This protocol adheres to the standards of the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) Statement [
12], and is registered with the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42019119129) [
13]. A populated PRISMA-P checklist for this protocol is provided as an additional file (see Additional file
1). Any post hoc modifications to the plans presented within the protocol will be recorded and described in the publication of the final report to ensure transparency. The final report will be developed in consultation with the PRISMA Extension Statement for NMA to ensure all aspects of methods and findings are fully reported [
14].
The search strategies will be developed and tested through an iterative process by an experienced medical information specialist in consultation with the review team. The strategies will be peer reviewed by another senior information specialist prior to execution using the PRESS Checklist [
16] Using the OVID platform, we will search Ovid MEDLINE®, including Epub Ahead of Print and In-Process & Other Non-Indexed Citations, and Embase Classic+Embase. We will also search the Cochrane Library on Wiley.
Strategies will utilize a combination of controlled vocabulary (e.g., “Meniere Disease,” “Endolymphatic Hydrops”) and keywords (e.g., “auditory vertigo,” “endolymphatic hydrops,” “Meniere’s”). Results will be filtered using headings for systematic reviews, randomized controlled trials, and non-randomized controlled trials as applicable for each database. Vocabulary and syntax will be adjusted across databases. There will be no language or date restrictions on any of the searches, but when possible, animal-only and opinion pieces will be removed from the results. Potentially relevant articles published in languages other than English will be included in an Additional file. We will document any unforeseen limitations pertaining to search strategy upon its completion.
A gray literature search of targeted clinical trial registries,
ClinicalTrials.gov, and the International Clinical Trials Registry Platform will also be undertaken. Studies included in existing reviews will be inspected to confirm no relevant studies have been missed. Content experts will be contacted to obtain information on unknown or ongoing studies. The proposed database search strategies are provided in Additional file
3.
Screening and data extraction
Screening will be performed in two stages via two reviewers working independently and in duplicate against eligibility criteria established a priori, using an online systematic review software program (Distiller Systematic Review (DSR) Software; Evidence Partners Inc, Ottawa, Canada). Stage 1 screening will be based on review of the abstracts and titles identified from the electronic search, while stage 2 screening will consider full-text review of the articles deemed potentially relevant during stage 1. Screening at both stages will commence with a calibration exercise to ensure consistent application of eligibility criteria. A screening pilot will be performed prior to full screening of titles and abstracts (25 titles and abstracts) and full-text screening (25 studies). At stage 1, two reviewers (NA and LE) will independently assess the titles and abstracts for eligible studies using the liberal accelerated method [
17] where only one reviewer is required to include citations for further assessment at full-text screening and two reviewers are needed to exclude a citation. At stage 2, the full-text articles of potentially relevant citations will be retrieved for full-text screening and the same two reviewers (NA and LE) will independently assess the article for relevancy. Disagreements between reviewers will be resolved via consensus or third-party adjudication. The study selection process will be reported using a PRISMA flow diagram [
18] in the final publication. References of all included studies will be scanned for inclusion, by one reviewer. At least one content expert will be consulted for additional studies. Study authors will be consulted where necessary for verifying eligibility and for missing or unclear information on studies (and information will be included if received in a timely manner). A list of the excluded studies alongside the rationale for their exclusion will be provided in an additional file to the completed review. With regard to duplicate publications, companion documents, or multiple reports of a primary study, we will collate all available data and use the most complete set and exclude the duplicate version or companion documents with no additional data.
