The efficacy and safety of proton-pump inhibitors in treating patients with non-erosive reflux disease: a network meta-analysis

Proton-pump inhibitors (PPIs) have been proved as safe and effective ways to treat patients with non-erosive reflux disease (NERD). However, less is known about the comparisons among different PPIs and their best dosage. We aimed to synthesize the available evidence through network meta-analysis to investigate the efficacy and safety of different PPIs in treating patients with NERD. Fifteen studies with 6309 patients were included in the meta-analyses. For the rate of symptomatic relief, compared with control groups, all interventions except rabeprazole 5 mg significantly increased rate of symptomatic relief. Among the comparisons of different interventions, omeprazole 20 mg group was associated with a higher rate of symptomatic relief in contrast to omeprazole 10 mg group (odds ratio, OR: 1.89, 95% confidence interval, CI: 1.34, 2.67; p-value: 0.0005) or rabeprazole 5 mg group (OR: 2.51, 95%CI: 1.16, 5.42; p-value: 0.019); dexlansoprazole 30 mg therapy significantly improved the rate of symptomatic relief compared with rabeprazole 5 mg group (OR: 2.64, 95%CI: 1.08, 6.43; p-value: 0.03). For the rate of adverse events, there was no significant difference among all interventions.

Scientific RepoRts | 6:32126 | DOI: 10.1038/srep32126 for the following reasons: improper patients, interventions and comparisons; inappropriate study design; no interested outcomes and reviews. Finally, we included 15 studies with 16 trials in the meta-analyses 15-29 . Study and patient characteristics. Table 1 shows that the publication year ranged from 1997 to 2011.
Most RCTs were conducted in Europe and USA. The duration of follow-up ranged from 1 month to 6 months. The total number of patients across the studies was 6299, with an average of 394 patients per trial. The proportion of males in the included studies ranged from 28.9% to 55.8%. The positive rate of helicobacter pylori tests ranged from 13.4% to 52.7%.

Risk of bias.
Only a small number of trials adequately described random sequence generation (37.5%) and allocation concealment (31.25%). Six trials had a low risk of bias in blinding of participants and personnel. Similarly, seven trials had a low risk of bias in blinding of outcome assessment. Most trials had a low risk of bias in incomplete outcome data (81.25%). For selective reporting domain, seven studies were judged as low risk of bias. Majority of included studies received commercial fund (75%), so they were judged as high risk of bias. All details is shown in Fig. 2. Rate of symptomatic relief. A total of 315 patients (1 trial) were assigned to dexlansoprazole 30 mg therapy, 315 (1 trial) to dexlansoprazole 60 mg therapy, 555 (4 trials) to omeprazole 10 mg therapy, 555 (4 trials) to omeprazole 20 mg therapy, 276 (1 trial) to lansoprazole 15 mg therapy, 277 (1 trial) to lansoprazole 30 mg therapy, 782 (5 trials) to esomeprazole 20 mg therapy, 523 (3 trials) to esomeprazole 40 mg therapy, 93 (1 trial) to rabeprazole 5 mg therapy, 445 (3 trials) to rabeprazole 10 mg therapy, 197 (2 trials) to rabeprazole 20 mg therapy and 1929 (15 trials) to placebo therapy. The network plot is shown in Fig. 3.
Compared with placebo groups, all interventions except rabeprazole 5 mg significantly increased rate of symptomatic relief. Among the comparisons of different interventions, omeprazole 20 mg group was associated with a higher rate of symptomatic relief in contrast to rabeprazole 5 mg group (OR: 2.51, 95%CI: 1.16, 5.42; p-value: 0.019) or omeprazole 10 mg group (OR: 1.89, 95%CI: 1.34, 2.67; p-value: 0.0005); dexlansoprazole 30 mg therapy significantly improved the rate of symptomatic relief compared with rabeprazole 5 mg group (OR: 2.64, 95%CI: 1.08, 6.43; p-value: 0.03). The details of all comparisons are indicated in Table 2. No inconsistency was found through global (p = 0.36) and loop-specific approach ( Figure S1). Because the majority of included studies had a high risk of bias, we could not perform sensitivity analyses. To test the robustness of the results, we made meta-regressions about mean age, percentage of male and duration of follow-up. The results of meta-regressions showed that these factors had no effect on the results of network meta-analyses. The funnel plot was indicated in Fig. 4.
There was no significant difference among all interventions. The details of all comparisons is indicated in Table 3. No inconsistency was found through global (p = 0.68) and loop-specific approach ( Figure S3). Because the majority of included studies had a high risk of bias, we could not perform sensitivity analyses. To test the robustness of the results, we performed meta-regressions about mean age, percentage of male and duration of follow-up. The results of meta-regressions showed they had no effect on the results of network meta-analyses. The funnel plot was indicated in Fig. 5.

Rank of the interventions.
For rate of symptomatic relief, dexlansoprazole 30 mg therapy ranked the first and placebo therapy ranked the last ( Figure S4). For rate of adverse events, omeprazole 20 mg therapy had the smallest incidence and lansoprazole 30 mg therapy was associated with the highest incidence ( Figure S5). The cluster plot ( Figure S6) shows omeprazole 20 mg might be the best intervention in treating patients with NERD. However, we did not make a conclusion that omeprazole 20 mg is the best option because omeprazole 20 mg did not show advantage over the other interventions.

