Resumption of Warfarin After Intracranial Hemorrhage in Patients With Mechanical Heart Valves: A Systematic Review and Meta‐Analysis

Abstract

Background

Resuming anticoagulation in patients with mechanical heart valves after intracranial hemorrhage presents a clinical dilemma. The optimal timing must balance the risk of thromboembolism against hemorrhagic expansion. Prior studies have addressed this issue, but most were limited by small sample sizes. This systematic review and meta‐analysis aimed to determine the optimal timing for anticoagulation resumption in this population.

Methods

We conducted a systematic review and meta‐analysis using PubMed, Embase, Web of Science, and Scopus from database inception to August 23, 2024. The target population included adults with intracranial hemorrhage who required anticoagulation for mechanical heart valves.

Results

Thirteen studies involving 788 patients were included: 12 retrospective and 1 prospective observational study. While off anticoagulation, 32 patients developed ischemic stroke, with a pooled event rate of 5.23% (95% CI, 3.80–7.20%) and negligible heterogeneity (I ² = 0%). The pooled average time to stroke was 8.08 days (95% CI, 1.99–14.18). After anticoagulation resumption, 73 patients experienced hemorrhagic recurrence, corresponding to a pooled event rate of 10.95%, with nonsignificant heterogeneity (I ² = 40.4%).

Conclusions

Withholding anticoagulation for up to 7 days after intracranial hemorrhage in patients with mechanical heart valves appears relatively safe. Hemorrhagic recurrence after resumption was more common than ischemic stroke during interruption. However, the included studies were observational and carried serious or critical risk of bias. Prospective multicenter studies, and ideally randomized trials, are needed to provide more definitive evidence.

Nonstandard Abbreviation and Acronym

ICH

intracerebral hemorrhage

Clinical Perspective.

What Is New?

• This meta‐analysis provides the most comprehensive pooled assessment to date, integrating data from 13 studies and 788 patients to better quantify early thromboembolic and hemorrhagic risks.

• The findings newly support a 7‐day withholding window as a reasonable balance between preventing early ischemic stroke and minimizing hemorrhagic recurrence.

What Are the Clinical Implications?

• Clinicians may consider delaying anticoagulation resumption for up to 7 days after intracranial hemorrhage in patients with mechanical valves, given the low short‐term thromboembolic risk.

Patients with mechanical heart valves require long term anticoagulation to prevent ischemic stroke, systemic embolism, and valve thrombosis. On the other hand, anticoagulation increases the risk of intracranial hemorrhage (ICH) 8‐fold. ¹

In the event of an ICH, patients require anticoagulation reversal and discontinuation. The optimal timing for resuming anticoagulation involves balancing the risk of thromboembolic complications associated with delayed resumption against the risk of ICH recurrence with early resumption.

Several studies ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ have addressed this issue, but their findings are limited by their observational design, small sample sizes, and the infrequent occurrence of thromboembolic and hemorrhagic complications.

To fill this gap in knowledge, we conducted a systematic review and meta‐analysis of the literature to evaluate the incidence and timing of ischemic and hemorrhagic complications in patients with mechanical valves who have experienced ICH.

Methods

Search Strategy

Our study adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. We conducted a systematic literature review using the Nested Knowledge AutoLit software. ¹⁵ This review involved a comprehensive search across multiple databases, including PubMed, Embase, Web of Science, and Scopus, covering articles from the inception of each database up to August 23, 2024.

We developed tailored search strategies for each database, employing a wide range of keywords and medical subject headings, including “cerebral hemorrhage,” “hemorrhagic stroke,” “intracranial hemorrhage,” “anticoagulants,” “heart valve prosthesis,” and “mechanical heart valve.” The complete search strategy can be found in Appendix S1.

We will make the data, methods used in the analysis, and materials used to conduct the research available to any researcher for purposes of reproducing the results or replicating the procedure. The data supporting the findings of this study are available from the corresponding author upon request.

The study was approved by an institutional review committee. Subjects gave informed consent.

