Study oversight
The study was investigator initiated and conducted in accordance with the principles of the Declaration of Helsinki, International Conference for Harmonisation Guidelines for Good Clinical Practice and relevant national regulations. The protocol (Additional file 2) was approved by the ethics committees of each participating centre. Furthermore, the efficacy, safety, integrity and feasibility of the trial was monitored by a data safety monitoring board (DSMB) consisting of three independent, non-participating clinicians. All patients provided written informed consent before any study procedures or assessments were done.
Study population and procedures
Subjects were screened at 6 German academic outpatient clinics, 4 of these investigational sites (German Center for Vertigo and Balance Disorders at the university hospital Munich; Department of Neurology of the general hospital Celle, University of Essen, and community hospital Altötting-Burghausen) allocated patients between June 20, 2012 (first patient in) and April 10, 2017; last patient last visit was on January 3, 2018.
The patient population consisted of male or female patients aged 18 years and above diagnosed as having probable or definite vestibular migraine according to the Neuhauser criteria [2] (see Additional file 2 for details). For enrolment, patients had to experience a frequency of 6 to 30 VM-related attacks per 3 subsequent months prior to screening visit (information retrospectively collected at in-person interviews), be capable to follow the study instructions and likely to complete study visits.
Exclusion criteria were a diagnosis of other co-existing vestibular disorders such as Menière’s disease, phobic postural vertigo, benign paroxysmal positional vertigo (BPPV) and vestibular paroxysmia. Patients were also excluded if they had central disorders such as paroxysmal brainstem attacks and transient ischemic attacks. Other exclusion criteria were given by contraindications for the treatment with metoprolol such as known allergic reaction to the trial drug or other beta receptor blockers, shock, acidosis, any bronchospastic disease (e.g. bronchial asthma), sick sinus syndrome, known sino-atrial- or atrio-ventriclar block, bradycardia of less than 50 bpm at rest, systolic blood pressure less than 100 mmHg, end grade peripheral arterial disease, known severe coronary heart disease or heart failure and concurrent treatment with monoamine oxidase inhibitors, sympathomimetic drugs, catecholamine-depleting drugs, digitalis glycosides. Other patient factors leading to exclusion included poorly controlled diabetes mellitus, pheochromocytoma, suspicion of developing thyrotoxicosis, disorders of homeostasis, porphyria, psoriasis; pregnancy or breastfeeding; persistent hypertension with systolic blood pressure higher than 180 mmHg or diastolic blood pressure higher than 110 mmHg (mean of 3 consecutive arm-cuff readings over 20 to 30 minutes) that cannot be controlled by antihypertensive therapy; life expectancy of less than 12 months; other serious illness that may confound treatment assessment; currently receiving beta-blockers; enrolment in another clinical trial; exposure to any investigational medication within 30 days prior to baseline visit.
Study procedures
This study consisted of a screening visit, a 6-month treatment period, and a final evaluation at month 9 after a 3-month post-treatment follow-up period. Both the study examinations and treatment were performed in an outpatient setting. Based on the information from the screening visit patients were randomized or were excluded if they did not meet eligibility criteria. At the day of inclusion, patients received their study medication together with a paper-based diary to document VM-related symptoms on a daily basis over the 9 month observation period. Patients were seen at 5 scheduled clinic visits for protocol-specified evaluations at screening, baseline (day of inclusion), and at months 1, 3, and 6 (end of treatment period); 3 standardized telephone visits were performed after 2, 4, and 5 months post-randomization for compliance checks with respect to treatment and diary documentation, and safety assessment.
All enrolled patients underwent a physical examination at baseline visit, and at every clinic visit post-randomization non-invasive neurological and neuro-otological and -ophthalmological examinations such as video-oculography including bithermal caloric testing, assessment of pursuit eye movement, gaze-holding, saccades and subjective visual vertical (SVV). For more details refer to the original protocol in the Additional file 2 and Additional file 4. Further, they had to complete the paper-based self-administered Dizziness Handicap Inventory (DHI), a 25-item validated questionnaire with a 3-point response scale to rate the self-perceived impact of dizziness on health-related quality of life [22, 23]. Possible DHI total scores range from 0 to 100 points, with higher values reflecting greater impairment. Results from the trial assessments were recorded in paper-based case report forms filled by the study personnel at each clinical site.
