Betahistine in Ménière’s Disease or Syndrome: A Systematic Review

Ménière’s disease is characterized by recurrent episodes of vertigo, hearing loss, and tinnitus, often with a feeling of fullness in the ear. Vertigo attacks can occur without warning, and their intensity varies, which may lead to psychological suffering and a reduction in quality of life. The disorder may be subdivided into 2 categories: it may be secondary to a number of established inner ear disorders (Ménière’s syndrome) or idiopathic (Ménière’s disease). Ménière’s disease is known to be associated with endolymphatic hydrops, that is, raised endolymph pressure in the membranous labyrinth of the inner ear [Hallpike and Cairns, 1938]. However, hydrops per se does not explain all its clinical features. Nonetheless, both categories may be considered as 1 entity, as in both, endolymphatic hydrops is the pathophysiological hallmark of the disease.

The diagnostic process may be difficult, as there is great variability in clinical presentation, and no reference standard exists. The American Academy of Otolaryngology – Head and Neck Surgery (AAO-HNS) has produced diagnostic guidelines [Alford, 1972], which have been revised twice [Pearson and Brackman, 1895; Committee on Hearing and Equilibrium, 1995]. The AAO-HNS 1995 guidelines formulate that a “definite” diagnosis can be made on the basis of at least 2 spontaneous episodes of rotational vertigo lasting at least 20 min, audiometric confirmation of sensorineural hearing loss, and tinnitus and/or a perception of aural fullness (see online suppl. Table 1; see www.karger.com/doi/10.1159/000515821 for all online suppl. material). More recently, diagnostic criteria have also been proposed by the Bárány Society [Lopez-Escamez et al., 2015].

In a recent study in the USA, the prevalence of Ménière’s disease was estimated at 200 per 100,000 people per year [Alexander and Harris, 2010]. Ménière’s disease is most common between 40 and 60 years of age [Harcourt et al., 2014]. Vertigo episodes tend to occur in clusters with a period of remission that may last for several months in between the clusters [Perez-Garrigues et al., 2008]. Episodes have been observed to occur with increasing frequency over the first few years after presentation and then decrease in association with a sustained deterioration in hearing [Moffat and Ballagh, 1997]. In most cases, vertiginous episodes eventually cease completely [Silverstein et al., 1989]. The fluctuating, progressive, and unpredictable natural history of Ménière’s disease makes investigation of any treatment effect difficult; studies, therefore, need to compare interventions with placebo over an adequate time period. The aim of treatment is to reduce the number, severity, and duration of attacks of vertigo; to prevent progression of the disease and the loss of hearing; and to alleviate any chronic symptoms (e.g., tinnitus and aural fullness).

Betahistine dihydrochloride (betahistine) is an oral drug that has been prescribed to an estimated 130 million people worldwide since its first launch [Jeck-Thole and Wagner, 2006]. Although betahistine has been used for vestibular vertigo in general [Murdin et al., 2016], it is thought by some clinicians to be specifically effective for Ménière’s disease [Martinez, 1972; Nauta, 2013]. The recommended daily dose of betahistine is 24–48 mg per day divided into 2 or 3 single doses containing 8, 16, or 24 mg [Jeck-Thole and Wagner, 2006]. Although gastrointestinal side effects are cited in many formularies, the rate of adverse effects in patients taking betahistine is not significantly different from those taking placebo in comparison studies [Murdin et al., 2016].

Betahistine is a weak histamine H1 receptor agonist and a potent histamine H3 receptor antagonist. The mechanism of action of the drug may be via the reduction of endolymphatic pressure through improved microvascular circulation in the stria vascularis of the cochlea [Meikle et al., 2012]. In addition, inhibition of activity in the vestibular nuclei may contribute to rebalancing neural activity and expedite the recovery process [Timmerman, 1994; Lacour et al., 2007]. Studies have shown that betahistine reaches a peak plasma concentration in about 1 h, and it has a plasma half-life of approximately 3.5 h. The maximal vestibular therapeutic effect will last approximately 3–4 h [Electronic Medicines Compendium, 2015]. The washout period can be calculated as 4 times the drug effect [Senn and Ezzez, 1999]. These pharmacological characteristics are thought to reduce the intensity and duration of vertigo symptoms in the short term (under 3 months) and additionally prevent attacks in the longer term (over 3 months).

A previous review found insufficient evidence of a benefit from the use of betahistine [James and Burton, 2001]. Despite this, it is still widely used and studied in clinical practice, especially in Europe. Reassessment of the effect of betahistine in the treatment of Ménière’s disease is therefore now warranted.

Randomized controlled trials (RCTs), including cluster RCTs, were included. We excluded quasi-randomized studies. Crossover trials were eligible only if data before the crossover were extractable to avoid the potential for a carryover phenomenon.

Patients with Ménière’s disease or syndrome were included. We classified studies according to the diagnostic criteria used for Ménière’s disease. We rated studies using the AAO-HNS or the Japanese Society of Equilibrium Research criteria to define probable, definite, or certain Ménière’s disease as class “I” studies and studies using other diagnostic definitions as class “II.” We rated studies including patients with “possible” Ménière’s disease as class “III.” Studies including participants who had received treatment with betahistine in the past were also eligible for inclusion.

Betahistine of any dose regimens or formulations and for any duration of treatment was included. The sole comparison was betahistine versus placebo. Concurrent use of other medication or other treatment was accepted if used equally in each group; for example, betahistine with an additional intervention versus placebo with an identical additional intervention. Where an additional intervention was used equally in both groups, we analyzed this as a separate comparison. None of the selected studies evaluated the effect of betahistine by concurrent use of other treatment.

We analyzed the following outcomes in the review, but these were not used as a basis for including or excluding studies. Based on the pharmacological properties of the drug described above, we assessed outcomes as short term (3 months or under) or long term (3 months or over).

The primary outcomes were the proportion of patients with a reduction in vertigo symptoms (considering the intensity, frequency, and duration of those symptoms altogether) and significant adverse effects including upper gastrointestinal discomfort.

