NEW YORK – Translational Software's announcement last week that it will stop providing pharmacogenetic decision support services in the US by the end of the year after failing to achieve regulatory clearance for its platform has left nearly 100 laboratory customers scrambling for alternative tools for reporting test results and the entire space looking for more clarity as to what the US Food and Drug Administration's expectations are for other similarly situated companies.
The software company had sought 510(k) clearance for its PGxPortal, a cloud-based solution that takes the genetic analysis from lab partners performing PGx testing, assigns a diplotype, and interprets how the results may affect patients' ability to respond to various drugs. According to Translational Software CEO Don Rule, the FDA did not clear PGxPortal due to concerns about its diploid calling feature and because the company provides treatment recommendations using guidelines from expert bodies the agency doesn't recognize, like the Clinical Pharmacogenetics Implementation Consortium (CPIC).
Before Translational Software's projected Dec. 31 date for stopping services, its 98 customers — mostly labs and a few healthcare organizations — will have to find another way of reporting PGx test results. According to Rule, his company services reference labs and labs providing PGx tests to guide psychiatry, cardiology, pain, and surgery medications, as well as drugs used in pediatrics.
Resolve Molecular Diagnostics, a Franklin, Tennessee-headquartered firm that provides PGx testing largely to independent physicians, relies on Translational Software's rules engine to translate a patient's genotype into a phenotype. In light of Translational Software's announcement, Resolve MDx CEO Wes Warrington has been trying to figure out if the FDA took particular issue with data submitted by Translational Software or if the agency's decision suggests that all similar clinical decision support companies are not in compliance. If it's the former, then Resolve MDx will have to find another rules-based engine to report PGx test results, but the latter has more severe consequences for the company.
"To say, 'Doctor, your patient is a CYP2D6 ultra-rapid metabolizer,' and that's it, that does not provide value to the physicians," Warrington said, noting most doctors lack genomics know-how and need guidance on what drugs to give patients based on PGx results.
He estimated that Resolve MDx has approximately 100 customers for its PGx test, which comprises around 25 percent of the firm's business. But "without having that translation" from genotype to phenotype, Resolve MDx will be "unable to provide a viable test," Warrington said.
In PGx testing, determining diplotypes, or haplotype pairs, based on detected genetic variants is a critical step to establishing how those changes might impact gene function and patients' ability to respond to drugs. Bronwyn Ramey, founder of Phoenix Laboratory Consulting, said most labs where she has helped set up PGx programs had "absolutely no idea" how to determine a haplotype from detected SNPs.
The leading genetic test platform providers for PGx are Thermo Fisher Scientific and Agena Bioscience, and Ramey is aware of software from Agena that can generate diplotypes. She noted there are a few other companies, such as Polytech PGx, GenXys, ActX, and Coriell Life Sciences, that offer PGx reporting tools and services with varying capabilities. She has also pointed labs she has worked with to PharmGKB, an organization that collaborates with CPIC on advancing PGx guidelines and has a free report generator. Still, Translational Software, a company Ramey has advised on quality and regulatory matters, has extensive and customizable PGx test reporting capabilities, she said, and with its shuttering, "there will be a vacuum in the market."
'Warning shots' for PGx software?
In light of Translational Software's experience with the FDA, other companies with clinical decision support platforms are undoubtedly trying to parse the potential regulatory implications for them. The agency has said it wants to regulate lab-developed tests and certain software as medical devices, but providers of PGx products, even those that have engaged with the FDA like Translational Software, say they are unsure how to meet the agency's regulatory expectations and come into compliance.
The 21st Century Cures Act, signed into law in 2016, excluded certain software functions from the FDA's statutory definition of a medical device, but in a final guidance last year, the agency explained it would still consider software a medical device, unless it meets four criteria: if it does not acquire a signal from an in vitro diagnostic device; if it is used to display medical information about a patient or data from peer-reviewed studies or practice guidelines; if it provides information that supports doctors in diagnosing and treating patients; and if doctors can independently review the basis of the recommendations presented by the software.
