Comparative efficacy and safety of urate-lowering therapy for the treatment of hyperuricemia: a systematic review and network meta-analysis

The prevalence of hyperuricemia and gout has been increasing, but the comparative effectiveness and safety of different treatments remain uncertain. We aimed to compare the effectiveness and safety of different treatments for hyperuricemia using network meta-analysis methodology. We systematically reviewed fifteen randomized controlled trials (involving 7,246 patients through January 2016) that compared the effects of different urate-lowering drugs (allopurinol, benzbromarone, febuxostat, pegloticase and probenecid) on hyperuricemia. Drug efficacy and safety, as outcomes, were measured by whether the target level of serum urate acid was achieved and whether any adverse events occurred, respectively. We derived pooled effect sizes expressed as odds ratios (ORs) and 95% confidence intervals (CIs). The efficacy and safety of the drugs were ranked by cumulative ranking probabilities. Our findings show that febuxostat, benzbromarone, probenecid, pegloticase, and allopurinol were all highly effective at reducing the risk of hyperuricemia compared to placebo. Febuxostat had the best efficacy and safety compared to the other drugs. Furthermore, febuxostat 120 mg QD was more effective at achieving urate-lowering targets (OR: 0.17, 95% CI: 0.12–0.24) and safer (OR: 0.72, 95% CI: 0.56–0.91) than allopurinol.

synthesis and the enzyme uricase. Among uricosuric agents, probenecid is commonly used 16 , whereas benzbromarone has been withdrawn in most European countries since 2003 due to serious hepatotoxicity 17 . Despite its adverse effects, benzbromarone is still applied commonly in clinics in several countries in Asia, including China 18 . Allopurinol, febuxostat and in particular Xanthine oxidase inhibitors (XOIs) are recommended as first-line drugs 16 . However, allopurinol has been reported to be associated with severe cutaneous adverse reactions 19 . In fact, humans lack urate oxidase, an enzyme that catalyses the oxidation of uric acid to allantoin, consequently resulting in hyperuricemia if accumulated in the blood 20 . Pegloticase, a recombinant polyethylene glycol conjugate of uricase (PEG-uricase), has been approved for the treatment of refractory chronic gout in the US and European Union 21 .
In 2014, a panel of 78 international rheumatologists raised ten key clinical questions pertinent to the diagnosis and management of gout, and one of these questions was how to determine the efficacy, cost-efficacy and safety of ULT (allopurinol, benzbromarone, febuxostat, peg-uricase and probenecid) in the treatment of gout 22 . Two previous pairwise meta-analyses analysed available individual studies and suggested that febuxostat may be associated with better urate lowering efficacy than allopurinol 23,24 . However, traditional meta-analyses can only draw interactions between comparisons among treatments with valid head-to-head trials 25,26 . Currently, several network-based approaches have been applied to potential drug discovery and biological information mining, e.g., drug-target interaction identification 27,28 , drug similarity calculation 29 and genome-disease function inference 30 . As an extension of a pairwise meta-analysis, a network meta-analysis provides a method for assessing the relative effectiveness of two treatments when they have no direct comparison in randomized trials 31 . Herein, we performed a network meta-analysis to evaluate the comparative efficacy and safety of five urate-lowering drugs, focusing on their ability to achieve target serum urate acid levels and the risk of adverse events.