The secondary outcomes were the proportion of patients with a progression of hearing loss (>15 dB), based on the 4-tone average of thresholds at 0.5, 1, 2, and 3 kHz, as measured by a pure-tone audiogram; the proportion of patients with a reduction in tinnitus, measured with patient-reported questionnaire scores such the Tinnitus Handicap Index (THI) ([Kleinstäuber et al., 2015]; see online suppl. Table 2), the Tinnitus Functional Index [Meikle et al., 2012], the Tinnitus Handicap Questionnaire [Kuk et al., 1990], the Tinnitus Questionnaire [Hallam, 1996], the Tinnitus Reaction Questionnaire [Wilson et al., 1991], and the Tinnitus Severity Scale [Sweetow and Levy, 1990]; the proportion of patients with a reduction in aural fullness, measured by patient-reported questionnaire scores (e.g., visual analog scale); other adverse effects (headache and allergic skin reactions [pruritus and rashes]); and well-being and disease-specific health-related quality of life: overall changes as reported particularly on the Functional Level Scale (FLS) (see online suppl. Table 3), the Ménière’s Disease Patient-Oriented Symptom-Severity Index (MPOSI), and the Dizziness Handicap Inventory (see online suppl. Table 4). The FLS will be used as defined by the AAO-HNS 1995 guideline [Committee on Hearing and Equilibrium, 1995]. The questionnaires are validated and often used in trials to assess the changes in dizziness-related and Ménière’s disease-related quality of life [Duracinsky et al., 2007]. We anticipated that various non-validated tools (e.g., questionnaires) were used. We included validated tools only to ensure that the outcomes were as reliable as possible.

The Cochrane Ear, Nose and Throat Disorders Group (CENTDG) Information Specialist conducted systematic searches for RCTs and controlled clinical trials. There were no language, publication year, or publication status restrictions. The date of the search was 29 January 2019.

Published, unpublished, and ongoing studies will be identified by searching the following databases from their inception: the Cochrane ENT Register (searched via Cochrane Register of Studies [CRS] to date); the Cochrane Central Register of Controlled Trials (CENTRAL) (searched via CRS 16 January 2018, rerun on 29 January 2019); Ovid MEDLINE® Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE® Daily and Ovid MEDLINE® (1946 to 16 January 2018, rerun on 29 January 2019); Ovid Embase (1974 to 16 January 2018, rerun on 29 January 2019); LILACS (searched on 16 January 2018 and rerun on 29 January 2019); Web of Knowledge and Web of Science (1945 to 16 January 2018 and rerun on 29 January 2019); ClinicalTrials.gov (www.clinicaltrials.gov,searched via the CRS 16 January 2018 and rerun on 29 January 2019); World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (searched on 16 January 2018 and rerun on 29 January 2019). The Information Specialist modeled subject strategies for databases on the search strategy designed for CENTRAL. Where appropriate, they were combined with subject strategy adaptations of the highly sensitive search strategy designed by the Cochrane Collaboration for identifying RCTs and controlled clinical trials (as described in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0, Box 6.4.b. [Handbook 2011]). Search strategies for major databases including CENTRAL are provided in online suppl. Table 5. This paper was written according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement (see PRISMA checklist) [Moher et al., 2009].

We scanned the reference lists of the identified publications for additional trials and contacted trial authors where necessary. In addition, the Information Specialist searched Ovid MEDLINE to retrieve existing systematic reviews relevant to this systematic review, so that we could scan their reference lists for additional trials, and run non-systematic searches of Google Scholar to retrieve gray literature and other sources of potential trials.

Two authors (B.E. and H.Z.-L.) independently selected studies to identify studies that appeared to meet the inclusion criteria. Both authors then reviewed the full-text articles of the retrieved trials and applied the inclusion criteria independently. We resolved any discrepancies by discussion or, failing that, by consultation of one of the other authors (T.B. and P.P.B.).

Two authors (B.E. and H.Z.-L.) independently extracted data from the studies using standardized data forms. We extracted data so as to allow an intention-to-treat analysis. If necessary or if insufficient data were provided in the paper, we contacted the authors for further information.

With regard to subgroup analysis, we extracted data to allow grading of the diagnostic accuracy of the methods used to define the study population (see Types of Participants), along with the duration of disease and treatment protocol (dose and duration of drug treatment). For the outcome “proportion of patients with a reduction in vertigo symptoms,” we sought to independently dichotomize these into “improved” or “not improved.” If we found studies with >2 groups (e.g., 2 or more active treatments compared to placebo), we extracted data from the intervention and placebo groups, but we made a note of the additional arm(s). If betahistine doses differed among the intervention groups within a study, we extracted data on the highest dose and compared this to placebo. Extraction of data on comorbidity involved, for example, the presence of migraine and benign paroxysmal positional vertigo. For each study, we extracted the following information: study design, duration of study, randomization, allocation concealment, number of participants, setting of study, diagnostic criteria, exclusion criteria, age and sex distribution of participants, country of recruitment, date of study, number of intervention groups, generic name of intervention, total dose per day (mg), method of administration, outcomes measured and definition of outcomes, missing data and final sample size, funding, conflict of interest (any author), and concomitant treatment.

B.E. and H.Z.-L. assessed the risk of bias of the included studies independently, as guided by the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011). The “Risk of bias” tool addresses the following domains: sequence generation, allocation concealment, blinding, selective outcome reporting, incomplete outcome data, and other sources of bias (e.g., improper statistical analysis).

The 2 authors judged these domains using the Cochrane “Risk of bias” tool in RevMan 5.3 [RevMan, 2014], which involved describing each of these domains, as reported in the trial, and then assigning a judgment about the adequacy of each entry: “low,” “high,” or “unclear” risk of bias. We resolved differences of opinion by discussion. If no consensus was reached, one of the other authors was consulted.

The primary outcome in this review was the proportion of participants with a reduction in vertigo symptoms, which is a dichotomized measure. For this type of data, we aimed to calculate the risk ratio (RR). For intervention effect measures using continuous data, we planned to calculate the mean difference (MD) between groups, provided that the selected studies used the same scale of measurement and a validated tool. If different scales were used, we planned to calculate the standardized mean difference (SMD). For studies with ordinal data, we planned to dichotomize these data wherever possible.

We planned to include cluster RCTs with the cluster as the unit of analysis. However, none of the included studies were cluster RCTs.

In Ménière’s disease, it is unlikely that symptom activity returns to its baseline level after the first treatment period. Therefore, we only used data from crossover trials only if the data prior to the crossover could be obtained.

In the event that we found studies with >2 groups (e.g., 2 or more active treatments being tested against placebo), we established which of the comparisons were relevant to the systematic review. We found only 1 multi-arm study that used independent groups of participants. As a result, participants were not included in >1 group and were treated as independent comparisons.