A standardized data extraction form in Microsoft Excel (Microsoft Corporation, Seattle, WA, USA) will be used for collecting key study information that includes all pre-specified data items (see Additional file
2). After piloting the data extraction form on a small number of studies, two reviewers (NA and LE) will extract the data independently and any discrepancies will be resolved by discussion or a third person. Information from each study will be recorded that will include (but not be limited to) the following: publication characteristics (e.g., authors’ names, publication year, and journal), study design traits (cited trial design, clinical setting, duration of follow-up, number of patients randomized and number analyzed for each outcome, occurrence of dropouts, funding source, and authors’ conflict of interest etc.), study population details (patient inclusion and exclusion criteria, age, sex, body mass index (BMI), race, comorbidities, prior treatments, and relevant baseline data, such as initial PTA, WRS/SDS, hearing loss in the opposite ear previously affected with MD, aural fullness in the opposite ear, time from onset to treatment, prior otologic surgery, presence of tinnitus and vertigo, etc.), intervention and comparator specifics (type, dose, unit, duration, frequency, route of administration, strategy of administration, and co-intervention, etc.), and outcome data (including reported outcome definitions and summary data related to treatment effects (e.g., mean change and the corresponding standard error for continuous outcomes, and number of events and number of total patients for dichotomous outcomes), and reported scales for evaluating the outcomes). A complete list of pre-specified data items is presented in Additional file 2. Means and measures of dispersion will be approximated from figures in the primary studies using online tools. When available, data from both intention to treat and per protocol analyses will be extracted. We will contact authors for any missing or additional data of interest. Authors’ defined pre-specified outcomes of interest will be extracted and grouped accordingly for analyses.
Outcomes and prioritization
We explored the Core Outcome Measures in Effectiveness Trials (COMET) initiative [
19] but did not locate a core outcome set for Meniere’s disease [
20]. As such, the endpoints of interest for this review were selected via consultation with our clinical experts. The primary outcome of interest will be vertigo, while the secondary outcomes of interest will include changes in hearing, tinnitus, quality of life, aural fullness, and harms. With regard to outcomes definitions, we will gather outcomes with any definitions provided by the primary study authors. We will group together the data with similar definitions across the studies. Priority will be given to established definitions where possible. For instance, for vertigo control, we will consider the definition from the American Academy of Otolaryngology-Head and Neck Surgery (AAO-HNS) [
9]. However, in the event of insufficient outcome data using this definition, we will consider other definitions in consultation with our clinical context experts, and the set with the most available data and clinical relevance will be given priority in analysis. In terms of time of assessment for the endpoints of interest, we will consider data reported according to the AAO-HNS recommendation on reporting the treatment results [
9] when available. For instance, for frequency of definitive attacks, outcome data from the period 6 months before treatment should be compared with interval occurring between 18 and 24 months post-treatment. However, if insufficient data are reported based on AAO-HNS criteria across primary studies, we will collect outcome data at various time points such as baseline, 6 months, 12 months, and 24 months post-treatment if possible. In the absence of sufficient data, we may consider other reported time points in consultation with our clinical content experts to ensure clinical relevance.
Risk of bias assessment
We will use the Cochrane Risk of Bias Tool for RCTs [
21] to evaluate the risk of bias of each included RCT and quasi-randomized trial. Two reviewers will carry out assessments independently and resolve disagreements via consensus or third-party adjudication. All domains of the Cochrane Risk of Bias tool for RCTs will be considered, including selection bias (sequence generation and allocation sequence concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective reporting), and other biases considered relevant to the review topic. We will also evaluate baseline imbalances between groups with respect to comorbidities and factors that may impact our outcomes of interest, including history of falls more than once in the past year, older age, white race, female sex, higher BMI, current tinnitus, prior therapies received, allergies, immune dysfunction (e.g., ankylosing spondylitis, systemic lupus erythematosus, psoriasis), autonomic dysfunction, poor mental health [
18,
20,
22], arthritis [
23], history of hearing loss or episodic vertigo [
24], familial history of MD, and diabetes mellitus [
25].
Approaches to evidence syntheses
A.
Criteria for quantitative synthesis
To assess the assumptions for conducting NMA (i.e., the transitivity assumption, relating to similarity amongst studies in an evidence network), a variety of information related to study methods, patient demographics, and eligibility criteria will be collected. These will include the following: mean age at onset, percent female patients, mean disease duration, frequency/severity measures of vertigo at baseline, percent with history of migraine headaches, percent with previously identified tinnitus, measures of average tinnitus intensity, and other factors. These study features will be reviewed with our clinical experts using a combination of table summaries, box plots, and bar plots to identify potential outlier studies that may warrant exclusion from formal analyses.
