Study selection
On the basis of the above-mentioned search strategy, 190 possible related references were initially identified from the following electronic databases: 54 from EMBASE, 73 from Web of Science, 40 from PubMed, 8 from Ovid, and 15 from the Cochrane Library. After excluding 94 replicates, 72 unrelated articles were excluded on the basis of the title and abstract comments. When the full text of the remaining 24 studies were examined, 12 articles without useful data were excluded. Finally, 12 trials that involved 2159 patients between 2009 and 2018 met the selection criteria and were included in the present meta-analysis (Figure 1).
Study characteristics
The baseline characteristics of the studies are shown in Table 1. No significant differences in preoperative haemoglobin levels were found in 9 studies. Nine studies reported no significant difference in ASA score between the two groups. Eight studies included the use of chemoprevention for thromboembolic complications, such as warfarin, low-molecular-weight heparin, or rivaroxaban. The route of TXA administration was different, and most regimens were administered intravenously (IV). The blood transfusion protocol was repeated in 9 studies, of which only one RCT had a wider blood transfusion indication[1] than the other 8 observational studies (haemoglobin level, <10 g/dL; haematocrit level, <30%). In summary, the pre-recorded demographic data and blood variables showed no statistically significant differences between the TXA and control groups.
Quality assessment
Table 2 summarises the methodological quality assessments of the 12 selected studies. Only one study was a RCT with a Jadad score of 5, which showed high quality. In the other 12 non-RCTs (2 prospective cohorts and 9 retrospective cohorts), the NOS scores ranged from 6 to 8, indicating that the quality of these studies was also acceptable.
Primary outcomes
1. Allogenic blood transfusion rate
Eleven trials [13-15, 20-27] were used in the meta-analysis of the transfusion requirements (Figure 2.1). Significant heterogeneity was observed, and a random-effects model was used (I2 = 73%, P < 0.0001). The transfusion rate was lower in the TXA group than in the control group (OR = 0.24; 95% CI, 0.14–0.41; P < 0.00001). Both the revision THA subgroup with five studies and the TKA subgroup with 7 studies had a significantly lower allogenic blood transfusion rate than the control group (OR = 0.24; 95% CI, 0.10–0.61; P = 0.003; OR = 0.24; 95% CI, 0.11–0.50; P = 0.0001).
2. Reduction in haemoglobin level
Of the 12 studies [13-15, 20-28], 5 trials with 1022 patients [14, 21, 22, 24, 26] compared preoperative and postoperative Hb levels in patients undergoing revision TJA with or without TXA (Figure 2. 2). High heterogeneity was observed, and a random-effects model was used (I2 = 76%, P = 0.002). The pooled MD in the Hb levels in the experimental group was less decreased than that in the control group (MD = −0.84; 95% CI, −1.28 to −0.41, P = 0.002). Compared with TKA (MD = −0.63; 95% CI, −0.97 to −0.30, P = 0.0002), TXA was more effective in the THA group regarding the reduction in haemoglobin level (MD = −1.28; 95% CI, −2.00 to −0.57; P = 0.0004).
3. Number of RBC units transfused per patient
Nine studies that involved 1,467 subjects [14, 15, 20-26] provided data for this result (Figure 2.3). As no significant heterogeneity was observed, a fixed-effect model was used (I2 = 24%, P = 0.22). The number of RBC infusions per patient in the TXA group was significantly lower than that in the control group (MD = −0.49; 95% CI, −0.61 to −0.38; P < 0.00001). The number of RBC units infused per patient in the revision THA subgroup was significantly lower as compared with that in the control group (MD = −0.81; 95% CI, −1.20 to −0.43; P < 0.0001). Similarly, the number of RBC units infused per patient in the revision TKA subgroup was also lower (MD = −0.46; 95% CI, −0.58 to −0.35; P < 0.00001).