Event-driven diary documentation
Dizziness event data were captured by means of the patient’s diary with entries made whenever symptoms associated with migrainous vertigo occurred. Patients were instructed to document the following items: time of onset, duration, severity and type of the vertigo symptom (rotatory or postural vertigo, gait unsteadiness, or light-headedness); occurrence of accompanying symptoms (headache; nausea, vomiting, photo- or phonophobia, diplopia, other visual symptoms, tendency to fall); any action taken including any medication use. A diary template (original German version together with an English translation) is provided in Additional file 3.
Randomization, concealment, and blinding
Patients who met the eligibility criteria for enrolment were randomized in a 1:1 ratio to receive either metoprolol succinate or placebo for six months (Figure 1 CONSORT). Each study site received a pool of study medication including the treatment assignment in an opaque, sealed emergency envelope. If an eligible patient dropped out before the study medication had been delivered he or she was replaced by the next eligible patient enrolled at the same site. The concealed allocation was performed by an internet-based randomization schedule stratified by study site (https://wwwapp.ibe.med.uni-muenchen.de/randoulette). The fixed block size was four (starting with 6) which was not disclosed during the trial. The random number list was generated by a person with no clinical involvement in the trial. Patients and site personnel including outcome assessors, data analysts and statisticians remained blinded to treatment allocation.
Study treatments
Metoprolol-succinate sustained-release tablets (Beloc-Zok® mite 47.5 mg) manufactured by AstraZeneca, Wedel, Germany) were encapsulated for blinding purposes. Hard gelatine capsules containing the active ingredient were refilled from original pharmacy packaging into relabelled blisters by the pharmacy of the university hospital in Heidelberg, Germany. Placebo was an identically appearing inactive capsule filled with mannitol and aerosil but not containing any active ingredient, and was packed in identically appearing blisters as the investigational drug. Randomized patients were instructed to take one capsule per day starting as soon as possible after the receipt of the trial medication kit dispensed at baseline visit. The treatment procedure included a one week run-in period of 47.5 mg metoprolol succinate or placebo once a day (up-titration), a six month maintenance dosage with 95 mg metoprolol succinate or placebo once a day, plus tapering with 47.5 mg metoprolol succinate or placebo once a day for two weeks before stopping the prophylactic therapy (down-titration). Placebo treatment was justified due to lack of well-designed placebo-controlled trials for any drug therapy in VM. The six month treatment duration was deemed necessary to reliably assess a long-term prophylactic effect of the drug treatment on the incidence of VM-related vertigo attacks. If the patient was on prophylactic drug treatment of migraine, a washout period of at least one month was required before enrolment. Topiramate, valproic acid, lamotrigine, tricyclic antidepressants and other beta-blockers were considered as prohibited concomitant medication and thus, a protocol violation. Acute medical treatment of VM-related attacks like in episodic migraine with aura using non-opioid analgesics, non-steroidal anti-inflammatory drugs or triptans was allowed serving as added rescue medication. We aimed to assess the comparative effectiveness of the assigned prophylactic treatment regardless of whether or not switching to rescue medication had occurred which can be denoted as ‘treatment policy estimand’ according to the ICH E9 addendum [24].
Statistical methodology and planned analyses
Protocol-defined efficacy outcomes, and changes after trial commencement
The primary objective was to assess whether metoprolol was superior to placebo with respect to both disease domains ‘vertigo’ and ‘headache’. For the purpose of the study, the target estimate was based on the overall monthly mean incidence of vertigo and headache attacks during a 3-month long assessment period at the end of the double–blind 6-month treatment period, i.e., month 4 to 6 (day 91 to day 180) was defined as the time period of primary interest assuming that the maximum treatment effect emerges after being on study medication for more than 3 months. The pre-specified primary efficacy outcomes were the patient-reported number of vertigo attacks and the number of headache attacks per 30 day interval (starting from time point 1 defined as the date of first intake of the study medication). According to the protocol, superiority was to be claimed based on the vertigo outcome domain alone. Thus, the incidence of headache attacks per 30 days was defined as a co-primary outcome. In case of claimed superiority with respect to the outcome domain ‘vertigo’, the comparison of the monthly incidence of headache attacks between both groups was to be considered next important.
