As in other studies, our results show the gastrointestinal tract as the most frequent origin of major hemorrhages. However, the location within the gastrointestinal tract has been different, with a lower percentage of upper digestive hemorrhages compared to what has been published by other groups.7 In the RELY study, 47% of major gastrointestinal hemorrhages were lower, compared with 53% of upper gastrointestinal hemorrhages.8 In a subanalysis of the ARISTOTLE study, upper digestive hemorrhages accounted for 59% of digestive hemorrhages, compared to 41% of lower gastrointestinal tract, hemorrhoids, and rectum,9 similar figures to those of the ENGAGE-AF TIMI 48 study.10 A subanalysis of the ROCKET study showed the following distribution: 48% upper gastrointestinal tract and 52% lower gastrointestinal tract and rectum.11 Previous studies performed with VKA showed higher rates of upper gastrointestinal hemorrhages, up to 75%,12 and thus treatment with proton pump inhibitors was recommended.

The fear of digestive hemorrhages in anticoagulated patients leads to a very generalized prescription of proton pump inhibitors. There is evidence supporting efficacy in prevention of upper gastrointestinal bleeding in patients treated with anti-inflammatories, antiplatelets, and vitamin K antagonists,13 even of a change in the proportion of digestive hemorrhages, with reduction of the upper and a relative increase of those originated in the lower digestive tract.14 However, theoretically this measure would nor result in a reduction of low digestive hemorrhages, and they have been the most frequent in our study.

Undoubtedly, intracranial bleeding is the most feared complication in patients anticoagulated. It is responsible for the greatest number of deaths and also for the majority of functional sequelae in anticoagulated patients.15 It is estimated that oral anticoagulation increases 2 to 5 fold the risk of presenting an intracranial hemorrhage,16 although these data probably reflect more the figures of vitamin K antagonists than those of direct-acting anticoagulants, since the latter have consistently shown a reduction in intracranial hemorrhage compared to warfarin,10,17–19 with figure more similar to those of simple antiplatelet therapy with aspirin.20

Although the severity of intracranial hemorrhage is evident in all studies, its frequency may be underestimated, especially in publications in which both major and minor hemorrhages are evaluated. The analysis of hemorrhages associated with hospitalizations yields the proportion of intracranial hemorrhages in relation to other relevant hemorrhages, removing from the denominator some minor bleeding episodes that are resolved in the emergency department or even in outpatient consultations. In our study, intracranial hemorrhages represented 30.6% of the total. The main randomized clinical trials show lower figures of intracranial hemorrhage compared to the rest of major hemorrhages, with values ranging from 8% to 25%.10,17–19 We attribute this difference to the patient selection bias inherent to clinical trials and to the comparison with major hemorrhages, a percentage of which may have been managed on an outpatient basis. In the comparison with other observational registers, we find values as different as 2%, 5% or 17%,21,22 in most cases figures lower than ours, differences that we also attribute to the fact that our results reflect the percentage against the total of bleedings causing hospitalization.

In our population, in-hospital mortality was 15.2%, a figure consistent with previous publications in anticoagulated patients. A recent retrospective series showed that mortality in patients treated with warfarin and admitted with excess of anticoagulation was 9.4%, although after excluding patients without bleeding, mortality among patients admitted with bleeding and anticoagulation with warfarin increased to 12%.23 The data published in the literature show considerable variability, probably reflecting differences in the selection of the study population. A meta-analysis of patients anticoagulated for deep vein thrombosis or pulmonary thromboembolism showed a mortality of 13.4% after an episode of major hemorrhage.24 In the ATRIA study, which included patients anticoagulated for NVAF, a mortality of 23.5% was reported 30 days after admission for intracranial hemorrhage or major extracranial hemorrhage. In this study, 88% of deaths corresponded to intracranial hemorrhages, and the entire population was treated with warfarin.15 On the contrary, another study in patients anticoagulated with warfarin admitted for major hemorrhage or with hemorrhage during admission for another reason, in-hospital mortality was 4.9%, significantly lower.21

The mortality after a bleeding episode is high, not only for the risk directly attributed to the hemorrhage, but also to the frailty of patients who survive the acute phase of the major bleeding. An analysis performed in patients from the RE-LY and ACTIVE studies showed that a patient surviving an episode of intracranial hemorrhage had a 27-fold higher risk of mortality compared to patients with similar characteristics, and in nonfatal extracranial hemorrhages the excess of mortality in the follow-up increased 5 times.25 In a similar publication from the ARISTOTLE study, this increase was higher, with mortality 121 times higher in case of nonfatal intracranial hemorrhage and 12 times higher in the case of nonfatal extracranial hemorrhage.26

The increase in mortality observed in patients who survive a severe hemorrhage is due to a combination of factors. Hemorrhages appear in patients with greater frailty and the bleeding itself weakens the patient even more, especially when it causes neurological deficits or when it requires surgical interventions or prolonged hospitalizations. The almost constant withdrawal of anticoagulants and antiplatelets, which often are not restarted, is another key factor that conditions the prognosis. In a subanalysis of the ARISTOTLE study, a 12-fold increase in the risk of myocardial infarction or stroke was detected after suffering a major nonfatal hemorrhage, in addition to the increased risk of bleeding in patients who had already overcome a major hemorrhage.26 Other factors possibly involved in the high mortality of patients who survive bleeding are the adverse effects of transfusions, anemia, subclinical organ damage during acute hemorrhages, and the prothrombotic state that accompanies major hemorrhages.27

Mortality in patients anticoagulated for NVAF is mostly due to cardiac causes: myocardial infarction, heart failure, and sudden death, while mortality directly related to thrombotic and hemorrhagic episodes represents only 6%.28 The appearance of hemorrhagic phenomena is a predisposing factor for cardiac mortality. In the ENGAGE-AF TIMI 48 study, the low-dose randomized treatment arm of edoxaban is a clear example of the importance of bleeding prevention: despite a lower reduction in stroke compared with warfarin, a significant reduction in mortality was obtained because of the marked reduction of major hemorrhages.10

We must acknowledge some limitations. This study was observational and may be subject to confounding factors. Because of the nature of the study the diagnoses are drawn from the hospital-determined codes in the medical records, allowing the possibility of misclassification, and in some cases imprecision (unspecified digestive hemorrhages). Data were obtained from an information services company; the main limitation is that only a small number of variables were available. Much relevant information like the type of anticoagulant treatment, concomitant meditation and type and severity of comorbidities was not available. The International Classification of Diseases, Ninth Revision is not the current diagnostic tool recommended, as has been replaced by the Tenth Revision.