Screening Process and Eligibility Criteria

Studies were selected according to predefined Patients, Exposure, Comparison, Outcome criteria. The target population comprised adult patients with intracranial hemorrhage, including subdural, intraparenchymal, intraventricular, or multicompartmental, who required anticoagulation therapy due to mechanical heart valves. The exposure of interest was the timing of anticoagulation resumption. The primary outcomes were the incidence of ischemic stroke during the period of anticoagulation discontinuation, the incidence of hemorrhagic recurrence after resumption of anticoagulation, mortality associated with these events, the hospital length of stay, and disability measured by the modified Rankin Scale at maximal follow‐up. The included studies reported 2 related outcomes: hemorrhage expansion, typically defined as rebleeding at the original ICH site, and hemorrhage recurrence, which referred to any bleeding event during follow‐up, including at the original site. Given the substantial overlap between these definitions, and to avoid confusion with hematoma expansion that occurs within the first 24 hours after ICH onset, we use the term hemorrhagic recurrence as our primary outcome. This term encompasses all reported expansion and rebleeding events across studies after anticoagulation resumption. Studies were excluded if they were nonoriginal research (eg, reviews, systematic reviews, meta‐analyses), case reports, case series with <5 patients, animal studies, studies with overlapping data, duplicates, meeting abstracts or were not available in English.

Two independent reviewers (T.E. and Y.V.) conducted a blinded screening of titles and abstracts to identify eligible studies. Studies that met the inclusion criteria were then reviewed in full text. Any disagreements during the selection process were resolved through discussion with senior authors (A.S. and S.G.).

Data Extraction

Data extraction was performed independently by 2 reviewers (T.E. and Y.V.) using AutoLit software. Extracted data included study characteristics (eg, study design, sample size, setting), patient demographics, comorbidities, and outcomes of interest (ischemic stroke incidence, hemorrhagic recurrence, and mortality rates). To ensure data accuracy and consistency, a third author (A.S.) conducted a secondary review of the extracted data. Discrepancies were resolved through consensus among all authors.

Risk of Bias Assessment

The risk of bias in included studies was assessed independently by 2 reviewers (T.E. and Y.V.) using the revised tool for assessing risk of bias in nonrandomized studies of interventions. This assessment evaluated potential biases across domains including confounding, selection, and outcome measurement. Any disagreements in risk of bias assessments were adjudicated by 2 additional authors (A.S. and S.G.).

Statistical Analysis

The statistical analysis was conducted using R version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria) with the "meta" packages. ¹⁶ For continuous outcomes, the meta‐ analysis was conducted using the metagen function, pooling means across studies. Study variances were calculated as the square of the SD divided by the sample size. Random‐effects models were fitted using inverse‐variance weighting. Between‐study variance (τ ²) in the random‐effects model was estimated with the DerSimonian–Laird method, with Q‐ profile CIs. ¹⁷ Heterogeneity was assessed using Cochran's Q statistic and quantified with the I ² statistic, with P value <0.05 or I ²>50% indicating substantial heterogeneity. We also report results from a fixed‐effects model to allow readers to examine our findings when each study is treated as having its own fixed, nonstochastic effect, without modeling between‐study variability.

Given the small number of events and 0 cells in several studies, we reported the crude overall prevalence (total events ÷ total participants) with exact binomial CIs and described between‐study variability in the same units as the effect (SD and variance of raw study proportions). ¹⁸

Results

Study Selection and Evaluation

The initial database search yielded 311 records, with contributions from various sources: 43 from Embase, 69 from PubMed, 45 from Scopus, and 154 from Web of Science. Following a thorough screening of titles and abstracts and the removal of duplicates, we refined the list to 194 studies. During this screening phase, 184 records were excluded, leaving 10 that underwent full‐text assessment for eligibility. One paper was excluded due to not being written in English.

Additionally, 4 records were recommended by experts, resulting in 13 studies ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ included in our quantitative synthesis (Figure 1). These studies were published between 1995 and 2024 and were conducted in the United States, United Kingdom, Canada, Germany, China, Japan, and Belgium.

Figure 1. Study flow diagram.

Study Characteristics

A detailed summary of the studies, including study design, sample size, age distribution, sex representation, location of mechanical valve, type and severity of ICH, anticoagulation reversal, traumatic cause, and surgical procedures for ICH, is depicted in Table S1.

