Genetic Variants in OUD Risk Algorithm Do Not Meet Standards in Identifying Risk

A recent study suggested that the genetic variants in the opioid use disorder (OUD) risk algorithm do not meet reasonable standards in identifying OUD risk.1

“We found no evidence to support the clinical utility of the 15 [single nucleotide variants] purported to predict OUD risk,” wrote investigators, led by Christal N. Davis, PhD, from the Center for Studies of Addiction at the University of Pennsylvania Perelman School of Medicine.

In 2022, 6.1 million US residents aged ≥ 12 years met the criteria for OUD; 94.8% acknowledged misusing prescription analgesics and 40.9% reported misusing medication from a physician. Since 2015, opioid overdose deaths have risen steadily, prompting efforts to identify individuals at risk.2

Research has shown that genetic variation (3.74%), such as single nucleotide variants across the genome, accounts for less OUD risk variance than sociodemographic factors (41.32%).1 Despite that, investigators have developed genetic risk algorithms for OUD, including causal gene candidates based on their presumed effect on neural reward systems.

The US Food and Drug Administration (FDA) recently gave pre-marketing approval to an OUD risk algorithm incorporating 15 single nucleotide variants. However, the package insert for the algorithm states, “…15 detected polymorphisms are involved in the brain reward pathways that are associated with OUD” but includes no citations to support the associations; all the gene variants were identified through candidate gene studies.

The clinical relevance of these genetic variants remains undetermined, with most lacking validation in genome-wide association studies. Genetic predictive models are also susceptible to bias stemming from variations in patterns of genetic similarity, such as individuals from different geographic origins.

Investigators conducted a case-control study to assess whether the 15 genetic variants were individually linked to OUD, how much variance in OUD risk the genetic variants accounted for collectively, whether the genetic variants were associated with genetic similarity rather than OUD risk, and whether basic demographic characteristics including age and sex more accurately predicted the OUD risk than the genetic variants.

The study included 452,664 participants in the Million Veteran Program across the US with opioid exposure. OUD cases were identified using diagnostic codes, with data collected from electronic health record data, such as pharmacy records.

Participants had a mean age of 61.15 years and 90.46% were male. The sample included a mix of European (67.46%), African (20.90%), mixed American (9.50%), East Asian (0.81%), and South Asian (0.07%) ancestries.

Models that did not account for genetic similarity showed that 13 of 15 variants were linked to OUD risk. Accounting for genetic similarity dropped this number to 3.

In analyses of local genetic similarity, 3 genetic variants were associated with OUD risk, but only in individuals genetically similar to the European superpopulation.

Logistic regression analyses revealed that the 15 candidate genes accounted for 0.40% of the variation in OUD risk (AUROC, 0.54). Among location-specific models, 7 genetic variants were linked to OUD in the European group, 2 in the African group, and 1 in the American group.

The ensemble machine learning model performed better at identifying the OUD risk with the 15 variants than random guessing. The model correctly classified 52.83% of participants with OUD (95% CI, 52.07 – 53.59%).

However, accuracy for location-specific analysis did not exceed random guessing: European (50.65%; 95% CI, 49.67% - 51.62%), African (50.53%; 95% CI, 49.09%-51.96%), and admixed American (49.69%; 95% CI, 47.21%-52.16%).

Age and sex alone accounted for 3.27% of the variation (AUROC, 0.66). In combined genetically inferred ancestry models, age and sex yielded more accurate estimates of OUD risk (59.49%; 95% CI, 58.82%-60.16%) than the 15 genetic variants.

The study ultimately showed that the candidate genetic variants included in the OUD risk algorithm do not meet reasonable standards, suggesting that using this algorithm in clinical care would result in high rates of false-positive and false-negative results.

“Notably, clinicians could better predict OUD risk using an individual’s age and sex than the 15 genetic variants,” investigators wrote. “Although the test approved by the FDA is intended to complement standard clinical assessment, its use is unlikely to confer additional benefits and may instead give clinicians and patients false and potentially harmful information.”

References

1. Davis CN, Jinwala Z, Hatoum AS, et al. Utility of Candidate Genes From an Algorithm Designed to Predict Genetic Risk for Opioid Use Disorder. JAMA Netw Open. 2025;8(1):e2453913. doi:10.1001/jamanetworkopen.2024.53913
2. Ahmad F, Cisewski J, Rossen L, Sutton P. Provisional drug overdose death counts. Updated August 14, 2024. Accessed January 14, 2025. https://www.cdc.gov/nchs/nvss/vsrr/drug-overdose-data.htm