Use of serotonin reuptake inhibitor antidepressants and the risk of bleeding complications in patients on anticoagulant or antiplatelet agents: a systematic review and meta-analysis

Abstract Background Serotonin reuptake inhibitor (SRI) antidepressants are implicated in increasing the risk of bleeding among users; however, the comparative increase in bleeding risk with concurrent antithrombotic therapy (anticoagulant or antiplatelet) remains unclear. As such, we performed a systematic review and meta-analysis of all available evidence to evaluate the effects of SRI and the risk of bleeding complications among patients receiving antithrombotic therapy. Methods We searched Medline, Embase, PubMed, PsycINFO, Cochrane Library, Web of Science, Scopus, CINAHL, and grey literature (Google Scholar and preprint reports) up to 26 November, 2020, with no language restrictions (updated on 31 July 2021). The primary outcome of interest was major bleeding. Secondary outcomes included intracranial haemorrhage, gastrointestinal bleeding, and any bleeding events. We used a random-effects model meta-analysis to estimate the odds ratios (ORs) and 95% confidence intervals (CIs). Results We did not identify any randomised studies but found 32 non-randomized studies (cohort or case–control) with 1,848,285 patients that fulfilled the study selection criteria and were included in the meta-analysis. Among individuals receiving anticoagulants (13 studies), SRI users experienced a statistically higher risk of major bleeding compared to non-SRI users: pooled OR was 1.39 (95% CI, 1.23–1.58; p < .001; moderate heterogeneity). Among individuals receiving antiplatelet therapy (2 studies), SRI users were associated with an increased risk of major bleeding: pooled OR was 1.45 (95% CI, 1.17–1.80; p = .001; low heterogeneity). For secondary outcomes, the use of SRI among individuals treated with antithrombotic therapy revealed a higher risk of gastrointestinal bleeding or any bleeding events, whereas only anticoagulant use was illustrated an increased risk of intracranial haemorrhage. Conclusions The use of SRI antidepressants among patients treated with antithrombotic therapy (either anticoagulant or antiplatelet) is associated with a higher risk of bleeding complications, suggesting that caution is warranted in co-prescription. PROSPERO Registration CRD42018083917 KEY MESSAGES In this meta-analysis of 32 non-randomized studies, SRI users were associated with the risk of bleeding complications compared to non-SRI users, with concurrent antithrombotic use (either anticoagulant or antiplatelet). The risk was consistently elevated across types of bleeding events (major bleeding, gastrointestinal bleeding, or any bleeding events), whereas only anticoagulant use was associated with intracranial haemorrhage. To promote the rational use of medicines, our findings suggest that the risk-benefit ratio must account for the clear efficacy of SRI against safety concerns in terms of bleeding risks.


Introduction
Serotonin reuptake inhibitors (SRIs), including selective serotonin reuptake inhibitors (SSRI) and serotonin-norepinephrine reuptake inhibitors (SNRI) are the most widely prescribed antidepressants that are used in various psychiatric settings including cardiac patients [1]. With respect to the favourable safety profiles compared to older generations of antidepressants, SRI antidepressants and antithrombotic agents (anticoagulants and antiplatelet) are often prescribed together as depression and anxiety often coexist with cardiovascular/cerebrovascular diseases, atrial fibrillation, myocardial infarction, and other thromboembolic disorders [2,3]. Besides the risk of bleeding complications among antithrombotic therapy, recent accumulating evidence suggests that SRI use may be associated with an increased risk of bleeding, intracranial haemorrhage, and in particular, gastrointestinal tract bleeding [4][5][6][7]. In addition, concurrent use of SRI may potentiate this risk of bleeding complications further via pharmacokinetics or pharmacodynamics drug interactions. Specifically, concurrent use of SRI and antithrombotic appear to have the potential to inhibit cytochrome P450 (CYP) isoforms metabolism and impair serotonin platelet function [8].
Although several existing epidemiological studies have recognized the increased risk of bleeding complications among patients who received SRI, the safety of their use concomitant with antithrombotic therapy has not been fully elucidated. Moreover, previous systematic reviews have focussed on the use of SRI concomitant with non-steroidal anti-inflammatory drugs (NSAIDs), with the majority of those studies investigating gastrointestinal tract bleeding risk [6,7,9]. To the best of our knowledge, no comprehensive systematic review and meta-analysis has yet been conducted to quantify the effects of SRI use concomitantly with antithrombotic therapy and the risk of bleeding complications. To address this knowledge gap, we aimed to systematically review and summarize all available evidence to evaluate the effects of SRI use and the risk of bleeding complications among patients who received antithrombotic anticoagulants or antiplatelet therapy.