Among the 13 studies ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ included in this meta‐analysis, 9 were single‐center observational studies, ² , ⁴ , ⁵ , ⁷ , ⁹ , ¹⁰ , ¹¹ , ¹³ , ¹⁴ 3 were retrospective multicenter observational studies, ³ , ⁸ , ¹² and 1 was a prospective single‐center observational cohort study. ⁶ A total of 788 patients with mechanical heart valves and ICH were included. The median age of patients across the studies was 68 years, and male patients predominated, comprising 65.8% of the sample.

Mechanical valve location was aortic in 446 patients, mitral in 233, and both aortic and mitral in 82. Seven patients had mechanical tricuspid valves, and the location was not reported in 20 subjects. Intraparenchymal hemorrhage was the most frequent ICH type (n = 388), followed by subdural hemorrhage (n = 267), subarachnoid hemorrhage (n = 53), and other types such as multicompartmental or intraventricular (n = 80). Most studies included different ICH types ³ , ⁴ , ⁵ , ⁷ , ⁹ , ¹¹ , ¹² , ¹³ , ¹⁴ with a higher proportion of subdural hemorrhage. However, 1 study that focused mostly on intraparenchymal hemorrhage (n = 133) ⁸ contributed to the higher overall number of these cases.

Six studies did not report measures of ICH severity, ² , ⁵ , ⁶ , ⁷ , ⁹ , ¹⁴ 7 studies ³ , ⁴ , ⁸ , ¹⁰ , ¹¹ , ¹² , ¹³ provided the median Glasgow Coma Scale score (14), and 2 studies ³ , ⁸ reported the median ICH volume (10.5 and 14.7 mL, respectively). Six studies did not specify whether the ICH was spontaneous or traumatic. ³ , ⁴ , ⁵ , ⁶ , ⁸ , ¹¹ , ¹⁴ Of the cases with reported cause (n = 277), 199 were spontaneous and 78 traumatic.

Only 1 study ¹¹ did not report whether patients received anticoagulation reversal. In the remaining 12 studies, most patients were reversed on admission (80.7%). Four studies ⁵ , ⁶ , ⁸ , ¹¹ did not report whether patients required surgery. In 9 studies that provided these data, 36% of patients underwent neurosurgical intervention.

Risk of Bias

Overall, the risk of bias was either serious or critical in 11 of the 13 studies. Eight studies exhibited serious overall risk, ² , ³ , ⁴ , ⁵ , ⁶ , ⁸ , ⁹ , ¹⁰ and 3 studies had critical risk of bias, ⁷ , ¹¹ , ¹⁴ primarily due to confounding and selection bias (Figures 2 and 3). Moderate risk was more frequently noted in the domains of missing data and reporting bias, and low risk was consistently observed in the measurement of outcomes.

Figure 2. Overall risk of bias.

Figure 3. Risk of bias domains.

Duration of Anticoagulation Interruption, Number, and Timing of Ischemic Stroke Events

All but 1 study ⁷ reported the duration of anticoagulation interruption, with a median range from 2 to 17 days (Table S2). No study reported a significant association between the duration of anticoagulation interruption and the occurrence of ischemic strokes.

The analysis of ischemic stroke incidence included 13 studies with a total of 788 patients who had 32 ischemic stroke events, the crude overall prevalence was 4.06% (32/788 [exact 95% CI, 2.79%–5.68%]). Study‐level proportions had a mean of 2.72% with SD 3.74% (variance 0.0014), indicating absolute variability of ~4 percentage points.

Only 3 studies reported the timing of ischemic stroke occurrence (n = 23). ³ , ¹⁰ , ¹² The pooled mean time of stroke occurrence was 8.08 days (95% CI, 1.99–14.18), with a substantial heterogeneity (I ² = 89%, τ ² = 22.648, P < 0.001), suggesting significant variability among studies (Figure 4). When analyzed using a fixed‐effects model, the pooled mean time to stroke occurrence was 3.51 days (Figure S1).

Figure 4. Timing of ischemic stroke development while off anticoagulation.

Recurrent Hemorrhage

Ten studies ² , ³ , ⁴ , ⁵ , ⁶ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ including 677 patients reported recurrent ICH. Across the included studies, there were 74 events among 592 participants, corresponding to a crude overall prevalence of 12.5% (74/592 [exact 95% CI, 9.9–15.4%]). The study‐level proportions had a mean of 13.1% with an SD of 13.6% (variance = 0.0185), indicating substantial variability in absolute terms.

