Warfarin versus direct oral anticoagulants for treating left ventricular thrombus: a systematic review and meta-analysis

To compare DOAC vs warfarin treatment in LVT patients by calculating the pooled effect estimates for:

We systematically searched PubMed/Medline (https://​pubmed.​ncbi.​nlm.​nih.​gov), Google Scholar, LILACS virtual health library ((https://​lilacs.​bvsalud.​org/​en/​), and Cochrane library (cochranelibrary.​com) databases from inception to 14th August, 2020 to identify and retrieve relevant studies using the following terms: “(left ventricular thrombus) AND (treatment); (left ventricular thrombus) AND (warfarin); (LV thrombus) AND ((anticoagulation) OR (vitamin K antagonist)); and (left ventricular thrombus) AND (direct oral anticoagulant).” Two authors (S.L. and T.D.) independently reviewed 12,176 citations by their titles, of which 6311 were duplicates and were excluded. The authors screened 3798 articles by titles, of which 3355 were excluded as they were not relevant to the outcome of interest, leaving 443 full-text articles to be assessed for eligibility. A total of 433 articles consisted of irrelevant patient populations (pediatric patients and young adults), case reports/ series, letters & editorials, reviews, and irrelevant study end-points (not pertaining to our pre-specified objectives), and thus were excluded. Ten studies were included in a qualitative synthesis, of which two studies were excluded because one study compared warfarin vs enoxaparin and not DOACs, while the other one was a duplicate. Finally, 8 articles met the criteria and were quantitatively evaluated (Fig. 1). Any conflicts pertaining to study selection were resolved by a mutual consensus [16]. All the published full-text articles and abstracts comparing warfarin with DOACs for the treatment of LVT were included in this systematic review, whereas studies that did not report outcomes stratified according to warfarin and DOACs cohort were excluded. We also excluded articles (case-reports and -series) comparing two treatments on a case-by-case basis.

S.L. extracted the data from studies included in the quantitative analysis and recorded data on following variables of interest: author(s) name(s); region; year; duration; sample size; primary end-point; bleeding complications; stroke or systemic embolization; thrombus resolution; and mortality. T.D. did the review and quality appraisal. The quality appraisal was performed by using MINORS (Methodological Index for Non-Randomized Studies) scale that incorporates eight methodological parameters scored from 0 to 2 [17]. The overall score of included studies varied from 5 to 11. The results are shown in e-Table 2, Supplement–2.

Categorical variables between the two groups were summarized using the Mantel-Haenszel risk ratio (RR) and odds ratio (OR) along with their corresponding 95% CI and p-values [18, 19]. The pooled estimates were calculated by using a random-effects model for meta-analysis. For calculating tau-square (τ²), we used Hartung-Knapp-Sidik-Jonkman (HKSJ) method as it is known to perform better with fewer number of studies and has lower type-I error rates even when combining studies with unequal sample size [20]. Descriptive values reported as median [interquartile range] were converted to mean ± standard deviation by using Wan method [21]. Publication bias was evaluated by constructing funnel plots and Galbraith plots (as shown in eFigure in Supplement-2) and by performing Egger’s linear regression test of funnel-plot asymmetry. Between study heterogeneity was quantified with Higgins I² statistic. We also constructed Baujat plots (see eFigure in Supplement-2) and performed an influential analysis to identify the presence of any outliers. A two-sided p-value < 0.05 was considered for defining statistical significance. We used meta and metafor packages for performing our meta-analyses [22, 23]. All statistical analyses were conducted in R (v3.6.3).

We identified a total of 8 studies [12‐14, 24‐28] with a pooled sample size of 1955 patients comparing the therapeutic effectiveness and risk profile of warfarin vs DOACs in patients with LVT. The mean age of patients was 61 years and 59.7 years in warfarin and DOACs group, respectively. The proportion of males was similar between the two groups ranging from 57 to 62%. The prevalence of hypertension (16–75%) and smoking (21–60%) was similar between two groups, but there was higher variability in the prevalence of diabetes in DOACs group (8–86%) compared to warfarin group (16–40%). The studies included and baseline characteristics are shown in Table 1.

A total of 6 studies measured the occurrence of bleeding events in their analysis [12, 13, 24‐27] using varying criteria to identify and record bleeding incidents: Bass et al. used The Global Use of Strategies To Open Occluded Coronary Arteries (GUSTO) criteria and blood product administration [29]; Jones et al. used Bleeding Academic Research Consortium (BARC) criteria [30]; Jaidka et al. classified bleeding as major and minor; while no criteria was specified by Robinson et al. and Yunis et al. We found that the pooled risk ratio (RR) of bleeding complications in patients treated with warfarin to those treated with DOACs was 1.15 (95% CI 0.62–2.13) (p = 0.57) as shown in Fig. 2. All included studies looked into occurrence of stroke or systemic embolization. The pooled RR for stroke or systemic embolization was — 1.04 (95% CI 0.64–1.68) (p = 0.85) for warfarin vs DOACs group (Fig. 3). No statistical significance was reached in either of above outcomes between two groups.

Seven studies had reported their rates of thrombus resolution [12‐14, 25‐28]. Daher et al., Robinson et al., Jaidka et al., and Ali et al. relied on transthoracic echocardiogram (TTE) to measure LVT resolution; Jones et al. used TTE or cardiac magnetic resonance (CMR); and Iqbal et al. used CMR to diagnose LVT at the baseline with subsequent assessments made by TTE. The odds of thrombus resolution in warfarin group was 11% higher compared to DOACs group, with a pooled odds ratio of 1.11 (95% CI 0.51–2.39) but it did not reach a statistical significance (p = 0.76), as shown in Fig. 4.

We also calculated the pooled risk for overall mortality between the two groups. The risk of mortality was 9% higher in patients who received warfarin for LV thrombus compared to those who were treated with DOACs — RR 1.09 (95% CI 0.70–1.70), however the difference was not statistically significant with a p-value of 0.48 (Fig. 5).

Our study has certain limitations that need to be addressed. First, largely the retrospective design of studies and lack of randomized trials included in the quantitative analysis. Second, the inclusion of individual studies with a relatively higher influence on the overall effect estimates and heterogeneity. The follow-up duration varied among various single-centered studies. Third, there was a paucity for data on the comparison of warfarin and DOACs, which could have been integral in our inability to reach a statistical significance. Fourth, due to limited information on duration to resolution of thrombus unable to conclude how long the patients should be treated. Readers of this meta-analysis should consider these limitations while interpreting the results and applying them to clinical practice.