Materials and methods
This systematic review and meta-analysis were performed and reported in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [10] and the Meta-analysis of Observational Studies in Epidemiology statement [11].
The pre-specified protocol was registered in the PROSPERO International prospective register of systematic reviews (CRD42018083917).

Data sources and search strategy
In collaboration with an experienced medical librarian, we searched electronic databases, including Medline, Embase, PubMed, PsycINFO, Cochrane Library (CENTRAL), Web of Science, Scopus, and CINAHL from inception to 26 November 2020 with no language restrictions. Grey literature from Google Scholar and the preprint reports (medRxiv, bioRxiv, and PsyArXiv) were supplemented to the electronic database searches to identify all relevant articles. We used combinations of Medical Subject Headings and search terms including pharmacological class or individual drugs (i.e. "antithrombotic" or "anticoagulant" or "antiplatelet", AND "serotonin uptake inhibitor" or "SSRI" or "SNRI") and bleeding complications (i.e. "bleeding" or "haemorrhage" or "blood transfusion"). The full search strategy for each database is available in the Supplementary, eTable 1. Relevant articles were also browsed from the reference lists of the included studies, previous systematic reviews, and major international pharmacoepidemiology/cardiology/psychiatry congresses. To update the search, a targeted manual search of relevant articles was performed through to 31 July 2021.

Study selection and outcomes
Eligible titles and abstracts of articles identified were screened independently by two reviewers (SN and CR). Then, potentially relevant full-text articles were assessed against the selection criteria for the final set of included studies. Potentially eligible articles that were not written in English were translated before the full-text appraisal. Any disagreement was resolved by a team discussion.
We included both randomized controlled trials (RCTs) and non-randomized studies (cohort or casecontrol) that (i) investigated the association between the use of SRI and risk of bleeding complications among adult patients (aged 18 years or more) receiving antithrombotic therapy (anticoagulant or antiplatelet agents) for any indications; (ii) consisted of two or more groups in which one group represented the use of SRI concomitant with antithrombotic therapy; (iii) consisted of SRI users including SSRI (i.e. citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, and sertraline), SNRI (i.e. desvenlafaxine, duloxetine, milnacipran, and venlafaxine), or mixed action antidepressant agents (i.e. bupropion, mirtazapine, and trazodone); (iv) reported bleeding complications or provided sufficient data to calculate the risk estimate. We excluded studies that (i) were case series/case reports, N-of-one, cross-sectional, reviews, or studies with small sample sizes (less than 50 patients); and (ii) had no control group. Details of the selection criteria are provided in the Supplementary, eTable 2. For the companion study that included overlapping patients and study periods, the study with the most detailed and relevant information was included.
The primary outcome of interest was major bleeding, defined according to the International Society on Thrombosis and Haemostasis [12,13]. Secondary outcomes of interest included intracranial haemorrhage, gastrointestinal bleeding, and any bleeding events. Additional secondary outcomes included blood transfusion, endoscopy-refractory bleeding, rebleeding, and bleeding-related mortality. Based on clinical relevance, we defined the outcomes according to each included study. For instance, gastrointestinal bleeding events that required hospitalisation or related to mortality were considered major bleeding events.

