Intra-articular vs. intravenous administration: a meta-analysis of tranexamic acid in primary total knee arthroplasty

Background The optimal dosage and administration approach of tranexamic acid (TXA) in primary total knee arthroplasty (TKA) remains controversial. In light of recently published 14 randomized controlled trials (RCTs), the study aims to incorporate the newly found evidence and compare the efficacy and safety of intra-articular (IA) vs. intravenous (IV) application of TXA in primary TKA. Methods PubMed, Embase, Web of Science, and Cochrane Library were searched for RCTs comparing IA with IV TXA for primary TKA. Primary outcomes included total blood loss (TBL) and drain output. Secondary outcomes included hidden blood loss (HBL), hemoglobin (Hb) fall, blood transfusion rate, perioperative complications, length of hospital stay, and tourniquet time. Result In all, 34 RCTs involving 3867 patients were included in our meta-analysis. Significant advantages of IA were shown on TBL (MD = 33.38, 95% CI = 19.24 to 47.51, P < 0.001), drain output (MD = 28.44, 95% CI = 2.61 to 54.27, P = 0.03), and postoperative day (POD) 3+ Hb fall (MD = 0.24, 95% CI = 0.09 to 0.39, P = 0.001) compared with IV. There existed no significant difference on HBL, POD1 and POD2 Hb fall, blood transfusion rate, perioperative complications, length of hospital stay, and tourniquet time between IA and IV. Conclusion Intra-articular administration of TXA is superior to intravenous in primary TKA patients regarding the performance on TBL, drain output, and POD3+ Hb fall, without increased risk of perioperative complications. Therefore, intra-articular administration is the recommended approach in clinical practice for primary TKA.


Background
Total knee arthroplasty (TKA) is a common major orthopedic surgery, and the demand is still increasing due to human longevity and large population suffering from knee osteoarthritis (OA) around the world [1,2].
TKA is an effective choice for end-stage OA [3]. But it is a major operation especially for the geriatric population, and the postoperative reduced hemoglobin (Hb) might require blood transfusion and potentially result in delayed physical rehabilitation, longer hospital stay, and higher medical cost [4].
Tranexamic acid (TXA) has been widely used in many orthopedic surgeries for controlling blood loss [5]. Its safety and efficacy has been validated by many studies [6][7][8]. However, the optimal administration approach for primary TKA remains to be investigated. Oral administration and intravenous (IV) administration have been validated as an effective approach, but there are potential risks of thromboembolic complications [9,10]. Besides, intra-articular (IA) administration provides a maximum concentration at the bleeding site with limited systemic influence [11].

Methods
Our meta-analysis was conducted in accordance with the guidelines of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement) [27]. We did not publish a protocol for this study.

Literature search
Four electronic databases including PubMed, Embase, Web of Science, and Cochrane Library were searched. Searching was conducted until April 20, 2020, with the following search terms: ("tranexamic acid" OR "TXA") AND ("total knee arthroplasty" OR "total knee replacement" OR "TKA" ). Literatures were limited to English publication. All studies were full text available. Unpublished investigations were not included.

Selection criteria
Two independent reviewers performed the search, removed duplicate records, reviewed the titles and abstracts, and identified studies as included, excluded, or uncertain. Full-text articles were reviewed to determine eligibility if identified uncertain. Disagreements were discussed with a third reviewer.
We retrieved all RCTs that compared IA with IV administration of TXA in patients receiving primary TKA. Inclusion criteria were (1) patients who underwent primary TKA, (2) comparative studies of IA vs. IV administration of TXA, (3) availability of full text, and (4) English publications. Exclusion criteria were (1) noncohort studies, (2) retrospective cohort studies, (3) reviews, and (4) unpublished studies.

Data extraction
The following data were extracted: characteristics of study (design, country, no. of patients, age, sex, body mass index, follow-up, and conclusion), method of administration and operation (IV or IA dosage, type of operation, and surgical approach), and surgical protocols (thromboprophylaxis, DVT screening, prosthetic properties, blood transfusion protocol, tourniquet application, and drainage).
Primary outcomes included total blood loss (TBL), which was calculated by the Gross formula or Hb Was the method used to assess adverse effects described?
No 0 Was the method of statistical analysis described?
No 0 balance method [28,29], and drain output. Secondary outcomes included hidden blood loss (HBL), Hb fall, blood transfusion rate, and perioperative complications including deep vein thrombosis (DVT), pulmonary embolism (PE), wound infection, and other vascular events. The duration of tourniquet application and length of hospital stay were also recorded and analyzed. Missing data were obtained from corresponding authors if possible.

