The PRISMA flow diagram is illustrated in Figure 1. From an initial set of 52,894 studies, 134 studies were short-listed for extraction following full-text screening. Following the extraction process, an additional 29 studies were excluded given concerns related to pooled samples (i.e., individual samples recruited for different studies that differed in selection criteria), and sample overlaps. In the latter case we retained comorbidity data from whichever reference reported on the largest sample from a given study, for each comorbid diagnosis. We chose to include comorbidities that had at least 5 studies reporting their prevalence within their samples. Among the 105 studies finally identified, 6 had community based recruitment. Pooling results from clinic- with community-based studies may suffer from biases given the conceptual and design differences in these two kinds of research. We have used only clinic-based studies, which reported lifetime comorbidities (n = 91), in our meta-analysis. Findings from community based studies have been qualitatively summarised separately. All selection decisions were reached by consensus.