Discussion
Summary of the results. 1. Compared with control groups, all interventions except rabeprazole 5 mg significantly increased rate of symptomatic relief. Among the comparisons of different PPIs, omeprazole 20 mg group was associated with a higher rate of symptomatic relief in contrast to rabeprazole 5 mg group or omeprazole 10 mg group; dexlansoprazole 30 mg therapy significantly improved the rate of symptomatic relief compared with rabeprazole 5 mg group; 2. For the rate of adverse events, there was no significant difference among all interventions; 3. According to overall rank results, omeprazole 20 mg might be the best intervention in treating patients with non-erosive gastroesophageal reflux.
Comparison with previous literatures. For rate of symptomatic relief, our results were similar with previous pair-wise meta-analysis except rabeprazole 5 mg therapy 13 . The difference may be attributed to the random bias (only 1 trial with 93 patients) or the low dose (the recommended dose is 10 mg/20 mg). On the other hand, the results of previous indirect meta-analysis were different from ours 14 . Previous one included dexlansoprazole 30 mg, dexlansoprazole 60 mg, esomeprazole 40 mg and esomeprazole 20 mg. Among the four PPIs, dexlansoprazole 30 mg significantly increased symptom relief compared with esomeprazole 20 or 40 mg. However, our results showed there were no significant results between dexlansoprazole 30 mg and esomeprazole 20 or 40 mg. The possible explanation is that the sample size of our study outnumbers those of previous meta-analyses (6 studies with  Limitations. Several limitations should be noted to help clinicians reasonably clarify the results. First, the sample size of one study was rather small (n = 37), which might cause potential small sample biases 20 . Thus, we performed a sensitivity analysis by excluding the study and the result remained stable. Then, there was only one study for some interventions (e.g., dexlansoprazole 30 mg therapy, dexlansoprazole 60 mg therapy, lansoprazole 15 mg therapy, lansoprazole 30 mg therapy, esomeprazole 40 mg therapy and rabeprazole 5 mg therapy) and we need to interpret these results cautiously. Additionally, we did not find study with pantoprazole for patients with NERD, which indicated our study did not comprehensively analyze all kinds of PPIs. Finally, most of included studies received funds from commercial companies, which might affect the imprecision of the results.

Conclusions
Compared with control groups, all interventions except rabeprazole 5 mg significantly increased rate of symptomatic relief. In addition, omeprazole 20 mg group was associated with a higher rate of symptomatic relief in contrast to omeprazole 10 mg group or rabeprazole 5 mg group; dexlansoprazole 30 mg therapy significantly improved the rate of symptomatic relief in contrast to rabeprazole 5 mg group. For the rate of adverse events, there was no significant difference among all interventions. To make conclusions more reliable and applicable, future RCTs should focus on dexlansoprazole, lansoprazole and pantoprazole and be conducted in Africa, South America and Oceania.   Interventions and comparisons: We included studies which had at least two of the following interventions: placebo, dexlansoprazole, omeprazole, lansoprazole, esomeprazole, rabeprazole and pantoprazole. We did not restrict doses of the PPIs.
Outcomes: rate of symptomatic relief (heartburn) and rate of adverse events. Study design: we included randomized clinical trials (RCTs). For cross-over studies, we used the data before wash-out period. We excluded cluster randomized clinical trials.
Information sources and literature search. We searched PubMed, EMBASE and Cochrane library from inception until February 11, 2016. Additionally, we searched clinicaltrials.gov for the unpublished researches. Finally, we manually searched the reference lists of included studies and meta-analyses with similar topic. Study selection process. Initially, two authors screened the title and abstract of citations. After completing the preliminary screening, they screened potential included studies by reading their full-text. If any disagreement could not be solved by discussion, a third author was consulted.      Data items, data abstraction process and risk of bias. Two authors abstracted study characteristics (e.g., first author, publication year and sample size), patient characteristics (e.g., mean age and gender), outcome definitions, and inclusion and exclusion criteria.
Two authors used risk of bias tool (Rob tool) to evaluate the quality of included RCTs independently 30 . The tool includes seven domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other bias (e.g., baseline imbalance, commercial funds and early drop-out). For each domain, two authors judged the quality of each RCT as high or low or unclear risk of bias. Data analysis. Before performing data synthesizing, we assessed the transitivity assumption and consistency assumption. For the transitivity assumption, we compared the study characters and patient characters across treatment comparisons 31 . For the consistency assumption, we use two ways to evaluate the inconsistency: global methods (the 'design-by-treatment' inconsistency model) and loop-specific approach 32 . Either the global P value in the global methods was less than 0.05 or the percentage of inconsistency loop was larger than 20%, we considered there was significant inconsistency in the study 33 . We conducted network meta-analyses with random model if there was a good transitivity across treatment comparisons and no significant inconsistency 34 . All data-analyses were conducted through STATA V.13.1.
If we find any significant inconsistency, we performed subgroup analyses and meta-regressions (e.g., mean age and proportion of male) to investigate the possible sources 35 . Then, we excluded studies with a high risk of bias as a form of sensitivity analysis. For the publication bias, comparison-adjusted funnel plot was used 36 . Finally, the surface under the cumulative ranking curve analysis (SUCRA) was conducted to rank all interventions. For any intervention, a larger value of SUCRA means a higher rank 36 . Additionally, we performed a cluster plot including both rate of symptomatic relief and rate of adverse events to observe the possible best intervention 36 .