Quality assessment
We assessed the qualities of included studies according to the criteria of the Cochrane Handbook for Systematic Reviews of Interventions [30]. The strength of evidence for each major outcome was evaluated according to the 8-point modified Jadad scale (Table 1) [31]. A study scoring above 4 was considered qualified. A study scoring above or equal to 7 was considered as high-quality evidence.

Assessment of bias
The risk of bias in individual studies was divided into five parts: selection bias (random generation sequence and allocation concealment), performance and detection bias (blind), attrition bias (incomplete data), reporting bias (selective reporting), and other biases. Publication bias across studies would be shown by funnel plot if necessary.

Statistical analysis
We analyzed continuous data by mean difference (MD) and its corresponding 95% confidence interval (CI). Odds ratio (OR) and its corresponding 95% CI were calculated for dichotomous data. We assessed heterogeneity by using the I 2 statistic. I 2 value above 50% was considered as high heterogeneity and a random-effects model would be used, while a value below 50% was considered as low heterogeneity and a fixed-effects model would be adopted [32]. Subgroup analyses would be considered  when meeting high heterogeneity. Statistical analyses were performed using Review Manager 5.3 software. Forrest plots were used to describe the primary results of the meta-analysis. Funnel plots for primary outcomes (TBL and drain output) were generated to evaluate the potential publication bias. P value < 0.05 was considered statistically significant. Formal ethical approval was deemed not necessary in our meta-analysis.

Result
Search results Figure 1 shows detailed steps of the literature search, in which 773 studies were reviewed: 698 studies were excluded after screening titles and abstracts, and the remaining 75 studies were reviewed in full text. After excluding 41 studies according to selection criteria, 34 studies encompassing 3867 patients were included in our study [13-26, 29, 33-51].

Study characteristics and quality assessments
As shown in Table 2, the sample size of the included studies ranged from 25 to 320, and the mean age of patients ranged from 57 to 73. Nine of the studies (9/34, 26.5%) favored IA administration, while four of the studies (4/34, 11.8%) preferred IV administration.

Meta-analysis of outcomes
All the results are listed in Table 6, including primary outcomes, secondary outcomes, three subgroup analyses, and three low heterogeneity analyses.

Total blood loss
Eighteen studies provided valid data of TBL on 1656 patients. Given the presence of significant heterogeneity among studies (P < 0.001, I 2 = 81%), we used a randomeffects model for analysis. IA administration showed a significant advantage compared to IV administration (MD = 63.99, 95% CI = 27.81 to 100.16, P < 0.001). Concerning about the high heterogeneity, we performed a sensitivity analysis based on the risk of bias and got another lower heterogeneity result (Fig. 2) by analyzing 13 studies (P = 0.34, I 2 = 11%) with a fixed-effects model, which still revealed a significant superiority of IA administration (MD = 33.38, 95% CI = 19.24 to 47.51, P < 0.001). Publication bias is shown by a funnel plot (Fig. 3).

Drain output
Seventeen studies involving 1494 patients provided valid data of drain output. Due to significant heterogeneity among studies (P < 0.001, I 2 = 93%), we used a randomeffects model for analysis. IA administration showed a significant advantage (Fig. 4) compared to IV administration (MD = 28.44, 95% CI = 2.61 to 54.27, P = 0.03). Drainage volume was analyzed in subgroup based on the duration of tube clamping. For studies in which the drainage tube was clamped postoperatively less than two hours, a significant superiority was shown in the IA group compared to the IV group (MD = 51.47, 95% CI = 6.02 to 96.92, P = 0.03). Considering the high heterogeneity (P < 0.001, I 2 = 92%), a random-effects was used for analysis. There was no significant difference (MD = 12.40, 95% CI = − 24.85 to 49.65, P = 0.51) for studies in which the drainage tube was clamped postoperatively over 2 h with high heterogeneity (P < 0.001, I 2 = 89%).

Hidden blood loss
Only six studies including 640 patients reported HBL. Since there existed significant heterogeneity among studies (P = 0.006, I 2 = 69%), we used a random-effects model for analysis. There existed no significant difference between the IV and IA groups (MD = 7.57, 95% CI = − 60.34 to 75.47, P = 0.83) on HBL.

