1. Selection of studies and evaluation of quality
The primary search identified 2,171 records after excluding duplicates. Subsequently, 45 records were left after excluding 2,126 records by reviewing titles and abstracts carefully according to the principle of PICOS. After reading full-text, 33 records were excluded for specific reasons listed in Fig. 1 Finally, 12 studies with 6,943 patients in total met the inclusion criteria, which included 2 RCTs [3, 15], 1 non-randomized clinical trial [16] and 9 cohort studies [17–25].
Both of the two RCTs were evaluated as high quality (Supplementary Fig. 1), the CCT had global ideal score being 19(>16) (Supplementary Table 4), and the whole cohort studies were considered of high quality because of the scores ranging from 6 to 9, with an average of 7.30 (Supplementary Table 5).
2. Studies characteristics and baseline characteristics of patients
Patients’ characteristics were displayed in Supplementary Table 6. The common demographic and baseline characteristics, such as mean age (with average of 82 years old), body mass index (BMI) and the percentages of female, diabetes mellitus, hypertension were similar between NOACs group and VKAs/APT group. The coronary artery disease (CAD), previous hemorrhagic or ischemic stroke, previous venous or arterial thromboembolism, permanent pacemaker and chronic obstructive pulmonary disease (COPD), which may have important impact on the procedure and prognosis of TAVR; the CHA2DS2-VASc score and HAS-BLED score that can affect the selection of antithrombotic therapy and the outcomes of studies; the glomerular filtration rate (GFR) and the percentage of chronic renal failure, which reflected kidney function and related to the choice of NOACs dose[9], were similar between the above two groups. A total of 6,943 post-TAVR patients (5,299 in 10 studies had indications for OAC; 1,644 in 2 studies didn’t have indications for OAC) were covered in this study. Indeed, GALILEO-4D was a substudy of GALILEO trial. After reading protocols and supplementary appendixes of two RCTs, we divided the patients included in the GALILEO into two parts: those who participated in the GALILEO-4D and those who did not. Then data extraction was performed in two parts respectively. The detailed data of outcomes in studies were shown in Supplementary Table 7.
3. NOACs therapy versus standard care (VKAs/APT for patients with/without indications for OAC).
3.1 The primary outcome (all-cause mortality)
Mantel-Haenszel random-effects model was utilized to pool the data of 4006 patients with indications for OAC (1459 received NOACs versus 2547 received VKAs) and 1644 patients without indications for OAC (826 received NOACs versus 818 received APT) from ten eligible studies. As shown in Fig. 2, no significant differences existed between NOACs and VKAs (RR = 0.85, 95% CI: [0.61, 1.18], p = 0.32), but NOACs were associated with higher risk of all-cause mortality than APT (RR = 1.66, 95% CI: [1.12, 2.45], p = 0.01).
CI (confidence intervals), NOACs (non-vitamin K antagonist oral anticoagulants), OAC (oral anticoagulation)
Subgroup analysis was performed because of significant high heterogeneity (I2 = 68%, p = 0.002) in studies with indications for OAC, and we demonstrated that NOACs were connected with the lower risk of all-cause of mortality than VKAs after more than 1-year follow-up (RR = 0.64, 95% CI: [0.42, 0.96], p = 0.03, Fig. 3)
CI (confidence intervals), NOACs (non-vitamin K antagonist oral anticoagulants), OAC (oral anticoagulation)
Cumulative analyses were supplemented with O’Brien-Fleming sequential monitoring boundaries because of significant high heterogeneity in the subgroup with follow-up time no more than 12m (I2 = 58%, P = 0.05). As shown in Fig. 4, Z-curve and O’Brien-Fleming futility boundaries intersect at the last point, which indicated that NOACs and VKAs were associated with similar all-cause mortality if follow-up time was no more than one year and this conclusion was stably. In the future, more clinical trials should be followed up for more than one year.
TSA (Trial Sequential Analysis)
Contour enhanced funnel plot was completed to inspect possible publication bias, and significant publication bias was further explored by Egger's test. As a result, there were no significant publication bias (p = 0.2949, Fig. 5).
RR (risk ratio)
3.2 The secondary outcomes
1. The efficacy outcomes
For the efficacy outcomes including all stroke, valve thrombosis and a composite endpoint, Mantel-Haenszel random-effects model was utilized to pool the data from nine, four and seven studies. As shown in Fig. 6, no significant differences existed in all the efficacy outcomes between NOACs and VKAs for patients with indications for OAC. For patients without indications for OAC, no significant differences existed in all stroke and a composite endpoint between NOACs and APT, but NOACs were better than APT to prevent valve thrombosis (RR = 0.19, 95% CI: [0.04, 0.83], p = 0.03).
Cumulative analyses were supplemented with O’Brien-Fleming sequential monitoring boundaries because of significant high heterogeneity in a composite endpoint (with indications for OAC) (I2 = 66%, P = 0.02). As shown in Fig. 7, the results may be false negative and more clinical trials were needed.
CI (confidence intervals), NOACs (non-vitamin K antagonist oral anticoagulants), OAC (oral anticoagulation)
TSA (Trial Sequential Analysis)
2. The safety outcome (major/life-threatening bleeding)
We utilized Mantel-Haenszel random-effects model to analyze the data of 4,005 patients with indications for OAC (1459 received NOACs versus 2546 received VKAs) and 1,644 patients without indications for OAC (826 received NOACs versus 818 received APT) from ten eligible studies. No significant differences were discovered between NOACs therapy and standard care (Fig. 8).
CI (confidence intervals), NOACs (non-vitamin K antagonist oral anticoagulants), OAC (oral anticoagulation)