Data extraction and quality assessment
Two reviewers (SN and RA) independently extracted the following pre-specified data using a standardized approach to gather information on (i) the study characteristics (the first author's name, study design [RCTs, cohort, case-control], study population, sample size, study country, study period, analysis method, and factors controlled for analysis); (ii) patient characteristics (mean or median age of study population, the proportion of females, and comorbid conditions); (iii) specific exposure and control groups (definition of SRI users and non-SRI users, SRI dosage, and concomitant medications); and (iv) predefined outcomes of interest (including assessment outcome definitions and outcome measurements). Studies with incomplete data or unclear information were clarified by the corresponding author. In cases where authors did not respond after two attempts of contact, we used information reported to calculate the required data or excluded the study in the analyses. The final set of data was cross-checked independently by one reviewer (CP and WC).
A pair reviewer (SN and CR) independently assessed and appraised the methodological quality of each included study using the Cochrane revised tool for assessing the risk of bias in randomised trials (RoB 2) [14] and the Newcastle-Ottawa Scale (NOS) for assessing the quality of included non-randomised studies [15]. The overall risk of bias of included studies was then classified into low, high, or some concerns for randomized trials (RoB 2), and the highest quality, if the summary score of the NOS was 8 or more points, for non-randomized studies. Moreover, we also used the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool to assess the risk of bias and categorized the overall judgement as low, moderate, serious, or critical risk of bias [16]. To interpret our findings, the strength of evidence for each outcome was critically appraised independently by a pair of reviewers (SN and CR) using the Grading of Recommended Assessment, Development, and Evaluation (GRADE) guidelines [17]. The strength of a body of evidence findings was then classified into very low-, low-, moderate-, or high-quality. Any discrepancies were addressed by team discussion.

Statistical analysis
Two-tailed with a P-value of less than .05 was considered statistically significant. All analyses and generated forest plots of the summary pooled effects estimate were performed using Stata software version 16.0 (StataCorp, College Station, TX, USA). Inter-rater agreements were tested using kappa (j) statistics to assess the agreement between reviewers in the study selection and risk of bias assessment processes. Based on the common risk estimates across the included studies, we used the aggregate odds ratios (ORs) with the greatest degree of adjustment for potential confounding factors as the summary effect estimates of association for each outcome of interest. As the methodological approach varied across included studies, we employed the random-effects models using the DerSimonion-Laird method for estimating the pooled ORs with corresponding 95% confidence intervals (CIs) to account for heterogeneity between studies [18].
Heterogeneity was assessed using the Cochran Q test, with a P-value of less than 0.10. The degree of inconsistency was investigated using I 2 and tausquared (s 2 ) statistics, [19,20] in which the heterogeneity was estimated as low (I 2 ¼25.0%, s 2 ¼0.01), moderate (I 2 ¼50.0%, s 2 ¼0.06), and high (I 2 ¼75.0%, s 2 ¼0.16). We tested publication bias using Begg's and Egger's tests for each specific outcome of interest (P-value of less than .10 indicated statistical publication bias) [21,22]. The visual inspection of funnel plots was also performed where there was sufficient data to explore for asymmetry of the funnel graph. Moreover, the trim and fill method was then performed to calibrate for publication bias and account for the number of studies with null effects which were missing from the meta-analysis [23].
Pre-planned subgroup analyses were conducted based on (i) patient characteristics (i.e. age, proportion of males, history of bleeding events, comorbid conditions [atrial fibrillation, diabetes, chronic heart failure, coronary artery disease, renal failure, cancer, and Helicobacter pylori infection], and concomitant medications [use of NSAIDs, corticosteroids, and gastroprotective agents]); and (ii) study characteristics (sample size [less than 5000 vs. 5000 or more], study design (RCTs, cohort, or case-control), and study location (North America vs. non-North America). If data were available, individual SRI use and dosage were also assessed to establish the evidence of a dose-response and duration-response relationship.
A set of sensitivity analyses were conducted to assess the robustness of primary findings, including (i) restricting analysis to studies that adjusted for key confounding factors (age, sex, and history of bleeding); (ii) restricting the analysis to studies with high quality; (iii) limiting the analysis to studies with the directness of effect estimates; (iv) removing unpublished studies; (v) removing individual study approaches (leave one out analysis); and (vi) using the fixed-effects models if the I 2 index less than 25.0%. Additionally, a random-effects univariate meta-regression was also performed according to the level of risk of bias, study characteristics, and baseline patient characteristics to explore the pre-specified effects on the meta-analytic estimates.