The unit of analysis was the participant. We did not anticipate that by-ear reporting was available, but data per ear were preferred in cases of bilateral Ménière’s disease. We regarded bilateral Ménière’s disease patients as “improved” if any ear showed no deterioration of hearing loss and the proportion of patients who had a reduction in tinnitus or aural fullness increased. If studies evaluated the effect over a longer time period, we recorded the results at multiple time points. To avoid unit of analysis error when combining study results in a single meta-analysis (and therefore counting the same participants in >1 comparison), we defined different outcomes related to the periods of follow-up, and we performed separate analyses.

Where necessary and where sufficient data from the study were not provided, we contacted the authors of the study requesting further details about missing data and reasons for the incompleteness of the data, in all those cases in which an email address was reported. We were alert to potential mislabeling or non-identification of standard errors and standard deviations. Our methods for imputation were according to chapter 7.7.3 of the Cochrane Handbook for Systematic Reviews of Interventions [Higgins and Green, 2011]. If data were missing, we used available case analysis using all data (as reported) for all randomized patients available at the end of the study/time point of interest, regardless of the actual treatment received. We considered the quality of outcome assessment as a study limitation (GRADE) and not as a stratifying factor. Unfortunately, we did not receive a useful response in any of the cases in which we contacted the authors. We did not impute missing data, as it remained unclear whether data were missing “ at random” or “ not at random.”

We determined whether the selected studies suffered from clinical, statistical, or methodological heterogeneity. We planned to quantify statistical heterogeneity using the I² statistic and the χ² test. With respect to the I² statistic, an approximate guide to interpretation is provided in the Cochrane Handbook for Systematic Reviews of Interventions [Higgins and Green, 2011]. If the I² value was 50% or higher, we considered the data to suffer from substantial or considerable heterogeneity. For the χ² test, we used the indicator that if the χ² was greater than the degrees of freedom, then heterogeneity was likely to be present. We considered heterogeneity to be statistically significant if the p value was <0.10. Subsequently, we performed the meta-analysis using fixed-effect (in the absence of heterogeneity) and random-effects modeling (in the presence of heterogeneity).

If an outcome was reported by at least 10 studies, we planned to assess publication bias using a funnel plot and Egger’s test. Unfortunately, none of the outcomes were reported in this number of studies.

We planned to analyze treatment differences as an RR, calculated using the Mantel-Haenszel method. Unfortunately, none of the selected studies analyzed the outcomes by means of comparable or validated tools.

There were insufficient data available for subgroup analyses. Although we planned to perform the following subgroup analyses, we were not able to do so for: stage of disease, as defined by the AAO-HNS 1995 guidelines (see online suppl. Table 6); type of Ménière’s disease (see Types of Participants); and dose of betahistine administered (minimum daily dose of 8 mg to a maximum of 148 mg).

We planned to conduct a sensitivity analysis by excluding those studies with a high risk of bias, thereby checking the robustness of the conclusion from the studies included in the meta-analysis. In addition, we planned to use sensitivity analyses for studies in which data were imputed. However, all but one study carried an unclear or high risk of bias, and in none of the studies, data were imputed.

Two authors (B.E. and H.Z.-L.) independently used the GRADE approach to rate the overall quality of evidence. The quality of evidence reflects the extent to which we are confident that an estimate of effect is correct, and we applied this in the interpretation of results. There are 4 possible ratings of quality: high, moderate, low, and very low. A rating of high quality of evidence implies that we are confident in our estimate of effect and that further research is very unlikely to change our confidence in the estimate of effect. A rating of very low quality implies that we are very uncertain about any estimate of effect obtained.

The GRADE approach rates evidence from RCTs that do not have serious limitations as high quality. However, several factors can lead to the downgrading of the evidence to moderate, low, or very low. The degree of downgrading is determined by the seriousness of these factors: study limitations (risk of bias), inconsistency, indirectness of evidence, imprecision, and publication bias. We included a “Summary of findings” table for our comparison of betahistine versus placebo, constructed according to the recommendations described in chapter 11 of the Cochrane Handbook for Systematic Reviews of Interventions [Higgins and Green, 2011], for the following outcomes: the primary outcomes vertigo (the proportion of patients with a reduction in vertigo symptoms) and significant adverse events (upper gastrointestinal discomfort) and the secondary outcomes hearing loss, tinnitus, aural fullness, other adverse effects (headache and allergic skin reaction), and well-being and disease-specific health-related quality of life.

The electronic database search was performed by the Cochrane ENT Information Specialist on 29 January 2019 and identified 1,130 records in total. No additional records were identified through other sources. This number dropped to 733 after the removal of duplicates. We screened the 733 records and found 710 to be irrelevant. We were left with 23 potentially eligible studies. We excluded 13 of these studies with reasons (see Excluded Studies). We identified 10 studies meeting the inclusion criteria in terms of study design, participants, and interventions. No further eligible records were identified from a handsearch of the reference lists. There are no studies awaiting assessment, and we identified no ongoing studies. The study selection process is shown in Figure 1.

We included 10 RCTs, the details of which are shown in the characteristics of included studies table (Tables 1-10). One of the included studies included >2 treatment arms [Adrion et al., 2016]. The study of Adrion et al. [2016] was a 3-arm study, which compared high-dose betahistine, low-dose betahistine, and placebo. This was also the only study to highlight no financial conflict of interest. We identified no unpublished industry studies.

In 5 out of 10 studies, a prospective, crossover comparison design was used [Burkin, 1967; Frew and Menon, 1976; Meyer, 1985; Schmidt and Huizing, 1992; Mira et al., 2003]. In 2 of these 5 studies, data prior to crossover were extractable. In the remaining 5 studies, a parallel group design was used. All studies were described as being double-blinded.

The sample size ranged from 10 [Ricci et al., 1987] to 221 [Adrion et al., 2016]. A total of 402 patients had results reported across the 10 included studies. No additional results from unpublished studies were included in this review.

All studies were conducted in otorhinolaryngology departments within hospitals. The majority of the studies were single-centered. The studies of Mira et al. [2003] and Adrion et al. [2016] were multicenter studies. The selected studies took place in Germany [Meyer, 1985; Adrion et al., 2016], the UK [Burkin, 1967; Frew and Menon, 1976], the USA [Elia, 1966], Italy [Salami et al., 1984; Ricci et al., 1987; Mira et al., 2003], Japan [Okamoto et al., 1968], and the Netherlands [Schmidt and Huizing, 1992].

