NEW YORK (GenomeWeb) – Having tested close to 15,000 samples within its pharmacogenomics program, Vanderbilt University is planning to increase the variants reported to patients within their electronic medical records, and assess how testing has impacted their outcomes.
"We're getting to the point now where we have enough individuals and enough time in follow up that we're looking for clinical outcomes in our genotyped patient cohort," Sara Lynn Van Driest, assistant professor of pediatrics and part of the team at Vanderbilt running the Pharmacogenomic Resource for Enhanced Decisions in Care and Treatment (PREDICT) program, said at the Festival of Genomics in Boston last week.
Launched in 2010, the PREDICT program previously used the Illumina VeraCode platform to genotype patients for 184 polymorphisms in 34 genes; Van Driest and her colleagues recently switched to Thermo Fisher Scientific's Life Technologies QuantStudio real-time PCR platform. So far, expert committees within Vanderbilt have vetted and approved four gene-drug interactions for incorporation into patients' EMRs, including testing for CYP2C19 variants for response to the antiplatelet drug Plavix (clopidogrel); CYP2C9 and VKORC1 polymorphisms for response to the anticoagulant warfarin; TPMT variants for response to thiopurine agents; and CYP3A5 alleles for response to the immunosuppressive drug tacrolimus.
When doctors go to prescribe one of these drugs using the EMR, the decision support system alerts them if the patient's genotype might impact his or her ability to respond to treatment. Variants the committee has deemed to have insufficient evidence to report in the EMR are stored in a separate database for research or future clinical implementation. For markers lacking evidence on their relationship to drug response, researchers can utilize Vanderbilit's DNA biobank, BioVU, containing de-identified data on more than 200,000 consented patients, to conduct genomic investigations.
Dan Roden, assistant vice chancellor for personalized medicine at Vanderbilt, told GenomeWeb that experts involved with PREDICT are thinking of expanding the program by reporting new variants associated with drug response or by adding drugs that are influenced by already reported variants. The PREDICT team is particularly eager to assay CYP2D6 in coming months, which "will open up a whole new set of drug/gene interactions that we can assess," Van Driest said at the meeting.
CYP2D6 is a highly variable gene that codes for the enzyme involved in metabolizing or activating 25 percent of prescribed drugs. But it sits right next to a pseudo-gene, CYP2D7, with a nearly identical sequence to CYP2D6. "CYP2D6 is probably the gene we want to go after the most and probably the most difficult to assay," Roden said.
Comparatively, the markers of interest in genes like CYP2C19 are single SNPs that are easy to assay. "With CYP2D6 there are multiple variants, some of which are not single SNPs," he said. Moreover, patients with the poor metabolizer phenotype are either homozygous or compound heterozygous for loss-of-function alleles in this gene.
Using the Illumina platform, researchers didn't have the ability to gauge copy number variations, which are phenotypically important when genotyping patients for CYP2D6 variants."[VeraCode] met our needs for a restricted platform where we would sort of get the assays we wanted but we didn't get a huge amount of extraneous information," Roden said.
By comparison, the QuantStudio platform is cheaper and is customizable for different analyses, according to Roden. Vanderbilt recently switched to QuantStudio after Illumina stopped supplying the necessary reagents for VeraCode. Because of this hiccup and after Vanderbilt ceased funding PREDICT, there was a period of about a year when researchers had to stop genotyping new patients within the program.
During that time, PREDICT leaders were figuring out which platform to use and ironing out laboratory and billing issues to make it a self-supported PGx program. "So, over the last year, the program has not added new patients, but we're watching what doctors are doing with the information, how they like the information delivered to them, and what patients think when they see that kind of information in their charts," he noted. Now that the group has moved to the QuantStudio platform, "the [PGx] test is like any other test and people will be able to order it."
As Roden's group works to expand PGx testing within Vanderbilt, it is hoping to report PGx markers that will help heart and kidney transplant candidates, and HIV patients, who tend to be on a lot of drugs and are at higher risk for adverse events. "You could roll it out in a clinic where people get a single drug, but really the value of this will be in patients exposed to multiple drugs that have multiple pharmacogenetic stories," Roden said.
Do doctors care?
One way to measure PREDICT's success so far is to see if doctors are actually paying attention to the PGx alerts popping up in patients' EMRs. According to Van Driest, the data collected to date suggests doctors are taking heed of this information.