Methods
Search strategies and selection criteria. This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) extension statement for network meta-analyses of health care intervention studies 32 . The PubMed, Medline, Embase, Cochrane Library databases and ClinicalTrials.gov were searched from inception to Jan 16, 2016. Following the PICOS (Participants, Interventions, Comparisons, Outcomes and Study design) principle 33 , the key search terms included (P) hyperuricaemia, hyperuricemia, gout, (I) urate-lowering therapy, uric acid, urate, (C/O) allopurinol, benzbromarone, febuxostat, pegloticase, probenecid, and (S) random*, and randomized controlled trial.
Studies meeting the following criteria were included: (a) Patients: adults (age > 18 years old) with hyperuricemia with or without chronic gout; (b) Intervention: established ULT with at least one of five agents (allopurinol, benzbromarone, febuxostat, pegloticase or probenecid); (c) Comparator: placebo or another agent of the five mentioned above; (d) the outcome of efficacy was defined as a failure to achieve the sUA treatment target level, i.e., ≤ 6 mg/dl (or 360 μ mol/l) with ULT, and the outcome of safety was defined as any adverse events during the period of the trial, including abnormal liver function, renal impairment, hyperlipidaemia, diarrhoea, gastrointestinal disorders, joint-related signs and symptoms; and (e) Study design: randomized controlled trial (RCT). The exclusion criteria were as follows: (a) trials comparing different doses of the same medication only; (b) studies without a designated intervention/comparator arm; (c) animal experiments; and (d) studies reported in a language other than English.
Three researchers (S.L., H.X.Y. and Y.N.G.) independently screened all records according to the inclusion and exclusion criteria. Any inconsistencies were resolved by discussion among the three authors. Finally, we identified fifteen qualified RCTs that were included in the current analysis 1,34-47 . The complete process and the exclusion reasons are shown in Fig. 1.
Data extraction and quality assessment. Two investigators (S.L., H.Y.) reviewed the full text of the eligible studies and extracted information into an electronic database. The information included study design, patient characteristics, inclusion/exclusion criteria, treatment protocols, and outcomes (the number of patients with/without successful treatment and occurrence of adverse events, which were available as binomial counts (successes/total)). The information was double checked by referring to the original articles when an inconsistency existed.
The quality of the included studies was reviewed and assessed by two investigators (S.L., Y.G.) independently using the Cochrane Collaboration's tool for assessing the risk of bias 48 . A risk of bias graph displays the grade of bias as high risk, unclear risk and low risk (see Supplementary Figure 1A,B). Five studies were considered to have a high risk of bias due to the lack of implementation of blinding 37,41-43,47 . Data synthesis and analysis. We calculated the odds ratios (ORs) and 95% confidence intervals (95% CIs) of the drugs for their failure to achieve the sUA treatment target level and related adverse events. Head-to-head meta-analysis was used to generate direct evidence (from studies directly comparing A to B). The pairwise meta-analysis with random-effect models was performed, and statistical heterogeneity was estimated using I 2 statistics, which describe the percentage of variability across studies caused by heterogeneity rather than chance 49 .
In addition to direct evidence, we also drew inferences between two intervention arms, such as A versus B, from indirect evidence (from combining studies through an intermediate comparator C, e.g., A vs. C and B vs. C studies) 50,51 . With the use of the adjusted indirect comparison method and inverse variance method, the effect estimates between treatments without direct comparisons and the combined results of direct and indirect evidence were obtained, respectively. Thus, even if there are no known comparisons for the investigated drug, a network meta-analysis still can estimate the potential effect of this drug based on existing head-to-head trials. As a result, a synthesized effect size and mean rank could be estimated for all the interventions.
Scientific RepoRts | 6:33082 | DOI: 10.1038/srep33082 When conducting a network meta-analysis, three assumptions need to be met, including homogeneity, transitivity, and consistency. The treatment effects together with their predictive intervals (PrIs) are examined to illustrate the magnitude of heterogeneity. A predictive interval plot is drawn to make comparisons between the 95% CIs and the 95% PrIs. The transitivity (or named similarity) assumption refers to the balance between the relative treatment effects and covariates across trials that are comparing different sets of interventions 52 . The inconsistency accounts for disagreements between direct and indirect evidence 53 . It is generally recommended to evaluate the consistency assumption using both global and local approaches. To assess the assumption of consistency in the entire network, we inferred the presence of inconsistency from any source in the entire network based on a Chi-square test. To evaluate the presence of inconsistency locally, we used the loop-specific approach to evaluate the inconsistency factor (IF, the difference between the direct and indirect estimate for one of the comparisons in a particular loop). We identified inconsistency as yielding a lower 95% CI limit that does not reach the zero line.
To rank the treatments based on efficacy and safety, we calculated the probabilities of the surface under the cumulative ranking curve (SUCRA). SUCRAs can illustrate the outcome percentages of every treatment relative to an ideal treatment, which always ranks first without uncertainty. Thus, the greater the SUCRA score, the more effective or safer the drug.
We performed the network meta-analysis using a frequentist model. Stata version 13 was used to make calculations. The metan and network commands were used for the pairwise and network meta-analyses, respectively. In the network meta-analysis, zero cells were corrected with the command "network setup" in Stata.

