In the United States, more than 400,000 hospital admissions per year for upper GI bleeding are estimated to occur, and mortality ranges between 3% and 14%; this has not changed in the past 10 years and increases with increasing age. Known risk factors for peptic ulcer bleeding are non-steroidal anti-inflammatory drugs use and Helicobacter pylori infection. More recently, selective serotonin reuptake inhibitors have been identified as another risk factor, and since then, 15 additional studies �Cincluding this one�C addressing this topic have been carried out. Albeit four studies found a strong significant association between SSRIs and upper GI bleeding, others found no association at all; thereby, the association remains a matter of controversy. The widespread use of antidepressants, particularly SSRIs, makes even small risks account for a large number of cases, converting this problem into an important public health issue. This fact, along with the lack of consistency of the findings in the studies carried out so far, has aroused a great interest on this subject. Our study has been conducted with carefully collected information to further understand the relationship between SSRIs and upper GI bleeding in the framework of a Perifosine clinical trial general study on risk factors of upper GI bleeding. In the construction of the model, patients were taken as level one, strata as level two, and hospitals as level three. In the estimation of the models we used the lmer function, implemented in the context of the lme4 R package. This function performs the fit by using the Laplace approximation, and the correlation structure within the random effects has only an additional constraint: the variance-covariance matrix must be symmetric and positive semidefinite. To construct these models, the absence or lowest level of exposure was taken as the reference category. We performed a bivariate analysis, using the CX-4945 variables of exposure and potential confounders; and a multivariate analysis, including those independent variables which yielded a statistical significance of less than 0.2 in the bivariate analysis. The independent variables with the highest level of statistical significance were successively eliminated from the original model, provided that the coefficients of the principal variables of exposure changed by no more than 10% and Schwarz��s Bayesian Information Criterion. In addition to each antidepressant group, the following variables were included in the model: alcohol and caffeine consumption, past history of GI disorders, family history of GI bleeding, osteoarthritis, number of medicines taken and the use of NSAIDs, salicylates, proton pump inhibitors, H2 antihistamines, antacids, antiplatelet agents and anticoagulants. The protocol of the study was approved by the corresponding Ethics Committees of the participating hospitals. All patients were asked for an informed consent and, to be included in the study, they had to give a written informed consent. When our study protocol was first prepared, four studies had been published; three of those studies found odds ratios of 3 or more and only one, clearly heterogeneous in design, found no risk; thus, our study had power to detect an odds ratio of 2 and failed to detect an association between SSRI exposure and occurrence of upper GI bleeding of this magnitude; in it, we have found a clear association with NSAIDs in the range of that reported in the literature. When considering the different groupings of antidepressants there were no differences.