All of the included studies described the recruited patients as having Ménière’s disease, but different inclusion criteria and definitions for the disease were used. In the study of Adrion et al. [2016], the internationally recognized criteria for “definite” Ménière’s disease were applied and it was therefore classified as class “I” (see Types of Participants). Both the studies of Schmidt et al. [1992] and Mira et al. [2003] used other diagnostic definitions, including patients with probable/possible Ménière’s disease according to the AAO-HNS criteria and the Utrecht working definition, and we therefore classified them as class “II.” We classified the studies of Elia [1966], Burkin [1967], Okamoto et al. [1968], Frew and Menon [1976], Salami et al. [1984], Meyer [1985], and Ricci et al. [1987] as class “III” since no specific predefined diagnostic criteria were provided or details on how vertigo attacks, hearing loss, and tinnitus were evaluated.

All included studies evaluated the effect of betahistine. The daily betahistine dose that was used in the included studies varied: 16 mg [Elia, 1966; Burkin, 1967], 24 mg [Salami et al., 1984], 2 mg [Frew and Menon, 1976; Mira et al., 2003], 36 mg [Okamoto et al., 1968; Meyer 1985] (2 times daily with 3 pills), 72 mg [Schmidt and Huizing, 1992], and 144 mg [Ricci et al., 1987]. One study compared high-dose betahistine (144 mg per day, in 3 doses) and low-dose betahistine (48 mg per day, in 2 doses) to placebo [Adrion et al., 2016]. Schmidt and Huizing [1992] used a slow-release formulation. Assessment with regard to compliance was only reported in detail by Adrion et al. [2016]. None of the selected studies evaluated the effect of betahistine with concurrent use of other treatment. All studies used a placebo as the comparator.

Most of the selected studies only evaluated short-term effects (<3 months), except for Schmidt et al. [1992], Mira et al. [2003], and Adrion et al. [2016]. Adrion et al. [2016] evaluated the effects of all 3 interventions arms after 9 months, whereas Schmidt et al. [1992] defined a follow-up period of 8 months. Mira et al. [2003] assessed the effects after 3 months. All included studies used one of our prespecified outcome measures (see Types of Outcome Measures).

All of the included studies included vertigo as one of their outcomes in the follow-up analyses. None of the included studies used the AAO-HNS diagnostic guideline to classify the frequency of vertigo attacks. In 3 studies, the frequency of attacks was used as the main outcome to measure the effect of betahistine after a long-term follow-up (3 months or more), in which all studies used different definitions to quantify the attack frequency, namely the log-transformed number of attacks per 30-day interval based on daily diary reports, the number of vertigo attacks per month, and the imbalance scores based on the number of attacks multiplying the number by 1, 4, or 9 for a mild, moderate, or severe attack, respectively [Schmidt and Huizing, 1992; Mira et al., 2003; Adrion et al., 2016]. Burkin et al. [1967] quantified whether patients experienced dizziness or not, while Elia et al. [1966] based the effect of treatment on a subjective scale, which ranged from 0 to 3. The remaining studies used different ordinal scales to quantify the severity/intensity of the vertigo attacks by means of a 4-point scale [Frew and Menon, 1976], a 5-point scale [Meyer, 1985], a 3-point scale [Okamoto et al. 1968], and a vertigo maximum intensity of the episode and the mean duration of each vertigo episode [Salami et al., 1984]. Ricci et al. [1987] used the AAO classification in which both the effects on vertigo and hearing were combined and classified into 4 groups (A–D).

The incidence of upper gastrointestinal discomfort was reported by 2 studies [Schmidt and Huizing, 1992; Mira et al., 2003], which both assessed the effect of betahistine in the long term (3 months or more).

The effect of betahistine on hearing loss was assessed in 7 studies in variable ways. Adrion et al. [2016] reported results of pure-tone audiometry per frequency (250, 500, 1,000, and 2,000 Hz) and reported the adjusted mean change for placebo; these were compared with the adjusted MD for the low-dose and high-dose betahistine. Frew and Menon [1976] reported the amount of deafness by means of a 4-point scale without any further details. Meyer et al. [1985] reported the mean frequency scores with standard deviation based on the 3-point threshold of 0.5 Hz, 1.0 Hz, and 2.0 kHz. Okamoto et al. [1968] used a 3-point scale by which subjective changes in hearing were assessed. The mean threshold for the frequencies of 0.5, 1.0, and 2.0 Hz were classified by the ANSI in the study of Ricci et al. [1987], resulting in 6 classes (0–25 dB = normal, 26–40 dB = mild hearing loss, 41–55 dB = moderate hearing loss, 56–70 dB = moderately serious hearing loss, 71–90 dB = serious hearing loss, and 91 dB = very serious hearing loss). Salami et al. [1984] used the mean threshold at frequencies of 0.25, 0.5, 1.0, and 2.0 kHz, but no mean and standard deviations were reported. Schmidt and Huizing [1992] used the mean threshold scores based on the frequencies from 0.25 to 2 kHz.

All but one study reported changes in tinnitus symptoms before and after treatment [Burkin, 1967]. Adrion et al. [2016] used the MiniTF questionnaire, whereas Elia [1966] used a subjective scale that ranged from 0 to 3 (3 = incapacitating, 2 = severe, 1 = moderate, 0 = not present). Frew and Menon [1976] used a 4-point scale, Meyer [1985] a 5-point scale, and Okamoto et al. [1968] a 3-point scale. Mira et al. [2003] reported tinnitus as part of the “associated symptoms,” which all together were scored with aural fullness, nausea, and vomiting by means of a 4-point scale (0 = absent, 1 = mild, 2 = severe, 3 = disabling). Both Salami et al. [1984] and Ricci et al. [1987] used a scale ranging from 0 to 6, whereas Schmidt and Huizing [1992] used a 4-point scale and the minimum masking level in dB with mean and standard deviations to assess the effect on tinnitus.

Aural fullness was reported by 7 of the selected studies, except for Burkin [1967] and Okamoto et al. [1968]. Adrion et al. [2016] reported that participants were instructed to record coexisting symptoms such as aural fullness but data were not shown in the results section. In line with previous outcomes, Frew and Menon [1976] used a 4-point scale and Meyer [1985] a 5-point scale. In line with the tinnitus outcome, Mira et al. [2003] reported aural fullness as part of the “associated symptoms” questionnaire. Both Salami et al. [1984] and Ricci et al. [1987] again used a scale ranging from 0 to 6. Aural fullness was evaluated in Schmidt and Huizing [1992] by means of a scale ranging from none to mild, moderate, or severe, similar to tinnitus.

