We assumed that one pack/bag/can of rolling tobacco weighs 50 g, on the basis of the available data in 2008, when the information on sales was available in both bags/cans and in kilograms.15 16 We estimated this figure by dividing the total grams sold in 2008 by all the bags/cans sold that same year, resulting in 46.85 g. Using the rounded

figure of 50 g per unit of pack/bag/can, we were able Navitoclax Phase 2 to estimate the sales of rolling tobacco in kg of the product for the whole study period (1991–2012). We also collected data of the Spanish population ≥16 years old for the period 1991–2012, using the population censuses and the official intercensuses data (available up to 2012) from the National Statistics Institute.17 This information allowed us to estimate the average number of manufactured and RYO cigarettes per year and person.18 19 Since this information is public aggregated data and it does not contain data on individuals, ethical approval was not required. Since the amount of tobacco included in a unit of RYO cigarette is variable as it depends on the way the smoker makes the roll,12 we considered three possible weights to estimate the number of cigarettes: 0.5, 0.8 and 1 g of tobacco. For each option, we calculated the annual per cent of change (APC) of the number of cigarettes

per person and year for manufactured cigarettes, RYO cigarettes, and both type of cigarettes taken together. In order to assess changing trends during 1991–2012, we used joinpoint regression. According to the procedure developed by Kim et al,20 and based on the shape of the time trend of the daily cigarette consumption per capita, we assumed a maximum number of four joinpoints. To predict trends, we fitted an autoregressive Bayesian log-linear Poisson model to the observed data in 1991–2012. This model allows better predictions in situations where other models may fail20 and gives more weight to data from recent periods, especially when changing trends arise through the study period.21 In this line, the temporal trend was modelled through a random

walk (RW). We assessed the performance of the model comparing an RW of order 1, which assumes constant rate of changes, with an RW of order 2, which is a moving average that changes in time and allows for smoothing GSK-3 of the trend.21 We found that the model with RW of order 2 showed less variability in the within-sample prediction of the observed cigarettes per capita in 1991–2012, and then the RW of order 2 assumption was used (see online supplementary figure S1). Once the model was fitted, we predicted the cigarette consumption for the period 2013–2020, based on the time trend estimated with this Bayesian model. Results The daily consumption per capita of manufactured cigarettes decreased from 7.6 units in 1991 to 3.8 units in 2012, with an average APC of −3.0 (figure 1).