Accuracy of Alcohol Use Disorders Identification Test for detecting problem drinking in 18-35 year-olds in England:...

To assess the accuracy of Alcohol Use Disorders Identification Test (AUDIT) scores for problem drinking in males and females aged 18-35 in England.

A method comparison study with 420 primary care patients aged 18-35. Test measures were AUDIT and AUDIT-C. Reference standard measures were (a) Time-Line Follow-Back interview for hazardous drinking; World Mental Health Composite International Diagnostic Interview for (b) DSM-IV alcohol abuse, (c) DSM-IV alcohol dependence, (d) DSM-5 alcohol use disorders.

Area under the curve (AUC) was (a) 0.79 (95% CI 0.73-0.85; males) and 0.84 (0.79-0.88; females); (b) 0.62 (0.54-0.72; males) and 0.65 (0.57-0.72; females); (c) 0.77 (0.65-0.87; males) and 0.76 (0.67-0.74; females); (d) 0.70 (0.60-0.78; males) and 0.73 (CI 0.67-0.78; females). Identification of threshold cut-point scores from the AUC was not straightforward. Youden J statistic optimal cut-point scores varied by 4-6 AUDIT scale points for each outcome according to whether sensitivity or specificity were prioritized. Using Bayes' Theorem, the post-test probability of drinking problems changed as AUDIT score increased, according to the slope of the probability curve.

The full AUDIT scale showed good or very good accuracy for all outcome measures for males and females, except for alcohol abuse which had sufficient accuracy. In a screening scenario where sensitivity might be prioritized, the optimal cut-point is lower than established AUDIT cut-points of 8+ for men and 6+ for women. Bayes' Theorem to calculate individual probabilities for problem drinking offers an alternative to arbitrary cut-point threshold scores in screening and brief intervention programmes.