Introduction

Approximately 1% to 2% of the population in Western countries is on long-term anticoagulation (1). Direct oral anticoagulants (DOACs) have overtaken vitamin K antagonists in market share, currently being prescribed in 68% to 79% of new anticoagulated patients in the European Union and the United States (2,3).

Despite being safer than vitamin K antagonists, the introduction of DOACs has been paradoxically associated with an increase in the number of urgent admissions for bleeding complications in some countries, probably due to the increase in the number of candidates for anticoagulation (4).

Adequate supportive care and discontinuation of the DOACs are essential for management of these bleeding complications, whereas additional measures, such as the administration of procoagulants and/or specific antidotes, may be needed to control bleeding (5). Various procoagulant agents, mainly 4-factor prothrombin complex concentrates (4PCC) and others have been tested for the treatment of DOAC-related bleeding with variable success, but can cause thromboembolic events (TEs) (5). Two specific agents for DOAC reversal are currently available: idarucizumab and andexanet alfa (6,7). These new agents are effective in neutralizing the anticoagulant effects of the DOAC, but no prospective trials in comparison with 4PCC are currently available.

We conducted a systematic review and meta-analysis to investigate the clinical outcomes associated with the use of nonspecific reversal of DOAC with 4PCC and specific reversal of DOAC with idarucizumab (for dabigatran) or andexanet (for oral direct FXa inhibitors) in patients with severe or uncontrolled major bleeding.

Methods

Protocol and registration

The protocol was registered in the International Prospective Register of Systematic Reviews PROSPERO, registration number: CRD42018100252.

Eligibility criteria

We included studies evaluating 4PCC, idarucizumab, or andexanet for the treatment of severe/uncontrolled bleeding associated with DOAC use (e.g., potentially life-threatening bleeding with signs or symptoms of hemodynamic compromise; major bleeding associated with a fall in hemoglobin >2 g/dl; or bleeding in a critical area or organ). Case-series with <10 patients were excluded. We excluded subgroups of patients who did not receive a reversal agent, those treated with other less common reversal modalities (e.g., activated prothrombin complex concentrate, recombinant factor VIIa, tranexamic acid, and/or vitamin K) and those in which the administration of the reversal agent was not indicated to treat a major bleeding (see also Supplemental Appendix 4).

Study identification and data collection

We used Medline and CENTRAL for the main literature search (from January 1, 2010, to December 1, 2020). We conducted additional gray literature searches on Google Scholar, websites of regulatory agencies, clinical trial registries, and relevant conference proceedings (see also Supplemental Appendix 1). No language restrictions were applied. Two investigators separately assessed the studies for eligibility (A.G.-O. and A.I.T.-F.), data extraction (A.-G.-O. and P.A.), and risk of bias (A.-G.-O. and G.C.-R.). We contacted the principal investigators by email for additional information or clarification if necessary.

Study characteristics and quality assessment

The following data were collected: type of study, inclusion and exclusion criteria, patient characteristics, type of DOAC, indication for anticoagulation, type and dose of the reversal agent, and duration of the hospital stay and follow-up period.

The quality of nonrandomized studies was assessed using the Risk of Bias Assessment Tool for Nonrandomized Studies (RoBANS) (8). Inter-rater agreement was assessed with the Kappa statistic using GraphPad QuickCalcs software (9,10). In case of disagreement, we used the worst score of the 2 evaluations.

Outcome measures

Main outcome was all-cause death. A descriptive analysis of the causes of death was also performed.

Total TEs were investigated as a secondary safety outcome. An additional analysis was performed separately for venous thromboembolism (VTE), and arterial thromboembolism (ATE).

Effective hemostasis was also investigated and defined as “excellent/good” hemostasis using the Sarode scale and the Andexanet Alfa, a Novel Antidote to the Anticoagulation Effects of FXA Inhibitors (ANNEXA-4) scale and effective hemostasis “yes” in the International Society on Thrombosis and Haemostasis (ISTH) scale or assessed by other means (using other scales or through the clinical judgment of the investigators) (see Supplemental Table 1 for complete definitions) (11–13).