Results
Characteristics of eligible studies. Fifteen studies involving 7,246 adult trial subjects were included in the network meta-analysis. The characteristics of the included studies are summarized in Table 1. The earliest study was conducted in 1999, whereas the latest one was in 2016. The duration of the trials ranged from 4 to 52 weeks. Seven trials made a comparison between allopurinol and febuxostat 1,[35][36][37][38]46,47 , two trials between allopurinol and benzbromarone 41,43 , one trial between benzbromarone and probenecid 42 , three trials between febuxostat and placebo 34,39,40 , and one trial between pegloticase and placebo 45 . A three-arm trial compared allopurinol, febuxostat and placebo 44 . The dosage of febuxostat among the trials ranged from 20 mg/day to 240 mg/day. In general, all of the trial patients had an average age of 30 or more years, and males accounted for more than 80% of the subjects in the included trials. At baseline, these trial subjects had sUA concentrations > 8.0 mg/dl. A network graphical structure displays the available direct comparisons of the network of trials organized from the fourteen RCTs (Fig. 2). Comparisons with febuxostat (20/40/60/80/120/240 mg once daily) or pegloticase (8 mg every two/four weeks) were classified by dosage.

Direct treatment comparisons.
Pairwise meta-analysis. The pairwise meta-analysis showed that allopurinol, febuxostat 20/40/60/80/120/240 mg QD (20/40/60/80/120/240 mg once daily), and pegloticase 8 mg 2 W/4 W (8 mg every two/four weeks) were all highly effective at achieving the sUA treatment target compared to placebo (   Heterogeneity and inconsistency. The 95% PrI and 95% CI of each pairwise comparison are displayed in Supplementary Figure 2. There was no clear evidence suggesting inconsistency between the direct and indirect network effect values in the results of the traditional pairwise meta-analysis and the network meta-analysis (see Supplementary Table 1). Specifically, no inconsistency was found in either efficacy (P = 0.054) or safety (P = 0.819) within Chi-square tests. The loop-specific approach did not present any statistically significant inconsistency.
Ranking. Cumulative ranking plots of each treatment for efficacy and safety are shown in Fig. 4 Utilizing the SUCRA values, we displayed a clustered ranking plot of these treatments in the two dimensions of the x-axis (efficacy) and the y-axis (safety) in Fig. 5. Febuxostat was superior to the other drugs in both efficacy and safety, especially febuxostat 120 mg QD. Allopurinol took a medium position in the benefits and harms ranking. Compared with pegloticase 8 mg 2 W, pegloticase 8 mg 4 W showed better efficacy and safety. Benzbromarone and probenecid were likely to have similar rankings with an overall moderate benefit. However, probenecid ranked the worst for safety.

