Modelling alcohol consumption patterns to enable policy impact assessment

Abstract

Objective To prevent harmful effects of alcohol use, various countries implement policies preventing excessive and heavy episodic drinking. To enable the evaluation of the impact of such policies on (future) drinking behaviour, we aimed to develop a model that predicts alcohol consumption patterns. Methods The model predicts alcohol use in three stages. First, a logistic submodel predicts probabilities of drinking any alcohol. Second, for drinkers, a submodel predicts the weekly consumption through a negative binomial distribution for the number of beverages. Finally, based on the predicted weekly consumption, a logistic submodel predicts probabilities of heavy episodic drinking. The distribution for the weekly consumption was calibrated, targeted to predict the prevalence of excessive and heavy episodic drinking accurately. Model parameters were estimated using Dutch individual-level cross-sectional survey data covering the years 2008–2022. The characteristics age, sex, education, calendar time and their interactions were used as predictors and the model accounts for trend breaks in the data. Model performance was assessed by comparing population-level predictions with observed data on which the model was calibrated (2014–2022). Results A comparison between predictions of the calibrated model and observed data shows that the prevalences of excessive (error <0.2 percent point (pp)) and heavy episodic drinking (error <0.1 pp) align, averaged over the years 2014–2022. Visual inspection using qq-plots and within-sample validation over time further indicates that the model fits well for predicting excessive and heavy episodic drinking, based on the predicted distribution for the weekly consumption. Conclusions We developed a model for alcohol consumption patterns based on Dutch data. This model enables evaluation of the impact of interventions on the (future) prevalence of excessive and heavy episodic drinking.