Rebleeding after initial control of the bleeding episode was also investigated as a secondary outcome. We also collected data on disability after DOAC-related intracranial hemorrhage (ICH).

Quantitative data synthesis

This review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (14). We used the intent-to-treat population and the longest observation period for which data were available. The Freeman–Tukey double-arcsine transformation was performed before data pooling to stabilize the variance of the proportion of events in each study (15). We also performed a comparative meta-analysis of the relative risk (RR) of death in patients who did not achieve effective hemostasis after administration of the reversal agent compared to patients who did achieve effective hemostasis. We used the random effects method in both meta-analyses (proportions and RR) (16). Pooled proportions and RR were presented together with 95% confidence intervals (CIs). Heterogeneity was defined as Higgins index I² >50% (17). Categorical variables were reported as numbers and percentage, and continuous variables were reported as mean ± SD and range.

Subgroup analyses were performed according to: 1) study duration; 2) reversal agent; 3) type of bleeding; 4) type of study; 5) risk of bias; 6) presence/absence of significant exclusion criteria; 7) hemostatic scale used; and 8) study sponsorship. Meta-regression techniques as well as the aforementioned subgroup analyses were used to analyze potential heterogeneity in the results. All analyses were performed with the OpenMetaAnalyst software version for Windows 10 (18).

Results

Study selection

The bibliographic search identified 1,997 articles and 18 reports from other sources for a total of 2,015 reports (Figure 1, Supplemental Appendix 1). Of these, 1,888 reports were excluded after review of the title and abstract and 67 additional references were excluded after checking of full text (see Supplemental Appendix 2 for the list of excluded reports and causes of exclusion).

Study design and risk of bias

Sixty studies met the inclusion criteria (19–78). The majority were retrospective cohorts (n = 48), followed by prospective cohorts (n = 10) and clinical trials (n = 2) (Table 1).

The 60 studies included 8,636 patients, of which 4,735 had severe bleeding related to DOAC treated with 4PCC (n = 2,688), idarucizumab (n = 1,111), or andexanet (n = 936) (see Supplemental Appendix 4 for excluded patients and causes for exclusion).

Interrater agreement on the assessment of the risk of bias of individual studies was high (Kappa = 0.81; 95% CI: 0.65 to 0.96), indicating near-perfect agreement (see also Supplemental Appendix 3). A total of 45 studies were at high risk of bias, 13 were at moderate/unclear risk of bias, and only 2 were at low risk of bias (Table 1).

Ten studies had exclusion criteria potentially altering the rates of death and/or TE (Table 1, Supplemental Table 2).

Patient and treatment characteristics

A total of 4,735 patients were analyzed. The mean age was 77 ± 3.5 years (range: 68 to 86 years), 57% were males and 55% had ICH as the index event (Table 2). The reason for anticoagulation with DOAC was atrial fibrillation (82%), VTE (14%), or other pathologies (4%), and the DOAC type was rivaroxaban (36%), followed by apixaban (32%), dabigatran (31%), and edoxaban (1%) (Table 2).

The type of bleeding was reported in 4,579 patients (96.7%): 2,537 (55%) patients had an ICH and 2,042 (45%) had other types of bleeding (Table 2).

The 4PCC dose was high (50 U/kg or ≥35 U/kg) or mostly high (>50% of patients receiving a high dose) in 15 studies, and low (25 U/kg or <35 U/kg) or mostly low (>50% of patients receiving a low dose) in 14 studies (see Supplemental Table 3). In 18 studies with idarucizumab, patients received a fixed 5-g idarucizumab dose, and 10 patients received 2 or more doses. In 10 studies with andexanet, 77.3% of patients received the low dose of andexanet (400-mg bolus in 15 min followed by 4 mg/min for 120 min [480 mg]), and 21.7% of patients were treated with a high dose (800-mg bolus in 15 min followed by 8 mg/min for 120 min [960 mg]) (Supplemental Table 3).