4. Blood loss
Data from 8 studies [13-15, 20-22, 24, 25] that involved 1,116 patients were used to examine blood loss (Figure 2.4). Owing to the high statistical heterogeneity, a random-effects model was used (I2 = 91%, P < 0.00001). The use of TXA in revision TJA resulted in significantly less blood loss than that in the control group (MD = −260.87; 95% CI, −394.88 to −126.86; P = 0.0001). Compared with the control group, the blood loss in the TKA subgroup was significantly reduced (MD = −428.61; 95% CI, −754.37 to −102.84; P = 0.010). However, such statistical significance was not observed in the THA subgroup (MD = −225.59; 95% CI, −475.99 to 24.82; P = 0.08).
Secondary outcomes
5. Venous thromboembolic complications
Seven studies [15, 21, 23, 25-28] reported the postoperative incidence rates of deep vein thrombosis and PE (Figure 2.5). No significant heterogeneity was observed, and a fixed-effects model was used (I2 = 0%, P = 0.78). No statistically significant differences were found between the patients in the TKA subgroup (OR = 0.99; 95% CI, 0.31–3.14; P = 0.98). However, the patients with modified THA using TXA had higher venous thromboembolic complication rates, although no statistically significant difference was observed (OR = 1.56; 95% CI, 0.43–5.59; P = 0.50).
6. Non-thromboembolic complications
Four studies [21, 24, 25, 27] that included 802 subjects described other non-thromboembolic complications (Figure 2.6). Overall, 9 patients developed postoperative infections, 6 had recurrent dislocation, and 4 had periprosthetic fractures. Owing to the high statistical heterogeneity found, a random-effects model was used (I2 = 78%, P = 0.001). In rTJA, the incidence of other complications was reduced in the TXA group as compared with the control group, but the difference did not reach statistical significance (OR = 0.54; 95% CI, 0.18 –1.68; P = 0.29). The OR in the THA and TKA groups were 0.44 (95% CI, 0.03–5.44; P = 0.52) and 0.71 (95% CI, 0.32–1.54; P = 0.38), respectively.
7. Leng<href="#/javascript:;">th<href="#/javascript:;">of hospital <href="#/javascript:;">stay
Two studies [14, 26] that reported LOHS were included in the meta-analysis (Figure 2.7). High statistical heterogeneity was found, and a random-effects model was used (I2 = 75%, P = 0.05). A trend toward a reduction in hospital stay was observed in the TXA group as compared with the control group in the TKA subgroup (MD = −2.89; 95% CI, −4.85 to −0.93; P = 0.004).
Sensitivity analysis
We found a significant heterogeneity in several outcome indicators. To find the source of heterogeneity, we performed a sensitivity analysis to examine whether the presence of a single study had a decisive effect on the results by removing each study in turn (Figure 3). In the sensitivity analysis for the transfusion rate group, we found that when the studies by Reichel et al. [25] and Huerfano et al. [21] were removed, the results changed significantly, indicating that they were the main sources of heterogeneity. In addition, the studies by Mariani et al. [23] and Ortega et al. [14] may be potential sources of heterogeneity. In the sensitivity analysis for the altered haemoglobin group, we found that when the studies by Smi et al. [26] and Park et al. [24] were removed, the results changed significantly, indicating that they were the main sources of heterogeneity. In the sensitivity analysis for the blood loss group, we found that when the study by Peck et al. [28] was removed, the results changed significantly, indicating the study by Peck et al. [28] was the main source of heterogeneity. Heterogeneity was also observed in terms of complications and LOHS. However, owing to the inadequate number of articles included, the definition of complications and LOHS in different regions were different; thus, heterogeneity could not be avoided. We expect more randomised controlled studies with precise conclusions in the future.
Publication bias
We quantitatively analysed publication bias using the Egger method (Figure 4). Except for the outcome group of blood transfusion rate, the other P values were all >0.05, indicating no significant publication bias. In the transfusion rate outcome group, the P values were <0.01, indicating a significant publication bias. However, after removing the missing studies and adding them in the analysis, and then recalculating the effect size, the OR did not change significantly, indicating that our results were relatively robust.