Effectiveness and Safety of Interventions for Treating Adults with Displaced Proximal Humeral Fracture: A Network Meta-Analysis and Systematic Review

Purpose Network meta-analysis (NMA) is a comparatively new evidence-based technique in medical disciplines which compares the relative benefits associated with multiple interventions and obtains hierarchies of these interventions for various treatment options. We evaluated the effectiveness and safety of open reduction and internal fixation (ORIF), hemiarthroplasty (HA), reverse shoulder arthroplasty (RSA), intramedullary nailing (IN) and non-operative treatment (NOT) of displaced proximal humeral fractures in adults using Bayesian NMA of data from clinical trials. Method PUBMED, EMBASE and CENTRAL in July 2016 were searched and clinical trials that evaluated interventions for treating adults with displaced proximal humeral fractures were identified. Methodological qualities of studies were assessed by the Newcastle—Ottawa Scale and risk of bias using the Cochrane Collaboration tool. Result Thirty-four trials involving 2165 participants were included in the study. RSA had significantly the highest Constant score and lower total incidence of complications than ORIF, HA and IN. Moreover, RSA resulted in a lower incidence of additional surgery than ORIF and IN. The rank of treatments in terms high Constant score was: RSA, ORIF, IN, NOT and HA. The rank for reduction in total incidence of complications was: RSA, NOT, HA, IN and ORIF. For lowering the risk of additional surgery, the rank was: RSA, NOT, HA, IN and ORIF. Conclusion RSA had the highest probability for improving functional outcome and reduction in the total incidence of complications and requiring additional surgery among the five interventions for treating adults with displaced proximal humeral fracture.


Introduction
Fractures of the proximal humerus are the third most common in elderly patients after those of the hip and distal radius [1], accounting for 5% to 6% of all adult fractures [2]. Their incidence rapidly increases with age, and women are affected between two and three times as often as men [2][3][4]. Both non-operative and operative methods are used to treat these fractures. Non-operative treatment (NOT) involves a period of immobilization, such as an arm sling, followed by physiotherapy and exercise. It is generally the accepted treatment option for minimally displaced fractures and often used also for people with displaced fractures [3]. Operative treatment is recommended for displaced and unstable fractures and those with more complex fracture patterns to avoid painful and dysfunctional malunion [5]. The common operative treatments include open reduction and internal fixation (ORIF), hemiarthroplasty (HA), reverse shoulder arthroplasty (RSA), intramedullary nailing (IN). The debate among surgeons regarding which treatment should be used is still unresolved today and has been since the 1980s. Previous pairwise meta-analyses [3,[5][6][7][8] could not obtain hierarchies of these treatments because some had not been compared one by one.
We compared the effectiveness and safety of these five treatments (NOT, ORIF, HA, RSA, IN) for displaced proximal humeral fracture in adults by network meta-analysis (NMA). Our intention was to provide hierarchies of the comparative Constant score, total incidence of complications and need for additional surgery.
NMA is a powerful technique that has been used for more than a decade to rank treatment options with both direct comparisons of treatments in randomized controlled trials and indirect comparisons across trials with a common comparator. We ensured validity through careful review design and rigorous analysis of criteria for inclusion of the various studies.

Eligibility criteria and literature search
This study was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [9]. We searched the Cochrane Register of Controlled Trials (CENTRAL, Jul 2016), PubMed (Jan 1980 to Jul 2016), and EMBASE (1980 to Jul 2016) databases to identify all studies that discussed the effectiveness and safety of NOT, ORIF, HA, RSA, IN. Keywords and MeSH terms used in the search strategy included "proximal humeral fracture", "non-operative treatment", "open reduction and internal fixation", "hemiarthroplasty", "reverse shoulder arthroplasty" and "intramedullary nailing".
The inclusion criteria were: (1) target population: patient was aged 16 years or older and presented within 3 weeks after sustaining a displaced fracture of the proximal humerus; (2) intervention: NOT, ORIF, HA, RSA, IN; (3) methodological criteria: randomized controlled trials and clinical trials. The exclusion criteria were: (1) target population: patient was aged under 16 years or presented more than 3 weeks after sustaining a displaced proximal humeral fracture; (2) methodological criteria: case reports and cohort studies. The study selection was conducted by two independent reviewers. Any disagreement between review authors was resolved by discussion.

Outcome assessment
The primary outcome measure was Constant score [10] (activity, mobility, strength and pain). The secondary outcome measures included total complications (e.g. surgical site infection, symptomatic malunion, transient paresthesia and avascular necrosis of the humeral head) and incidence of additional surgery.

