where you have an economy of scale and the data is pre-loaded, it does make sense." Additionally, at a cancer center, McLeod noted that warfarin is managed for patients by subspecialists who are not coagulation experts. "I look at the genetic information as another way of trying to keep people safe. So, we're heading toward the point where we will use genetic testing," he said.

Lesko added that with the growing availability of point-of-care testing and INR, "the convenience of precision dosing is at our disposal." He acknowledged that warfarin has risks but asserted that genotype-guided dosing can address many of those risks.

Gage didn't answer a question from PGx Reporter about whether he would alter the use of PGx testing to dose warfarin for his patients. But he did indicate that it will be important for healthcare providers to wait for the results of GIFT.

Meanwhile, NIH's Rosenberg maintained that the totality of published data currently provides "no indication" that PGx-guided dosing improves anticoagulation control within the first five days of warfarin therapy for patients, when other clinical information is available. An editorial accompanying the NEJM studies also said as much. Bruce Furie of Harvard Medical School wrote in the editorial that "despite the variation in trial design, these trials indicate that this pharmacogenetic testing has either no usefulness in the initial dosing of vitamin K antagonists or, at best, marginal usefulness, given the cost and effort required to perform this testing."

In the end, doctors are left to parse the data and figure out what to do in their own practices. Most, whether they are warfarin PGx-supporters or detractors, will likely continue to do what was working for their patients. "It's really a huge bummer," McLeod said, referring to the two NEJM studies. "I'd rather have two negative studies than our current situation, because at least I'd know what to do."

The Scan

Billions for Antivirals

The US is putting $3.2 billion toward a program to develop antivirals to treat COVID-19 in its early stages, the Wall Street Journal reports.

NFT of the Web

Tim Berners-Lee, who developed the World Wide Web, is auctioning its original source code as a non-fungible token, Reuters reports.

23andMe on the Nasdaq

23andMe's shares rose more than 20 percent following its merger with a special purpose acquisition company, as GenomeWeb has reported.

Science Papers Present GWAS of Brain Structure, System for Controlled Gene Transfer

In Science this week: genome-wide association study ties variants to white matter stricture in the brain, and more.