There were two steps to use prevalence estimates of cannabis use to model cannabis dependence. Step one used DisMod-MR to model cannabis use and step two modelled cannabis dependence. Region-, sex- and year-specific cannabis use prevalence DisMod-MR output for the 20 to 44 age groups from step 1 were inserted into the cannabis dependence dataset. Cannabis use estimates were restricted to these age groups so as not to inflate the ratio of use to dependence and thereby make it difficult for DisMod-MR to derive a plausible fit to the cannabis dependence estimates. We used a ‘cannabis use’ study-level covariate to adjust estimates of cannabis use downwards towards its corresponding level if the studies produced survey estimates of cannabis dependence (see Text S1 for an example of adjustment impacts). Values for incidence were set to zero before age 13 and after age 60 because the disorder is rare in children under 13 years and very rarely occurs for the first time in persons older than 60 years[47]. Incidence was derived using prevalence and remission and assuming no elevated mortality[31]. For examples of outputs see Figure S2 .