However, due to the poor documentation concerning headache-related symptoms and the diary focussing on the vestibular symptoms, derivation of a measurable variable for headache attacks was considered impossible and the co-primary efficacy endpoint had to be omitted. Further, the initially planned secondary outcomes duration and severity of vertigo episodes were omitted due to insufficient data quality. Instead, the number of vertigo days per 30 days (which was not preregistered in the protocol) was defined as a clinically meaningful key secondary efficacy outcome to assess the disease burden with respect to the outcome domain vertigo. A vertigo day was defined as a calendar day (0:00 to 23:59) demonstrating at least one documented vertigo episode of at least 5 minutes (regardless of severity and type). Derivation of this efficacy variable relies on fewer assumptions compared to a vertigo attack and also enables handling missing diary items.
On the contrary, a vertigo attack was endorsed applying the following decision rules: duration of at least 5 minutes and no longer than 72 hours, irrespective of vertigo type and severity; if time data (start and stop time) for a vestibular symptom were absent a duration of 24 hours was assumed; a vertigo episode which was interrupted by sleep or temporarily remits, was classified as one single attack, and not two; if applicable, patient-reported vertigo symptoms reported on two or more consecutive calendar days were summarized to one single vertigo episode lasting over consecutive calendar days (however, if the resulting duration extends 72 hours, these calendar days were considered free of vertigo attacks but counted as vertigo day).
A pre-specified diary-based secondary efficacy outcome was the number of monthly headache days (a calendar day where headache of any severity occurred according to the patient ratings). During the blind data review, this patient-reported outcome was refined to a migraine headache day (MHD), requiring at least one additional migraine-associated symptom such as nausea, vomiting, phono- or photophobia, disturbance of vision, or “migraine” provided as free text on the diary. However, features such as duration and severity of migraine headache or criteria as proposed by the International Classification of Headache Disorders-3 (ICHD-3) from 2018 [1] were not considered in order to derive definite MHD since these items were not requested on the diary.
All these changes to the latest protocol version concerning efficacy evaluation were made before breaking the treatment blind minimizing outcome reporting bias. Owing to the complexity of vestibular and migraine-associated symptoms, inaccurately documented episodes of vertigo (e.g. missing outcome items) and different individual perceptibility of both domains of the disease, the assessment of vertigo attacks, days and migraine headache days based on the raw daily diary recordings was very challenging. Therefore, a computer algorithm programmed in SAS was developed for the process of outcome adjudication and to derive all diary-based efficacy variables.
Protocol-defined observer-reported secondary efficacy endpoints included the proportion of patients achieving an improvement from baseline to month 6 in pursuit eye movement (change from state ‘saccadic (of any direction)’ to ‘smooth’ vs. change from state ‘smooth’ to ‘saccadic’, or no change); the proportion of patients achieving an improvement in SVV (change from state ‘abnormal’ to ‘normal’ vs. change from ‘normal’ to ‘abnormal’ or no change; state ‘abnormal’ was defined as the absolute deviation of more than 2.5° from vertical). Further, the absolute change from baseline to month 6 in the DHI mean total score was assessed.
Statistical efficacy analyses (including changes to protocol-specified analyses)
Efficacy analyses were conducted for the full analysis set (FAS) which included all randomized patients (ITT, intention-to-treat population) who did not fail to satisfy a major entry criterion, irrespective whether they were treated or not. Subjects who provided neither primary nor secondary efficacy data were excluded from efficacy analyses assuming missingness at random (MAR). The per protocol (PP) sample consisted of all subjects part of the FAS who did not substantially deviate from the protocol and who were on treatment for more than 90 days, counting from day of first intake. Safety analyses were done on all patients who received at least one dose of study drug.
