WASHINGTON – Molecular laboratories, health systems, vendors of electronic health records and creators of clinical decision support content seem to agree that pharmacogenomics is one of several necessary components of precision medicine, but the promise cannot be realized until it becomes an unobtrusive element of clinician workflow.
That was the overarching message of a lively panel discussion and Q&A session at the American Medical Informatics Association (AMIA) annual symposium here this week.
Marsha Fearing, a pediatrician, biochemical geneticist, and clinical geneticist who serves as a physician specialist at EHR vendor Meditech, discussed the slow uptake of clinical pharmacogenomics.
"Why aren't we doing this?" Fearing asked. "The reason is because it's hard. If was easy, we would have done it a long time ago."
Fearing lamented the lack of discrete genetic data coming out of labs and said that primary care physicians in particular don't have time to read multipage reports in PDF files, which are simply digital images rather than computable data.
"You cannot do decision-making on a PDF. You cannot do it if you can't find the data," she said. The results have to come in as discrete data without the need for manual entry, ideally as a direct application programming interface with the reference lab.
Westwood, Massachusetts-based Meditech offers a genomics module called Expanse Genomics, which is designed to deliver actionable results based on genetic tests to clinicians within standard clinical workflows. However, it has limited utility in pharmacogenomics without the kind of standardization that exists in medical imaging and nonmolecular lab testing.
"The issue here is figuring out what genes need to be on a panel if you're going to be doing pharmacogenomics," said Ray Lorenz, director of medical science liaisons in neurology and pharmacogenomics for Quest Diagnostics. He said that such information should come from clinical literature about the utility of genetic markers and gene-drug interactions.
That is where clinical decision support systems should come in.
"Every week, new mutations are being discovered, so this is an impossible science to keep up with. It’s not an area of comfort, of familiarity for most practicing physicians," Fearing said. "You can see how it's an incredible cycle of doom to put all this together."
Anna Dover, a pharmacist and product manager at clinical decision support content developer First Databank, said that pharmacogenomics decision support is not in many EHR installations yet, particularly for physicians who mostly prescribe common drugs. "Clinicians aren't aware of the information [in genetic reports] or what to do with it," she said.
"Ideally, we are able to leverage the patient's test result or the absence of that test result to promote testing, suggest alternative therapies when appropriate, and even consider modifying doses, and provide useful information to the clinician," Dover said. "That is one of the greatest challenges in decision support, getting it at the right place in the workflow, at the right time, and in the right format so they can actually consume it."
Patricia Rice, clinical director for precision medicine and genetics at Frederick Health, a health system in Frederick, Maryland, that adopted Expanse Genomics this year, said that the organization's legal department told her to come up with a separate consent form for both somatic and germline testing. Extra consent forms complicate life for primary care physicians. "This is another thing the PCP has to do," she said.
The health system expects Meditech to fully support an organization-wide precision medicine program within a couple of years, though the rollout is happening in phases. Rice said that Frederick Health is working with the vendor to integrate the consent process into the EHR when alerts for genetic testing come up.
Not everything applies to a physician's specialty, either, creating conditions for overloading them with information on drugs they are not actually managing and exacerbating the ongoing problem of alert fatigue.
Lorenz said during the presentation that Quest reviewed more than 3,700 patient results for five genetic markers for pharmacogenomics and found that 99.7 percent had variants that could potentially impact medication metabolism and efficacy. He also noted that a 2019 paper studying pharmacogenomics at the US Department of Veterans Affairs suggested that more than 99 percent of patients had at least one pharmacogenomically actionable variant.
Those with psychiatric diagnoses were more likely to have clinically meaningful variations than those who had not been diagnosed with a psychiatric condition, according to Quest's internal research. "Maybe those are the types of patients we should be testing more often" to reduce the need for trial and error, Lorenz mused.
One issue is that many drugs are metabolized by multiple enzymes. Clinical decision support helps unravel some of the complexity and allows pharmacists and front-line clinicians to offer better care, Lorenz said.
However, sources of information do not always agree. Genetic panels vary from lab to lab, and vocabulary standards are lacking in pharmacogenomics, according to Dover, making it difficult to share relevant information across EHRs.
Even if the terminology issue is worked out, Lorenz said that pharmacogenomics should not replace a physician's judgment, Lorenz said. For example, genetics may suggest that a patient would respond well to GlaxoSmithKline's antidepressant Wellbutrin (bupropion), but that does not account for the fact that the drug is contraindicated for individuals with seizure disorders.
"While we can automate a lot of this genetic information and the combination of all of these genomic factors, it still needs to have review by an actual person," Lorenz said.
"This is a very esoteric science," said Fearing, who is tasked with engaging and educating clinicians about Expanse Genomics. "We have to agree on genetic markers" to include genetic panels. The information also has to be organized in ways such as separating somatic and germline variants in lab reports.
Another issue, according to Lorenz, is that the US Food and Drug Administration in 2018 recommended that labs stop putting medication guidance on reports.
That makes reports less useful for clinicians, according to Lorenz. He said that it is likely that lab reports and clinical decision support will remain separate in the future, though this creates an opportunity for software vendors to partner with labs to create clinical decision support systems to send actionable reports to clinicians.
During the Q&A portion of the 90-minute session, physician and audience member Thomas Payne, medical director of information technology services at the University of Washington's UW Medicine branch, said that this is not a good time to add new alerts for frontline practitioners, who are already suffering from alert fatigue.
Dover responded that the purpose of decision support is to educate providers and that companies like FDB are looking for ways to reduce alert fatigue. "This is meant as a fail-safe, preventing harm when it matters," she said.
"Pharmacogenomics is not a silver bullet. Pharmacogenomics won't help you find a needle in a haystack, but it removes some of the hay," Lorenz said.