The incidence of other adverse effects was reported by 4 studies [Okamoto et al., 1968; Schmidt and Huizing, 1992; Mira et al., 2003; Adrion et al., 2016].

The effect on well-being was evaluated in 2 studies. Adrion et al. [2016] used the DHI, whereas Mira et al. [2003] used the DHI, the vestibular disorders activities of daily living (VDADL), and the disease-specific health-related quality of life questionnaire.

We excluded 13 studies for several reasons: duplicate publication (based on the available information, full texts were checked), wrong study design, wrong comparator, and wrong patient population (see characteristics of excluded studies in tables).

Two authors (B.E. and H.Z.-L.) critically reviewed the studies for risk of bias. Where necessary, authors were contacted if we felt more detailed information on the methodology was required. In general, random sequence generation, allocation concealment, and blinding of participant and personnel and outcome assessment were not reported clearly. This can be seen in the number of unclear scores regarding these matters (see Fig. 2). All studies were reported to be double-blinded, whereas only Okamoto et al. [1968] and Adrion et al. [2016] reported in detail how blinding was accomplished. Many studies had incomplete outcome data and other sources of bias, resulting in high risk of bias scores. The characteristics of each trial are listed in the “characteristics of included studies” table, and results on risk of bias are summarized in Figures 2 and 3.

We considered the risk of selection bias due to inadequate method description on sequence generation to be unclear in 7 studies [Elia, 1966; Burkin, 1967; Frew and Menon, 1976; Salami et al., 1984; Meyer, 1985; Schmidt and Huizing, 1992; Mira et al., 2003] and low in the remaining 3 studies [Okamoto et al., 1968; Ricci et al., 1987; Adrion et al., 2016]. In the study performed by Adrion et al. [2016], a 1:1:1 ratio was used, creating a high-dose betahistine, low-dose betahistine, and placebo groups. Okamoto et al. [1968] used a table of random numbers created by a third party independent from the medical institution. Likewise, Ricci et al. [1987] assigned patients to the betahistine or placebo group based on a random list.

The allocation concealment was rated as unclear in all but 3 studies [Elia, 1966; Okamoto et al., 1968; Adrion et al., 2016]. Elia [1966] defined that a fifth person who was not involved in the study coded the tablets. The treating physician, the statistician, the nurse, and the patients were not aware of the given drug, whereas the code was not broken until the final draft of this report. Adrion et al. [2016] described in detail that allocation concealment was performed by means of an Internet-based randomization schedule, which was generated by an investigator with no clinical involvement in the trial. The patients, clinicians, core laboratories, and trial staff were all described as blinded to treatment allocation. Finally, Okamoto et al. [1968] described that drug bottles were labelled with serial number according to the random layout list. The list was created at random by a third party.

In 2 studies [Elia 1966; Frew and Menon, 1976], no details on baseline characteristics were reported. Both studies were rated as “class III” with regard to the diagnostic criteria applied to include patients with Ménière’s disease. Although Okamoto et al. [1968] described the sex distribution among the population, no information on age was given and unclear diagnostic criteria were used to describe the population of studies (class III). With regard to the robustness of diagnostic criteria used to include patients with Ménière’s disease, 7 studies were rated as “class III” [Elia, 1966; Burkin, 1967; Okamoto et al., 1968; Frew and Menon, 1976; Salami et al., 1984; Meyer, 1985; Ricci et al., 1987], 2 as “class II” [Schmidt and Huizing, 1992; Mira et al., 2003], and 1 [Adrion et al., 2016] as “class I.” No significant differences were found in the studies that presented baseline characteristics for age and sex distribution [Salami et al. 1984; Ricci et al., 1987; Schmidt and Huizing, 1992; Mira et al., 2003; Adrion et al., 2016]. Only Salami et al. [1984], Ricci et al. [1987], Schmidt and Huizing [1992] and Adrion et al. [2016] reported the duration of disease before the start of the trial. The effect of betahistine on hearing loss was objectively assessed by Salami et al. [1984], Ricci et al. [1987], Schmidt and Huizing [1992], and Adrion et al. [2016], although specific hearing score outcomes were only given by Schmidt and Huizing [1992] and Adrion et al. [2016].

Due to inadequate blinding in 7 out of the 9 studies [Elia, 1966; Burkin, 1967; Okamoto et al., 1968; Frew and Menon, 1976; Salami et al., 1984; Meyer, 1985; Ricci et al., 1987; Schmidt and Huizing, 1992; Mira et al., 2003], there was a risk of performance bias and detection bias in most studies. Although Elia [1966] described that a fifth person coded the tablets given during trial execution, the same sequence was repeated (A–D) in all patients. As a result, the intervention could be predicted by the patients, physician, or statistician and was therefore considered to be of high risk. Ricci et al. [1987] described that a random list was used to divide participants, but no information on blinding was provided in the methods section. Therefore, we considered that there was still a considerable risk of inadequate blinding in both studies.

We considered only 2 studies to have a low risk of attrition bias [Schmidt and Huizing, 1992; Adrion et al., 2016], as concrete reasons of non-completion of the trial were given. In the studies performed by Burkin [1967], Frew and Menon [1976], Salami et al. [1984], and Ricci et al. [1987], there was no mentioning of dropping out or discontinuation of trial participation for any reason. But as it remained unclear how many patients were analyzed per outcome and only the level of significance was given, we assessed the risk of attrition bias to be unclear. The risk of attrition bias due to incomplete outcome data was high in Elia [1966], Okamoto et al. [1968], Meyer [1985], and Mira et al. [2003]. In the study performed by Elia [1966], 4 of 20 participants dropped out due to noncompliance to the trial and migration of participants. In 2 patients, it remained unclear whether they had received betahistine or placebo. Meyer [1985] reported a lower number of participants in some outcomes (for instance, disturbed walking pattern) than in other outcomes, but no information was reported on this matter in the manuscript. The participants studied by Mira et al. [2003] were not balanced across groups, for which they did not correct in the analyses. Last, Okamoto et al. [1968] reported that 4 patients out of 36 dropped out (11%), not due to adverse effects of the drug use, but any other reason for dropout was not clarified.

