Application of postoperative autotransfusion in total joint arthroplasty reduces allogeneic blood requirements: a meta-analysis of randomized controlled trials

A systematic review was performed in accordance with guidelines [25]. All holdings of PubMed, the Cochrane Library, and Embase were searched for relevant trials published until February 2016 using the following terms: (1) “autotransfusion,” “autologous blood transfusion,” “blood transfusion,” or “autologous transfusion”; (2) “total knee arthroplasty,” “total knee replacement,” “TKA,” “total hip arthroplasty,” “total hip replacement,” or “THA”; and (3) “postoperative,” or “post-operation.” Only English studies with full texts were included in the final analysis. We chose the most recent study if several publications reported on the same set of patients. The search strategy is provided in Additional file 1.

Eligible literature was carefully identified and selected according to the flow chart in Fig. 1. Studies were included on the basis of the following criteria: (1) a randomized controlled trial (RCT) was designed; (2) the comparison was between a postoperative autotransfusion system and no drainage/regular drainage (i.e., suction drainage, vacuum drainage, hemovac drainage); (3) the results included key data such as transfusion rate; and (4) patients underwent total joint replacement. The exclusion criteria were as follows: (1) an anticoagulant was added to the autotransfusion system; (2) patients underwent bilateral surgery or revision surgery; and (3) patients underwent TKA without a tourniquet because its use would have affected blood loss [26].

Two authors (Weiping Ji and Xianfeng Lin) reviewed all titles and abstracts independently to determine which articles met the inclusion criteria. All variables and outcomes of interest were extracted independently. The authors solved disagreements by in-person discussion.

The study group referred to the autotransfusion group and the control group encompassed the regular drainage group and no drainage group. We extracted the following characteristics for all studies: year, country of study, type of surgery, population information (age, sample size and gender), autotransfusion systems, and funding sources. The outcomes included the numbers of patients needing allogeneic transfusion, postoperative Hb levels on day 1, day 2, day 3, and day 5, and adverse effects. In this meta-analysis, the transfusion rate was the primary outcome and defined as the rate of patients receiving allogeneic transfusion in both the study group and regular drainage/no drainage group postoperatively during hospitalization. The adverse events included complications such as wound infection and febrile reactions recorded in-hospital. Data relating to Hb levels and adverse effects were extracted to assess recovery after total joint arthroplasty. Blood loss volume, amount of blood transfusion per patient, length of hospital stay, and adverse reactions such as deep vein thrombosis (DVT) and swelling were extracted (data not shown), but the data were insufficient to be included in the meta-analysis.

The Cochrane Collaboration’s tool was used to assess the bias of all eligible RCTs [27]. The assessment criteria were as follows: (1) the method of randomization was adequate; (2) the treatment allocation was concealed; (3) the groups were similar in the most important prognostic indicators at baseline; (4) the patients were blinded to the intervention; (5) the caregivers were blinded to the intervention; (6) the outcome assessors were blinded to the intervention; (7) co-interventions were controlled; (8) compliance was acceptable in all groups; (9) the dropout rate was described and acceptable; (10) the timing of assessment in all groups was the same; and (11) intention-to-treat analysis was performed. We awarded a score of 1 for each item if it was completely met; scores of 0.5, 0, and N/A were awarded for partly met, not met, and unclear, respectively. The total score of each study was then calculated and the validity scores of >6 were considered to represent a low degree of bias.

Statistical analyses were performed with STATA Version 12.0 (StataCorp LP, College Station, TX, USA). Relative risks (RRs) and corresponding 95% confidence intervals (CIs) were estimated by fixed-effect or random-effect meta-analyses. Weighted mean difference was used to perform meta-analysis for continuous outcomes. The significance of the pooled RRs was evaluated by a Z-test and p < 0.05. The I² statistic was used to evaluate heterogeneity. A random-effect model was used when significant heterogeneity was detected among studies (p < 0.1, I² > 50%). Otherwise, a fixed-effect model was adopted (I² < 50%, p > 0.1). The fixed-effect model was adopted if a small number of studies (< 5) were included in the meta-analysis. Considering the source of heterogeneity, the studies were classified by control intervention. To explain some of the clinical heterogeneity observed in the included trials, a subgroup analysis by type of control intervention was performed for the primary outcome. Two sensitivity analyses were performed to assess the stability of pooled effects for TKA and THA. The RRs calculated after excluding a single study were compared with initial effects (before excluding). The results could be considered stable if the RRs did not change significantly (not beyond the initial CI). Egger’s test and Begg’s test were used to test for publication bias by assessing the association between intervention effects and a measure of study size. Egger’s test used actual standardized effect size while Begg’s test used ranks of effect sizes and variances. Funnel plots were used to depict asymmetry on visual inspection. A value of >0.05 for Pr > |z| (continuity correct) suggested no publication bias [28]. All p-values were two-sided.

Seventeen RCTs [2, 7, 12‐18, 22‐24, 29‐33] were assessed and 2314 patients were included in the meta-analysis (Table 1). Sample sizes ranged from 36 to 231. Three studies included two control groups [13, 18, 33]. Another three studies included two types of surgery: TKA and THA [2, 17, 23]. The criteria for transfusion thresholds were varied and based on authors’ experiences, anesthetists’ protocols, or hospital policies (data not shown). The data of tranexamic acid use were rarely provided in the included studies. All of the included studies were published in 1992 or later. Validity scores are shown in Table 2 and two studies with validity scores of <6 were considered as having a high degree of bias [16, 33].

The autotransfusion group had a significantly lower requirement for postoperative ABT (p < 0.001; RR = 0.446 [95% CI = 0.287, 0.693]; I ² > 50%) than the regular drainage/no drainage group (Fig. 2). Although the autotransfusion group did not show superior reductions in ABT requirements when compared with the no-drainage group (p = 0.456; RR = 0.804 [95% CI = 0.453, 1.426]; I ² < 30%), a significant reduction in ABT requirements was noted when compared with the regular drainage group (p < 0.001; RR = 0.377 [95% CI = 0.224, 0.634]; I ² > 50%). The sensitivity analyses revealed stable results (Additional file 2: Figure S1), namely that the excluded studies did not influence the pooled RRs. The pooled results of post-operation Hb level (four studies included) suggested that the autotransfusion group had higher post-operative Hb levels than the regular drainage/no drainage group (Additional file 3: Figure S2). Six studies reported wound infections (three studies included) and febrile reactions (three studies included). We found no significant differences between the autotransfusion group and the regular drainage/no drainage group (Additional file 4: Figure S3). The reliability of these results was limited by the small number of studies included (< 5).

The autotransfusion group had a significantly reduced need for postoperative ABT (p = 0.020; RR = 0.757 [95% CI = 0.599, 0.958]; I ² < 30%) than the regular drainage/no drainage group (Fig. 3). Similar results were found when the autotransfusion group was compared with the regular drainage group and no-drainage groups. Specifically, ABT requirements significantly decreased when the autotransfusion group was compared with the regular drainage group (p < 0.001; RR = 0.536 [95% CI = 0.379, 0.757]; I ² < 30%). However, the autotransfusion group did not demonstrate a significant reduction in ABT requirements compared with the no-drainage group (p = 0.904; RR = 1.020 [95% CI = 0.740, 1.405]; I ² < 30%). Sensitivity analyses proved the results to be stable (Additional file 5: Figure S4). The autotransfusion group also had higher Hb levels (two studies included) compared with the regular drainage/no drainage group (Additional file 6: Figure S5). The pooled results for adverse effects (four studies reporting wound infections and two studies reporting febrile reactions) showed no significant differences between the groups (Additional file 7: Figure S6). The small number of included studies (< 5) limited the reliability of these results.

Egger’s and Begg’s tests suggested a lack of publication bias with respect to ABT rate (Additional file 8: Table S1). According to this analysis, the included studies were relatively comprehensive and yielded statistically reliable results. The funnel plots did not reveal obvious asymmetry of the included studies (Additional file 9: Figure S7 and Additional file 10: Figure S8).