This Drug's for You
By Robert Freimuth, PhD; Liewei Wang, MD, PhD; and Richard Weinshilboum, MD
For The Record
Vol. 28 No. 10 P. 20
Mayo Clinic is out in front of both the research and the implementation of pharmacogenomics.
Genetic data have been used to inform clinical decision-making for decades, but recent technological advances and an expanded understanding of the genome have dramatically increased our ability to use genomics to improve patient care. Those advances, however, also present new challenges to health care institutions faced with managing large patient genomic data sets and knowledge bases that change on a daily basis.
To achieve the vision of genomic-based precision medicine, solutions are needed that enable the efficient clinical implementation of genomic medicine within complex clinical electronic environments and workflows. The field of pharmacogenomics has led the way in developing methods for successfully integrating genomics into clinical practice, and the lessons learned from those experiences have identified opportunities for the development of novel solutions that will enable the widespread adoption of genomic medicine.
Pharmacogenomics is the study of the role of inheritance in variation in response to treatment with the powerful and effective drugs available to modern physicians. The development of entirely new classes of drugs has resulted in a therapeutic revolution with regard to a clinician's ability to treat, control, and even cure disease—diseases ranging from childhood leukemia to hepatitis C—that would have been inconceivable just a few years ago.
However, there can be very large individual variation in response to these agents. Most patients experience the desired therapeutic effect, but the same drug can fail to display the desired benefit in some patients, and, rarely, serious and even life-threatening adverse drug reactions can occur.
It is now clear that genetics is an important factor contributing to individual variation in drug response. Pharmacogenomics represents an attempt to use genomic information, ie, variation in the patient's DNA sequence, to predict the appropriate drug and dosage for individual patients. Also referred to as precision drug therapy, this practice is based, in part, on genomics.
The term pharmacogenomics is not new. The conceptual basis for the use of genetic information to predict and optimize drug therapy dates back over half a century. In fact, some of the examples that are now being implemented clinically and for which alerts are firing in EHRs have been known for decades. However, it was only after completion of the Human Genome Project in 2003 that the convergence of developments in genomic science and DNA sequencing, together with the increasing adoption of EHRs, that it became possible to consider bringing this aspect of clinical genomics to the bedside on a broad scale.
The Human Genome Project was often depicted in the popular press as a race to the finish line. In fact, it was really a race to the starting line, in terms of bringing this aspect of modern medical science into routine clinical practice. At Mayo Clinic and other academic medical centers, that is happening today.
Clinical Implementation at Mayo Clinic
At Mayo Clinic, under the leadership of the Center for Individualized Medicine (CIM), the CIM Pharmacogenomics Program began several years ago to lay the groundwork needed to make this aspect of clinical genomics available to the facility's physicians and patients. Specifically, the Pharmacogenomics Task Force was established within the Drug Formulary Committee to select drug-gene pairs, drugs for which variation in specific genes had been shown to have clinical utility for helping physicians tailor drug therapy by taking into account variation in the ability of that patient to respond to that drug treatment.
When a consensus existed that this information would be helpful to physicians, alerts were implemented in the EHR to inform physicians at the time that they wrote prescriptions of the availability of a genomic test for individual variation in response to treatment with that drug. Currently, 18 of these drug-gene pair alerts fire across all Mayo campuses for the approximately 1.4 million patients treated each year by the organization's physicians. However, this is only an initial step toward the ultimate goal. The current alerts require that the physician order a genomic test and await the result if he or she thinks that testing may be helpful. The eventual goal would be to have the genomic information for that patient already present in the EHR.
As a step toward making that a reality, a pilot study was conducted with 1,013 Mayo Clinic Biobank DNA samples from local patients who had provided consent. In the study, called the RIGHT study (the RIGHT drug at the RIGHT dose for the RIGHT patient), the samples were sequenced for 84 pharmacogenes—genes that show common variation associated with clinically important variation in drug response.
For only the top five of those 84 genes, 99.1% of those 1,013 patients had clinically actionable DNA sequence variation in at least one of the five genes. That finding emphasizes the fact that pharmacogenomics represents an aspect of clinical genomics that may ultimately be indicated for every patient. If this type of genomic information were available preemptively, it would be unnecessary to fire alerts for patients who do not carry DNA sequence variants that might present a problem.
Currently, the alerts fire for all patients for whom a prescription is written for a drug included in pharmacogenomic drug-gene pairs. However, in the future, when a preemptive alert would fire, it would not require the physician to order a test or wait for the test result—the test result would already be in the EHR. For that ideal situation to become the standard of care, it will be necessary for payers to conclude that the cost-benefit ratio is acceptable, which will necessitate evidence that supports the value of preemptive pharmacogenomic testing.
As the next step in testing the hypothesis that pharmacogenomics would be of broad value, the Mayo Clinic CIM and the Mayo Clinic Center for the Science of Health Care Delivery—in collaboration with the Baylor College of Medicine Human Genome Sequencing Center—have recently initiated the RIGHT 10K study in which 10,000 consenting Mayo Clinic Biobank patients will have 77 pharmacogenes sequenced. The genomic information will be placed in the Mayo EHR preemptively to make it possible to both retrospectively—using EHR data that extend back for well over a decade—and prospectively test the hypothesis that having the genomic information immediately available would make it possible to individualize drug therapy for that patient.