Data extraction and quality assessments
Study type, country, sample size, length of follow-up and interventions data were gathered from each trial. Data on random sequence generation, allocation concealment, blinding, selective reporting and incomplete outcome data were gathered from randomized controlled trials. Data on representativeness of cases, selection of controls, definition of controls, comparability of cases and controls, ascertainment of exposure, equivalent methods of diagnosis and determination of response rate for cases and controls were gathered from controlled clinical trials. In addition, the following clinical data were extracted if available: Constant score, total number of complications and incidence of additional surgery.
The Cochrane Collaboration tool for assessing risk of bias [11] was used to assess the quality of randomized controlled trials, and the Newcastle-Ottawa Scale [12] was used to assess the quality of case-control trials in terms of selection and comparability of the study groups and determination of outcomes. In evaluating randomized controlled trials by the Cochrane Collaboration tool, quality of the studies was assessed using the following criteria: (1) randomization sequence generation: assessment for selection bias; (2) allocation concealment: assessment of selection bias; (3) level of blinding (blinding of participants and blinding of outcome assessment): assessment for performance bias and detection bias; (4) incomplete outcome data: assessment for attrition bias; and (5) selective reporting: assessment for reporting bias [11]. For case-control studies, the total Newcastle-Ottawa Scale score was calculated with a maximum of nine points using the criteria listed in S1 Table [12].

Data synthesis and analysis
Two researchers extracted data independently according to the prespecified selection criteria. Disagreements were resolved by discussion. In each study, the relative risk (RR) was calculated for dichotomous outcomes (e.g. total complications and incidence of additional surgery), and treatment effects for continuous outcomes (e.g. Constant score) including mean differences (MDs) for studies with comparable outcome measures used a 95% confidence interval (CI).
We performed conventional pairwise meta-analyses for all outcomes and comparisons, using a random effects model by STATA (version 12.0, Stata Corp, College Station, TX). The pooled estimates of RRs or standardized MDs and 95% CI of three outcomes (Constant score, total incidence of complication and additional surgery) were shown. NMA combines direct and indirect evidence within a Bayesian framework and was implemented using WinBUGS statistical software (version 1.4.3) using Markov Chain Monte Carlo (MCMC) methods. The models, codes and software used in this study are available free online [13]. We performed surface under the cumulative ranking curve (SUCRA) probabilities to rank the five interventions for treating displaced proximal humeral fractures [14]. SUCRA is a proportion, expressed as a percentage of the efficacy of an intervention on the outcome that would be ranked first without uncertainty, and so equals 100% when the treatment is certain to be the best and 0% when certain to be the worst [14].
Inconsistencies of this NMA were assessed by the Higgins model. Significance levels smaller than 0.05 were interpreted as evidence of inconsistency. The sensitivity analysis was performed by comparing the results of different effects models (random effects and fixed effects model).

Study selection
Fig 1 shows the study selection process according to the PRISMA statement. This search strategy retrieved a total of 692 studies: 94 studies were from CENTRAL, 315 studies from PUBMED and 283 studies from EMBASE. Titles and abstracts of these references were examined by two reviewers, and 34 studies  were identified for further analysis. One study  [49] was excluded because the operative treatment included two interventions (HA and ORIF). We also found that Fjalestad (2010) [50], Fjalestad (2012) [51] and Fjalestad (2014) [24] reported on the same patients at different follow-up times, so we included only one article of the multiple studies. Ten randomized controlled trials and twenty four controlled clinical trials remained and were considered for primary relevant studies, which were all included in this NMA. Table 1 provides a summary of the studies in the review. A total of 2165 participants (mean age: 69 years) were included. For each intervention, 748 patients (mean age: 66.06 years) were assigned to ORIF therapy, 803 (mean age: 71.3 years) to HA therapy, 191 (mean age: 77.3 years) to RSA therapy, 267 (mean age: 61.8 years) to IN therapy and 153 (mean age: 73.9 years) to NOT therapy. Study sample size ranged from 18 to 368. All 34 studies directly compared one treatment with another. These studies were published between 1984 and 2014. Twenty four studies reported Constant score as an outcome. Thirty one studies used total complications as an outcome and twenty studies reported incidence of additional surgery as an outcome.
The network of comparisons on Constant score is shown in Fig 3. Table 3 provides hierarchies of effect size on Constant score. Ranking graphs of the distribution of probabilities of Constant score are displayed in Fig 4. The direct and indirect comparisons indicated that HA significantly decreased the Constant score compared with the other groups and RSA was responsible for a significantly higher Constant score compared with the other groups. Based on SUCRA, HA ranked first (0.9675), the second was NOT (0.5905), IN was third (0.4805), the fourth was ORIF (0.4520) with RSA being last (0.0095).

