Safety of Intravenous Thrombolysis Among Patients Taking Direct Oral Anticoagulants

A Systematic Review and Meta-Analysis

Shima Shahjouei, MD, MPH; Georgios Tsivgoulis, MD, PhD, MSc, FESO; Nitin Goyal, MD; Alireza Sadighi, MD; Ashkan Mowla, MD; Ming Wang, MS, PhD; David J. Seiffge, MD; Ramin Zand, MD, MPH


Stroke. 2020;51(2):533-541. 

In This Article


This article adheres to the AHA Journals' implementation of the Transparency and Openness Promotion Guidelines. The authors declare that all supporting data are available within the article and in the online-only Data Supplement. In case of special data requested, the corresponding author will provide the additional data. The protocol of conducting and reporting this study was performed according to items by the PRISMA (Preferred Reporting Items for Systemic Reviews and Meta-Analyses) and MOOSE (Meta-analysis of Observational Studies in Epidemiology).[18–20]

Search Strategy

We searched PubMed, EMBASE, Scopus, Cochrane Library, International Clinical Trials Registry Platform, and ClinicalTrials.Gov with no limitations in date/time, language, document type, and publication status. The selected keywords were obtained by review of primary search results, experts' opinions, and controlled vocabularies—Medical Subject Headings (MeSH) in Pubmed and Excerpta Medica Tree (EMTREE) in Embase. Names, synonyms, and Chemical Abstracts Service registry number of each medication were retrieved from TOXNET databases (Toxicology Data Network).[21] The search filters for adverse effects and prognosis were selected based on the recommendations and guidelines of ISSG (InterTASC Information Specialists' Sub-Group) search filters resource[22] and McMaster University.[23] Details of search strategies are available in the online-only Data Supplement (Methods; Document I). Search results were duplicated in Mendeley[24] for screening. Forward and backward citation tracking of the identified included articles were also conducted to reveal further relevant studies.

Eligibility Criteria for Meta-analysis

The candidate studies were retrieved through search strategy in Search Strategy and were included in meta-analysis if they provided us (either in the main article or through personal communication) the details of the case and control series of patients according to the following criteria. Studies reporting <3 patients on DOACs who received IVT were not included in meta-analysis. Inclusion/exclusion criteria, description of control groups, and subgroup meta-analysis have been explained in detail in the online-only Data Supplement (Methods, Document III).

Data Abstraction and Management

After removing the duplicates, the title and abstract of retrieved studies were screened by 2 independent reviewers (Drs Shahjouei and Sadighi) according to prespecified inclusion/exclusion criteria. The full texts of the selected studies were reviewed, and data were extracted. Disagreements on the process of selection and data extraction were resolved by discussion between reviewers and supervision of a third reviewer (Dr Zand). In the case of ongoing or similar trials in which the exact desired outcome was not reported, we further contacted the authors. Studies reported as abstracts only were included in case series if they contained all the required information. The following data element group were extracted for each included study: (1) authors and publication information, (2) study methodology, (3) definition of the adverse effect, (4) study inclusion and exclusion criteria, (5) sample size, (6) cases and controls and their demographics, and (7) rate of sICH and fetal ICH.

One of the included studies was the case series of patients taking DOAC who underwent thrombolysis in our multicenter study.[10] We retrospectively retrieved data of the patients who received thrombolysis on warfarin (INR <1.7) and without prior anticoagulation in the corresponding centers in the similar study period and used these data as the control group of our case series. Similarly, when appropriate control groups were not provided by a candidate study, we asked the authors to provide us with control groups from the same centers in the same interval.

Risk of Bias Assessment, Statistical Analysis, and Data Synthesis

Quality control and bias identification were performed independently by the 2 reviewers (Drs Shahjouei and Sadighi) who performed the literature search and a third reviewer (Dr Zand) in case of any conflicts. All emerging conflicts were resolved with consensus. We used ROBINS-E (Risk of Bias in Nonrandomized Studies—of Exposures) approved by Cochrane collaboration for quality appraisal of the observational studies (in the online-only Data Supplement; Methods, Document II). The outcome is summarized in Table as low, medium, or high risk of bias. We also considered guideline by The Joanna Briggs Institute for critical appraisal of case reports and small case series.[26,27] We assessed the heterogeneity among studies by the aid of the Cochran's Q test for heterogeneity, with a P value <0.1 to be considered significant. The heterogeneity among the studies was visualized via Forest plot. The percentage of observed total variation due to heterogeneity rather than chance were quantified using I2: I2=100%×(Q−df)/Q, where Q=Cochran's χ2 statistic and df=degree of freedom (0%–30%: inconsiderable heterogeneity; 30%–60%: moderate heterogeneity; 60%–80%: substantial heterogeneity; and 80%–100%: considerable heterogeneity). I2 importance and borderline heterogeneities were judged by considering the magnitude and direction of effects, in addition to the strength of evidence for heterogeneity.[28,29] Publication bias was assessed by Funnel plot analysis and the Egger test if >10 studies were included in a single analysis. Trim-and-fill method was used if the funnel plot showed significant asymmetry.

We used the mixed-effect models by assuming that studies are heterogeneous. Odds ratios (OR) and 95% CI were reported for each study and the pooled analysis in Forest plot. Note that a continuity correction of 0.5 was used if there were zero cell counts. Except in heterogeneity, the threshold of statistical significance was set at P value <0.05.[30]

Statistical analyses were conducted using R Studio, R Development Core Team (2008). Meta-analysis was performed by Metafor package.