A study protocol was available for the study performed by Adrion et al. [2016], published prior to the execution of the study, from which we found that predefined outcomes were evaluated in the published version of the final manuscript, reporting on study results. In 7 studies, the outcomes that were mentioned in the abstract and/or methods section were also reported in the results section. Therefore, we considered the risk of selective reporting to be low in these studies [Elia, 1966; Burkin, 1967; Okamoto et al., 1968; Salami et al., 1984; Ricci et al., 1987; Schmidt and Huizing, 1992; Mira et al., 2003]. The studies performed by Frew and Menon [1976] and Meyer et al. [1985] mentioned outcomes in the method section that were not shown or described in the results section without reasoning and were considered to suffer from a high risk of selective reporting.

None of the studies had a low risk bias on other potential sources of bias. Adrion et al. [2016] did not reveal data on pre-randomization attack frequency, although it was considered as an inclusion criterion. Data were not shown with respect to duration and age at the onset of disease, although groups were reported to be balanced based on these characteristics; thus, it remained unclear whether this was performed properly. Although Elia [1966], Burkin [1967], Salami et al. [1984], Meyer [1985], and Ricci et al. [1987] reported no details on how statistical analysis was performed, the authors concluded that a positive effect of betahistine on symptoms of Ménière’s disease was found; this was considered to be a high potential source of bias [Elia, 1966; Burkin, 1967; Salami et al., 1984; Meyer, 1985; Ricci et al., 1987]. Frew and Menon [1976] used 1-sided testing, which should have been 2-sided. Moreover, standard deviations were not reported, and we considered a high risk of selection bias due to a pretreatment period, in which the investigator was allowed to exclude placebo responders, thereby decreasing external validity of the study results. Sample size calculation performed by Mira et al. [2003] was done without referring to previous studies performed. In the outcome section, improvement of associated symptoms including tinnitus, fullness of the ear, nausea, and vomiting was summarized in 1 figure. However, it was unclear how it was performed and whether data were complete. The trial medication during the execution of the trial by Okamoto et al. [1968] was supplied by Eisai Co.; the role of this subsiding party remained unclear. We considered there was a high risk of bias in the study by Schmidt and Huizing [1992] since the intention-to-treat analysis was not correctly executed because 1 patient crossed over due to side effects earlier than the protocol stated. Furthermore, the data were analyzed per protocol. Moreover, in these analyses, the authors did not account for the loss of follow-up from dropouts.

The effects of the interventions are summarized in Table 11 (summary of findings).

All of the included studies evaluated the effect of betahistine on vertigo symptoms by means of different Likert-type scales or by using a mathematical formula, resulting in both dichotomous and continuous data; we therefore could not pool the data for this outcome. In addition, data from the first period could not be extracted from 4 crossover studies [Elia, 1966; Burkin, 1967; Frew and Menon, 1976; Meyer, 1985]. Ricci et al. [1987] combined the effect on vertigo and hearing loss to 1 outcome, and no numerical data were presented. No data could be extracted from Salami et al. [1984].

Okamoto et al. [1968] used a 3-point visual analog scale, from which the proportion of patients with an improvement in vertigo symptoms at short-term follow-up was quantified. The RR was 3.0 (95% confidence interval [CI] 0.97–9.30) in favor of betahistine (GRADE: low certainty) (Fig. 4).

Schmidt and Huizing [1992], Mira et al. [2003], and Adrion et al. [2016] all assessed the effect of betahistine after a long-term follow-up. Data could not be pooled because there was significant heterogeneity in outcomes between studies and no raw data to impute standard deviations were available. Mira et al. [2003] described a significant improvement in the monthly vertigo attack frequency without presenting absolute baseline and endpoint data for the placebo group. Schmidt and Huizing [1992] found no difference between the betahistine and placebo groups in the effect on imbalance scores. The study of Adrion et al. [2016] was the study with the lowest risk of bias; this study found no favorable effect after comparing high-dose and low-dose betahistine to placebo. In summary, 2 studies found no favorable effect for betahistine, which included 1 study with a high quality [Schmidt and Huizing, 1992; Adrion et al., 2016]. We assessed the certainty of the evidence for this outcome as moderate (GRADE).

Both Schmidt and Huizing [1992] and Mira et al. [2003] reported no significant difference in the incidence of upper gastrointestinal discomfort. The pooled RR was 0.86 (95% CI 0.13–5.83; 2 studies; 118 participants) in favor of placebo (Fig. 5) (GRADE: low certainty).

Hearing loss was evaluated in both the short and long term by means of both dichotomous data (proportion of patients with improvement) [Okamoto et al., 1968; Ricci et al., 1987] and continuous data based on means with corresponding 4-point thresholds for the frequencies from 0.25 to 2.0 kHz [Schmidt and Huizing, 1992]. Data from the 4 remaining studies could not be pooled because only data per frequency were reported and no mean 4-point threshold score could be calculated [Adrion et al., 2016], no pre-crossover data were available [Frew and Menon, 1976; Meyer, 1985], or no data were presented [Salami et al., 1984]. No significant difference between the betahistine and placebo groups could be found in the included studies.

In the short term, Okamoto et al. [1968] reported an RR of 3.00 (95% CI 0.34–26.19; 1 study; 36 participants) for the improvement of hearing (GRADE: low certainty) (Fig. 6).

The long-term effect on hearing loss was evaluated by Ricci et al. [1987], which reported an RR of 3.00 (95% CI 0.15–59.89; 1 study; 10 participants) (GRADE: very low certainty) (Fig. 7). Schmidt and Huizing [1992] found no difference between the betahistine group and the placebo group based on mean threshold scores at long-term follow-up (MD 10.10, 95% CI −0.97 to 21.17; 1 study; 35 participants) (GRADE: low certainty) (Fig. 8).

The effect of betahistine on tinnitus was evaluated at short-term follow-up by Okamoto et al. [1968], which reported the proportion of participants with an improvement in RR of 2.67 (95% CI 0.84–8.46; 1 study; 36 participants) (GRADE: low certainty) (Fig. 9). These results are not statistically significant or clinically relevant.

At long-term follow-up, Ricci et al. [1987] found no difference between the betahistine group and the placebo group based on the proportion of patients without deterioration of hearing (RR 1.00, 95% CI 0.71–1.41; 1 study; 10 participants) (GRADE: very low certainty) (Fig. 10). Long-term effect was reported as the SMD based on the MiniTF in Adrion et al. [2016], which found no difference between betahistine and placebo groups (SMD −0.16, 95% CI −0.48 to 0.17; 1 study; 144 participants) (GRADE: moderate certainty) (Fig. 11).

