Usefulness of the Korean Version of the CAGE-Adapted to Include Drugs Combined With Clinical Predictors to Screen for Opioid-Related Aberrant Behavior

BACKGROUND: As national opioid consumption in South Korea has soared, well-validated screening tools for opioid use disorder (OUD) have become indispensable. The aims of our study were to evaluate OUD using the Korean version of the CAGE-Adapted to Include Drugs (CAGE-AID) and the CAGE-Opioid (an alternative version of the CAGE-AID), and to investigate clinical predictors that might be useful to screen for OUD in conjunction with the CAGE-AID/Opioid questionnaires. METHODS: A single-center, prospective, observational study was performed. After linguistic validation of the Korean version of the CAGE-AID/Opioid questionnaires, we assessed OUD in patients with chronic opioid treatment. Multivariable logistic models of the CAGE-AID/Opioid questionnaires combined with relevant clinical predictors were established. Then, the receiver operating characteristic curve analysis of the multivariable CAGE-AID/Opioid models was conducted to assess diagnostic accuracy to screen for OUD. Next, we calculated predicted probability with >85% sensitivity and >50% specificity in each CAGE-AID and CAGE-Opioid model. Using the optimal value of the predicted probability, a cutoff score of the CAGE-AID/Opioid questionnaires combined with the relevant clinical factors was suggested to screen for OUD. RESULTS: Among 201 participants, 51 patients showed ≥1 OUDs. In the multivariable regression model, male sex, comorbid neuropsychiatric disorder, and current heavy drinking significantly remained as clinical variables to predict OUD combined with the scores of the Korean CAGE-AID/Opioid questionnaire. The area under the curve was 0.77 (95% confidence interval, 0.71-0.83) for the CAGE-AID model and 0.78 (95% confidence interval, 0.71-0.83) for the CAGE-Opioid model. The optimal predicted probability values to screen for OUD in the CAGE-AID/Opioid models were >0.135 (sensitivity, 0.86; specificity, 0.52) and >0.142 (sensitivity, 0.86; specificity, 0.53), respectively. When we used these predictive probabilities, the cutoff score of the CAGE-AID/Opioid questionnaires ranged from 0 to 3, which was dependent on the presence of the relevant clinical variables in each model. CONCLUSIONS: In this study, one fourth of the total participants with chronic opioid treatment showed OUD in the Korean population. The multivariable models of the CAGE-AID/Opioid with sex, comorbid neuropsychiatric disorder, and current heavy drinking are valid parameters to screen for OUD, with the cutoff scores of the CAGE-AID/Opioid questionnaires ranging from 0 to 3 depending on the presence of the clinical variables.