Background

Like all anticoagulants, the use of direct oral anticoagulants (DOACs; apixaban, rivaroxaban, edoxaban, dabigatran) for prevention and treatment of thromboembolism is limited by bleeding. About 2 to 4% of anticoagulated patients experience major bleeding each year and another 10 to 12% experience clinically relevant non-major bleeding [1,2,3,4,5,6,7,8]. Unlike intracranial bleeding which is a rare severe event associated with substantial mortality and long-term morbidity, gastrointestinal (GI) bleeding is the most frequent single site of anticoagulant-related bleeding representing up to 40% of bleeds [9,10,11].

As a class, DOACs increase rates of GI bleeds, the adverse outcomes of which, including mortality, have been less well characterized compared to intracranial bleeds [12, 13]. This is due, at least in part, by a lack of standardized and validated definitions for fatal GI bleeding which may not reflect the potential severity of events. Overall mortality incorporates bleeding and non-bleeding causes of death and appears to be as high as 10% after major GI bleeds [14,15,16,17]. In a sub-study of the ARISTOTLE (Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation) trial, all-cause mortality was 9.2% within 30 days of major extracranial bleeding, the majority of which was GI bleeding [14]. Further, patients with major extracranial bleeding had a 12-fold increased risk of all-cause death in comparison with those without such bleeding [14]. Similar rates of 30-day all-cause mortality (10%) were reported among DOAC-treated patients with acute major GI bleeding treated with specific DOAC reversal agents [15,16,17].

By summarizing the available evidence regarding mortality after DOAC-related GI bleeding, this systematic review and meta-analysis will provide key prognostic information for shared decision-making between patients and their providers, and provide researchers with prognostic estimates to support prospective prevention and treatment studies.

Objective

The primary objective is to determine the risk of all-cause mortality within 30 days of major DOAC-related GI bleeding.

Methods

This protocol has been developed following the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) statement [18] and prospectively registered with PROSPERO (CRD42022295815). Important amendments made to the protocol will be documented and published alongside the results of the final systematic review. Reporting of the final systematic review will follow the PRISMA statement [19].

Search strategy and study selection

The search strategy was developed and peer-reviewed by at the University of Ottawa [20] and includes MEDLINE (Ovid), Embase (Ovid), Cochrane Central Register of Controlled Trials (Ovid), CINAHL (EBSCO), and Web of Science (Core Collection) (see Appendix 1 for full search details) with no limits to language or publication date. The main search concepts included terms related to atrial fibrillation or venous thromboembolism, GI hemorrhage, and DOACs. Search results will be exported to Covidence (Melbourne, Australia) and duplicates will be removed. Title and abstract screening and full-text review of potentially eligible studies will be conducted independently by two reviewers. Disagreements will be resolved by discussion, or involvement of a third reviewer if consensus cannot be reached among the two reviewers.

Study inclusion criteria

Studies will be eligible for inclusion if they meet the following criteria.

Population

Adults (18 years or older) treated with an oral anticoagulant for atrial fibrillation or venous thromboembolism who experienced a major GI bleeding event. Venous thromboembolism will be defined as lower limb deep vein thrombosis (DVT), pulmonary embolism (PE), or both. Studies which included individuals treated for non-limb venous thrombosis (e.g., cerebral vein, splanchnic vein) in addition to DVT and/or PE will be eligible. Studies which included only patients with non-limb venous thrombosis will not be eligible. For the purpose of this review, major bleeding will be defined as per individual studies. Hospitalization for GI bleeding will also be considered major GI bleeding.

Intervention and comparator

The review will include studies wherein participants received anticoagulation with an approved DOAC (i.e., rivaroxaban, apixaban, dabigatran, and edoxaban). Studies will be excluded if they exclusively evaluate anticoagulants other than DOACs. Studies including participants treated with antiplatelet agents in addition to anticoagulants (double or triple therapy) will be eligible for inclusion. Since the aim of the review is to determine the prognosis (mortality) of patients with a major GI bleeding event on DOACs, a comparator is not applicable.

Outcome

Studies that report all-cause mortality after a major GI bleed will be eligible.

Study design

Randomized controlled trial (RCT), and observational (prospective and retrospective) studies will be eligible. Case reports, cross-sectional studies, and case series will be excluded.

Primary outcome

The primary outcome is 30-day all-cause mortality after major GI bleeding; measured by the number of deaths within 30 days following a major GI bleeding event relative to the number of major GI bleeding events.

Secondary outcomes

Secondary outcomes include the number of deaths (a) during the entire follow-up and (b) within 90 days following a major GI bleeding event, and number of fatal GI bleeding events (as defined in individual studies) relative to the number of major GI bleeding events.