Hemoglobin fall
In all, 19 studies involving 1749 patients reported the data of postoperative Hb fall. Because different studies reported Hb of postoperative day (POD) 1 to 5 with high heterogeneity (P < 0.001, I 2 = 87%), we conducted subgroup analyses based on POD1, POD2, or POD3+.
Ten studies involving 1052 patients reported the POD1 Hb fall. The random-effects model (P < 0.001, I 2 = 91%) was used for analysis, and there was no significant difference between the IV and IA groups (MD = − 0.34, 95% CI = − 0.70 to 0.02, P = 0.07). Regarding the high heterogeneity, a sensitivity analysis was performed and two studies were excluded [14,48], then we got a lower heterogeneity result (P = 0.14, I 2 = 36%) by analyzing the rest of 8 studies including 839 patients with a fixed-effects model. No significant difference was shown between the IV and IA groups (MD = − 0.01, 95% CI = − 0.11 to 0.13, P = 0.86).
Eight studies involving 701 patients reported the POD2 Hb fall. Considering the significant heterogeneity among studies (P < 0.001, I 2 = 82%), we used a randomeffects model for analysis. There existed no significant difference between the IV and IA groups (MD = 0.17, 95% CI = − 0.20 to 0.53, P = 0.37). We also performed a sensitivity analysis based on the risk of bias and excluded two studies [23,39] and got a lower heterogeneity (P = 0.11, I 2 = 44%) result by analyzing the rest of six studies involving 531 patients with a fixed-effects model. No significant difference was shown between the IV and IA groups (MD = − 0.08, 95% CI = − 0.25 to 0.09, P = 0.36).
Six studies involving 637 patients reported the POD3+ Hb fall. Because of low heterogeneity among studies (P = 0.18, I 2 = 34%), a fixed-effects model was used for analysis. The IA group showed a significant advantage compared to the IV group (MD = 0.24, 95% CI = 0.09 to 0.39, P = 0.001).

Blood transfusion rate
Twenty-eight studies involving 3270 patients had data on blood transfusion. Transfusions were reported as 109/1664 (6.6%) in the IV group and 99/1606 (6.2%) in the IA group. Only 25 studies with 2950 patients were included in our meta-analysis, while the other three studies reported no transfusion event. The risk of a blood transfusion was similar between the two groups (OR = 0.93, 95% CI = 0.69 to 1.24, P = 0.62), and the data showed low heterogeneity (P = 0.54, I 2 = 0%).

Complications
In our study, certain complications were our concern, including DVT, PE, wound complications, and other adverse events. In all, 33 studies involving 3807 patients mentioned data of complications. The incidence of complications was mentioned as 77/1946 (4.0%) in the IV group and 77/1861 (4.1%) in the IA group. In these 33 studies, 13 of them reported no complication, so only 20 studies with 2594 patients were included in the metaanalysis. The risk was the same between the two groups (OR = 1.00, 95% CI = 0.72 to 1.39, P = 0.98) with low heterogeneity (P = 0.47, I 2 = 0%). In subgroup analysis, complications were classified into four types: DVT, PE, wound complications, and other adverse events. All subgroups showed insignificant differences between the IV and IA groups.
Wound complications included infection, necrosis, delay healing, and dehiscence. There were 58 wound complications reported in 14 studies. A fixed-effects model was used due to low heterogeneity (P = 0.39, I 2 = 6%), and a similar risk of wound complications was shown in two groups (OR = 0.95, 95% CI = 0.58 to 1.55, P = 0.83 ).  Other adverse events were reported in 65 patients of 13 studies. Zhang et al. [15] reported 14 patients with idiopathic venous thromboembolism, and Wang et al. [23] reported one patient with intramuscular vein thrombosis. Besides, Abdel et al. [20] reported one patient with a thrombotic cerebrovascular accident. Functional disorders, such as stiffness, vomiting, nausea, dizziness, constipation, and paresthesia, were also reported in several studies [36,43,46]. A similar risk was shown (OR = 1.10, 95% CI = 0.68 to 1.80, P = 0.69) with low heterogeneity (P = 0.42, I 2 = 2%) between IA and IV.

Length of hospital stay
Seven studies involving 748 patients reported data on length of hospital stay. Because of low heterogeneity (P = 0.35, I 2 = 11%), we used a fixed-effects model for

Duration of tourniquet application
Nine studies including 815 patients reported data of tourniquet time. A fixed-effects model was used for analysis due to the low heterogeneity (P = 0.74, I 2 = 0%). It did not show a statistical difference between the two groups (MD = − 1.22, 95% CI = − 3.06 to 0.62, P = 0.19).