The mean time from the last dose of DOAC to administration of the reversal agent was 13.8 ± 3.6 h (range: 8 to 21 h) (data available from only 9 studies). Anticoagulation was resumed after controlling the bleeding event in 57% of patients (range: 25% to 73% in studies), and time to resumption was 11 days (range: 5 to 27 days in studies) (Supplemental Table 3).

Concomitant use of hemostatic adjuncts or interventional procedures was reported in 32 studies. Hemostatic adjuncts included packed red blood cells (range: 2.6% to 53%), fresh frozen plasma (range: 2.1% to 35%), tranexamic acid (range: 0.7% to 35.7%), and platelet concentrates (range: 1.6% to 35.7%). Invasive procedures (range: 4.3% to 67%) included neurosurgery, endoscopy, embolization, pericardiocentesis, and others (Supplemental Table 4).

Outcomes

Mortality

There were 623 deaths in 4,169 patients evaluable for mortality (rate: 17.7%; 95% CI: 15.1% to 20.4%) with significant heterogeneity between trials (I² = 71.3%) (Figure 2). Study duration was a covariate significantly correlated with mortality rates in meta-regression analyses (p < 0.001) (Supplemental Figure 1). Mortality rate was higher in studies lasting ≥30 days (19.7%) than in studies <30 days (13.4%) (Table 3). Heterogeneity still persisted within each subgroup, so other sources of heterogeneity were also investigated. The type of bleeding also contributed to heterogeneity in mortality rates, which were higher in ICH patients (20.2%; 95% CI: 17.2% to 23.3%; I² = 45.1%) than in patients with extracranial bleeding (15.4%; 95% CI: 11.9% to 19.2%; I² = 78%) (Table 3). In addition, studies with significant exclusion criteria resulted in lower mortality rates (13.3%; 95% CI: 8.8% to 18.6%; I² = 73.6%) than studies without such exclusion criteria (19.2%; 95% CI: 16.2% to 22.2%; I² = 70.8%) (Table 3).

No relevant differences were found in death rates depending on the reversal agent used, the type of study, risk of bias, or study sponsorship (Table 3 and Supplemental Appendix 6).

Causes of death were available only for 146 cases. Of them, 99 (67.8%) were adjudicated to the underlying index bleeding event, only 2 (1.4%) were adjudicated to TE, and 45 (30.2%) were attributed to other causes: sepsis/multiorgan failure (n = 18), heart failure/cardiac arrest (n = 8), cancer (n = 4), and other (n = 15).

Thromboembolic events

TEs occurred in 159 of 3,092 patients (rate: 4.6%; 95% CI: 3.3% to 6.0%; I² = 44.7%) (Table 3, Supplemental Figure 2). The risk was high with andexanet (10.7%; 95% CI: 6.5 to 15.7%; I² = 37.9%) and relatively low with 4PCC (4.3%) and idarucizumab (3.8%), and the results were consistent regardless of the type of bleeding, the type of study, the risk of study bias, and the duration of the study (Table 3, Supplemental Appendix 6).

Pooled rates of VTE and ATE were 1.8% and 2.2%, respectively (Supplemental Figure 4). The risk was relatively high for andexanet (VTE 5.5% and ATE 5.0%) and lower with 4PCC (VTE 2.4% and ATE 2.2%) and idarucizumab (VTE 1.7% and ATE 3.3%) (Supplemental Appendix 6).

