The discovery of germline pharmacogenomic variants that affect drug response or toxicity has exploded in recent years, and a number of these markers now have amassed a decisive body of clinical evidence sufficient to deem them clinically actionable. Despite this evidence—and despite the calls of numerous stakeholders, including U.S. legislators—translating pharmacogenomic testing into clinical use has not kept pace.
A number of barriers, many relevant to clinical laboratories, present unique challenges to implementing pharmacogenomics into practice. This mini-review will focus on such barriers and present an overview of our institutional efforts to surmount obstacles to preemptive pharmacogenomic testing and results delivery.
Preemptive Versus Reactive Genotyping
A first practical challenge related to implementing pharmacogenomic testing is the issue of when to genotype. Arguments have been made for both preemptive genotyping, which involves testing prior to prescribing so that results are available before the time of prescription consideration, and reactive genotyping, when a clinician orders a test only after he or she decides to prescribe a particular drug (but ideally in advance of the patient actually starting the medication) (1).
While reactive genotyping is directly applicable to specific patient situations and more likely to be reimbursed by insurance companies, many believe the advantages of preemptive genotyping outweigh the current benefits of a reactive approach.
In our large, institutional pharmacogenomic implementation program, we have chosen to genotype adult patients preemptively within our CLIA- and College of American Pathologists-accredited laboratory under an Institutional Review Board-approved research protocol. Patients who provide informed consent are genotyped across a custom broad panel, which includes germline polymorphisms that have been shown to impact toxicity or response for a significant number of commonly used medications. We also include a dedicated CYP2D6 panel with copy number assessment given the importance of this key pharmacogene (2).
We chose the variants on our custom panels based on systematic analysis of high-level pharmacogenomic publications (3) in addition to external pharmacogenomic references on actionable markers. Importantly, all interrogated variants are deemed potentially clinically actionable, as our focus does not lie in capturing variants of unknown significance.
Patients submit a single blood sample at the time they enroll in the project, and we test all variants regardless of the medications patients are currently taking. Our turnaround time is approximately 2 weeks, with results available in advance of the patient’s next visit to the medical center. The technical investment to establish these laboratory developed tests is not insignificant, and initially our technical validation steps required approximately 6 months (for the large preemptive panel) to fully validate these tests for clinical use.
The major advantage to this preemptive approach is that it removes the delay inherent in reactive approaches that has proven burdensome to adopting pharmacogenomics in clinical practice. Preemptive genotyping offers many other benefits, from technical efficiency to improved patient care. In fact, the technical efficiency of this approach may, in the long run, be most cost-effective. With the cost for genetic tests plummeting in recent years (4), we have found it beneficial, using an economy of scale, to interrogate hundreds of potentially clinically actionable variants at once as opposed to conducting tests for one variant or gene at a time.
Unfortunately, the current regulatory scheme creates a perverse incentive for clinical laboratories to do the opposite and receive reimbursement for each single variant or single gene assay, each ordered separately over a patient’s lifetime.
Another benefit of our broad preemptive approach is scalability. We update our laboratory developed tests on a regular basis allowing for inclusion (after validation) of newly discovered pharmacogenomic variants and have done so four times in the last 5 years. This approach enables us to report the most updated and current clinically actionable information, but does require that we re-validate the comprehensive panel after each update. It also requires that we store and retain a portion of each patient’s DNA (via explicit permission of the patient) for future re-testing.
A major hindrance, however, is the long standing distaste of payers for composite “screening” germline pharmacogenomic panels. In our own institutional work, we do not currently seek insurance reimbursement for any of our preemptive germline pharmacogenomic comprehensive tests; the costs are supported instead by grants, institutional support, and other means. The reimbursement landscape, however, could change rapidly, especially based on recent events in the coverage of laboratory developed assays in the oncology field, discussed further at the end of this article.
As an illustration of the potential clinical applicability of our preemptive approach, we found after examining more than 800 samples that 100% of patients have at least one actionable variant that could impact the use of one or more of the 40-plus drugs with clinical pharmacogenomic evidence (5).
We propose that having such results available at the time of prescribing will not only prevent high-risk medications from being prescribed to genotyped patients, but also will prove more efficient—and potentially cost-effective—for overall healthcare delivery. In support of the former, we have previously shown that in more than 4 years of delivering preemptive test results prior to prescribing visits, no pharmacogenomically high-risk medications have been prescribed when providers accessed our clinical decision support tool (6). We believe this to be the ultimate evidence in favor of preemptive genotyping and results delivery.