As noted earlier, separate sets of analyses pertaining to comparison of pharmacologic interventions and surgical interventions will be performed. Initially, we will inspect the characteristics of included studies such as patients’ clinical characteristics (age, sex, and clinical history, including duration of hearing impairment and baseline severity) and methodologic homogeneity (e.g., risk of bias, study design), and we will summarize them accordingly. A pairwise meta-analysis for each intervention comparison will be pursued to explore statistical heterogeneity (based upon the
I2 statistic) if data permit.
B.Planned quantitative analysis
If data permits, both fixed effects and random effects Bayesian NMAs will be performed to compare interventions contained within the included studies that are sufficiently connected for a specified clinical endpoint. A common between-trial standard deviation will be used for Bayesian NMAs as per established methods [
26‐
28]. Model fit will be assessed by comparing total residual deviance with the number of unconstrained data points [
29] and will be considered adequate if these quantities are approximately equal. The deviance information criteria (DIC) will be used for selection between models, with a difference of five points suggesting an important difference [
29] (with smaller values being preferred). The type of endpoint under analysis (e.g., continuous or binary) will determine the use of specific NMA models. We will use mean change per arm (between pre- and post-intervention) for the analysis of continuous endpoints measured in the same units, and the corresponding effect size will be the mean difference. It is common in general that studies report findings of the same continuous endpoint in different formats, some reporting mean changes with corresponding standard errors (SEs), while others report only mean values at baseline and post-treatment with corresponding standard deviations (SDs) for each treatment arm. For the latter scenario, we will consider the appropriateness of assuming a correlation between mean values at baseline and follow-up and calculate the mean changes and corresponding SEs when they are not reported. If we encounter continuous endpoints that are measured using different scales across studies (e.g., a visual analog scale from 0 to 100 versus an itemized, composite score scale to assess severity of vertigo attacks), a model for estimating the effect size as a standardized mean difference (SMD) will be considered to explore benefits across related scales and maximize usage of available data. Estimates of effect sizes for binary endpoints will be expressed as odds ratios. All pairwise comparisons between interventions will be expressed with 95% credible intervals. Key secondary measures of effect, such as the surface under the cumulative ranking curve (SUCRA) and average treatment rankings [
30], will be estimated to explore potential orderings of treatments. All NMAs will be carried out using OpenBUGS version 3.2.3 [
31] and the R2OpenBUGS package [
32] version 3.2-3.2 in R [
31].
We will rely on pairwise meta-analyses if the evidence networks of pharmacologic interventions or surgical interventions are not well-connected. Whenever data permit for each pairwise comparison, we will use funnel plots to assess for small-study effects as a signal of publication bias. Where protocols are available, they will be reviewed to inform evaluations for selective reporting within the set of included studies.
C.Proposed additional analyses
If feasible (based upon quantity of available evidence and rigor or reporting of the included studies), we will explore subgroup analyses [
27,
33] to evaluate robustness of our findings and the impact of covariates. In consultation with our clinical experts, the subgroups will be chosen and may include (but not be limited to) gender distribution (e.g., percent females), age (older versus younger), BMI (higher BMI versus lower BMI), race (white versus others), presence of dizziness, number of days since initial treatment (or onset of MD), severity of initial hearing loss, type of MD (unilateral versus bilateral), and types of unilateral MD [
34]: (1) classic MD (sporadic MD, without migraine and autoimmune disorder), (2) delayed MD (hearing loss antedates vertigo episodes for months or years and is without migraine or autoimmune disorder in most cases), (3) familial MD, (4) MD with presence of migraine, and (5) MD comorbid with autoimmune disorder.
If excessive heterogeneity is identified and the research team feels meta-analysis is inappropriate, a narrative summary of findings with supporting tables and figures will be prepared.