Four studies ² , ¹⁰ , ¹² , ¹³ reported the timing of hemorrhagic recurrence after anticoagulation resumption. In one of the largest retrospective observational studies on this topic ¹² the mean time to hemorrhagic recurrence among 17 patients who developed that complication was 8 days. In contrast, in a smaller cohort study, ¹⁰ 2 patients suffered hemorrhagic recurrence at mean time of 1.5 days after anticoagulation resumption. The remaining 2 studies that reported the timing of hemorrhagic recurrence after the resumption of anticoagulation identified it as occurring on days 19 ² and 90, ¹³ respectively. In the 2 studies that reported the timing of hemorrhagic recurrence from the initial ICH event, the mean times to recurrence were 4.5 ¹⁰ and 18 days, ¹² respectively.

Length of Stay

The length of stay is presented in a descriptive manner due to inconsistent reporting. In the 6 studies that reported the length of stay, ² , ³ , ⁴ , ⁹ , ¹² , ¹³ means ranged between 7 and 19 days, with an average of 13.8 days.

Mortality

All but 1 study ⁷ reported mortality. Across the included studies, there were 95 events among 743 participants, corresponding to a crude overall prevalence of 12.8% (exact 95% CI, 10.5–15.4%). The study‐level proportions had a mean of 14.8% with an SD of 13.7% (variance = 0.0189), indicating substantial variability in absolute terms.

Follow‐Up

The mean follow‐up time among 9 studies reporting such data ranged from 30 to 1125 days. ² , ³ , ⁴ , ⁵ , ⁶ , ⁹ , ¹¹ , ¹² , ¹³ Only 1 study ¹² reported the modified Rankin Scale score between 3 and 6 months after discharge, but only for patients who developed ischemic or hemorrhagic complications during hospitalization. The study found that patients with hemorrhagic expansion had higher mortality and a worse modified Rankin Scale score at 3‐ to 6‐month follow‐up compared with those who developed ischemic stroke while being off anticoagulation.

Discussion

For clinicians, the decision to resume obligatory anticoagulation in a patient who survived an ICH is “cutting a Gordian knot.” This systematic review and meta‐analysis evaluates the incidence and timing of ischemic strokes following anticoagulation discontinuation, as well as the hemorrhagic complications associated with the resumption of anticoagulation in patients with mechanical heart valves who experienced an ICH. Our analysis shows that ischemic stroke occurred between at a mean of 8 days after ICH, suggesting that withholding anticoagulation for up to 7 days is associated with a low risk of thromboembolic complications. Also, patients are more likely to experience hemorrhagic recurrence when anticoagulation is resumed as compared with ischemic stroke during anticoagulation withholding (12.5% [74/592; 95% CI, 9.9–15.4%] versus 4.1% [32/788; 95% CI, 2.8–5.7%]). This is particularly relevant because hemorrhagic recurrence may be associated with higher long‐term mortality and disability as compared with the occurrence of ischemic stroke. ¹² The timing of hemorrhagic recurrence following anticoagulation resumption is a key factor in clinical decision‐making, yet it remains insufficiently studied, with only 4 studies addressing this issue. They reported hemorrhagic recurrence between 1.5 and 90 days ² , ¹⁰ , ¹² , ¹³ after restarting anticoagulation and between 4.5 and 18 days ¹⁰ , ¹² from the index bleeding event. These rates are highly variable, likely reflecting heterogeneity among the original studies in terms of anticoagulation resumption timing and definitions of recurrent bleeding. Among the studies presenting detailed data on this outcome, the largest reported a mean time to local hemorrhagic recurrence of 8 days after anticoagulation resumption and 18 days from the index ICH. ¹²

These findings are relevant given the paucity of aggregated data available to inform clinical decision‐making in this high‐risk population. The American Heart Association guidelines do not provide strong recommendations regarding the timing of anticoagulation resumption in patients with mechanical heart valves and intracranial hemorrhage. ¹⁹ However, the guidelines do reference a retrospective study suggesting that, based on the composite of hemorrhage and thromboembolic risks, anticoagulation may be considered as early as day 6 post ICH in patients with mechanical valves. ⁸ The European Stroke Organisation Guidelines for the Management of Spontaneous Intracerebral Hemorrhage have been recently updated but do not address this issue. ²⁰

Our meta‐analysis key strength is that it consolidates data from 788 unique patients with mechanical heart valves and intracranial hemorrhage, an uncommonly reported and challenging clinical scenario. Despite the limitations identified in the individual studies, our combined analysis helps mitigate bias and provides valuable insights to inform clinical decision‐making and guide future research.