Total incidence of complications
For this outcome, 31 trials were included in the NMA. The network of comparisons on total incidence of complications is shown in Fig 5. Table 4 provides hierarchies of effect size on their incidence. The ranking graphs of distribution of probabilities of incidence can be seen in Fig 6. Direct and indirect comparisons indicate that RSA results in a lower incidence of

Incidence of additional surgery
For this outcome, 20 trials were included in the NMA. The network of comparisons on incidence of additional surgery is shown in Fig 7. Table 5 provides the hierarchies of effect size on incidence of additional surgery, with ranking graphs of the distribution of probabilities on incidence of additional surgery in Fig 8. Only indirect comparison indicates that RSA results in a lower incidence of additional surgery than ORIF and IN. Both direct and indirect comparisons suggest that HA results in a lower incidence of additional surgery than ORIF. Based on SUCRA, RSA ranks first (0.9450), NOT second (0.6870), the third was HA (0.5538), the fourth was IN (0.2272) with ORIF last (0.0870). Inconsistency and sensitivity analysis      Interventions for Treating Proximal Humeral Fractures

Discussion
The NMA provided hierarchies for the Constant score, incidence of total complications and additional surgery in adult with displaced proximal humeral fracture treated with different methods, which had advantages over comparison with traditional pairwise meta-analyses [3,[5][6][7][8]. The meta-analysis indicated that: (1)     2165 participants with the mean age of 69 years were included in this network meta-analysis. So the findings were suitable for the patients with mean age of more than 60 years. There are some particular strengths to the analysis in this NMA: (1) It was conducted using common methods and was designed to allow for reproducible research selection and inclusion; (2) a broad and extensive search strategy was used to minimize the possibility of publication bias; (3) the study overcomes a major limitation of conventional pairwise meta-analysis by combining direct and indirect evidence of the efficacy of treatment strategies; (4) the SUCRA and posterior probabilities of outcomes were used to distinguish the subtle differences among five treatments.
However, this analysis has several limitations. Firstly, randomized controlled trials and case -control studies were both included in the analysis, and the case-control studies may have reduced the significance of the conclusions. Secondly, the lack of any treatment-provider blinded studies may have introduced detection bias, in which the assessors are likely to have preferentially attributed the occurrence of injury to the control group. Thirdly, some of the study characteristics such as type of fracture, age and performance bias might be potential obstacles to the outcomes of our study. Finally, these interventions (NOT, ORIF, HA, RSA and IN) may have different indications, so the comparisons between treatments within trials and sample for each trial may have interacted in ways that this analysis would not reveal. However, our network meta-analysis can still provide useful information about effectiveness and safety of interventions for treating adults with displaced proximal humeral fracture to the surgeons.
Xie [52] previously reported that operative treatments (including ORIF and HA) did not significantly improve Constant score and led to a higher incidence of postoperative complications compared with NOT. Li [6] also indicated that ORIF did not improve the Constant score when compared with NOT. Dai [5] showed that ORIF resulted in better Constant score Interventions for Treating Proximal Humeral Fractures than HA, and HA could reduce the rate of revisions and the method-related complications significantly. Wang [7] stated that treatment with ORIF caused no significant difference in Constant score and total number of complications compared with IN. Compared with HA, Shukla [53] reported that RSA resulted in more favorable Constant scores and Zhang [8] demonstrated that RSA was associated with a lower rate of total complications. Our network analysis showed that RSA resulted in significantly higher Constant score compared with the other four interventions. We also found that RSA resulted in lower total incidence of complications than ORIF, HA and IN. RSA also resulted in a lower incidence of additional surgery than ORIF and IN. However, we also use the SUCRA and posterior probabilities of outcomes to distinguish the subtle differences among five treatments. For achieving higher Constant scores the rank on treatments was: RSA, ORIF, IN, NOT and HA. For reducing incidence of total complications, the rank on treatments was: RSA, NOT, HA, IN and ORIF. For lowering risk of additional surgery, the rank on treatments was: RSA, NOT, HA, IN and ORIF.

Conclusions
In summary, this Bayesian NMA of data from clinical trials demonstrated that RSA has the highest probability of improving the functional outcome and reducing the total incidence of complications and additional surgery among the five interventions for treating adults with displaced proximal humeral fracture.