Intra-articular Hyaluronic Acid in Treating Knee Osteoarthritis: a PRISMA-Compliant Systematic Review of Overlapping Meta-analysis

Numerous meta-analyses have been conducted aiming to compare hyaluronic acid (HA) and placebo in treating knee osteoarthritis (OA). Nevertheless, the conclusions of these meta-analyses are not in consistency. The purpose of the present study was to perform a systematic review of overlapping meta-analyses investigating the efficacy and safety of HA for Knee OA and to provide treatment recommendations through the best evidence. A systematic review was conducted based on the PRISMA guidelines. The meta-analyses and/or systematic reviews that compared HA and placebo for knee OA were identified. AMSTAR instrument was used to evaluate the methodological quality of individual study. The information of heterogeneity within each variable was fetched for the individual studies. Which meta-analyses can provide best evidence was determined according to Jadad algorithm. Twelve meta-analyses met the eligibility requirements. The Jadad decision making tool suggests that the highest quality review should be selected. As a result, a high-quality Cochrane review was included. The present systematic review of overlapping meta-analyses demonstrates that HA is an effective intervention in treating knee OA without increased risk of adverse events. Therefore, the present conclusions may help decision makers interpret and choose among discordant meta-analyses.

reported that HA is not recommended in the treatment of Knee OA 13 . However, Altman et al. 14 investigated ten guidelines regarding the use of HA for the treatment of knee OA and reported that the recommendations were highly inconsistent as a result of the variability in guideline methodology. Thus, the inconsistent recommendations make it difficult for clinical professionals to determine its appropriateness when treating knee OA.
The purpose of the present study is to perform a systematic review of overlapping meta-analyses determining the clinical effects of HA in treating Knee OA, to evaluate the mythological quality of included individual meta-analyses, and to take best evidence through the currently inconsistent evidence.

Materials and Methods
Search strategy. The present systematic review was conducted following the guideline of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analysis) statement 15 . PRISMA statement was used to guarantee high-quality reporting of systematic reviews or meta-analyses 16 . Electronic databases including MEDLINE, EMBASE and Cochrane library were searched for all meta-analysis or systematic review published through Nov 2015. The following MeSH items or free words were taken: osteoarthritis, knee, meta-analysis, systematic review, and hyaluronic acid. The references of searched studies were also reviewed to explore other meta-analyses or systematic reviews. No restrictions were made on the publication language.
Inclusive and exclusive criteria. Studies were considered eligible for inclusion if they met the following criteria: (1) Meta-analyses or systematic reviews only including randomized controlled trial (RCT); (2) Meta-analyses or systematic reviews comparing HA with placebo in treating knee OA; (3) Meta-analyses or systematic reviews reported at least one variable (such as pain, function, and safety).
Exclusion criteria included the following items: (1) Meta-analyses or systematic reviews including non-RCT; (2) Systematic reviews did not conducting meta-analysis or pooling data; (3) Abstract, commentary, methodological study, narrative review.

Meta-analyses/systematic reviews selection. Firstly, two reviewers assessed the titles and abstracts of
researched studies for the eligibility criteria independently. The two reviewers were not blinded to the journals, organizations, financial assistance, conflict of interest and researchers' information. Subsequently, the full text of the studies that potentially met the inclusion criteria was read to determine the final inclusion. Any disagreement was resolved by reaching a consensus through discussion.
Date extraction. Two reviewers independently extracted the data from each included literature by the use of a standard data extraction form. The following items were extracted: title, authors, original study design, database, total number of studies, level of evidence, the pooled results and methodological variables.
Assessment of methodological quality. The quality assessment was independently conducted by two authors. Disagreements were resolved by discussion or a third reviewer was involved. The Assessment of Multiple Systematic Reviews (AMSTAR) method was used to evaluate the methodological quality of included studies 17 . The AMSTAR was a measurement scale containing eleven items, and it was applied extensively in assessing methodological quality of published meta-analysis or systematic review 18 . Heterogeneity within included studies. Heterogeneity of each outcome (primary and secondary outcomes) was reported for the each included meta-analyses. The following two questions were also evaluated: whether sensitivity analysis was performed in meta-analysis and whether the included meta-analyses evaluated potential sources of heterogeneity across primary studies. Upon the Cochrane Handbook, Heterogeneity of each outcome between 0% and 40% is regarded as not important; between 30% and 60% is moderate; between 50% and 90% is substantial, and between 75% and 100% is considerable. Therefore, I 2 less than 60% are accepted in the present study.
Choice of best evidence. Treatment recommendations were made according to the Jadad decision algorithm 19 .
The methodological instrument confirmed the source of inconsistence between meta-analyses, including differences in clinical problem, inclusion and exclusion standard, extracted data, methodological quality assessment, data combining, and statistical analysis methods 19 . The application of algorithm was performed by two independent reviewers. Our evaluation group came to conformity as to which of included meta-analyses can provide best evidence based on the current information.
Scientific RepoRts | 6:32790 | DOI: 10.1038/srep32790 Table 1 presented the characteristics of included meta-analysis. The number of original studies in meta-analysis varied from 5 in that study published in 2006 to 89 that published in 2012 ( Table 2). All included meta-analyses conducted qualitatively data synthesis.
Search methodology. The literature search methodology which was adopted by included meta-analysis was present in Table 3. Most of the databases that the included studies searched were Medline, Embase or Cochrane database.