Nevertheless, it also has limitations worth mentioning. First, most of the included studies were assessed as having either serious or critical risk of bias, primarily due to their observational, single‐center, and retrospective design. This is particularly relevant, as few studies reported explicit criteria for resuming anticoagulation. Decisions were likely driven by individual clinical judgment, reflecting variability in practice rather than adherence to standardized protocols.

Second, although the type of hemorrhage (intraparenchymal hemorrhage, subarachnoid hemorrhage, and subdural hemorrhage) was reported, the associated ischemic and hemorrhagic risks were not stratified. This is a relevant limitation, as the risk of recurrence differs significantly among these conditions. ¹ Also, other clinical characteristics that influence the risk of recurrent hemorrhage, such as hematoma location (lobar versus deep), factors affecting thromboembolic risk, including malignancy ¹² and valve location (mitral versus aortic), as well as the incidence of systemic bleeding and systemic thromboembolism were not consistently reported across studies. As a result most analyses did not stratify outcomes by these variables, limiting the feasibility of a focused meta‐analysis and constraining the identification of patient subgroups that might benefit from earlier or delayed anticoagulation, as well as the assessment of the broader systemic consequences of treatment interruption or resumption. Third, only a minority of the included studies provided data on the precise timing of hemorrhagic recurrence, which limits our ability to analyze and make definitive conclusions on this important aspect. Fourth, 4 studies provided no follow‐up data after hospitalization ⁷ , ⁸ , ¹⁰ , ¹⁴ and only 1 ¹² provided data on long term disability. Consequently, we could not evaluate the effect and timing of anticoagulation resumption on long‐term functional outcomes. Lastly, although the estimation of the between‐study variance may be imprecise with only three studies, we retained the random‐effects model as our primary analysis for ischemic stroke timing given the very high heterogeneity (I ² ≈ 90%) and clear methodological and clinical differences among studies. These differences include sample size, inclusion criteria with varying proportions of intracranial bleeding subtypes, hemorrhage mechanism (spontaneous versus traumatic), outcome definitions, anticoagulation protocols, valve positions, and follow‐up durations, making it unlikely that all studies estimate a single underlying effect. Notably, Nishimura et al. ¹⁰ included only ICH cases (n = 25; 6.6% of the pooled sample) and reported just 3 ischemic events with an extremely early median stroke onset of 3.3 days compared with much larger cohorts by Barra and Sakusic et al. (94% of the pooled sample) ³ , ¹² reporting median timings of 11 and 12 days, respectively. Within this context, it is more plausible, and consistent with our clinical experience, to expect an 8‐ to 10‐day window for ischemic stroke occurrence following anticoagulation discontinuation, because the underlying physiological processes (drug washout, thrombus formation, embolization, and infarction) rarely unfold within such a narrow time frame. Although we also provide results from a fixed‐effects model for transparency, this approach treats each study's effect as a fixed, nonstochastic parameter and does not incorporate between‐study variance. As a result, it provides a stable descriptive mean for this limited data set but overstates precision and lacks generalizability beyond these specific studies, particularly when 1 small, methodologically distinct cohort exerts disproportionate influence on the pooled estimate.

Conclusions

Our meta‐analysis suggests that withholding anticoagulation and ICH for up to 7 days in patients with mechanical heart valves is relatively safe. Overall, the reported rates of hemorrhagic recurrence after anticoagulation resumption are higher than the rates of ischemic stroke while anticoagulation is being withheld. It is important to note that most included studies had either serious or critical risk of bias. Given the limitations of the available data, there is a pressing need for prospective multicenter registries and randomized clinical trials to better account for patient heterogeneity and generate high‐quality evidence to guide anticoagulation management in these complex clinical scenarios.

Sources of Funding

None.

Disclosures

Authors declare that they have no conflict of interest.

Supporting information

Appendix S1

Tables S1–S2

PRISMA Checklist