3 Results
3.1 Study selection and characteristics
A total of 286 records were screened using the aforementioned search criteria. After excluding 9 duplicates, we further excluded 250 records based on title and abstract, resulting in 27 studies for full text review. After the additional exclusion of 19 studies that did not meet the inclusion criterion (16 studies were one arm or didn’t have a thorough comparison between the groups of interest and 3 studies had less than 20 patients in either group17-19), 8 studies were included in the final meta-analysis7,8,13,20-24(Figure 1). Baseline study characteristics with the MINORS total score are shown in Table 1. A total of 1390 patients were allocated to two groups: a DOAC group of 466 patients and a VKA group of 924.
3.2 Outcomes
3.2.1 Days in hospital meta-analysis
5 Studies reported on days in hospital and were included in the meta-analysis13,20-23. A significantly longer in-hospital stay was found in the VKA group (SMD= -0.22, 95% CI -0.42 to -0.02, P= .03; Table 2). There was no heterogeneity among the studies (I2= 0%, 95% CI 0 to 71.8, P= .6). The fixed effects model was therefore applied.
3.2.2 Posterior epistaxis meta-analysis
6 Studies reported the rates of posterior epistaxis7,8,13,20,22,23 to be significantly more common in the VKA group (OR= .39, 95% CI .23 to .68, P= .001; Table 3). The heterogeneity was low (I2= 16.7%, 95% CI 0 to 61.5, P= .3). The fixed effects model was therefore applied.
3.2.3 Admission rate meta-analysis
With regard to the rate of admission to hospital as a percentage of all patients seen in the emergency room, data was extracted from 6 studies7,8,20,21,23,24. No statistical significance between the groups could be found (OR= .76, 95% CI .54 to 1.07, P= .12; Table 4). The heterogeneity was low to moderate (I2= 46.03%, 95% CI 0 to 78.63, P= .1). The fixed effects model was therefore applied.
3.2.4 Intervention meta-analysis
7 Studies reported on surgical intervention and/or embolization7,8,13,20-23. No statistical significance between the groups could be found (OR= 1.29, 95% CI .63 to 2.66, P= .38; Table 5). No heterogeneity was detected (I2= 0%, 95% CI 0 to 69.3, P= .5). The fixed effects model was therefore applied.
3.2.5 Recurrence rate meta-analysis
The recurrence rates were noted in 5 studies7,8,13,21,22. The rates were comparable between the groups with no statistically significant difference (OR= 1.13, 95% CI .75 to 1.7, P= .57; Table 6). No heterogeneity was noted (I2= 0%, 95% CI 0 to 41.03, P= .86) and the fixed effects model was therefore applied.
3.2.6 Haemoglobin level meta-analysis
The haemoglobin value on admission was reported in 5 studies7,13,20,22,23. No significant differences between the groups were noted (SMD= - .32, 95% CI - .89 to .28, P= .28; data not shown). The heterogeneity was high (I2= 81.6%, 95% CI 57.2 to 92.1, P= .0002). The random effects model was therefore applied.
3.2.7 The need for transfusion meta-analysis
5 Studies reported on the rate of transfusion8,13,20,22,23. The need for transfusion was very rare and no statistically significant difference between the groups was found (OR= 1.34, 95% CI .52 to 3.26, P= .57; data not shown). The heterogeneity was very low (I2= 0%, 95% CI 0 to 61.96, P= .73). The fixed effects model was therefore applied.
3.3 Publication bias
Publication bias was assessed by funnel plots (Tables 2-6, data for haemoglobin value and need for transfusion not shown). The plots were generally symmetrical for all investigated parameters, except for the haemoglobin value. This suggests a low risk of publication bias for the investigated criteria.