Predictors of alcohol screening quality in a US general population sample and subgroups of heavy drinkers

Abstract

Alcohol screening is one of the most cost-effective clinical preventive services and important for intervening in the development of alcohol problems. We examine predictors of the quality of alcohol screening, approximated by alcohol quantity screening, which is a prerequisite for appropriate counseling, and compare conventional regression approach with Classification and Regression Trees (CART). Data come from the 2020 National Alcohol Survey, a population survey of US adults aged 18 years and over. Analyses focus on those reporting any alcohol screening at all (N = 989). The primary outcome was whether a healthcare profession had ever asked how much they drink, which is necessary to identify heavy drinking. We examined 12 potential predictors of alcohol quantity: gender, age, race and ethnicity, education, marital status, having a usual source of primary care, insurance, and health conditions. Analyses were replicated in heavy episodic drinking (HED) and high intensity drinking (HID) subgroups, both warranting alcohol counseling. Logistic regression results show that having diabetes and not having a college degree predict missed alcohol quantity screening in the sample overall, and younger age predicts missed alcohol quantity screening in the HED/HID subgroups. CART identified Black and Hispanic respondents who had not attended college at high risk of missed screening for heavy drinking in the overall sample, and those with public insurance at high risk of missed screening for heavy drinking in the HED/HID subgroups. The quality of alcohol screening needs improvement in general, and to avoid unintended disparities in alcohol-related health services.

Keywords

• Alcohol screening
• High intensity drinking
• Heavy episodic drinking
• Machine learning
• Classification and regression trees