Discussion
Using a network meta-analysis approach, we found that febuxostat tended to have higher efficacy and safety than other urate-lowering drugs, especially at a dose of 120 mg once daily. There was no evidence suggesting that adverse drug events outweighed the benefits of any of the five categories of ULT other than probenecid.
Probenecid was introduced as a uricosuric drug in 1951, and it is generally applicable for patients who cannot tolerate XOIs or fail to achieve their target serum urate acid with them 15,16 . Allopurinol, a purine analogue, has been widely used as a hypouricemic drug since the 1960s and was approved by the US Food and Drug Administration (FDA) in 1965. However, patients taking allopurinol have a high risk of serious hypersensitivity syndromes, such as Stevens-Johnson syndrome and toxic epidermal necrolysis, which may have a strong association with the HLA-B*5801 allele, a genetic change more commonly observed in Asian populations 19,54 . Benzbromarone was introduced as a uricosuric drug in the 1970s. It was widely registered in countries throughout Europe, Asia and South America before being withdrawn from the European market in 2003 due to its serious hepatotoxicity 17 . Febuxostat, a non-purine selective inhibitor of xanthine oxidase, has been approved by the European Medicines Agency (EMA) since 2008 and by the US-FDA since 2009. Considering that febuxostat is far more expensive than allopurinol 15 , it is often used when allopurinol is contraindicated or not tolerated 55 . Pegloticase, a new anti-hyperuricemia drug, was introduced to markets in 2010 by the FDA, and only one report  including two placebo-controlled RCTs was reported in 2011 45 . To be noted, immunogenic responses to pegloticase should be monitored because it is a recombinant porcine-like uricase. Several meta-analyses of RCTs have attempted to address comparative effects of urate-lowering drugs. A Cochrane systematic review compared febuxostat against allopurinol in achievement of urate-lowering target levels (relative risk (RR) of febuxostat 80 mg vs. allopurinol: 1.5, 95% CI: 1.2-1.8) and (RR of febuxostat 120 mg vs. allopurinol: 2.6, 95% CI: 2.0-3.3), which outcome was measured by an opposite indicator from ours. Regarding the occurrence of adverse events, there was a lower rate when comparing febuxostat 80 mg and 120 mg against allopurinol (RR: 0.93, 95% CI: 0.87-0.99, and RR: 0.90, 95% CI: 0.84-0.96, respectively) 56 . Additionally, a previous meta-analysis including five trials compared febuxostat with allopurinol in urate-lowering efficacy (RR: 1.56, 95% CI: 1.22-2.00, its efficacy outcome by the proportion of patients meeting the therapeutic target for serum uric acid level) and risk of adverse events (RR: 0.94, 95% CI: 0.90-0.99) 24 . Using a more advanced approach, our study found that febuxostat had an advantage over allopurinol in urate-lowering efficacy and safety. According to the guidelines of ACR and the European League against Rheumatism (EULAR), XOIs such as allopurinol and febuxostat are recommended for use prior to uricosuric agents and uricase for ULT 16,57 . Therefore, it is also worthwhile to discuss the efficacy and safety of uricosuric agents and uricase.
According to our ranking of efficacy, benzbromarone was only second to febuxostat at achieving urate-lowering targets. Benzbromarone has performed excellently at promoting the excretion of uric acid despite life-threatening adverse events reported in the past 17 . Essential guidelines have been recommended to prevent benzbromarone hepatotoxicity such as regularly monitoring liver function 57 . Limited clinical trials have been carried out with benzbromarone and probenecid, partially owing to the impact of being withdrawn from the market, the development of new drugs, regional/ethnic differences, prescribing habits and cost. Uricase-based drugs can metabolize uric acid to allantoin, which reduces the risk of precipitate. In addition, short-term trials have shown their urate-lowering effectiveness. However, our study did not reveal any significant differences in comparisons of pegloticase against other drugs.
Our study has clinical implications. The prevalence of gout and hyperuricemia has increased around both developed and developing countries, presumably due to lifestyle changes 3,10 . Hyperuricemia is associated with metabolic syndromes such as hypertension, dyslipidaemia, obesity and diabetes [10][11][12] and with renal and cardiovascular diseases [13][14][15] . More and more patients need urate-lowering treatment. It is essential to know the comparative effects and safety of urate-lowering drugs available in the market. Our study pooled and ranked the efficacy and safety of these drugs using the data from individual RCTs, and thus our findings may be useful to clinicians in their decisions on which drug to use.
Our study has strengths. We designed our network meta-analysis as standardized by the PRISMA principle and conducted it carefully to minimize errors and ensure the validity of findings from all relevant studies identified. To our knowledge, our network meta-analysis is the first to address comparative effects of different ULTs with explicit rankings of efficacy and safety of different ULTs. We look forward to using this network-based statistical method to combine findings from individual studies and provide useful information for clinical decision-making. Finally, we analysed all of the trials of ULTs being used commonly, and we came to the conclusion that febuxostat had better urate-lowering effects than other drugs.
There are some limitations to our study. Firstly, this study included a limited number of trials. On the one hand, some drugs were only used in limited countries and areas, e.g., benzbromarone. On the other hand, we set language restrictions and excluded studies not in English. Secondly, some estimated results of the network meta-analysis relied on indirect comparisons. However, our results from direct comparisons were in accordance  with the indirect and mixed comparisons. No obvious evidence suggesting inconsistency was found by fitting the inconsistency model. Thirdly, medicines with specific indications and some new drugs under development were not considered. With the improvement and application of network-based approaches, we promise to implement further predictions for drug/genome-target interactions with known reachable paths in the network and provide better interpretations for decision-makers.

Conclusions
In conclusion, this systematic review and network meta-analysis provides clear evidence of the efficacy and safety of ULT. When comparing the ability to achieve sUA treatment targets and the occurrence of adverse events, febuxostat ranked first among the urate-lowering drugs. Benzbromarone and probenecid had moderate therapeutic effects, but they caused unpleasant side effects. Comprehensively considered, our findings support the recommendation of XOIs such as febuxostat and allopurinol. Pegloticase and similar new uricase drugs need further investigation through RCTs and meta-analyses.