Data on aural fullness could not be extracted from any of the 7 studies because first period, pre-crossover data could not be extracted [Frew and Menon, 1976; Meyer, 1985], no aural fullness data were presented [Adrion et al., 2016], no numerical data were presented [Salami et al., 1984; Schmidt and Huizing, 1992], data for the betahistine group and placebo group were not shown [Ricci et al., 1987], or results were reported only with a p value without data on baseline absolute values and endpoint values [Mira et al., 2003].

The incidence of “other” adverse effects was reported at both short and long-term follow-up, which were dull headache, tinnitus, ear discomfort, nervous system disorders, headache, heart burn, skin rash, increased diuresis, extrasystoles, and oral formication.

Okamoto et al. [1968] found no significant difference in other adverse effects between the betahistine and placebo groups (RR 1.67, 95% CI 0.47–5.96; 1 study; 36 participants) (GRADE: low certainty) (Fig. 12).

At long-term follow-up, Schmidt and Huizing [1992], Mira et al. [2003], and Adrion et al. [2016] found a lower RR in favor of placebo than for betahistine. The pooled RR was 2.58 (95% CI 1.21–5.49; 3 studies; 265 participants) (GRADE: moderate certainty) (Fig. 13).

Disease-specific health-related quality of life was evaluated by Mira et al. [2003], but because the results were reported only as percentage reductions without baseline absolute values and missing measures of spread, no useful data could be extracted. Adrion et al. [2016] evaluated disease-specific health-related quality of life by means of the DHI, which was reported as SMDs compared to placebo. No significant difference between the placebo and high-dose betahistine group could be demonstrated (SMD 0.08, 95% CI −0.25 to 0.40; 1 study; 144 participants) (GRADE: moderate certainty) (Fig. 14).

The current review includes 10 RCTs, which evaluated the effects of betahistine compared to placebo in a total of 402 adult participants with Ménière’s disease. For the primary outcome, the reduction in vertigo symptoms (considering together the intensity, frequency, and duration of those symptoms), there was clinical heterogeneity between studies due to differences in the outcome measured and methods used. We could, therefore, not perform data pooling for this outcome. One adequately powered study with low risk of bias did not find evidence of a difference between the betahistine and placebo groups for this outcome [Adrion et al., 2016]. We assessed the certainty of this evidence to be moderate (GRADE). No statistically significant or clinically relevant difference was found with respect to the significant adverse effect (upper gastrointestinal discomfort) in the 2 studies that reported this outcome [Schmidt and Huizing, 1992; Mira et al., 2003]. No differences in hearing loss, tinnitus, or well-being and disease specific health-related quality of life were found between the betahistine and placebo groups in any of the individual studies assessing these outcomes (low- to very low-certainty evidence). Aural fullness was evaluated by 1 study based a non-validated visual analog scale, which lacked information whether or not results were statistically better in the betahistine than in the placebo group. The other adverse effect that was seen in the short term was a dull headache. No significant difference between the betahistine and the placebo groups (low-certainty evidence) could be demonstrated. Adverse effects in the long term included tinnitus, ear discomfort, nervous system disorders, headache, heartburn, skin rash, increased diuresis, extrasystoles, and oral formication. The pooled RR demonstrated a lower risk in favor of placebo over betahistine. High-quality studies evaluating the effect of betahistine on patients with Ménière’s disease are lacking. However, 1 study with low risk of bias found no evidence of a difference in the effect of betahistine on the primary outcome, vertigo, in patients with Ménière’s disease when compared to placebo [Adrion et al., 2016].

Specific diagnostic criteria were used to select patients for trial participation in only one of the included studies [Adrion et al., 2016]. In the remaining studies, either rather vague diagnostic criteria were applied, including recruiting patients with “probable” Ménière’s disease, or no details were provided about how patients were diagnosed with Ménière’s disease. In particular, in the 6 studies involving “class III” rated participants (see Types of Participants), it remains disputable whether these patients can be considered to have Ménière’s disease. The applicability of the evidence in these studies is therefore limited. In none of the included studies were data provided on the previous duration of the disease, including the frequency and intensity of attacks. Generally, in Ménière’s disease, vertigo attacks stop after approximately 5–15 years. It is, therefore, of great importance that this information is collected before trials are started to allow the interpretation of any observed treatment effect.

The certainty of the evidence in this review ranged from moderate to very low, although 1 high-quality study was included [Adrion et al., 2016]. Since none of the studies used similar methods to evaluate the effect of treatment on vertigo, it remains hard to assess whether the reported estimates are true. Future research should aim to use more standardized and comparable methods to assess the effect on vertigo in order to increase the level of evidence and allow more concrete conclusions to be drawn from the data. The certainty of the evidence was mainly negatively affected by study limitations (risk of bias of bias), the low level of external validity, and imprecision due to the small sample sizes. Studies lacked information on the selection procedure used to identify participants, and methods were poorly reported, especially with respect to statistical analyses. In most studies, it remained unclear how randomization, allocation concealment, and blinding of personnel, participants, and outcome assessors were performed. Only one of the included studies had a prepublished protocol available for inspection.

We made no significant changes to our planned methods. We performed a comprehensive electronic database search. Language was not a barrier for inclusion, and we reviewed full-text articles in Japanese, German and Italian after these were translated. The roles of all authors were predefined before the start of the review process. Two authors selected studies for inclusion and judged risk of bias independently. Two authors independently extracted data to minimize personal bias. Both clinical and statistical heterogeneity were evaluated before considering quantitative analyses. The predefined outcome measures were as broad as possible, aiming to allow the summarizing of data or make pooling of data more feasible.