In addition, it will be determined whether it would be cost-effective to implement preemptive pharmacogenomics on a broad scale. Finally, any novel DNA variants, ie, mutations, identified during the sequencing effort will be studied in the laboratory to better understand their functional impact. That DNA sequence information could then be incorporated into clinical decision support (CDS) rules to assist physicians when making pharmacogenomic decisions in the future.
The clinical implementation of preemptive pharmacogenomics presents a series of technical challenges, but the fact that we are now so close to bringing this important aspect of clinical genomics to the bedside broadly tells us just how far we have come from the starting line that was represented by the completion of the Human Genome Project in 2003.
In order to make the clinical implementation of pharmacogenomics and projects such as the RIGHT study possible, systems must be modified and/or built to support the integration of genomic data into the EHR. Like many medical institutions, Mayo Clinic uses both homegrown and vended components in its EHR environment, but none of those components was designed to support the type of large-scale genomic sequencing data that now can be obtained through clinical-grade tests. As a result, institutions must make significant investments in upgrading their infrastructure to support the management of genomic data and the delivery of genomic-based clinical recommendations to health care providers.
To support the clinical implementation of pharmacogenomics at Mayo Clinic, many decisions were made regarding the representation, storage, and delivery of patient genomic data within EHR systems. In most cases, the capabilities of existing systems were used to the extent possible.
For example, genomic test results were described in a detailed, text-based report intended for human consumption, which was stored in the EHR as a scanned document much like any other lab test. However, computable forms of those results were required to serve as trigger criteria for drug-gene alerts, making it necessary to load key summary results into the EHR as discrete data elements. Those data were added to the patient problem and/or allergy lists within the EHR, where both could be accessed by the CDS rule engine.
Using published clinical guidelines and input from experts, the functional requirements of the CDS rules were defined by the CIM pharmacogenomics task force and were subsequently implemented as executable rules by enterprise CDS teams. The rules fire an alert when certain trigger criteria are met, such as when a specific drug is ordered for a patient who has a given pharmacogenomic test result. Those alerts are often "active" alerts, which interrupt the clinical workflow and present information to the provider to ensure a patient's pharmacogenomic test results are considered when making drug therapy decisions.
Opportunities for Clinical Informaticists Abound
Mayo Clinic's implementation of 18 drug-gene alerts—one of the largest in the world—demonstrates the feasibility of using patient genomic data to guide pharmacotherapy. The lessons learned from this effort identified many opportunities to enhance the organization's EHR infrastructure to more fully support pharmacogenomics and, more broadly, genomic-based precision medicine. In particular, there are clear needs for advances in EHR design and standards for the representation of genomic data and knowledge.
For example, EHRs are not designed to store the large, complex data generated by modern genomic sequencing systems. Specialized systems that can be easily integrated into existing EHRs are needed for storing those data, similar to the systems used for storing radiology images. Secondly, current methods for storing and displaying traditional lab test results, which are often clinically relevant for only a limited time, do not easily support genomic results that are applicable throughout a patient's lifetime.
Also, the problem and allergy lists cannot express the clinical context or degree of risk inherent to genomic results (eg, an increased risk of an adverse reaction if the patient is exposed to a given drug). Finally, unlike most lab tests that are static, the interpretation of genomic test results can change over time as our understanding of biology improves; therefore, EHRs must also support the dynamic reinterpretation of genomic test results (along with the evidence and provenance associated with reinterpretations).
Currently, there are few standards for representing clinical genomic data and knowledge. Although the Federal Register 45 CFR Part 170 ("Health Information Technology Standards, Implementation Specifications, and Certification Criteria and Certification Programs for Health Information Technology") specifies standards that EHRs must support to meet meaningful use requirements, data and terminology standards offer only limited support for genomics.
In particular, robust standards are needed for expressing the results of clinical genomic tests, including both raw results and interpretations, which are unambiguous and computable. Standards are also needed for expressing genomic knowledge, which is used to translate raw results into interpretations. Finally, standards are absent for expressing genomic CDS rules in a portable format that is HER agnostic.
Several organizations are working to fill these gaps, which, if filled, will improve interoperability and enable sharing of both data and knowledge among health care institutions.
The clinical implementation of pharmacogenomics at Mayo Clinic and elsewhere provided invaluable opportunities to explore methods for integrating genomic data into the EHR and developing systems to deliver those data to providers at the point of care. The lessons learned from those efforts also identified gaps in the capabilities of existing systems.
Improving EHR design and developing standards for the representation of genomic data and knowledge will help health care move past the "one-gene-at-a-time" model of implementation and allow us to scale our efforts to the level of the entire genome. This will bring us closer to translating the advances in understanding catalyzed by the Human Genome Project to routine clinical practice and realizing the promise of precision medicine.
— Robert Freimuth, PhD, is an assistant professor of biomedical informatics at the Mayo Clinic College of Medicine.
— Liewei Wang, MD, PhD, is a professor of pharmacology at the Mayo Clinic College of Medicine.
— Richard Weinshilboum, MD, is the Mary Lou and John H. Dasburg professor of cancer genomics, professor of medicine and pharmacology at the Mayo College of Medicine, and the pharmacogenomics program director at the Mayo Clinic Center for Individualized Medicine.