Since all these studies have been conducted during different periods, in different geographical areas and healthcare settings, with different designs and different ways of collecting information, the explanation of those different estimates may lie on this heterogeneity; nevertheless, retrospective or prospective character seems to be a remarkable factor of heterogeneity; thus, only 1 out of 4 prospective studies published so far found a significant risk while, in the retrospective studies, it was 9 in 12 which found a significant risk. It is possible that, since most of these retrospective studies have been conducted with pre-existing information �Cnot collected for the purposes of these particular studies�C some relevant confounding factors might be difficult to adjust for. For instance, in these studies neither there is information upon intake of medication, nor upon exposure to OTC drugs �Cdepressive patients are more prone to self-medicate, nor upon co-morbidities not stated that may lead to the intake of non-prescription NSAIDs, nor, in some studies, upon alcohol intake. In our study, we could control for self-medication with NSAIDs; thus, when restricted to non self-medicated patients, the risk slightly decreased denoting a certain influence of selfmedication. However, it cannot be ruled out that self-medication, with a differential distribution between cases and controls, might have some influence on other studies. Furthermore, it has been pointed out that Selumetinib observational studies, in which cases are collected from hospitals and controls are non-hospitalised 
patients, might be affected by a selection bias. In the study by de Abajo et al., the prevalence of current use for acid-suppressing drugs was 19.5% in case patients and 11.2% in controls; therefore, the authors state that these figures suggest an important confounding factor by indication; in a broader approach, they might be also interpreted as a risk marker denoting differential severity between cases and controls impossible to fully adjust for in that study. Selection bias is consistent with the fact that none of the studies in which cases and controls came from hospitals found a significant risk. Thus, a possible explanation for the risk detected in some studies would be that, since depression is EX 527 currently associated with more morbidity and also with past use of antidepressants, these drugs in turn might be spuriously associated with bleeding. Had the SSRIs caused upper GI bleeding through a mechanism related to serotonin reuptake inhibition, other bleeding complications might also appear; however, there is no consistency at this regard. While two epidemiological studies upon abnormal bleeding and perioperative blood transfusion, respectively, suggested an increased risk related to SSRIs, in other 6 studies upon hemorrhagic stroke, postpartum haemorrhage and hemorrhagic events in different locations, no association was found. Depletion of serotonin from platelets caused by SSRIs has been currently postulated as the most likely mechanism for bleeding. Accordingly, inhibition by SSRIs of serotonin reuptake by platelets is thought to lead to reduced platelet serotonin levels and it would lead to diminished serotonin release from platelets on activation and to decreased platelet aggregation. In the study by serotonin concentration in the platelets decreased.