Effective hemostasis and rebleeding

Effective hemostasis was achieved in 1,469 of 1,890 patients (rate: 78.5%; 95% CI: 75.1% to 81.8%) (Table 3, Supplemental Figure 5), and there was heterogeneity between studies (I² = 60.8%). The use of different definitions of effective hemostasis was the main source of heterogeneity, with the results being homogeneous between studies that applied the ISTH definition (I² = 26.3%) and the Sarode definition (I² = 0%) (11,13). Heterogeneity remained significant between studies that used the ANNEXA-4 definition (I² = 54.3%) or other definitions (I² = 74.6%) (Supplemental Figure 6) (12). Heterogeneity within studies that used the ANNEXA-4 definition disappeared after the removal of 1 outlier study (I² reduced from 54.3% to 16.63%) (74). The reason that study found a very low rate of hemostatic efficacy with andexanet (47.6%; 10 of 21 patients) remains unclear (74). The use of the ISTH definition tended to yield more conservative results on effective hemostasis (72.4%) than those obtained by applying the Sarode definition (79.6%) or the ANNEXA-4 definition (81.2%) (Table 3, Supplemental Appendix 6).

The rate of hemostatic efficacy was high with 4PCC (80.1%; 95% CI: 75.9% to 84.2%), idarucizumab (76.7%; 95% CI: 68.5% to 85%), and andexanet (80.7%; 95% CI: 73.5% to 87.9%) (Table 3, Supplemental Appendix 6, Central Illustration). Retrospective studies, those with high risk of bias, and those with significant exclusion criteria tended to produce more optimistic estimates of effective hemostasis (80.6%, 80.8%, and 80.7%, respectively) than prospective studies, those with low-moderate risk of bias, and those that did not apply significant exclusion criteria (73.4%, 74.1%, and 77.7%, respectively) (Table 3, Supplemental Appendix 6).

A total of 28 of the 232 evaluable patients from only 6 studies experienced rebleeding (rate 13.2%; 95% CI: 5.5% to 23.1%; I² = 68.19%) (Supplemental Figure 7). The mean time to rebleeding was 13.4 ± 9.3 days (range: 5 to 34 days) and 78% of rebleeds occurred after resumption of anticoagulation. A rebleeding event was described as an ICH in 82% of cases.

Correlation between failure to achieve effective hemostasis and risk of death

Failure to achieve effective hemostasis resulted in a more than a 3-fold increase of death compared to achieving effective hemostasis (RR: 3.63; 95% CI: 2.56 to 5.16; I² = 0%) (data from 12 studies) (Figure 3). The results were robust regardless of study type, risk of bias, hemostatic scale, and type of reversal (Supplemental Figures 53, 54, 55, and 56, Supplemental Appendix 6).

Disability-related outcomes

A total of 96 of 160 evaluable patients from 4 studies had moderate-severe disability at discharge (modified Rankin Scale [mRS]: 3 to 5) (rate: 52.6%; 95% CI: 26.6% to 77.9%; I² = 86.94%) (Supplemental Figure 8).

Five studies reported a good outcome (mRS 0–3) or bad outcome (mRS 4 to 6: moderate to severe disability or death). A total of 54 of 198 patients experienced a good outcome (33.8%; 95% CI: 12.1% to 59.6%; I² = 90.38%) (Supplemental Figure 9), whereas the remaining 144 patients experienced a poor outcome (66.2%; 95% CI: 40.4% to 87.9%; I² = 90.38%) (Supplemental Figure 10).

A total of 67 of 229 patients in 6 studies were discharged to a rehabilitation center (27.9%; 95% CI: 17.2% to 39.9%; I² = 68.28%) (Supplemental Figure 11).

Discussion

Our meta-analysis shows a relatively high risk of death (18%) in patients with severe bleeding associated with DOAC despite the administration of a reversal agent. The risk of death varied between studies, with the duration of the study (between 5 to 180 days) being the main determinant of this variability. The effective hemostasis rate was high (79%), being this the main predictor of survival. In fact, the death rate was more than 3 times higher in patients in whom effective hemostasis was not achieved. Therefore, a plausible conclusion is that, in the event of insufficient response, additional attempts and/or combination with other treatment modalities aimed at achieving effective hemostasis should be considered.