In the case of Plavix, the genotype of a poor or intermediate metabolizer can trigger a choice for the physician to give alternative therapies or Plavix at different does. The system asks doctors to explain their prescribing choice. Based on 12,000 genotyped patients, more than 500 out of 2,700 coronary stent patients had an actionable CYP2C19 variant, and their doctors received an EMR alert. Of these patients, 33 percent of intermediate metabolizers, 58 percent of poor metabolizers, and 8 percent of those with a non-actionable genotype received an alternate drug.
"What we took from this information is that … providers are using the genotype information to inform their prescription choice," Van Driest said. She cited further analysis suggesting that it's not just a few physicians that are prescribing alternate treatments, but doctors across the board seem to be changing their prescribing behavior prompted by the alerts.
"We have data now that doctors do pay attention," Roden said. "The more severe the genetic defect, the more likely it is that they will change their minds about what drugs to prescribe. So, doctors not only pay attention but pay attention to the details that are important in all this and that's a pretty gratifying result."
In reporting the drug/gene interactions for clinical decision making, Vanderbilt uses an internally developed EMR that places genotype information on the front page near where patients' allergies and adverse drug reactions are noted. Moreover, each alert has to be crafted specifically in the therapeutic context. For instance, the alert about thiopurine drugs is general, since there are multiple drugs given for a variety of indications, while the advice for warfarin includes specific dosing recommendations based on genotype and other patient data, such as age, weight, additional medications.
"These aren't just standard templates that you can pull out and use for all gene/drug interactions," Van Driest said, noting for example, that for warfarin and Plavix, there isn't enough evidence to provide guidance based on pediatric patients' genotypes. "You have to put careful thought and consideration into each one."
While there are lots of considerations that go into creating a PGx alert, ultimately what the doctor sees in the EMR has to be simple and straightforward. "Physicians don't want paragraphs. They want blurbs," Van Driest said, adding that doctors can click on a link to additional evidence if they want to learn more.
Proving value
Ultimately, performing a panel test and storing the information in the EMR is more efficient than developing individual tests for each indication, according to Van Driest.
The four drug/gene pairs have very different risk profiles. For example, only 10 percent of genotyped patients at Vanderbilt have TPMT alleles that placed them at risk for responding poorly to thiopurines — drugs such as azathioprine, mercaptopurine, and thioguanine that are prescribed to patients with leukemia, inflammatory bowel disease and autoimmune disorders. In comparison, warfarin has more actionable genotypes and 69 percent of tested patients at Vanderbilt have a variant that puts them at high risk for experiencing an adverse event to the drug.
Looking at any single drug/gene interaction one might get the impression that there is a small risk of an individual being a poor metabolizer, Van Driest said, but looking at the collective experience of all tested patients gives a more accurate picture of the impact of PGx testing. "In 88 percent of cases [tested within Vanderbilt,] you're going to find at least one actionable allele," she said.
According to data researchers have collected so far, 44 percent received information about one genetic risk factor, while 33 percent, 10 percent, and 1 percent of patients learned they had two, three, and four genetic risk factors associated with drug response. Only 12 percent of patients had no PGx risk factors reported in their EMR.
Genomics experts tend to agree that pre-emptive multiplex testing of the kind that Vanderbilt and University of Florida are implementing might be the most cost-effective way to incorporate PGx information into medical care, but the billing and reimbursement aspects of such programs remain uncertain. Insurance coverage can be spotty for PGx testing, Roden admitted.
Insurers will generally cover PGx testing in certain situations, such as when a patient is scheduled to get a stent in two weeks and would require Plavix. "But the question is then how do you figure out what to do with those data if the system accesses them again when the patient goes on azathioprine or tacrolimus or warfarin?" Roden posited. "That's an open question across the genomics community. No one quite knows how to deal with this idea of multiplex pharmacogenetics testing."
These are the questions that Roden's team is wrestling with as it advances PGx testing within Vanderbilt, as well as in "informatics-poor" healthcare systems. Within the National Institutes of Health's Implementing Genomics in Practice (IGNITE) program, Vanderbilt is running a demonstration project to show that providing PGx information through web-based interfaces can help doctors in other healthcare environments incorporate genetic information in their prescribing practices.
Through this project, some of the tools developed within PREDICT will be made available to Nashville Veterans Affairs Medical Center, Nashville General Hospital at Meharry Medical College, and Aurora Health Care in Wisconsin. "Such collaborations are important for exporting this type of service to less informatics-rich places, where there [isn't] an army of programmers and sophisticated web interfaces," Roden said.