Comparative efficacy and acceptability of antiepileptic drugs for classical trigeminal neuralgia: a Bayesian network meta-analysis protocol

Introduction Trigeminal neuralgia (TN) affects 4 to 28.9/100 000 people worldwide, and antiepileptic drugs such as carbamazepine and oxcarbazepine are the firstline treatment options. However, the efficacy and safety of other antiepileptic drugs remain unclear due to insufficient direct comparisons. Objective To compare the efficacy and acceptability of all currently available antiepileptic agents for the treatment of patients with classical TN. Methods We will search the PubMed, EMBASE, Cochrane Library and Web of Science databases for unpublished or undergoing research listed in registry platforms. We will include all randomised controlled trials comparing two different antiepileptic drugs or one antiepileptic drug with placebo in patients with classical TN. The primary outcomes will be the proportion of responders and the number of subjects who dropout during the treatment. The secondary outcomes will include the two primary outcomes but in the follow-up period, changes in the self-reporting assessment scale for neuralgia and quality of life assessment. In terms of network meta-analysis, we will fit our model to a Bayesian framework using the JAGS and pcnetmeta packages of the R project. Ethics and dissemination This protocol will not disseminate any private patient data. The results of this review will be disseminated through peer reviewed publication. PROSPERO registration number CRD42016048640.

What are the assessment results of transitivity and similarity?
It is not clear why and how the 9 antiepileptic drugs were selected in this project.
For indirect comparisons, a random effects model network metaanalysis will be developed. What random effects model does this refer to? There are two broad categories of methods for network meta-analysis: contrast-based and arm-based. Have the author ever considered the arm-based method (Zhang 2014

Reviewer #1
Comment 1: Introduction TN is not most common neuralgia. It may be worth mentioning that many of the studies are very old, there is lack of detail and their quality is poor, GRADE scores are low Clinical Evidence Zakrzewska Linskey 2014.
Answer: Many thanks for picking up this error. Accordingly, we have changed the sentence, as follows: "It is estimated that approximately 4 to 28.9 per 100,000 people worldwide suffer from TN, and the number affected tends to be higher among women at all ages and even increases with age." We've already quoted the following reference to indicate that "many of the studies are out-of-date with limited methodology, and were assessed as low GRADE scores" (Zakrzewska JM, Linskey ME. Trigeminal neuralgia. BMJ Clin Evid. 2014. pii:1207), please see page 3. Comment 3: Methods Acceptability in primary outcomes is only patients who drop out. Many patients will continue medications despite side effects because of severe pain. So any side effects should form part of the outcome measures as a separate measure.
Answer: Thanks for your professional suggestions. We've changed the acceptability in primary outcomes as follows: "Treatment acceptability is defined as the proportion of patients who have intervention related adverse events during the 4 to 12 weeks." Comment 4: Methods Concern that may get many Chinese articles that have not been through careful peer review.

Answer:
We've also considered your concern, thus, Chinese databases will not be searched. However, studies that could been searched on the English databases will be scanned or included, if the studies could meet our inclusion criterias.
Comment 5: Funding Funding nil but who is funding the publication, who is the sponsor, this is mentioned on the PROSPERO website.
Answer: This study has no sponsor and we have reported this at the end of the manuscript. Comment 6: References References 8 and 14 are the same and essentially 18 is the same group and same results. No reference is provided for oxcarbazepine. Use a more up to date reference instead of reference 2 (e.g. Zakrzewska Linskey BMJ) Answer: Many thanks for picking up this error. We've already deleted the repeated reference, and adjusted the order accordingly. We've quoted the following reference in the Background section of the manuscript instead of reference 2 (Zakrzewska JM, Linskey ME. Trigeminal neuralgia. BMJ. 2015; doi: 10.1136/bmj.h1238). In addition, we've also added the following reference in the Background section for oxcarbazepine (Zakrzewska JM, Patsalos PN. Long-term cohort study comparing medical (oxcarbazepine) and surgical management of intractable trigeminal neuralgia. Pain. 2002;95:259-66).

