Fact checked byHeather Biele

Read more

February 05, 2025
5 min read
Save

Use of genetic variants to predict opioid use disorder risk may lead to inaccurate results

Fact checked byHeather Biele

Key takeaways:

  • The 15 genetic variants assessed in the study accounted for 0.4% of the variation in OUD risk.
  • A new machine learning model composed of the variants correctly classified 52.83% of individuals.

Candidate genetic variants included in a genetic risk algorithm may not meet standards of reasonable clinical efficacy in identifying opioid use disorder risk, according to a study published in JAMA Network Open.

“The FDA recently gave premarketing approval to an algorithm (AvertD, SOLVD Health) that incorporates 15 single nucleotide variants to predict opioid use disorder (OUD) risk,” Christal N. Davis, PhD, postdoctoral fellow at the Corporal Michael J. Crescenz Department of Veterans Affairs Medical Center in Philadelphia, and colleagues wrote. “The package insert for the algorithm states that the ‘... 15 detected polymorphisms are involved in the brain reward pathways that are associated with OUD...,’ but it provides no citations to support the associations, all of which appear to have been identified through candidate gene studies.”

Psych0125Davis_Graphic_01
Data were derived from Davis CN, et al. JAMA Netw Open. 2025;doi:10.1001/jamanetworkopen.2024.53913.

Study methods, results

Davis and colleagues sought to examine the association between the 15 genetic variants AvertD includes — without using its proprietary algorithm — and the risk for OUD.

They used electronic health record data from Dec. 20, 1992, to Sep. 30, 2022, to conduct a case-control study of 452,664 U.S. veterans (mean age, 61.15 years; 90.46% men) in the Million Veteran Program with opioid exposure, including 33,669 who had OUD.

Genetically inferred ancestry was assigned based on patterns of similarity to reference genomes of individuals in the 1000 Genomes Project; ancestry groups included European (67.46%), African (20.9%), admixed American (9.5%), East Asian (0.81%) and South Asian (0.07%), with 1.25% unassigned.

Exposures included the number of risk alleles present across the 15 genetic variants.

Logistic regression models showed that the 15 single nucleotide variants collectively accounted for 0.4% of the variance in OUD status (area under the receiver operating characteristic curve [AUROC] = 0.54). Across the ancestry groups, the single nucleotide variants accounted for variances ranging from 0.04% in the African group to 0.16% in the admixed American group.

On the other hand, age and sex alone accounted for 3.27% of the variance in OUD status (AUROC = 0.66), according to the researchers.

Overall in the combined group, the researchers observed that an ensemble machine learning model using the 15 variants as predictive factors correctly classified 52.83% (95% CI, 52.07-53.59) of individuals, a rate they called “slightly greater than random guessing.” Age and sex alone had a higher estimation, at 59.49% (95% CI, 58.82-60.16).

Test data of the ensemble machine learning model revealed a sensitivity of 50.72% and specificity of 54.95%, compared with the manufacturer-reported figures of 82.76% for sensitivity and 79.23% for specificity.

The researchers also pointed out that when accounting for local variations in genetic similarity, only three of the original single nucleotide variants remained associated with OUD risk, all of which only showed such associations in the European population.

“Although the AvertD test uses a proprietary algorithm, the issues identified herein suggest that the manufacturer has a fundamental misunderstanding of genetic principles, particularly the impact of differences in population structure and allele frequency,” the researchers wrote.

To reflect the approach used by SOLVD Health to evaluate AvertD, the researchers also selected a subset of individuals with short-term documented exposure to opioids, defined as 4 to 30 days, results of which were similar to the prior analyses.

“These findings underscore the need for more robust and complete data, particularly given the complex nature of psychiatric conditions, including OUD,” Henry R. Kranzler, MD, professor of psychiatry and director of the Center for Studies of Addiction at University of Pennsylvania’s Perelman School of Medicine, and one of the study’s authors, said in a related university press release. “The potential harms deriving from a faulty genetic test for OUD include both false negatives and false positives.”

Response from SOLVD Health

In a statement from SOLVD Health, Ron McCullough, PhD, MBA, senior vice president of clinical operations, addressed Davis and colleagues’ findings and expressed confidence over “the clinical validity and rigor of AvertD,” emphasizing that the genetic risk test empowers physicians to make prescribing decisions for patients with acute pain as part of a complete clinical evaluation prior to the first prescription of oral opioids.

“The researchers in the recent JAMA publication did not have access to our technology (AvertD), instead applying their own unvalidated model; therefore, any comparisons or conclusions in the article to AvertD are invalid,” McCullough said in the statement. “The study used only 50% of the genetic markers included in the FDA-authorized AvertD test, while the remaining markers were estimated with a 20% error rate per genotype. Due to error propagation, four out of five patients in the study may have incorrect genotyping, significantly undermining the study’s accuracy.”

In an interview with Healio, McCullough also highlighted that in addition to the social and psychological components of opioid addiction, genetics play a large role in addictive behaviors.

“The current modalities for opioid risk assessment are often subjective, and that puts a lot of burden on physicians to really think about those things,” McCullough told Healio. “This is another arrow in the quiver that has a more objective result, so it’s important to let health care providers know about the complexity of addiction and that genetics play a large factor in it.”

In response to Davis and colleagues’ comment that the package insert for AvertD does not provide citations to support the associations between the detected polymorphisms and the brain reward pathways associated with OUD, McCullough told Healio that all supporting evidence for these genetic associations can be found in the FDA submission for AvertD.

The researchers also questioned the genetic diversity of the sample used to develop AvertD, but McCullough informed Healio that their 8-year post-approval study, which launched last year, focuses directly on the distribution of ethnic populations.

McCullough noted that the study published in JAMA Network Open had several limitations, confining its applicability.

“The research relied on nonvalidated data, biased study populations and methods inconsistent with established research and clinical standards,” he said in the statement. “The study also relied on ICD codes — which have a 29% inaccuracy rate in similar populations, meaning nearly one-third of patients were likely misclassified — leading to distorted results,” he told Healio. “The FDA-approved study for AvertD used clinically validated DSM-diagnoses — which are the gold standard in OUD risk assessment — ensuring an accurate representation of at-risk patients.

“These limitations undermine their study’s conclusions, which contrast sharply with the robust validation and regulatory review behind AvertD,” McCullough added. “Additionally, the researchers acknowledge numerous conflicts of interest, calling into question their motivations to publish this study. We encourage independent experts to evaluate these discrepancies and remain committed to advancing proactive health care solutions addressing the opioid crisis.”

Reference: