The Clinical Effect of Arthroscopic Rotator Cuff Repair techniques: A Network Meta-Analysis and Systematic Review

Rotator cuff tears are common and are associated with shoulder pain, disability, and dysfunction. Previous studies that have reported different arthroscopic techniques are controversial. A network meta-analysis with indirect and direct evidence was performed to compare the effectiveness of arthroscopic techniques for the treatment of rotator cuff tears. PUBMED, the Cochrane Register of Controlled Trials, EMBASE and Scopus were searched based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement, and related studies that evaluated arthroscopic techniques for the treatment of rotator cuff tears were identified in May 2018. The primary outcome measure was the retear rate. The secondary outcome measures included the constant score and the range of motion (forward flexion and external rotation). Twenty-one trials comprising 1815 shoulders were included in the study. This study indicated that single-row (SR) repair resulted in a higher retear rate than suture bridge (SB) and double-row (DR) repairs. Moreover, the SR and DR repairs resulted in higher incidences of retear than SB repair. The ranking of the treatments based on the constant score and external rotation was SB repair, SR repair and DR repair, whereas the treatment ranking according to forward flexion was SB repair, DR repair and SR repair. In summary, this network meta-analysis provides evidence that SB repairs might be the best choice to improve the postoperative recovery of function and decrease the retear rate.

The relative efficacies of SR repair, DR repair and SB repair for rotator cuff tears in patients were compared using both network meta-analyses and pairwise meta-analyses to provide recommendations based on the comparative retear rate, constant score, external rotation and forward flexion.

Methods
This systematic review was designed and performed with an a priori protocol (PROSPERO 42017071720) established according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement for network meta-analyses and systematic reviews for healthcare interventions 17 . Data extraction and quality assessments. The country, study design, patient sample size, different interventions and lengths of follow-up of the included studies were collected. Additionally, the primary outcomes were the retear rate, constant score, external rotation and forward flexion. Two researchers extracted the above data independently, and disagreements were resolved by discussion.
The Cochrane Collaboration tool 18 was used to assess the risk of bias for the RCTs, and the Newcastle-Ottawa Scale 19 was used to judge the quality of prospective and retrospective comparative trials. For RCTs, data were obtained regarding random generation, allocation concealment, blinding of the outcomes, incomplete outcomes, www.nature.com/scientificreports www.nature.com/scientificreports/ selective reporting and other biases 18 . For case-control studies, selection, comparability and exposure were obtained. A total score based on the Newcastle-Ottawa Scale was calculated as presented in Table S1 19 .
Outcome assessment. The primary outcome for analysis was the retear rate. The secondary outcome measures included the constant score and the range of shoulder motion (external rotation and forward flexion).

Data synthesis and statistical analysis.
Conventional meta-analyses were performed for all primary and secondary outcomes using a random effects model with Stata software (version 13.0). The pooled estimates of the odds ratios (ORs) or standardized mean differences (MDs) and 95% confidence intervals (CIs) of the four outcomes (i.e., retear rate, constant score, external rotation and forward flexion) are presented 20 . Heterogeneity was assessed by chi-square tests and I-square tests. Significance values less than 0.10 for chi-square tests or more than 50% for I-square tests were interpreted as evidence of heterogeneity. Egger's linear regression test were used for detectint the publication bias with Stata. Network meta-analyses using indirect and direct evidence based on a frequentist framework including network plot, forest plot and predictive interval plot were performed with Stata software (version 13.0). The codes, model and network graphs package used in this meta-analysis are free online 21,22 . The surface under the cumulative ranking curve (SUCRA) was used to determine the ranking of the three arthroscopic rotator cuff repair techniques in terms of rotator cuff retear 23 . The SUCRA was used to measure the therapies; 100% indicates the best treatment, and 0% indicates the worst treatment 23 . Dias model was used to assess the inconsistencies in this study 24 . Fixed effects models and random effects models were used to assess the sensitivity analyses.

Results
Study selection. The study selection flow is presented in Fig. 1 and was conducted according to the PRISMA statement. A total of 1407 studies (283, 321, 803 and 818 from the Cochrane Register of Controlled Trials, PUBMED, EMBASE and Scopus, respectively) were originally included in this study. After two researchers read the full-texts, twenty-one studies 6,25-44 (nine RCTs 6,[25][26][27]29,30,34,35,38 and twelve observational comparative studies 28,[31][32][33]36,37,[39][40][41][42][43][44] were considered relevant and included in this meta-analysis. Characteristics and qualities of the included studies. Twenty-one trials containing 1815 shoulders with evaluated treatments were included in the network meta-analysis. The characteristics of the included studies are presented in Table 1. Of the nine randomized controlled trials (RCTs) analyzed, the Cochrane Collaboration tool indicated that eight studies 6 A total of 608 shoulders underwent SR repair, 437 underwent DR repair, and 232 underwent SB repair. The network comparisons for the retear rate are presented in Fig. 3a. The effect size (OR with 95% CI) of the retear rate via direct comparison by pairwise meta-analysis is presented in Fig. 4a-c, The effect size (OR with 95% CI and 95% Prl) of the retear rate vianetwork meta-analysis is presented in Fig. 3b. Both the network meta-analysis and pairwise meta-analysis indicated that SR repair resulted in a higher retear rate than the SB [network: 0. 40 Constant score. Regarding the constant score, 12 trials 25,26,28,31,[33][34][35][36][39][40][41]44 including 1031 shoulders reported constant score data that met the criteria for inclusion in this network meta-analysis. The network used to compare the constant scores is illustrated in Fig. 5a. Hierarchies of effect size (MD with 95% CI) based on pairwise meta-analysis is presented in Fig. 4d-f. Hierarchies of effect size (MD with 95% CI and 95% Prl) based on www.nature.com/scientificreports www.nature.com/scientificreports/ network meta-analysis is presented in Fig. 5b. Both pairwise meta-analysis and network meta-analysis indicated no significant differences among SR repair, DR repair and SB repair in terms of the constant score. Based on the SUCRA probability, the ranking from first to third was SB repair (0.720), SR repair (0.539) and DR repair (0.242).
External rotation. Regarding external rotation, five trials [32][33][34]38,44 were included in this network meta-analysis. The analysis of the comparisons of external rotation is presented in Fig. 6a. Hierarchies of effect size (MD with 95% CI) based on pairwise meta-analysis is presented in Fig. 7a-c. Hierarchies of effect size (MD with 95% CI and 95% Prl) based on network meta-analysis is presented in Fig. 6b. Both pairwise meta-analysis and network meta-analysis indicated no significant differences among SR repair, DR repair and SB repair in terms of external rotation. Based on the SUCRA probability, the ranking from first to third was SB repair (0.700), SR repair (0.425) and DR repair (0.376).

Forward flexion.
Regarding forward flexion, six trials 29,32-34,38,44 were included in this network meta-analysis, and the analysis of the comparisons of forward flexion is illustrated in Fig. 8a. Hierarchies of effect size (MD with 95% CI) based on pairwise meta-analysis is presented in Fig. 7d-f. Hierarchies of effect size (MD with 95% CI and 95% Prl) based on network meta-analysis is presented in Fig. 8b. Both pairwise meta-analysis and network meta-analysis indicated no significant differences among SR repair, DR repair and SB repair in terms of forward flexion. Based on the SUCRA probability, the ranking from first to third was SB repair (0.610), DR repair (0.483) and SR repair (0.408).   www.nature.com/scientificreports www.nature.com/scientificreports/ Inconsistency, publication bias and sensitivity analyses. In general, the results based on the pairwise meta-analysis matched well with those based on the network meta-analysis. In this study, no inconsistencies for each outcome was found in the Dias model analysis ( Table 3). The publication bias was only detected in the comparison of DR repair versus SR repair for the outcome of forward flexion (Table 4). This may be due to the number of studies included in this analysis was small. Fixed effects and random effects models were compared to detect the sensitivity. The results based on the fixed effects model (effective number of parameters  Discussion. The network meta-analysis revealed hierarchies based on the primary (retear rate) and secondary outcomes (constant score, external rotation and forward flexion) for patients diagnosed with rotator cuff tears who underwent SR, SB and DR repair. The meta-analysis indicated the following: (1) SR repair resulted in a higher retear rate than DR and SB repairs; (2) the SR and DR repairs had higher incidences of retear than SB repair; (3) the ranking based on the constant score was SB repair, SR repair and DR repair; (4) the ranking for external rotation was SB repair, SR repair and DR repair; and (5) the ranking for forward flexion was SB repair, DR repair and SR repair.  www.nature.com/scientificreports www.nature.com/scientificreports/ There are some advantages of this type of network meta-analysis: (1) a network meta-analysis provides greater accuracy than a conventional meta-analysis 15 ; (2) the network meta-analysis provided indirect comparisons 45 , and SUCRA, which could improve the precision of even slight distinctions between SR, SB and DR repairs.    www.nature.com/scientificreports www.nature.com/scientificreports/ However, this study also has some limitations. First, some low quality RCTs and case-control studies were included in this study, which may have reduced the significance of the conclusions. Second, the sample size was small, which reduced the accuracy. Third, some potential biases in the data, such as the size of rotator cuff tear, patient age and performance bias, could have affected the accuracy. Finally, some studies did not provide detailed information, such as the standard deviations; therefore, the inadequate data were replaced with statistical methods based on the provided data.
Retear remains a major concern after arthroscopic rotator cuff repair. Hein 10 previously performed a qualitative review without quantitative analysis to report that both DR repair and SB repair resulted in a lower retear rate than SR repair in most tear size categories, and differences in retear rates were found between DR and SB repair. Spiegl 8 summarized eight reviews of DR and SR repair and found a higher retear rate following SR repair than after DR repair. Sobhy 46 also found that SR repair resulted in a higher radiological overall detected retear rate and radiological partial-thickness retear rate than DR repair. Moreover, Bedeir's study 47 suggested that DR and SB repair had a higher type 2 retear (medial cuff failure) rate than SR repair. Our network meta-analysis combined indirect and direct evidences to calculate the difference among multiple techniques for rotator cuff repair using quantitative method, which revealed that the DR and SB repairs resulted in lower retear rates than SR repair. Additionally, no statistically significant difference was found between the DR and SB repairs regarding the retear rate. Furthermore, the posterior probabilities and SUCRA were used to differentiate the slight distinctions among SB, DR and SR repair. In terms of achieving lower retear rates, the treatments were ranked as follows: SB repair, DR repair, and SR repair.
The constant score includes shoulder function, range of motion, pain and strength and is used to assess shoulder therapies 48 . In a previous review 8 , no significant difference was found between SR repair and DR repair in terms of the constant score. No previous evidence-based study has analyzed the difference between SB repair and SR or DR repair. In this network meta-analysis, no significant difference was observed among SR repair, DR repair and SB repair in terms of the constant score. Based on the SUCRA, the treatments were ranked as SB repair, SR repair, and DR repair in terms of the achievement of higher constant scores.
The range of motion is critical for assessing the therapeutic value of these three arthroscopic methods. Prasathaporn 14 previously reported that DR repair provides greater external rotation but found no statistically significant difference in forward flexion. Xu 49 found no significant differences in external rotation or forward flexion. This network meta-analysis revealed no significant differences among SR repair, DR repair and SB repair in terms of external rotation or forward flexion. Moreover, SB repair provided a greater range of motion (external rotation and forward flexion) than SR and DR repair according to the SUCRA.
In summary, this network meta-analysis provides evidence that SB repairs might be the best choice to improve the postoperative recovery of function and decrease the retear rate because of the greater tendon-bone contact area and the higher tendon-bone contact pressure.