Methodological quality of included meta-analyses. Methodological characteristics of included
Meta-analyses were presented in Table 4. All included meta-analyses only included RCTs and/or quasi-RCTs. The evidence degree of each meta-analysis was Level II. REVMAN, STATA, SAS, R and Comprehensive Meta-analysis software were used in meta-analyses. Subgroup and sensitivity analysis were used in some of the included studies. None meta-analysis used GRADE in their study. The AMSTAR results with each question of included   20 Wang et al. 21 Modawal et al. 22 Arrich et al. 23 Strand et al. 24 Bellamy et al. 25 Bannuru et al. 9 Rutjes et al. 10 Colen et al. 26 Miller et al. 11 Richette et al. 12 Strand et al. 6 Adams et al. 1995  20 Wang et al. 21 Modawal et al. 22 Arrich et al. 23 Strand et al. 24 Bellamy et al. 25 Bannuru et al. 9 Rutjes et al. 10 Colen et al. 26 Miller et al. 11 Richette et al. 12 Strand et al. 6 Huang et al.
Puhl et al. 1993 meta-analysis were shown in Table 5. The average score of AMSTAR of included meta-analyses was 7.25, ranging from 4 to 11. All included meta-analyses reported that there was no conflict of interest in making meta-analysis. One meta-analysis conducted by Bellamy et al. 25 was the highest quality study.
Heterogeneity Assessment. Table 6 presented the data of heterogeneity of each variable in each meta-analysis. The I 2 value was adopted to calculate the heterogeneity among original studies as a measurement aiming to ascertain the inter-studies variability in all included meta-analyses.

Primary Study
Lo et al. 20 Wang et al. 21 Modawal et al. 22 Arrich et al. 23 Strand et al. 24 Bellamy et al. 25 Bannuru et al. 9 Rutjes et al. 10 Colen et al. 26 Miller et al. 11 Richette et al. 12 Strand et al. 6 Sanofi-Aventis et al.  Results of Jadad Decision Algorithm. All outcomes reported in primary meta-analyses were reported in Fig. 2. According to the following three respects (the meta-analyses addressed the same clinical question, did not include the same original studies, and not have similar inclusion/exclusion criteria), the Jadad algorithm proposed that the eligible meta-analyses can be elected on account of the methodological quality and publication statue (Fig. 3). As a result, a Cochrane meta-analysis 25 with highest quality was selected. Bellamy et al. supported the use of the HA in the treatment of knee OA with beneficial effects on pain, function and patient global assessment.

