(PS1-40) Differences in Alcohol Consumption at Classifying Alcohol Use Disorder by Age and Sex: A NESARC-III Investigation

Purpose: A small number of studies examining the relationship between alcohol consumption and alcohol use disorder (AUD) status suggest that indicators of drinking quantity and frequency hold promise as AUD screening tools. However, there remains a gap in research concerning whether these metrics vary based on key demographic factors such as age and sex. The aims of this study were threefold: (1) identify consumption variables that perform better than chance at classifying AUD status among various demographic groups (i.e., young adults, adults, and older adults, females, males) and (2) examine which quantity/frequency variables are significant predictors at classifying AUD status when included in a model collectively, and (3) explore model efficacy at classifying AUD from consumption variables.

Methods: This study used data from the National Epidemiological Study of Alcohol Related Conditions III (NESARC-III). It included participants who reported drinking over the past year, responded to past-year AUD diagnostic criteria, and provided past year alcohol consumption information (N = 25,778, Female = 54%; AUD prevalence = 20%, M_age = 43). The following consumption variables were examined: Number of Drinking Days (N-DD), Drinks per Drinking Day (D-DD), Largest Number of Drinks per Drinking Day (LD-DD), Frequency of Drinking Five or More Drinks in a Day (DD-5), Frequency of Drinking Five or More Drinks in Two Hours or Less (DD-5-2), Frequency of Drinking Eight or More Drinks in a Day (DD-8), Frequency of Drinking Twelve or More Drinks in a Day (DD-12), and Frequency of Exceeding Daily Low-Risk Drinking Limits (FLRD). First, Receiver Operating Characteristic curves were generated for each variable within each demographic group. Next, significant variables were entered into a binomial logistic regression. Then, F1 scores were calculated to ascertain model efficacy. Analyses were conducted separately for young adults (aged 18-25), adults (aged 26-64), older adults (aged 65+), females, and males.

Results: Area under the curve (AUC) was statistically significant for all variables across demographic groups, except DD-12 for older adults (AUCs = 0.60 - 0.89, ps < .0001). Regressions revealed that while there was heterogeneity in statistically significant variables depending on sex and age, N-DD and DD-5 were consistently the strongest predictors across all demographic groups (B = 0.19 - 0.37). Models yielded the following F1 scores: 0.63 (young adults), 0.53 (adults), 0.23 (older adults), 0.47 (females), and 0.58 (males).

Conclusions: Taken together, this study reveals that while many consumption variables can classify AUD better than chance, when included in a model together, only a few key variables were of predictive importance. Further, models performed variably well at classifying AUD across demographic groups, with highest performance for young adult and male samples and lowest in older adults. Findings from this study suggest that practitioners using alcohol screening measures may need to exercise caution in interpreting screening measures, as differences in classification efficacy exist across different demographic groups.