According to the protocol, the principal analysis for the primary endpoint incidence of vertigo attack per 30 days during the 3-month time period of primary interest (month 4 to 6) was a robust nonparametric comparison between both treatment groups by means of the Wilcoxon rank-sum test. However, in the course of the trial it was evident that dropouts and incomplete diary documentation creating missing data could not be adequately handled by the intended test-based approach. In order to deal with the missing data structure over time we used a Poisson mixed effects model (Poi GLMM) which not only yields unbiased parameter estimates when missing observations are missing at random (MAR), but also provides reasonably stable results even when the assumption of MAR is violated [25-27].
For this longitudinal model-based approach, the log-transformed number of evaluated days per 30 days (defined as the number of calendar days with assessments recorded in the diary within a 30-day interval) was considered as offset term in order to reflect missing diary information (e.g. for withdrawals) and to standardize the monthly incidence of vertigo attacks to 30 days for that month. Time (range 1 to 6) and treatment-by-linear time-interaction was used as fixed effects, together with patient-specific intercepts and slopes for time as normally distributed random effects. Assuming no rapid onset of effect staying stable over time, the main fixed effect for treatment group was omitted. The target estimates consist of the decay rate for the placebo group (fixed effect for time) as well as the incidence rate ratios (IRR) for the metoprolol group (treatment-by-time interaction) to assess if the magnitude of the difference between treatment groups varies over time. The latter can be interpreted as ‘speed of efficacy’ [28], that is, whether the active agent may be distinguished from placebo by how quickly reduction in incidence of attacks was achieved. The same longitudinal model approach was applied for vertigo days serving as supplementary analysis.
Migraine headache days per 30 days were analysed with a negative binomial model (with offset term for the corresponding number of evaluated days during the 90-day assessment period) using self-reported symptoms documented within month 4 to 6 only. An analysis of covariance (ANCOVA) for absolute change from baseline in DHI mean total score at month 6 was performed which used a factor for treatment group and the baseline value as covariate.
For the binary response measures, change state from baseline in SVV and pursuit eye movement at month 6, a logistic regression analysis was conducted (1, from abnormal to normal; 0, otherwise).
Safety and tolerability
Adverse events (AEs) and tolerability were systematically assessed by evaluating reported AEs, physical examinations and concomitant medication use. The safety population included data from all randomized patients who received at least one dose of the investigational medicinal product during the double-blind treatment phase. Serious AEs (SAEs) were coded and summarized by the Medical Dictionary for Regulatory Activities (MedDRA) system organ class and preferred term. For some AEs, the exact starting date (day and or month) was partially or completely missing. In order to deal with this different input accuracy or partial date issues, AEs were classified with respect to their occurrence (AEs emerging while on treatment vs post-treatment AEs) assuming the AE was experienced at the earliest possible date.
Sample size considerations
A fixed sample size calculation was performed for the primary efficacy outcome (number of vertigo attacks). A sample size of 106 patients in each group will have 80% power to detect a probability of 0.389 that an observation XM is less than an observation XP using a Mann-Whitney test with a 5% two-sided significance level. The probability of P(XM < XP)=0.611 was calculated with a presumed normal distribution and difference in means of 1 and a standard deviation of 2.5 (nQuery Advisor 7.0). On the basis of our experience with patient compliance in previous studies and routine treatment, we assumed a drop-out rate of about 20%. Thus, the fixed target sample size was a total of 266 patients (133 in each treatment group) to be allocated. Further detailed descriptions on how the sample size was calculated are provided in the Additional file 2.
The study database was stored in SAS (Unix Version 9.2, SAS Institute, Cary, NC). Statistical analyses were performed using the statistical software package R version 3.5.1 [29]. For the efficacy analyses we used the R package lme4 (version lme4_1.1-18-1) to fit frequentist generalized linear mixed effects models [30, 31]. All statistical tests were 2-sided, with a significance level of 0.05.