Cross-Tabulating Epidemiological Covariates with AUDIT-C Data in Large-Scale Biobanks

Abstract

Background: The Alcohol Use Disorders Identification Test-Consumption (AUDIT-C) is a widely used screening tool in large-scale electronic health record (EHR) biobanks. However, its categorical, range-based survey responses pose a major challenge for epidemiological research, particularly when continuous quantitative variables are preferred. Standard workarounds, such as assigning categorical midpoints or using aggregate ordinal scores for regression, may create spurious mathematical precision or obscure important behavioral differences between drinking frequency and quantity. Objectives: To develop and validate a novel two-dimensional cross-tabulation framework and systematic bounding algorithm for analyzing and presenting categorical alcohol survey data. Methods: Two complementary descriptive techniques were developed: 1) a two-dimensional cross-tabulation matrix that preserves the interaction between drinking frequency and typical quantity and 2) a systematic bounding algorithm that applies time-interval correction factors to calculate strict lower and upper estimates of average daily alcohol consumption. To demonstrate real-world utility, these methods were applied to three analytical scenarios within a European ancestry (EUR) cohort in the All of Us Research Program: generalized anxiety disorder (GAD) prevalence (n = 104,893), minor allele frequency (MAF) for the rs1229984 genetic variant (n = 104,890), and self-reported active duty military service history (n = 104,893). Results: Application of the cross-tabulation matrix revealed patterns across all three descriptive scenarios. For example, participants reporting the highest drinking frequency ("4 or more times a week") combined with the highest quantity ("10 or more" drinks) had a GAD prevalence of 13.5%, compared with 5.8% among those reporting the same frequency but a low quantity ("1 or 2" drinks). Anxiety generally increased with drinking quantity but decreased with drinking frequency. Participants who reported having served on active duty in the US Armed Forces reported more frequent drinking and higher average daily consumption estimates (0.339 to 0.875) than those who had not (0.297 to 0.770). Conclusions: This bounding and mapping framework provides researchers with an additional method to complement traditional midpoint and aggregate scoring approaches. By explicitly defining the uncertainty inherent in categorical survey instruments and visualizing cohort distributions across intersecting behavioral axes, this methodology enhances the resolution, reproducibility, and interpretability of lifestyle exposure data.