Translating Genotyping Results
Another barrier to more streamlined pharmacogenomic implementation is translating raw genotype results into a format that is easier for providers to understand and incorporate into their busy workflows (1). It is important, especially for laboratories, to have a standardized language for translating genotypes to phenotypes. Seminal efforts have been undertaken to develop these standards such as the Clinical Pharmacogenetics Implementation Consortium, PharmGKB, ClinGen, and the Centers for Disease Control and Prevention (7).
As an illustration of this, key pharmacogenes such as CYP2D6 and CYP2C19 employ the star nomenclature for describing allelic haplotypes. Each patient’s diplotype (i.e. CYP2C19 *1/*2) can then be translated easily to a phenotype that represents the drug metabolizer status (i.e., intermediate metabolizer, using this example).
An added layer of complexity lies in the necessary connection between laboratories and those who populate results into the electronic health record (EHR) or other pharmacogenomic results delivery systems. Currently most institutions customize these workflows, although some commercial vendors are now developing such capabilities. While some institutions have developed machine-readable platforms that allow direct entry of pharmacogenomic laboratory data into the EHR (8), typically data transformation must occur with human review and the involvement of informatics teams.
We have leveraged existing public databases to aid in our genotype-to-phenotype translations. As recently published by our group (3), such translations currently occur for a subset of genotype data reported by our lab from known variants within five key pharmacogenes (CYP2D6, CYP2C19, CYP2C9, SLC22A1, and TPMT). In contrast to this gene-level data, our variant-level annotations do not require such genotype-to-enzymatic phenotype conversions and are thus handled in a more straightforward manner. PharmGKB, a publicly available pharmacogenomic knowledge database, provides gene-specific haplotype tables for translating genetic variants to star alleles. Additional tables show the translation for star allele designations to phenotype (corresponding enzyme status).
We derived our own genotype-to-phenotype translation algorithms from these publicly available tables. They differ in that our algorithms only include variants available within our validated genotyping panels. Some examples have been previously published (3). Notably, we have developed and re-validated up-to-date iterations of our algorithms, as appropriate, when updates are made to our panels, such as when additional variants are validated and clinically made reportable.
We posit that adapting available resources, such as the ones mentioned above, provides an opportunity to standardize genotype-to-phenotype translations among laboratories, a critical step for widespread clinical implementation.
Results Delivery Including Clinical Decision Support
Perhaps one of the most crucial steps in implementing preemptively-obtained pharmacogenomic results lies in how results are delivered to clinicians. Key considerations include EHR integration, the storage and security of large amounts of patient data, whether the results delivery system is active or passive, and how results are depicted or displayed. Our own results delivery system, the Genomic Prescribing System (GPS), was designed with these challenges in mind. The system has been previously described in detail (3), with highlights described below (See Figure 1). A key consideration when we were initially designing the GPS was ensuring that the system could be easily integrated into provider workflows. This meant surmounting the challenge of presenting complex genomic data to individuals who may not have had prior formal genetics or genomics training. Given this potential barrier of lack of provider knowledge, we designed the GPS with universally known iconography: the traffic light. This provided an easily recognizable signal with intuitive meaning.
Each drug associated with a pharmacogenomic variant for which we test is assigned a traffic light; red lights correspond to high genomic risk of non-response or side effects, yellow lights represent cautionary—but not necessarily contraindicatory—genomic information, while green lights represent favorable genomic signals. In addition to using traffic lights, we also designed the GPS to have concise clinical decision support (CDS) summaries, which provide, when applicable, pharmacogenomically favorable alternative medications or patient-specific dose recommendations, alongside links to the primary literature from which the CDS summary was derived.
Importantly, the GPS data only conveys drug-genetic relationships (not, for example, drug-drug information, nor information about use of drugs in certain disease states like renal dysfunction), as we designed it to supplement other pertinent clinical data sources that providers already consider when prescribing medications. Further, we devised GPS to be a passive system, meaning that it does not employ pop-up alerts regarding patient results. We felt this was important to avoid alert fatigue, although we are considering for the future active alerting, which involves push notifications to providers about the highest risk-genomic signals.
