Importance: Recently, the US Food and Drug Administration gave premarketing approval to an algorithm based on its purported ability to identify individuals at genetic risk for opioid use disorder (OUD). However, the clinical utility of the candidate genetic variants included in the algorithm has not been independently demonstrated.
Objective: To assess the utility of 15 genetic variants from an algorithm intended to predict OUD risk.
Design, setting, and participants: This case-control study examined the association of 15 candidate genetic variants with risk of OUD using electronic health record data from December 20, 1992, to September 30, 2022. Electronic health record data, including pharmacy records, were accrued from participants in the Million Veteran Program across the US with opioid exposure (n = 452 664). Cases with OUD were identified using International Classification of Diseases, Ninth Revision, or International Classification of Diseases, Tenth Revision, diagnostic codes, and controls were individuals with no OUD diagnosis.
Exposures: Number of risk alleles present across 15 candidate genetic variants.
Main outcome and measures: Performance of 15 genetic variants for identifying OUD risk assessed via logistic regression and machine learning models.
Results: A total of 452 664 individuals with opioid exposure (including 33 669 with OUD) had a mean (SD) age of 61.15 (13.37) years, and 90.46% were male; the sample was ancestrally diverse (with individuals of genetically inferred European, African, and admixed American ancestries). Using Nagelkerke R2, collectively, the 15 candidate genes accounted for 0.40% of variation in OUD risk. In comparison, age and sex alone accounted for 3.27% of the variation. The ensemble machine learning. The ensemble machine learning model using the 15 variants as predictive factors correctly classified 52.83% (95% CI, 52.07%-53.59%) of individuals in an independent testing sample.
Conclusions and relevance: Results of this study suggest that the candidate genetic variants included in the approved algorithm do not meet reasonable standards of efficacy in identifying OUD risk. Given the algorithm's limited predictive accuracy, its use in clinical care would lead to high rates of both false-positive and false-negative findings. More clinically useful models are needed to identify individuals at risk of developing OUD.