Can we use routine data to optimize screening for hazardous alcohol consumption in primary care?

Aims: The present study examined whether frequently routinely collected data, such as demographic characteristics and health status/behaviours of patients as well as traditional laboratory parameters, may be used to substitute or to optimize questionnaire-based screening strategies for hazardous drinking. Methods: Univariate tests as well as Classification and Regression Tree (C&RT) analyses were performed in a sample of 2940 German primary care patients to create practitioner-friendly algorithms that predict hazardous drinking, as defined by the Alcohol Use Disorders Identification Test (AUDIT), from routinely collected data (demographic characteristics, health status/behaviours, laboratory parameters). Several different data basis scenarios and models were tested and compared with each other as well as with reference models. Results: In univariate analyses male sex, younger age, living without a partner, higher education, unemployment, good physical status, smoking, and increased values in laboratory markers (mean corpuscular volume, gamma glutamyltransferase, aspartate aminotransferase, and alanine aminotransferase) were associated with an increased probability of hazardous drinking. C&RT analyses led to models defined by age, sex and smoking status. All routine data-based models proved to be significantly inferior to screening the total population with the AUDIT questionnaire. Conclusions: In spite of the associations of routinely collected demographic, health status/behaviour and laboratory data with hazardous drinking, screening strategies that are based solely on such information are not as accurate as the administration of screening questionnaires.