The first criteria in the guidance "that keeps you away from being a medical device is that you don't take raw lab data and convert it to genotype and phenotype calls, which is part of the service that Translational Software provided," said Kristine Ashcraft, medical affairs director of pharmacogenomics at Invitae, which in 2020 bought a PGx clinical decision support company she led called YouScript.
Based on the FDA's guidance, Ashcraft said YouScript avoided falling into the medical device category by starting with lab reports with genotype and phenotype calls that a medical director had already signed off on. The company then provided drug and dose optimization guidance supported by clickable references to published literature or drug labels, allowing healthcare providers to look up the evidence themselves and make an independent judgment. (Recently, Invitae, amid restructuring and layoffs, decided it would no longer provide PGx testing and sell YouScript, but Ashcraft said the decision was purely a financial one and not due to regulatory pressures.)
Coriell, another player in the PGx space, sent a letter to its customers last week acknowledging that the news about Translational Software "has left … customers, and the industry as a whole, with some uncertainty" about how to keep providing PGx insights to providers and researchers. In the same letter, Coriell also differentiated its services from Translational Software's by stating that it "does not provide services in the production of clinical laboratory reporting."
In an email, Coriell CEO Scott Megill maintained that the firm "does not do any diplotyping for laboratories as part of any laboratory clinical reporting" and that labs it works with are "contract bound" to process samples and produce reports that adhere to federal lab standards under the Clinical Laboratory Improvement Amendments. "If diplotypes are part of their CLIA report, they need to perform that alignment of variant call to diplotype on their own," he said.
Megill described Coriell as a professional services research organization where experts provide their opinion on the consensus and strength of evidence for variants identified by labs, but it doesn't provide a clinical diagnosis. "Users of our research reviews are provided all references for consideration and independent analysis," he said. "When required for our own internal literature review, we align any input data provided to publicly available, generally accepted tables for diplotypes."
For these reasons, Coriell told customers it isn't seeking 510(k) clearance for its research services. Other software companies contacted for this story did not reply to a request for comment before press time.
Phoenix's Ramey believes, however, that based on Translational Software's experience with the FDA, "all PGx [service] providers are at risk." Companies in the PGx space remember all too well the FDA's warning letter to Inova Health in 2019, asking the Virginia healthcare system to stop marketing PGx tests without the agency's approval, which it did. Subsequently, the FDA began contacting other PGx labs and asking why they were providing testing without its approval.
FDA enforcement actions like this against a single company in the past have been "warning shots" to the entire field to change its practices, Ramey observed. "We still, however, have zero guidance from the FDA as to what their evidentiary standards for PGx are or what they would deem acceptable," she said. "All we have is Don [Rule's] story, which is not actionable."
Looking for lessons in a failed 510(k)
Robert Boorstein, medical director of Brooklyn-based Lenco Diagnostic Laboratories, agrees that Translational Software's regulatory experience and shuttering of services could impact a "major" portion of the lab community doing PGx testing. "It is not clear what is going to happen to this entire sector," he said.
Outside of the FDA's efforts to bring certain clinical decision support platforms under its medical device definition, pure-play software services that analyze or interpret data from clinical labs aren't explicitly included in CLIA regulations, though workgroups within the Centers for Disease Control and Prevention's Clinical Laboratory Improvement Advisory Committee are considering updating regulations to specify oversight of such companies.
Boorstein, who also inspects labs for accreditation by the College of American Pathologists, said there are many standards for software validation and documentation in general checklists, including for software applied off-site to lab data, but there may be limited requirements in molecular testing checklists. "It is not clear to me the degree to which Translational Software and users complied with those standards, and the degree to which CAP enforced compliance," he said.
According to Rule, Translational Software's lab customers have used its software to answer questions for CAP's proficiency tests, but "the software validation is not considered, to my knowledge," he said. "Arguably, maybe it should be."
