NEW YORK (GenomeWeb) – Prescient Medicine, a firm offering toxicology, pharmacogenomic, and other testing, has made available a new genetic testing service that predicts patients' risk for opioid addiction, which it recently described at the annual meeting of the American Association for Clinical Chemistry with its collaborator AutoGenomics.
The company also published an article in the most recent issue of Annals of Clinical & Laboratory Science, reporting 97 percent sensitivity and 87 percent specificity for the test — initially called the NeurR score and now called LifeKit Predict — in distinguishing opioid addicts from normal controls in a cohort of 138 individuals.
The pharmacogenomics of pain medicine has been a tricky area for other companies, most notably with Proove Biosciences' controversial launch last year of a test to help physicians determine whether or not patients are at risk of abusing opioid medications.
This June, federal investigators searched Proove's offices as part of an investigation related to "healthcare fraud," though it is unclear whether FBI scrutiny was related to the company's test quality, or to alleged kickbacks to doctors for enrollment in research studies.
Yale University professor Joel Gelertner, an expert in genetics and addiction who raised questions regarding Proove's test, expressed similar concerns about Prescient this week.
In particular, he called into question the value of the company's published data considering the small sample size studied. He also expressed skepticism that the reported predictive power would hold up when applied to larger datasets, and argued that in the absence of better validation, physicians should not use this type of testing.
Keri Donaldson, CEO and Medical Director of Prescient Medicine, said that the approach Prescient has taken with its own test is significantly different from that of Proove, despite the fact that the two products assay several of the same genes — Prescient's panel incorporates all but two of the genes in the Proove opioid risk test, along with five other unique targets.
Prescient has different business arms — collaborating in diagnostics development with other companies, and offering clinical testing services, which before the development of the new addiction assay included toxicology and drug testing and pharmacogenomics.
"We identified the need to help identify patients with a high likelihood of dependence before they get into that drug monitoring space," Donaldson explained.
Autogenomics, which Prescient has worked with before, had developed an algorithm in that vein, and Prescient worked with that company to further validate it.
"We worked with them to sharpen the knife, and now we are taking it forward," he said.
The company has also had a pharmacogenomics test on the market for some time already, called Lifekit Prescript.
At the time that Proove gained attention for its opiate addiction test, academic researchers in the field of addiction genetics expressed skepticism that the genes on its list could be incorporated into a polygenic score as predictive as the company claimed.
Proove had also not published data on its test validation at the time, but has since done so, reporting in the Journal of Addiction Research and Therapy.
According to Donaldson, skepticism about the predictive power of the genes in question is irrelevant to the accuracy of Prescient's own test because the algorithmic approach is agnostic of the actual genes involved.
Questioning Proove's test last year, experts raised that fact that risk loci and weighted genomic risk scores are best developed using an unbiased approach like GWAS, rather than by picking candidate genes that have a supposed biological relevance, as is the case in both Proove and Prescient's tests.
However, Donaldson argued, the fact that there may not be great science supporting the individual predictive ability of the genes or SNPs in question doesn't mean that they can't be predictive when assayed in the context of the right algorithmic method.
In other words, he said, in the end, it is the discriminatory ability of the classifier that is important.
Sherman Chang, AutoGenomics' vice president of research and development, reiterated this, saying that based on his understanding, Proove's test grades patients straightforwardly for each SNP in the panel based on their wild-type, heterozygous mutated, or homozygous mutated status. In contrast, the approach designed by AutoGenomics and now commercialized by Prescient weighs each variant differently.
The two companies each authored an abstract at the annual AACC meeting earlier this month discussing the test. In one, AutoGenomics authors describe their initial selection of the 16 genes in the panel based on relevance to brain reward pathways.
Company researchers then trained an algorithm using the panel in 70 patients diagnosed with prescription drug-induced opioid/heroin addiction and 68 normal control subjects.
The team designed a risk model that computes a score from one to 100, with any score over 52 representing an elevated risk of addiction. Of the 70 addicts tested in that study, 53 had an addiction risk score greater than 52. Among the healthy controls, 49 of 68 healthy scored under 52, and the resulting positive predictive and negative predictive values were both calculated at 74 percent.
In its own evaluation, Donaldson said that Prescient refined the algorithm. In the poster, authors report that they applied it to 37 patients with prescription opioid or heroin addiction and 30 age- and gender-matched individuals with no history of addiction. As also reported in the group's published study, the team then tested the assay on an additional 138 samples to determine its generalizability, and calculated a 97 percent sensitivity and 87 percent specificity.
But Yale's Gelernter reiterated some of the same concerns in regard to Prescient he did for Proove, despite the company's abstracts and published data.
He said in an email that based on his scientific experience, the report Prescient published in Annals of Clinical & Laboratory Science does not represent a large enough sample size to draw valid genetic-prediction conclusions.
"If they really want to assess their purported classification methods, they could download my GWAS data from dbGAP and all other relevant samples. Putting everything together they'll end up with [more than] 10,000 subjects … [and] they can then extract the marker data based on their algorithm and test if they are predictive in a sample size large enough to draw reasonable conclusions," he challenged.
Without that strength of evidence, he said, as a physician, "I would not use this test to classify genetic risk, and I would advise my colleagues to steer clear of it."
In addition, regardless of whether the predictive ability of the company's algorithm might hold true in a larger sample set, the data only speak to the test's ability to discriminate between known opiate addicts and normal individuals. To prove clinical utility — that the test can actually help at-risk patients avoid harmful drugs by predicting addiction before it happens — would require prospective data tracking the test results in patients as they are treated with these medications.
Donaldson's position, in contrast, is that the results of the company's published study do speak adequately to the ability of the test to classify patients.
He also said that Prescient has conducted several other studies evaluating the generalizability of the assay across different cohorts, with the parameter estimates remaining within the confidence interval that the company published in its recent paper.
This, he argued, is strong evidence that the algorithm does indeed distinguish opiate addicts from normal controls within the accuracy defined by the company's published study.
Donaldson agreed however, that the "meat on the bone" is the actual clinical questions that the test can answer. And the firm does have a number of studies in the works in that vein, he added, though they have not yet been published.
"We have partnered with payors and providers at some institutions to implement [the test] in the context of diagnostic workflows prior to certain surgeries," he explained. Another use the company is exploring with early adopting customers is the use of the test in patients who are seeking help with addiction, as a tool to help cement or shore up the rationale for treatment.
Finally, he added, Prescient is also investigating whether the same panel of genes can be used for addicts in other context like alcoholism, and what its use might be in more complicated contexts — for example patients with not only substance abuse disorders but also other mental health diagnoses.