While we created GPS initially as a stand-alone system, its integration with our EHR over 2 years ago has provided the opportunity for more seamless incorporation into provider workflows. Within the EHR, a link enables providers to launch the GPS user interface easily as an internal browser window, using institutional authentication credentials to log in.A final, critical functionality of the GPS is the dynamic drug and disease search field. This functionality is vital when a large number of variants have been tested preemptively, as it allows providers to search any disease or drug with return of patient-specific pharmacogenomic results.
In fact, we tested the utility of this functionality among a cohort of more than 1,000 genotyped patients. We found that over 90% of the top approximately 20 diseases identified in this population (and at least 93% of patients) could be treated with at least 1 drug with actionable pharmacogenomic information (9).
Overview of Project Results
We have found that delivering preemptive pharmacogenomic results to participating study providers leads to measurable benefits for providers and patients. We analyzed data comprised of more than 2,200 outpatient clinic visits over a 3-year period for 547 unique preemtively genotyped patients enrolled in our institutional implementation program (6). Of these evaluated visits, clinicians accessed the GPS 69% of the time, with providers being more likely to log in to the system at visits where medication changes were made (Odds Ratio (OR) = 1.6 [95% Confidence Interval [CI], 1.2-2.1], p < 0.0001). (They did not access GPS at some visits, for example, due to time constraints.)
Importantly, more than one-third of medications that patients were currently taking (at the time of the visits in question) were associated with pharmacogenomic information. We determined whether these patient-specific pharmacogenomic results influenced prescribing by analyzing medication change rates and found that both red and yellow light medications were changed more often than medications with no pharmacogenomic information. The OR of a red light medication being changed was 26.2 (CI, 9.0-75.3) (p < 0.0001), while that of a yellow light drug was 2.4 (CI, 1.7-3.5)
(p < 0.0001). Green light medications, on the other hand, were not changed significantly more often than medications without pharmacogenomic information, suggesting favorable genomic signals may have confirmed providers’ prescribing choices—a perhaps underappreciated aspect of pharmacogenomics.
Alongside analyzing provider behavior in response to pharmacogenomic results availability, we also studied patient perspectives through surveys and focus groups. We found that when providers considered pharmacogenomic results, patients perceived increased doctor-patient empathy and privacy, a better understanding of medical decision-making, and most significantly, a greater sense of personalized care from their providers (10). Results from patient focus groups also showed that those who were previously genotyped were open-minded about the use of their pharmacogenomic results in the clinic (11). Regardless of whether patients had been genotyped, they expressed concerns about employment discrimination and insurance coverage based on pharmacogenomic results.
Taken together, we believe our results on patient perspectives represent vital considerations for implementation efforts, as patients are key stakeholders in widespread adoption of pharmacogenomic testing. Importantly, our efforts would not have been possible without early engagement and support from key institutional stakeholders, including insitutional leadership laboratory personnel, the research informatics and hospital informatics teams, and the willing early-adopter physicians and patients.
The last major challenge to wider adoption of pharmacogenomic testing is the currently limited scope of insurance reimbursement for germline pharmacogenomic tests. At this time, most major carriers cover only a handful of gene/drug pairs. Payors may be awaiting additional prospective data (including perhaps cost-effectiveness analyses) before deciding about expanded reimbursement for additional drugs and genes/variants. However, recent analogous decisions about panel-based “composite” tumor-based genomic approaches in the realm of oncology provide one illustration of how preemptive germline testing could be evaluated going forward.
Recently, the Food and Drug Administration (FDA) approved Foundation Medicine’s tumor genomic profiling test, FoundationOne CDx, which interrogates more than 300 somatic genes “preemptively” in one panel. Simultaneously, the Centers for Medicare and Medicaid Services issued a preliminary national coverage determination for the test. Other similar tumor-based “panel” tests have also recently received FDA approval. All of these tests offer the potential to inform the use of multiple drugs at once by interrogating multiple genes simultaneously and preemptively. We wonder whether this sea change in approach could eventually be extended to germline pharmacogenomic testing.
One could argue that—not dissimilarly to oncology—a critical mass of clinically actionable germline pharmacogenomic variants now exists. With these variants informing the use of various different drugs, panel-based preemptive testing for a set of germline pharmacogenomic markers—especially in certain populations at risk for near-future or lifetime poly-pharmacy—could be a sound and cost-effective strategy to evaluate in future implementation efforts.