Results
The search strategy retrieved 2505 records. From these, 594 duplicate records were removed, and 1911 records remained. Based on the title and abstract screening, we identified 211 articles that seemed to be relevant to the study question. Of these, 32 non-randomized studies fulfilled the study selection criteria and were included in the meta-analysis, while we did not identify any clinical randomised trials ( Figure 1). The inter-rater agreement between reviewers on the study selection and data extraction was 0.87 and 0.79, respectively. Table 1 summarizes the characteristics of all the included studies. In total, 1,848,285 patients were identified with a mean age ranging from 52.4 to 82.4 years, proportion of male sex ranging from 22.5% to 79.0%, and most of the included studies not providing a specific indication of the use of SRI and antithrombotic therapy. Detailed measurement and definition of bleeding events, methodology for the study, comorbid conditions, and concomitant medications of the included studies are described in Supplementary, eTables 3, 4, and 5, respectively. According to the risk of bias assessed in 32 non-randomised included studies (Supplementary, eTable 6), summary scores ranged from 3 to 9 points, with 19 studies (59.4%) having the highest quality (NOS of 8 or more). Based on the ROBINS-I tool, we found that most included studies had a moderate risk of bias (21 studies, 65.6%); however, no study with critical risk of bias was observed (Supplementary, eTable 6). The summary results and strength of evidence findings are provided in Table 2. Details of evidence synthesis by the GRADE system are provided in Supplementary, eTable 7.
Evidence for blood transfusion (one study [54]), endoscopy-refractory bleeding (one study [55]), and rebleeding (one study [55]) are inconclusive owing to the limited data available (Supplementary, eTable 8). However, no study has reported bleeding complications in terms of bleeding related to mortality. Moreover, risk estimates according to individual SRI use, as well as a dose-and duration-relationship cannot be established due to lack of information.

Subgroup analyses
Several pre-planned subgroup analyses according to baseline patient characteristics and secondary  Figure 1. Study selection flowchart.  outcomes, could not be performed due to the small sample size and limited information on outcomes across the included studies. Based on study characteristics, among individuals receiving anticoagulant therapy, the effect estimates between SRI use and the risk of bleeding complications was no longer statistically significant when the sample size was less than 5000 (for gastrointestinal bleeding), while the risk of intracranial haemorrhage was sensitive to sample size, study design, and study location (Supplementary, eTable 9). With a small number of included studies for individuals receiving antiplatelet therapy, no further association was observed, particularly among studies with a sample size less than 5000 or case-control study design (Supplementary, eTable 9).

Sensitivity and meta-regression analyses
The set of sensitivity analyses was robust and did not change substantially from the main findings

Anticoagulants (non-specified)
Renoux et al (2017) 38 Kharofa et al (2007) 37 Zhang et al (2020) 34 Nguyen et al (2014) 46 Kurdyak et al (2005) 41 Gaist et al (2020) 40 Vitry et al (2011) 47 Lin et al (2013) 48 Gaist et al (2020) 40 Gaist et al (2020) 40 Kharofa et al (2007) [50], resulted in no further association between the use of SRI and gastrointestinal bleeding risk among individuals who received antiplatelet therapy. A univariate meta-regression was suitable for the primary outcomes (Supplementary, eTable 16). For individuals receiving anticoagulant therapy, the increased risk of bleeding complications was associated with the baseline proportion of NSAIDs use and the proportion of male sex for major bleeding and intracranial haemorrhage, respectively. Nonetheless, the heterogeneity of the included studies was not explained by any of the study characteristics, baseline patient characteristics, and the risk of bias among individuals receiving antiplatelet therapy.

Publication bias
For individuals receiving anticoagulant therapy, evidence of publication bias related to the sample size was observed in the results of major bleeding, intracranial haemorrhage, and any bleeding, with the P-values tested for asymmetry less than .10. Asymmetry tests were observed in the results of intracranial haemorrhage and gastrointestinal bleeding among those who received antiplatelet therapy (Supplementary, eTable 17). The visually inspected funnel plots for each outcome are provided in the Supplementary, eFigure 1. However, after calibration for publication bias using the trim and fill method, the main findings were not substantially different. Notably, there was no longer an association between SRI use and the risk of intracranial haemorrhage and gastrointestinal bleeding among those who received anticoagulant and antiplatelet therapy, respectively, after the analysis was calibrated for publication bias (Supplementary, eTable 17).