Several years ago, Lenco tried to contract with Translational Software for PGx test reporting but was thwarted by the New York State Department of Health, which said the software company was operating as a lab and would need to hire a lab director with the necessary qualifications, which the firm lacked. The Lenco deal fell through. "If [Translational Software] is going to give clinical recommendations, then it seems they have to be regulated by something, either under CLIA or by the FDA or by state laws on the practice of medicine," said Boorstein. It would not have been a trivial expense, he acknowledged, for the firm to hire a lab director and get licenses in multiple states, "but why should it be trivial? [Lab testing] is a big business," Boorstein said.
Rule pointed out that Translational Software has worked with other labs licensed in New York, but in those instances, he's not sure whether the difference was working with a different inspector or a change in rules or the labs provided some validation data on the software. Still, that NYSDOH experience with Lenco now seems like a presaging of the FDA's determination that Translation Software's platform is a medical device.
In meetings with Translational Software, the agency seemed particularly concerned about how its platform handles differences in diplotype calls made for different labs. PGx tests, even for the same indication and gene, may not gauge the same alleles, and while Translational Software pre-configures translation tables for labs based upon the specific alleles in their panels, sometimes a particular combination of variations may be explained by two different diplotypes.
When these ambiguities come up, the lab director has the option of reviewing the result and is responsible for electronically signing off on the report before releasing it to the ordering clinician. Such issues arise in less than 1 percent of cases, according to Rule, but the FDA felt it wasn't reasonable to expect that lab directors would be able to independently verify the accuracy of Translational Software's diplotype calls.
Based on the diplotype calls, the firm's platform also makes treatment recommendations using a curated knowledgebase, which contains information on gene-drug associations from FDA-approved drug labels, the published literature, and guidelines issued by expert groups such as CPIC. Translational Software described its knowledgebase as functioning like a library of gene-drug association data, but the FDA disagreed with this characterization, Rule said, since the company was providing patient-specific reports for multiple variants. Moreover, Rule said the FDA didn't recognize any of the expert databases the company was using and said the firm could only provide information that aligned with FDA-approved drug labeling.
The problem with this requirement, according to Lauren Marcath, clinical pharmacogenomics manager at Translational Software, is that the drug-gene associations that the medical community and even the FDA deem medically actionable may not be reflected in drug labeling. A few years ago, the FDA released a list of PGx variants that it has determined based on evidence in the published literature and data in FDA drug labels to impact patients' drug response, but experts have found drug labels that don't contain the PGx variant information that's in the agency's list.
For example, the FDA's list notes that CYP3A5 intermediate or normal metabolizers may have difficulty metabolizing Astellas' immunosuppressive drug Prograf (tacrolimus) and need dose adjustments. However, Prograf's label doesn't contain this information, and doctors looking at the label and not the FDA's list "are missing out on that safety warning," Marcath said. It becomes increasingly difficult to update labels of drugs that have gone generic, experts said.
Based on the FDA's feedback, Translational Software tried to validate the gene-drug associations in expert guidelines by going back to the primary literature and scoring the evidence, but Marcath said that wasn't sufficient. "The agency had concerns about making recommendations to drug management even as basic as '[this diplotype] may increase exposure to X drug,'" she said. "We're not entirely certain what threshold of evidence FDA was looking for, but it often felt like they wanted you to recreate the process that CPIC and other guidelines bodies had done."
The FDA has said that its PGx variant list is not comprehensive and is not meant to be used for regulatory purposes, so product developers are uncertain how to use it. The agency has recognized databases like Clinical Genome Resource consortium's ClinGen, which allows sponsors to cite variant data from the repository to support the clinical validity of their tests in regulatory submissions.
FDA recognition of PGx databases could potentially ease the path for companies seeking regulatory approval for tests and software products. CPIC has applied for FDA recognition of tables it publishes describing the functional impact of PGx variants, and PharmGKB is also seeking FDA recognition for its variant and clinical annotations.
However, according to some experts who asked to remain anonymous due to the sensitivity of these interactions, the FDA is unlikely to recognize CPIC's guidelines on how PGx variants inform treatment decisions since the agency doesn't see a role for itself in recognizing clinical dosing guidelines.