We have highlighted important considerations for preemptive pharmacogenomic testing, many of which are relevant to clinical laboratorians. Choosing when to genotype, along with how to translate and deliver results, are critical considerations for any implementation effort. We have taken an individualized approach to overcome common barriers, including leveraging public resources to aid in genotype-to-phenotype translations, creating a dynamic results-reporting system that is accessible both outside of and within our institutional EHR, and delivering CDS in a way that is compatible with provider workflows.
Our implementation program has positively impacted provider prescribing in a way that is aimed at reducing patient risk of non-response or medication toxicity. The doctor-patient relationship, along with both provider and patient attitudes toward pharmacogenomic testing, has facilitated this process, and our program has received overwhelmingly positive feedback. We believe our model is one example of how to realize the era of precision medicine.
We thank our collaborators in the department of pathology at the University of Chicago for their support of our institutional implementation efforts, specifically K-T Jerry Yeo, PhD, director of the pharmacogenomics laboratory, and Edward Leung, PhD, associate director, who oversee the clinical laboratory. We thank Xun Pei, MB(ASCP), for her technical expertise in developing the assays, and Keith Danahey for his invaluable assistance in developing the clinical decision support technologies. We also sincerely acknowledge the key role of Mark J. Ratain, MD, director of the Center for Personalized Therapeutics, for his institutional vision for this project and his support.
We also wish to acknowledge the funding sources for this work: NIH K23 GM 100288-01A1, NIH/National Heart, Lung, and Blood Institute grant 5 U01 HL105198-09, The Conquer Cancer Foundation of the American Society for Clinical Oncology, The William F. O’Connor Foundation, The University of Chicago Comprehensive Cancer Center support grant, The University of Chicago Bucksbaum Institute for Clinical Excellence Pilot Award, and the Central Society for Clinical and Translational Research - Early Career Development Award.
Dr. O’Donnell is named as a co-inventor on a pending patent application for the Genomic Prescribing System.
Brittany A. Borden, MA, is a clinical research coordinator and senior member of the Center for Personalized Therapeutics at the University of Chicago. +Email: [email protected]
Peter H. O’Donnell, MD, is an assistant professor of medicine in the section of hematology/oncology and the associate director for clinical implementation for the Center for Personalized Therapeutics at the University of Chicago. +Email: [email protected]
- Relling MV, Evans WE. Pharmacogenomics in the clinic. Nature 2015;526:343-50.
- Leung EKY, Agolini E, Pei X, et al. Validation of an extensive CYP2D6 assay panel based on Invader and TaqMan copy number assays. The Journal of Applied Laboratory Medicine 2017;5:471-82.
- Danahey K, Borden BA, Furner B, et al. Simplifying the use of pharmacogenomics in clinical practice: Building the genomic prescribing system. J Biomed Inform 2017;75:110-21.
- Schildcrout JS, Denny JC, Bowton E, et al. Optimizing drug outcomes through pharmacogenetics: A case for preemptive genotyping. Clin Pharmacol Ther 2012;92:235-42.
- O’Donnell PH, Danahey K, Ratain MJ. The outlier in all of us: Why implementing pharmacogenomics could matter for everyone. Clin Pharmacol Ther 2016;99:401-4.
- O’Donnell PH, Wadhwa N, Danahey K, et al. Pharmacogenomics-based point-of-care clinical decision support significantly alters drug prescribing. Clin Pharmacol Ther 2017;102:859-69.
- Kalman LV, Agundez J, Appell ML, et al. Pharmacogenetic allele nomenclature: International workgroup recommendations for test result reporting. Clin Pharmacol Ther 2016;99:172-85.
- Nishimura AA, Shirts BH, Dorschner MO, et al. Development of clinical decision support alerts for pharmacogenomic incidental findings from exome sequencing. Genet Med 2015;17:939-42.
- Hussain S, Kenigsberg BB, Danahey K, et al. Disease-drug database for pharmacogenomic-based prescribing. Clin Pharmacol Ther 2016;100:179-90.
- McKillip RP, Borden BA, Galecki P, et al. Patient perceptions of care as influenced by a large institutional pharmacogenomic implementation program. Clin Pharmacol Ther 2016;102:106–14. doi:10.1002/cpt.586.
- Lee YM, McKillip RP, Borden BA, et al. Assessment of patient perceptions of genomic testing to inform pharmacogenomic implementation. Pharmacogenet Genomics 2017;27:179-89.