Discussion
According to the above mentioned methodology, the meta-analysis conducted by Bellamy et al. 25 is with highest quality compared with others. The best available evidence hints that HA is an effective intervention in treating knee OA without increased risk of adverse events. Therefore, the current evidence supports the use of the HA in the treating knee OA. Meta-analyses or systematic reviews are commonly regarded as the highest level of clinical evidence 35 . Clinicians can make meaningful clinical decisions with the help of meta-analyses or systematic reviews. However, a larger number of meta-analyses involving in the same clinical question have been published with conflicting results. This phenomenon was also occurred in the evidence-based study in HA injections for knee OA. Although numerous meta-analyses or systematic reviews have been written in treating knee OA via HA, there was still in controversy. Such discrepancy results in some difficulties for decision makers (including clinicians, policymakers and patients, depending on the context) who rely on this synthesized evidence to help them make decisions among pharmacological interventions when the results of trials are not unanimous.
Jadad et al. 19 concluded the following potential sources of inconsistency among meta-analyses, including the clinical topic, eligible criteria, data extraction, quality assessment, assessment of the ability to combine studies, and statistical methods for data synthesis. Furthermore, Jadad et al. 19 provided a decision methodological tool which summarizes the process for identifying and resolving causes of discordance. The ultimate purpose was to   20 Wang et al. 21 Modawal et al. 22 Arrich et al. 23 Strand et al. 24 Bellamy et al. 25 Bannuru et al. 9 Rutjes et al. 10 Colen et al. 26 Miller et al. 11 Richette et al. 12 Strand et al. 6 Was a prior design provided?  help clinical decision makers to select best evidence from inconsistency meta-analyses and systematic reviews. As recommended by Jadad et al., decision algorithm, a widely used tool [36][37][38] , is a useful instrument for differencing between meta-analyses or systematic reviews. Although Jadad decision algorithm choose comprehensive meta-analysis among discordant reviews, more empirical evidence is required to establish the effect of these elements on the validity of the review process, their relative importance and their effect on the results of a review. According to the decision algorithm, the Cochrane meta-analysis conducted by Bellamy et al. was selected in the present study. Bellamy et al. 25 reported that HA was an effective treatment for knee OA at different post injection periods but especially at the 5 to 13 week post injection period, and few adverse events were reported in the HA. However, there is considerable between-product, between-variable and time-dependent variability in the clinical response. Therefore, we concluded that HA is an effective and safety intervention in treating knee OA. Although the positive results were reported, effect size statistic was not used in the study. Thus, we did not have entire confidence in the extent of symptomatic improvement.
Rutjes and his colleagues 10 published a high-quality systematic review (AMSTAR score: 10) in Annals of Internal Medicine. This study used effect size statistic and demonstrated that OA was associated with a small and clinical irrelevant benefit and an increased risk of adverse events. However, the use of the effect size statistic to infer clinically meaningful changes in efficacy outcomes is frequently misinterpreted. Rutjes et al. reported an effect size of 0.37 and then erroneously state that this is equivalent to an improvement in knee pain of 0.9 cm on a 10 cm scale. As showed in Table 1, Rutjes meta-analysis included largest number of RCTs (including published RCTs and grey literature). However, the conclusions in this paper were heavily influenced by inclusion of unpublished, unverifiable data.
The conclusion of the present study is consistent with the finding published in 2015 by Richette and his colleagues 12 . They performed a meta-analysis only including low bias and high-quality RCTs (adequate randomization and concealment and double-blind design) and showed that HA provided a moderate but real benefit for patients with knee OA. Recently, Strand et al. 6 conducted a meta-analysis to investigate the safety and efficacy of UA-approved HA for knee OA. Strand's study was the only known report to cite the pretreatment to Outcomes Lo et al. 20 Wang et al. 21 Modawal et al. 22 Arrich et al. 23 Strand et al. 24 Bellamy et al. 25 Bannuru et al. 9 Rutjes et al. 10 Colen et al. 26 Miller et al. 11 Richette et al. 12 Strand et al. 6 Overall pooled effect size +    Thus, it reported that US-approved HA is safe and efficacious through 26weeks in treating knee OA. The primary limitations of this meta-analysis include the following: (1) English language studies were included in the present overlapping meta-analyses. Although numerous meta-analyses were included in the present study, it is possible that we have omitted non-English language reviews. (2) Several factors of primary trials, such as study design, publication bias and clinical heterogeneity, may influence interpretation. (3) The selected meta-analysis was published in 2006, which will influence the stability of the results. Newest published high-quality meta-analyses are needed to confirm the present evidence.
To sum up, the present systematic review of overlapping meta-analyses investigated efficacy and safety of HA in treating Knee OA. Currently, the best evidence suggested that HA is an effective intervention in treating knee OA without increased risk of adverse events. Therefore, the evidence supports the use of the HA in the treating knee OA. Further studies with effect size statistic are still required to qualify the clinical efficacy.