Reviewer #2
Comment 1: Interesting study, deals with a major problem regarding the anti-epileptic drugs and their use in classical trigeminal neuralgia. Mainly due to the heterogeneity, very challenging protocol. If this adds to the current data in terms of safety, efficacy and toxicity, could be a plus in the pharmacological management.
Answer: Many thanks for reviewing our manuscript. The results of this study will be produced in the next year or two (see also reply to first comment by Reviewer #3).

Reviewer #3
Comment 1: This paper conducted a systematic review and network meta-analysis to compare the efficacy and acceptability of antiepileptic drugs for trigeminal neuralgia, which has not been done before. However, it is not clear what results were obtained and what conclusion were drawn.
Answer: This study will assess the comparative efficacy and acceptability of 9 antiepileptic drugs for the classical trigeminal neuralgia. We anticipate the findings of this study will be produced in the next year or two.
Comment 2: What are the assessment results of transitivity and similarity? Answer: The assumption of transitivity and similarity will be assessed mainly base on clinical and methodological characteristics. We will assume that intervention effects are transitive in this network meta-analysis because we only include antiepileptic drugs, and we will investigate similarity based on clinical characteristics, such as antiepileptic drug dose, period of treatment, and severity of pain symptoms at baseline, as well as according to methodological characteristics such as study quality. Please see the 'Assessment of transitivity and similarity' section in page 9.
Comment 3: It is not clear why and how the 9 antiepileptic drugs were selected in this project.

Answer:
We chose a group of 9 antiepileptic drugs looking at the drugs which were licensed for neuralgia in many countries and which were frequently used in clinical practice.
Comment 4: For indirect comparisons, a random effects model network meta-analysis will be developed. What random effects model does this refer to? There are two broad categories of methods for network meta-analysis: contrast-based and arm-based. Have the author ever considered the armbased method (Zhang 2014)?
Answer: Thanks for your professional suggestions. In the Statistical analysis section of manuscript, the model refer to arm-based parameterization random effects model. Accordingly, we've already quote the following reference (Zhang J et al. Network meta-analysis of randomized clinical trials: Reporting the proper summaries. Clinical Trials. 2014;11:246-262), please see page 8.

REVIEWER
Jing Zhang University of Maryland, USA REVIEW RETURNED 12-Aug-2017

GENERAL COMMENTS
This seems a nice protocol. But its significance is weakened because it is only a protocol and no results are obtained yet.
Though the authors claimed that they will use the arm-based method to do the analysis. It is not clear what exact models they will use. The 2014 Clinical Trials paper proposed models for binary outcomes. Other papers, for example, "Detecting outlying trials in network meta-analysis" and "A Bayesian missing data framework for generalized multiple outcome mixed treatment comparisons" may provide ideas for continuous outcomes and multiple outcomes.
It is not clear why the authors want to conduct analysis using both winbugs and STATA. Do the authors expect inconsistency of the results between these two softwares? There are some existing R packages that the authors may want to consider.

VERSION 2 -AUTHOR RESPONSE
Comments from Reviewer #3 Comment 1: Though the authors claimed that they will use the arm-based method to do the analysis. It is not clear what exact models they will use. The 2014 Clinical Trials paper proposed models for binary outcomes. Other papers, for example, "Detecting outlying trials in network meta-analysis" and "A Bayesian missing data framework for generalized multiple outcome mixed treatment comparisons" may provide ideas for continuous outcomes and multiple outcomes.
Answer 1: We will use Chaimani model for network meta-analysis, which could either calculate the continuous outcomes and binary outcomes. We have added the information in the statistical analysis section: ' For indirect comparisons, network meta-analysis will be developed in a Bayesian framework using Markov chain Monte Carlo simulation methods in WinBUGS (Medical Research Council's Biostatistics Unit, Cambridge, UK) with a Chaimani model.' Comment 2: It is not clear why the authors want to conduct analysis using both winbugs and STATA. Do the authors expect inconsistency of the results between these two softwares? There are some existing R packages that the authors may want to consider.
Answer 2: We use winbugs software to calculate the data, after that, we will use Stata software to draw the pictures. We have clarified the role of Stata software in Statistical analysis section: 'The effectiveness of each treatment among all available treatments will be ranked by calculating the OR in order, and plots of the surfaces under the cumulative ranking curves (SUCRAs) will be generated to rank the various treatments for each outcome using Stata software.' Please see reference 44.