The use of narcotics and opioid use disorder (OUD) are both critical public health issues. In 2020, 6.1 million Americans aged 12 and older were diagnosed with OUD. Of these, 94.8% admitted to abusing prescription analgesics, and 40.9% misused prescribed drugs from medical professionals. The heightening overdose mortality rates stemming from opioids underline the attempt to identify high-risk opioid misusers. Common genetics account for only a fraction of variation in the liability to OUD. Scores based on commensurate genetic variation in common risk loci (i.e. single nucleotide variants (SNVs)) across the genome account for considerably less OUD trait variance (3.74%) than do sociodemographic traits (41.32%).
Despite this, efforts have been made to develop and commercialize genetic risk algorithms for OUD. Such models typically specify a couple of SNVs in reputedly causal gene candidates based on their presumed impact on neurological reward circuits. Even though these candidate variants often boast minuscule effects, very few have been validated by genome-wide association studies (GWAS), which are the gold standard for identifying risk alleles.
This case-control study aimed to investigate 15 candidate genetic variants associated with opioid use disorder (OUD) using electronic health record data from December 20, 1992, to September 30, 2022. The data was collected from participants enrolled in the Million Veteran Program within the range of the USA with opioid exposure (n = 452,664). The presence of OUD in the cases was ascertained by the International Classification of Diseases, Ninth Revision, or International Classification of Diseases, 10th Revision diagnostic codes, while controls were those with no such diagnosis of OUD.
The list was revised according to research studies that found that 452,664 opioid-exposed individuals should include 33,669 cases of OUD. The mean (SD) age was 61.15 (13.37) years, wherein 9.54% were female and 90.46% were male. This group had a genetically inferred ancestry (GIA) composition that was assigned based on patterns of similarity to the reference genomes of individuals in the 1000 Genomes project. It was 67.46% European, 20.90% African, 9.50% admixed American, 0.81% East Asian, and 0.07% South Asian, with 1.25% unassigned. A total of 125,514 individuals, including 3704 OUD cases, exposed to opioids for the short term had a mean (SD) age of 59.98 (14.84) years, 9.63% were female, and 90.37% were male. The GIA composition of this subset of the entire population is around 67.59% European, 19.22% African, 10.36% admixed American, with 1.25% East Asian 0.11% South Asian, and 1.47% unassigned.
In single-SNV models that did not account for global genetic similarity, 13 of 15 SNVs were significantly associated with OUD risk, as per Bonferroni corrected significance, which dropped to three after including measures of global genetic similarity. Five of 10 SNVs no longer associated with OUD risk demonstrated opposing considerations in uncontrolled analyses. When analyzed by GIA, three SNVs did not have any impact on OUD risk within certain genetic similarities and were statistically associated only with European superpopulations. Analyses of individuals who had short-term exposure to opioids produced similar results.
The study found no evidence to support the clinical utility of the 15 candidate SNVs purporting to predict the risk of OUD in a diverse sample of over 450,000 opioid-exposed patients, including 33,669 individuals with OUD in a case-control study. Altogether, the SNVs explained 0.40% of the variation in the risk of OUD, equating to the small individual effects that common genetic variants sometimes have on complex traits. A high incidence of false-positive or false-negative results occurred during independent testing in a subgroup: 47 out of 100 predicted cases and controls were misclassified. False-positive results contribute to stigma, undue patient concern, and biased health decisions.
For example, the Million Veteran Program (MVP) is predominantly male, allowing for analysis in more than 40,000 cases of women (more than 2500 women with OUD), and the numbers are all well in excess of the entire sample from which the mean deviation was calculated and on which the genetic algorithm was trained and tested (female participants: n = 1,762, OUD cases: n = 653). Also, there are high prevalences of OUD and pain, with an older age distribution than is seen typically in documents as we urge that the 15 SNVs be evaluated in additional datasets. Third, we used this GIA group as a population descriptor.
As genetic risk models in psychiatry continue to evolve, some may prove clinically useful, so it would make sense for research and regulatory reviews to establish rigorous validation and evaluation before considering an application in clinical settings. When considering a genetic risk algorithm using advanced statistical approaches for regulatory review, the guide should be left to scientific advisers while always testing the credibility by independent validation. An excellent ethical evidence-based approach uses high standards to minimize bias.
Reference: 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