Discussion
The most important finding in our study is that the difference of TBL and drain output between IV and IA administration is supported by newly added RCTs. Based on available evidences, the IA group shows significant superiority over the IV group regarding TBL, drain output, and POD3+ Hb fall. Besides, this study suggests that there exists no statistical difference on HBL, POD1 and POD2 Hb fall, incidence of blood transfusion, length of hospital stay, and time of tourniquet application between the two groups.
As an antifibrinolytic agent, TXA is a synthetic derivative of the amino acid lysine which competitively blocks the lysine-binding sites in the plasmin and plasminogen activator molecules, thereby preventing dissolution of the fibrin clot [54]. A previous study [6], which included 23,236 patients undergoing primary TKA, proved that TXA application was associated with decreased blood loss and transfusion risk without noticeably increased risk of complications. Besides, it could also reduce the risk of venous thromboembolism [6]. Several previous studies have compared IV and IA administration in TKA: Xie et al. [55] included 18 RCTs and found no significant difference between IV and IA. Gianakos et al. [12] included 18 RCTs and 5 non-RCTs, and they found significant differences regarding TBL and drain output between IV and IA. However, it was a study of high heterogeneity. Therefore, we performed this meta-analysis with more newly published RCTs. Moreover, subgroup analysis and sensitivity analysis were performed to reach a more convincing conclusion.
In our study, IA administration shows significant superiority on the TBL to IV group (MD = 33.38, P < 0.001). A previous study indicated easier administration of topical TXA with a maximum concentration at the bleeding site and minimal systemic absorption [53], and therefore, topical application may deliver better blood loss control theoretically. The IA group also shows significant superiority on drain output (MD = 28.44, P = 0.03). The difference is more significant when the drainage tube is clamped postoperatively less than 2 h (MD = 51.47, P = 0.03). However, when the drainage tube is clamped over 2 h after surgery, there exists no statistical difference between them (P = 0.51). It is possibly due to a higher concentration of TXA and longer contact time in the IA approach.
There exists a significant difference on POD3+ Hb fall (MD = 0.24, P = 0.001), while POD1 (P = 0.86) and POD2 Hb fall (P = 0.36) show no noticeable difference between the two groups. POD3+ Hb fall is usually caused by HBL [55]. However, due to the limited data,  [56]. Therefore, it is a reasonable explanation of similar effects on POD1 and POD2, and a better result in the IA group on POD3+.
Fillingham et al. [5] published a clinical guideline of TXA application in joint replacement, but no optimal approach was recommended. In contrast, in our study, IA was found to be of superior value in light of the recently published RCT results. Although we have not compared IA with oral or combined administration, future clinical trials might validate our findings and possibly influence the revision of the clinical guideline of TXA. Besides, Fillingham et al. [5] also admitted dosage amount and multiple doses of TXA did not significantly affect the blood loss. However, several recent studies had different conclusions. Tzatzairis et al. [57] made a comparison between one to three doses of 15 mg/kg TXA intravenously and concluded that the three-dose group displayed better outcome. Lei et al. [58] reach the same conclusion by comparing 20 mg/kg and 60 mg/kg TXA intravenously. Moreover, Zhang et al. [52] even reported a better outcome of six-dose IV TXA. Besides, Tammachote et al. [59] compared high dosage (3 g) with low dosage (0.5 g) for IA TXA and also found a better outcome of high dosage. All the results of recent RCTs favor high-dose administration of TXA. Although TXA dosage and timing were popular topics, there is no meta-analysis about them by now. In our meta-analysis, there existed no standard dosage protocol for included studies (Table 3): In the IV group, 52.9% of the studies (18 studies) used a weight-based dosage (10 to 20 mg/kg) and the rest 47.1% of the studies (16 studies) chose a standard dosage (0.5 to 1.5 g). In the IA group, only 5.9% of the studies (2 studies) used a weight-based dosage (15 mg/kg) and the rest 94.1% of the studies (32 studies) chose a standard dosage (0.75 to 3 g). However, restricted by limited data, we did not perform a subgroup analysis for TXA dose and timing.
Advantages of our study include substantial highquality RCTs (Table 5) and adequate analysis. 73.5% of the studies (25 studies) have detailed random generation sequence, and 55.9% of the studies (19 studies) have adequate allocation concealment. Besides, 73.5% of the studies (25 studies) are recent studies (published after 2015). Our analyzing methods are subgroup analysis and sensitivity analysis when the previous analysis has high heterogeneity.
There are several limitations in our studies. Firstly, the inherent bias in different studies because of the inconsistent threshold for blood transfusion cannot be overlooked. Besides, the DVT rate might be influenced by the inclusion criteria, and the RCT of TXA in a DVT high-risk population might be required to validate our findings. Furthermore, repeated dose seemed a better choice than a single dose in both IV and IA administration [52,[57][58][59], and therefore, different methods of administration may influence the result. Lastly, data for HBL, length of hospital stay, and duration of tourniquet application are limited for analysis, and costeffectiveness remains to be investigated.

Conclusion
IA administration of TXA is superior to IV TXA in patients receiving primary TKA regarding the performance on TBL, drain output, and POD3+ Hb fall, without noticeably increased risk of complications. Therefore, IA administration should be the preferred approach in clinical practice.