Clinical Performance of a Gene-Based Machine Learning Classifier in Assessing Risk of Developing OUD in Subjects Taking Oral Opioids: A Prospective Observational Study

OBJECTIVE

To reduce the incidence of Opioid Use Disorder (OUD), multiple guidelines recommend assessing the risk of OUD prior to prescribing oral opioids. Although subjective risk assessments are available to help classify subjects at risk for OUD, we are aware of no clinically validated objective risk assessment tools. An objective risk assessment based on genetics may help inform shared decision-making prior to prescribing short-duration oral opioids.

METHODS

A multicenter, observational cohort of adults exposed to prescription oral opioids for 4-30 days was conducted to determine the performance of an OUD classifier derived from machine learning (ML). From this cohort, the demographics of the U.S. adult opioid-prescribed population were used to create a blinded, random, representative group of subjects (n=385) for analysis to accurately estimate the performance characteristics in the intended use population. Genotyping was performed via a qualitative SNP microarray on DNA extracted from buccal samples.

RESULTS

In the study subjects, the classifier demonstrated 82.5% sensitivity (95% confidence intervals: 76.1%-87.8%) and 79.9% specificity (73.7-85.2%), with no statistically significant differences in clinical performance observed based on gender, age, length of follow-up from opioid exposure, race, or ethnicity.

CONCLUSION

This study demonstrates an ML classifier may provide additional objective information regarding a patient's risk of developing OUD. This information may enable subjects and healthcare providers to make more informed decisions when considering the use of oral opioids.

Interference From Anti-Streptavidin Antibody

Immunoassays are commonly used for clinical diagnosis, although interferences have been well documented. The streptavidin-biotin interaction provides an efficient and convenient method to manipulate assay components and is currently used in several immunoassay platforms. To date, there has been no report in the literature of interference from endogenous anti-streptavidin antibodies; however, such antibodies would potentially affect multiple diagnostic platforms. We report results from a patient being treated for thyroid dysfunction who demonstrated a T-uptake result of less than 0.2 and a nonlinear thyroid stimulating hormone dilution that suggested an immunoassay interference. Protein-A sepharose pretreatment corrected the nonlinear dilution and revealed an interference trend of falsely decreased results, as measured by sandwich assay, and falsely elevated results, as measured by competitive assay. The results of streptavidin-agarose adsorption were comparable to adsorption with protein-A sepharose. To our knowledge, this is the first published description of an endogenous anti-streptavidin antibody interfering with clinical laboratory assays.