Discussion
This systematic review and meta-analysis of 32 included non-randomized studies showed low certainty evidence that SRI users experienced a statistically higher risk of bleeding complications compared to non-SRI users, particularly among patients treated with anticoagulant therapy. We found very low certainty of evidence on the association between SRI use and the risk of intracranial haemorrhage and gastrointestinal bleeding among patients who received anticoagulant and antiplatelet therapy, respectively.
Several mechanisms have been proposed to explain the association between SRI use and the risk of bleeding complications. Theoretically, it has been demonstrated that serotonergic antidepressants increase bleeding complications via inhibiting platelet aggregation [8]. Another possible explanation for SRI in relation to bleeding risk is increased gastric acid secretion directly by increasing the vagal tone, subsequently leading to potential ulcerogenic effects and gastrointestinal bleeding [8,56,57]. As expected, the use of SRI concomitant with antithrombotic therapy either anticoagulants or antiplatelet agents can aggravate the risk of bleeding via both pharmacokinetics or pharmacodynamics interactions [8]. Among individuals receiving SRI and warfarin, for instance, proposed drug-drug interactions that increase bleeding risk may include impairing platelet aggregation and CYP 450 inhibition of warfarin metabolism; the potency of CYP inhibition varied among SRI [8]. Furthermore, SRI may further decrease platelet or endothelial activation and reduce the efficiency of haemostasis beyond that associated with concomitant antiplatelet agents such as aspirin or clopidogrel [56].
Our findings expanded previous meta-analyses by providing insight into the impact of SRI use concomitant with antithrombotic therapy (both anticoagulants and antiplatelet agents) and the bleeding risk, which has not been fully addressed previously. With regard to the credibility of the evidence, a previous umbrella review by Dragioti et al. (2019 [58]) was based on highly suggestive evidence, which indicates an increased risk of bleeding complications among individual use of SSRI or SNRI users. The summary ORs of severe bleeding at any site and upper gastrointestinal bleeding were 1.41 (95% CI, 1.27-1.57) and 1.55 (95% CI, 1.35-1.78), respectively [58]. These findings are also supported by our results that the use of SRI among individuals who received antithrombotic therapy (particularly anticoagulation), included the risk of major bleeding, gastrointestinal bleeding, and any bleeding events. However, it is unclear whether the use of SRI among individuals receiving antithrombotic therapy may have additional intracranial haemorrhage. Several existing meta-analyses with substantial heterogeneity have illustrated an increased risk of intracranial haemorrhage among individuals receiving SRI that did not particularly focus on patients treated with antithrombotic therapy [59][60][61]. On the other hand, other studies have not supported this finding when restricting analyses to high-quality data [58,62]. Furthermore, our findings did not confirm this association when sensitivity analysis and publication bias were considered. Given the statistical power and the imprecision of our findings, evidence for the risk of intracranial haemorrhage among individuals' use of SRI antidepressant concomitant antithrombotic therapy remains inconclusive.

