Diagnostic Validity of Drinking Behaviour for Identifying Alcohol Use Disorder: Findings from a Nationally Representative Sample of Community Adults and an Inpatient Clinical Sample

Background and Aims: Alcohol consumption is an inherent feature of alcohol use disorder (AUD), and drinking characteristics may be diagnostically informative. This study had three aims: (1) to examine the classification accuracy of several drinking quantity/frequency indicators in a large representative sample of U.S. community adults; (2) to extend the findings to a clinical sample of adults; and (3) to examine potential sex differences. Design: In cross-sectional epidemiological and clinical datasets, receiver operating characteristic (ROC) curves were used to evaluate diagnostic classification using area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Measurements: Classifiers included measures of quantity/frequency (e.g., drinks/drinking day, largest drinks/drinking day, number of drinking days, and heavy drinking frequency). The clinical criterion (reference standard) was AUD diagnostic status per structured clinical interview (community sample) or a symptom checklist (clinical sample). Setting and Participants: Two samples were examined: a large, nationally representative random sample of U.S. community adults who reported past-year drinking (N=25,778, AUD=20%) and a clinical sample from a Canadian mental health and addictions inpatient treatment centre (N=1,341, AUD=82%). Findings: All drinking indicators performed much better than chance at classifying AUD (AUCs=0.60-0.92, ps<.0001). Heavy drinking frequency indicators performed optimally in both the community (AUCs=0.78-0.87; accuracy=0.72-0.80) and clinical (AUC=0.85-0.92; accuracy =0.77-0.89) samples. Collectively, the most discriminating drinking behaviors were number of heavy drinking episodes and exceeding drinking low-risk guidelines. No substantive sex differences in optimal cut-offs or variable performance were observed. Conclusions: Quantitative drinking indices performed well at classifying AUD in both a nationally representative and large inpatient sample, robustly identifying AUD at rates much better than chance and above accepted benchmarks, with limited differences by sex. These findings broadly support the potential clinical utility of quantitative drinking indicators, such as routine patient assessment via electronic medical records.