The bleeding site is important in decision-making as the different types of bleeds vary in their treatment and outcome. We found a mortality rate of 20.2% in patients with ICH related to DOAC, which compares favorably with the ICH case fatality rate of 37.5% to 49% reported 10 years ago during the main trials with DOAC when the use of reversal agents was anecdotal (79,80).

There was a high rate of patients (52.6%) who had moderate/severe disability at discharge (mRS: 3 to 5), and a 66.2% of patients with poor outcomes (mRS: 4 to 6). Therefore, although the mortality rates from DOAC-related ICH appear to have decreased with the introduction of reversal agents over the past decade, the rates of moderate/severe disability and poor prognosis remain very high.

The overall rate of TE in our review was 4.6%, which was particularly high with andexanet (10.7%), but in the absence of a direct prospective comparison with other reversal strategies it should be considered only as a signal that deserves to be confirmed in further studies. A prothrombotic rebound effect cannot be ruled out with andexanet within the first days after reversal since andexanet has shown to transiently increase thrombin generation (81). TE is generally manageable and most episodes are probably related to a return to baseline risk of thromboembolism when anticoagulation with DOAC is withdrawn in the context of an incomplete and late resumption of anticoagulation. A recent consensus paper recommends resumption of anticoagulation after major bleeding as soon as the thrombotic risk exceeds the rebleeding risk, in most cases within 1 week (82). In our review, anticoagulation was resumed on average 11 days after admission. The rebleeding rate was 13.2%, and 78% of rebleeds occurred after resumption of anticoagulant therapy. Eighty-two percent of rebleeds were described as an ICH. Therefore, caution with anticoagulant resumption is advised.

The main strength of this review is that the results are based on a large sample size of more than 4,700 patients from 60 studies, half of them published in 2020, which were identified through an exhaustive systematic search. Four previous reviews focused on the safety and effectiveness of 4PCC or and andexanet and idarucizumab, but they had limitations, as the sample size was small or did not analyze patients with major bleeding separately from other indications (83–85,86).

Study limitations

The main limitation of our review is that 47 of the 60 studies were retrospective cohorts and 45 had a high risk of bias. Retrospective studies and those at high risk of bias tended to report more optimistic hemostatic effectiveness results than prospective studies and studies at low-moderate risk of bias, but the analysis of deaths and TE yielded similar results regardless of the type of study and risk of bias. A limitation of the analysis of mortality in patients with and without effective hemostasis is that it is based in a selected population of patients with an assessment of effective hemostasis within 48 h, and it may not be available in patients who die early (11–14). Another limitation is that, in general, there was poor reporting of some important clinical data in the studies (e.g., time from the last dose of anticoagulant to reversal that is difficult to obtain in practice, the administered dose of the reversal agent, post-bleeding anticoagulation management, exclusion criteria, etc.), implying caution in interpretation. A comparative clinical trial of andexanet versus standard is currently being conducted in patients with ICH associated with oral FXa inhibitors (n = 900) ( NCT03661528).

The arrival of DOAC-specific antidotes meets an unmet need and will increase confidence in the safe use of the DOAC. Some recent guidelines favor the use of specific reversal agents based on expert consensus and surrogate evidence from biomarkers (87). However, 4PCC may be associated to similar rates of effective hemostasis than specific antidotes and low rates of TE, and therefore could be a valuable treatment when a specific antidote is not available.

Conclusions

Our systematic review shows a high rate of effective hemostasis, around 80%, with 4PCC or specific reversal agents and a relatively high rate of deaths (17.7% in average). Failure to achieve hemostatic efficacy was correlated with a more than 3-fold increase in mortality. TEs occurred with a high frequency with andexanet. In the absence of prospective comparative trials, it cannot be determined whether specific reversal agents are more effective and/or safer than nonspecific reversal with 4PCC. Comparative studies are needed.

Footnotes