Strengths and limitations
The strengths of this study include a large sample size. We expanded and addressed the further risk of bleeding complications associated with the use of SRI among patients who received antithrombotic therapy, which had not been investigated by previous metaanalyses. From a methodological viewpoint, we used a rigorous and comprehensive systematic review approach, as well as extensive searching without language restriction. Furthermore, according to the set of sensitivity analyses, these results were consistent with the main analysis in most cases, suggesting the robustness of our findings.
This systematic review and meta-analysis have several limitations. First, despite conducting a comprehensive search strategy, data from RCT were not identified. Our findings relied on non-randomized observational studies, confounding by indication/ contraindication, and unmeasured confounders must be noted. As a result, the causality of the use of SRI among patients who received antithrombotic therapy and the subsequent risk of bleeding complications cannot be established. Most of the studies included in this review were based on routinely collected administrative data and electronic health records, which could be prone to information bias. Second, on the basis of the NOS summary score, the quality of the included studies was varied; most cases (18 studies [56.2%]) had high quality (NOS more than 8 points) and 14 studies had low quality (NOS ranged from 3 to 7). Of these, four studies [38,46,48,52] (12.5%) were reported as abstracts, which could lead to incomplete information. However, the results after restricting the analysis to studies with high-quality or removing unpublished studies, according to the mentioned sensitivity analysis methods, yielded main findings that were not substantially different. Therefore, we advocate that future studies with high methodological quality are required. Third, disparities of individual SRI exposure and bleeding outcome definitions were observed across the included studies, which could have contributed to the moderate heterogeneity of our findings. Although the degree of inconsistency improved in most cases when subgroup analysis was performed, several pre-planned subgroup analyses could not be conducted due to the small number of included studies. Fourth, data on the individual use of SRI, key patient characteristics, and several confounding factors related to bleeding complications, such as renal function, history of bleeding events, or use of NSAIDs were not gartered across all included studies. As a result, a dose-and durationrelationship and risk effects estimate, based on the different subpopulations, cannot be established due to lack of information. Fifth, information on both SRI and antithrombotic therapy in terms of treatment medication and adherence over time were also lacking; thus, misclassification bias should be stated. Moreover, the data on pre-specified additional secondary outcomes, including, blood transfusion, endoscopy-refractory bleeding, rebleeding, and bleeding-related mortality were insufficient, which is an emerging concern with respect to the increased SRI use among individuals who received antithrombotic therapy and is needed for further studies. Finally, it is possible that publication bias exists and might account for some of the effect estimates we observed. Moreover, our strength of evidence findings using the GRADE approach was based on a low or very low body of evidence. Therefore, the interpretation of our findings should be exercised.

Implications for practice and future research
Given the limited strength of the body of evidence, this systematic review and meta-analysis provides the best available evidence that can emerge as insight with respect to the use of SRI among individuals receiving antithrombotic therapy in general practice. In cardiac patients receiving antithrombotic drugs, the risk-benefit ratio must account for the clear efficacy of antidepressants against adverse health outcomes, which should be balanced with safety concerns in terms of bleeding risk. Thus far, individuals receiving combination therapy including SRI and antithrombotic therapy warrant proactive monitoring of bleeding complications, especially among individuals with a history of bleeding, or pre-existing risk of bleeding-that is, peptic ulcer disease, chronic liver disease, chronic kidney disease, or received concomitant medication that may further increase the risk of bleeding (i.e. NSAIDs). The findings from this review support the interventions or strategies that promote appropriate SRI prescriptions and minimise risk in relation to drug-drug interactions in real-world practice. In addition, patients should also be informed about the benefits and risks of concomitant SRI and antithrombotic therapy in terms of bleeding risks to promote the rational use of medicines.
Further research in RCTs alongside collaborative pharmacoepidemiology research and proactive realworld evidence surveillance systems are needed to reaffirm and clarify the potential causal association between SRI and risk of bleeding among individuals receiving antithrombotic therapy. Such research should elaborate on the use of individual SRI, clinical diagnoses and indications, pathogenesis and mechanistic processes, the severity of clinical and bleeding risk conditions, and dose-effect and duration-effect response.

Conclusions
This systematic review and meta-analysis revealed that SRI use among patients treated with antithrombotic therapy, especially anticoagulants may increase the risk of bleeding complications, including major bleeding, gastrointestinal bleeding, and any bleeding events. However, these findings were limited by the nature of non-randomised included studies and the low strength of the body of evidence. However, evidence for intracranial haemorrhage or those who received SRI concomitant with antiplatelet therapy are inconclusive. Further pharmacoepidemiologic research, including proactive longitudinal surveillance systems, is needed to clarify and confirm the safety of using SRI in concomitance with antithrombotic therapy and the subsequent risk of bleeding complications.

Ethical approval
Ethical approval was nor required as this study did not require use of patient identifiers.

Disclosure statement
All authors declare no competing interests. All the researchers involved performed this study in the context of their research.

Funding
This work was funded by the Faculty of Pharmacy and partially supported by the Pharmacoepidemiology and Statistics Research Centre (PESRC) through the Chiang Mai University, Thailand. The funder of the study had no role in the study design collection, analysis, or interpretation of the data, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit it for publication. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the funders.

Data availability statement
All data generated or analysed during this study are included in this article and its supplementary information files. The data that support the findings of this study are available from the corresponding author upon reasonable request.