Data extraction

Data will be extracted independently in duplicate using a standardized data extraction form, with discrepancies resolved by the same approach as for disagreements in the study selection process. Data extracted will include the information in Appendix 1. For studies which report bleeding outcomes within a combined cohort of DOAC and VKA-treated patients, we will use only DOAC treated patients.

Handling of duplicate and missing data

For studies which reported findings from the same study participants in more than one publication, data will be extracted from the most complete data set, including information from other publications where needed. If studies include both VKA- and DOAC-treated patients and DOAC-specific data cannot be extracted, study authors will be contacted. Studies for which DOAC-specific data cannot be provided will be ineligible. We will not contact study authors for other missing or incompletely reported data.

Risk of bias assessment

Risk of bias assessment will be conducted by two reviewers independently. As our analysis is concerned with prognosis (all-cause mortality after major GI bleeding) the risk of bias of included studies will be assessed using modified criteria proposed by Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) for assessing the risk of bias in prognostic studies (Tables 1 and 2) [21]. Individual arms of RCTs will be treated as prospective cohorts for the analysis. Studies with one or more domains marked as “definitely no” or two or more domains marked as “definitely no” or “probably no” will be judged as being at high risk of bias. In order to be considered at low risk of bias all domains would need to be judged as “probably yes” or “definitely yes”. Otherwise, the study will be considered as having an unclear risk of bias. Disagreements will be resolved by discussion and an attempt at consensus. Involvement of a third reviewer will be used if consensus cannot be reached among the original two reviewers.

Table 1 Summary of the bias domains and prompting items for risk of bias assessment
Full size table
Table 2 Definitions of levels of evidence for typical risk of broadly defined population used for risk of bias assessment [21, 22]
Full size table

Data synthesis

Using the aggregated number of outcomes for each included study, summary estimates for proportions (i.e., all-cause mortality per major bleeds and fatal GI bleeding events per major GI bleeds) will be calculated using the random effects inverse variance method. A continuity correction of 0.5 will be used in studies with zero cell frequencies. A double arcsine transformation will be conducted prior to pooling proportions to stabilize estimates as there may be a range of proportions. As we expect the event rate to be over 1% (rates of 10% have been reported in some studies [14]) we believe this method to be statistically appropriate. The Wilson score method will be used to estimate 95% confidence intervals. Statistical heterogeneity will be evaluated using the I2 statistic and by providing 95% prediction intervals. We will assess the degree of heterogeneity to determine the degree to which pooling appears appropriate, for subgroups where I2 > 75% representing considerable heterogeneity we will not report pooled estimates. The potential for non-reporting bias and small study effect will be assessed for the primary study outcome by funnel plots representing the proportion of patients that died following a major GI bleed on a logit scale versus the inverse of sample size for each study [23], when 10 or more studies will be included in the meta-analysis. When asymmetry will be suggested by visual inspection, we will investigate asymmetry by Peter’s test using a weighted regression model with the logit of event rate as outcome variable, the inverse of sample size as predictor variable and the number of events and non-events to compute weights [24]. Analysis will be performed in R using the meta package for meta-analysis [25].

Analysis of subgroups or subsets

Exploratory subgroup analyses will be conducted if feasible based on the following criteria: study design (prospective vs. retrospective), type of DOAC (apixaban, rivaroxaban, edoxaban, dabigatran), mechanism of DOAC (factor Xa inhibitors vs thrombin inhibitors), sex (male vs female), indication for anticoagulation (atrial fibrillation vs venous thromboembolism), concomitant antiplatelet use, and funding (i.e., industry vs non-industry vs none/not reported). We will also perform a sensitivity analysis based on the risk of bias for included studies as well as the information reported on GI vs extracranial bleeding.

Discussion

This meta-analysis will synthesize evidence from studies on the risk of all-cause mortality after GI bleeding in patients on DOACs. The quality of evidence will be evaluated using a risk of bias tool adapted for this purpose. We do acknowledge that there may be substantial heterogeneity within the available data, and this may limit the number of studies that can be included in an ultimate meta-analysis.

Clinical discussions around the risks and benefits of anticoagulation focus on both prevention of vascular and bleeding risks. Bleeding related risk remains understudied, and a true understanding of mortality associated with GI bleeding would be of utility in clinical decision making [26]. In addition to discussions of anticoagulation initiation, restarting anticoagulation after a GI bleed is an area of ongoing study with substantial practice variation [27]. We anticipate the resulting evidence will provide specific data regarding the consequences of the most common complication of DOAC treatment to inform shared decision-making in routine clinical practice and future research questions.