Unclear expectations
While FDA recognition of CPIC and PharmGKB's PGx variant data might ease the regulatory path for some PGx companies, the reality is that it takes time to achieve such recognition for a database, and at present, firms providing treatment recommendations based on detected PGx variants are unsure about the FDA's regulatory expectations. Translational Software's experience only seems to have caused more anxiety and confusion in the PGx space, especially as the FDA is intensifying its efforts to regulate lab-developed tests.
The agency recently proposed a rule to make explicit in regulations that LDTs are in vitro diagnostics that it can oversee as medical devices under the Federal Food, Drug, and Cosmetic Act. The rule, if finalized, will end the FDA's decades-long practice of enforcement discretion, under which labs can develop and perform tests on patients as long as they are CLIA-certified, and would impose risk-based requirements on LDTs, such as premarket review, adverse events reporting, and quality control measures.
Translational Software's Rule recognizes that the FDA wants to oversee LDTs out of concern that the present regulatory system isn't ensuring the accuracy of marketed tests and patients are being harmed. In the interest of public health, "some level" of additional oversight may be useful, he said, noting that Translational Software has also had to stop working with a lab customer that wasn't providing good quality testing.
But the point at which the FDA and Translational Software's regulatory discussions fell apart, according to Rule, is when the agency asked the company to guarantee the accuracy of the lab tests its software was pulling variant data from and wasn't willing to accept the competency of labs based on their CLIA certification or CAP accreditation.
"We're not averse to regulation," Rule said. Despite engaging with the FDA since 2016 to try to garner market authorization for Translational Software's platform, Rule still doesn't understand the agency's evidence expectations or if there's even a viable regulatory pathway for similarly situated software companies that analyze data from PGx tests — the vast majority of which are LDTs.
PGx labs are also closely tracking the FDA's efforts to regulate LDTs. At Resolve MDx, Warrington said there is "no chance in hell" the small-to-midsized regional lab would be able to pursue FDA review for its LDTs. "We are under such scrutiny from the payors. All our costs downstream are increasing, and we have no opportunity to raise our fees. … Healthcare laboratories are shutting down every single day," Warrington said. "The idea of us having resources or to even have a team of people go see the FDA is unrealistic."
Sara Rogers, a clinical assistant professor at Texas A&M University's college of pharmacy, agreed that the LDT approach allows PGx tests to remain up to date in a rapidly advancing field. "But we need standardization in certain areas," she said.
In 2020, in response to the FDA's warning letter to Inova and communications with other PGx labs about regulatory compliance, Rogers co-led the formation of Standardizing Laboratory Practices in Pharmacogenomics (STRIPE), a collaborative community that stakeholders in the PGx space formed to communicate their concerns to the FDA and establish consensus standards in the field. "The biggest challenge that stakeholders have expressed [in STRIPE] is that it's not always clear what the agency's expectations are," Rogers said, but added it is also well recognized in the industry that PGx testing and reporting differ among labs and the field would benefit from more alignment.
STRIPE will hold a conference in Washington, D.C., next year to try to build consensus around some of the challenges in PGx, including how the industry evaluates evidence on drug-gene associations. The FDA has advanced templates to help companies validate COVID-19 tests, and the STRIPE community similarly would like to work with the agency, Rogers said, to develop clear mechanisms that can ensure tests are safe and reliable. The group is also exploring if the National Institutes of Health's Genetic Testing Registry can be leveraged to identify PGx testing laboratories that are adhering to community standards.
It's not clear to what extent the STRIPE community will be able to influence the FDA's thinking on PGx test standardization. In the meantime, Invitae's Ashcraft is optimistic that most labs, if they follow the FDA's guidance on clinical decision support, will be able to figure out a PGx test reporting solution by year's end.
While Translational Software's regulatory experience "is not a good thing," and the FDA "should have created a path forward" for the firm, ultimately Ashcraft is hopeful that this only "a minor setback" for the PGx testing space, "not a death."