[Background]: Alcohol consumption has widespread effects on health and is linked to over 200 detrimental conditions. Reliable age- and sex-specific population estimates of trends in alcohol consumption are key for designing prevention activities and the efficient planning of treatment services. Surveys are the main source of information in this regard, but almost invariably underestimate the real consumption. [Methods]: We propose a novel rescaling method to adjust survey data for underreporting and present an application of the methods to the estimation of trends in alcohol use in South Africa between 1998 and 2016. We used a Bayesian meta-regression approach to combine information from 17 nationally representative population surveys with administrative data on total alcohol consumption at country level. We argue that our implementation provides improved quantification of the uncertainty associated with the data and the method itself and requires less stringent assumptions compared to previous approaches. [Results]: Between 1998 and 2016 the prevalence of drinkers in South Africa has substantially increased among young men and women and decreased in the older age groups Overall, the prevalence was slightly higher in 2016 than in 1998 for women (20% vs. 19%) and slightly lower for men (50.6% vs. 53.9%). The average daily consumption among drinkers decreased from 4.3 standard drinks per day to 3.6 drinks per day for men and from 2.7 drinks to 2.2 drinks for women. The most substantial decrease happened among men in the 25-34 age category, with a reduction by more than 35%, from 6.8 to 4.4 drinks per day. Since 1998, the proportion of heavy drinkers has constantly decreased in the older age groups, especially among women. Among the young of both sexes, after an initial decrease, the proportion of heavy drinkers has started increasing after 2010, at the expense of the proportion of light drinkers.