At least 2 other reviews have evaluated the effect of betahistine in the treatment of Ménière’s disease [Lacour et al., 2007; Nauta, 2013]. Both reviews concluded that there is a favorable effect of betahistine on vertigo. Lacour et al. [2007] is an expert opinion paper, which describes the definition of Ménière’s disease and its epidemiology and pathophysiology and the role of betahistine in its therapeutic management including the mechanisms of action, which are hypothesized to play a role in the potential positive effect of the drug. The favorable clinical effect of betahistine is evaluated by means of a narrative summary of the results found in the study of Mira et al. [2003]. In addition, comparative studies and the results of an as yet unpublished open trial study are discussed. No data pooling or meta-analysis was performed. The authors concluded that betahistine is an effective therapy for Ménière’s disease and related conditions. Nauta et al. [2013] is a review and meta-analysis on patients with vestibular vertigo or Ménière’s disease, which aimed to assess the “overall judgment of the investigator on the effectiveness of the drug treatment.” Statistical analyses were performed to combine ordered categorical data. The overall random effect, the average odds ratio, was 2.58 (95% CI 1.67–3.99). When restricted sub-analyses of Ménière’s disease patients only were performed, the average odds ratio was 3.37 (95% CI 2.14–5.29). No analysis of validity or risk of bias assessment was presented. Cochrane ENT has published 2 systematic reviews evaluating the effects of betahistine for other clinical indications than Ménière’s disease. One review evaluated the effect of betahistine on symptoms of vertigo, identifying 17 studies (1,025 participants) [Murdin et al., 2016]. Out of these 17 studies, 5 evaluated the effect of betahistine for Ménière’s disease, from which the pooled RR was 1.56 (95% CI 0.92–2.62; 3 studies; 139 participants). Similar to the current review, the authors stated that results need to be interpreted with caution, as the diagnoses differed between studies and did not necessarily meet the standard diagnostic criteria. Moreover, the incidence of adverse effects was similar for both betahistine and placebo groups. The second review evaluated the effect of betahistine on tinnitus and included 5 studies (303–305 participants) [Wegner et al., 2018]. This review concluded that there is no evidence to suggest that betahistine has an effect on subjective idiopathic tinnitus.

Recently, a clinical practice guideline for Ménière’s disease was published, which includes an advice concerning oral pharmacotherapy [Basura et al., 2020]. Both diuretics and betahistine are mentioned as an optional treatment. The guideline mentions that serious side effects are rare and patients should be clinically reassessed to monitor improvement and intolerance of side effects. In case the symptoms may subside, it is advised to titrate down or stop its use. We believe that it is of great importance that patients should be informed about the lack of efficacy found in the study by Adrion et al. [2016], since this study offers the best evidence to date. Nonetheless, periodic clinical evaluation regarding the effects, intolerance, or side effects may help patients and clinicians in their decision to either stop or continue treatment with betahistine.

In summary, previous reviews have concluded either that there is insufficient evidence to say whether betahistine has any effect on Ménière’s disease or that there may be a positive effect of betahistine based on low-quality studies, so further research is likely to have an important impact on the interpretation of the results. In line with the findings of the current review, previous work has also concluded that betahistine is generally well tolerated with a similar risk of treatment-related adverse effects to placebo. Moreover, all previously evaluated studies included in reviews or meta-analyses have suffered from significant heterogeneity with respect to participants, dose of betahistine, follow-up duration, and the methods of evaluation for outcomes.

High-quality studies evaluating the effect of betahistine on patients with Ménière’s disease are lacking. However, 1 study with high quality found no evidence of a difference in the effect of betahistine on the primary outcome, vertigo, in patients with Ménière’s disease when compared to placebo [Adrion et al., 2016]. Betahistine appears to be generally well tolerated, and the risk of gastrointestinal discomfort is comparable to that of placebo. Further studies with a low risk of bias (in particular, with respect to allocation and blinding) and rigorous inclusion criteria are required to independently verify the lack of evidence of a beneficial effect of betahistine for Ménière’s disease compared to placebo. Patients considering treatment options should be informed about the findings of this review, which found no evidence of a beneficial effect of betahistine on the primary outcome, vertigo. Patients should also be informed that betahistine is generally well tolerated and the risk of adverse effects is low and comparable to that of placebo. Based on this information, patients may still choose to start their treatment with betahistine, especially in the current absence of any other safe, noninvasive effective treatment that has high patient acceptability and relatively low cost and is well supported by high-certainty evidence. Nonetheless, it remains questionable whether prescription of betahistine is justifiable or cost-effective. If patients decide to proceed with betahistine, a trial period of around 3 months could be offered. This period is sufficient to assess whether the patient experiences any beneficial effects on their symptoms or any adverse effects. If any unwanted effects outweigh any benefit, or there is no apparent improvement, therapy can be withdrawn.

Future research into the effectiveness of betahistine in patients with Ménière’s disease should use rigorous methodology. Due to the subjective nature of most outcome measures, the risk of bias with respect to randomization and blinding needs to be low to avoid any placebo effect. Standardized diagnostic criteria should be rigorously applied. A standardized method of designing and reporting trial results such as the CONSORT statement should be used (CONSORT 2010). We recommend validated, patient-centered outcome measures for research in the field of Ménière’s disease. A core outcome set would be of particular value for this condition because of the multiple subjective symptoms that are characteristic. By means of a core outcome set, a standardized set of outcomes would be reported, which would facilitate direct comparison between studies and the ability to perform data pooling. Due to the highly variable and poorly understood natural history of Ménière’s disease, baseline characteristics and information on the natural course of the disease is of great importance for the interpretation of the treatment effects. For instance, information on the duration of disease, the frequency of vertigo attacks since the start of the disease, the duration and intensity of the vertigo attacks, age, and the amount of hearing loss may all be of value at the time of trial enrolment. Moreover, with the exception of the 1 high-quality study [Adrion et al., 2016], none of the included studies in this review carried out an adequate power calculation before the start of trial. Future trials should include a power analysis to make sure that the estimated difference in effect between treatment arms can indeed be identified by the number of included participants. Research into the natural history of the condition via prospective longitudinal studies or registries would also be valuable in planning future clinical trials of therapy for Ménière’s disease. However, in the light of limited means; the huge effort involved in conducting a trial on the part of patients, doctors, and researchers; and the very low estimated added value of betahistine in the treatment of Ménière’s disease found in this review, we anticipate that research on this topic may not be prioritized.

The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the Systematic Reviews Program, NIHR, NHS, or the Department of Health. Specific thanks to Martin Burton, Lee Yee Chong, and Jenny Bellorini from the Cochrane ENT for their extensive advice in setting up this protocol.

The paper is exempt from Ethical Committee approval since it involves analyses on previously published data.

The authors have no conflicts of interest to declare.

This research received no specific grant from any funding agency in the commercial or not-for-profit sectors.

All authors were involved in the drafting of the protocol. B.E. selected and obtained studies, extracted data, and assessed risk of bias. B.E. entered data into RevMan 5 and carried out and interpreted the analyses. B.E. drafted the final review and has responsibility for updating and maintaining the review. H.Z.-L. selected studies, extracted data, assessed risk of bias, and helped interpret the analyses. H.Z.-L. provided advice throughout the analyses and drafted the final review. T.B. provided advice and drafted the final review. P.P.B. initiated the revision of the review, provided advice, and drafted the final review.