Intended for healthcare professionals

CCBYNC Open access

Rapid response to:

Research

Efficacy and safety of betahistine treatment in patients with Meniere’s disease: primary results of a long term, multicentre, double blind, randomised, placebo controlled, dose defining trial (BEMED trial)

BMJ 2016; 352 doi: https://doi.org/10.1136/bmj.h6816 (Published 21 January 2016) Cite this as: BMJ 2016;352:h6816

Rapid Response:

Question about primary outcome definition in the BEMED trial: Author response

Dear Editor,

We thank Dr Doshi for his letter to our BMJ article [1] and would make the following points to clarify matters:

As stated in the trial registries, the protocol, the abstract or other parts of the main text, the pre-specified primary efficacy outcome measured for each study participant at each time interval was the absolute number (incidence) of Menière’s attacks per 30 days, hence a count outcome derived from the daily diary recordings. This means that the raw daily diary data (i.e., the documented vertigo symptoms evaluated according to prospectively defined rules) were summarized using absolute frequencies over the 30-day time intervals. Each patient’s diary was assessed for the whole study duration before unblinding.

As part of the principal outcome, the number of documented diary days per 30 days cannot be ignored in order to properly account for missing data concerning the patient’s vertigo status (see, e.g., the missingness map displayed in Figure S1 of the Web appendix 2). This implies the use of incidence rates (number of counts per day) for the time periods of interest. In this regard, we have to admit that the terminology we used could have been better.

The primary analysis approach was in line with the primary estimand including handling of missing data due to incomplete diary documentation or study treatment discontinuation.

Following the principle of intention to treat (i.e. all participants that satisfied the exclusion criteria were analysed as belonging to the treatment arms to which they were randomized, regardless of whether they received or adhered to the allocated intervention for the full duration of the trial), this MAR-based analysis was performed according to an "all observed data strategy" [2]. This likelihood based approach of all observed outcome data conducted such that it is valid under MAR is optimally statistically efficient, and therefore was considered to be a reasonable primary analysis for this trial [3, 4].

Principal model:
The primary efficacy analysis for the outcome variable "number of Menière’s attacks per 30 days" was performed using a negative binomial generalized linear mixed model (NB GLMM) with a log link function, random intercept and random slope associated with time (see methods section of the article). One of the underlying key parametric assumptions was a linear time trend. To account for incomplete diary documentation (incomplete vertigo profile) an offset variable for the log-transformed absolute number of evaluated days ("time at risk", "exposure time") per 30 days was defined. Therefore, the observation window was allowed to vary for each 30 day time unit of a patient. The merits of using a negative binomial model accounting for an explicit dispersion parameter as an alternative way of inflating the variance to allow for this variation from patient to patient are reported elsewhere [e.g. 5-7].

Since it is important to pre-specify the precise definition of the primary endpoint together with the method of statistical analyses that will be applied a statistical analysis plan (SAP) was prepared (unpublished document). The SAP was updated as a result of the blind review of the data and finalized before the blind was broken. There was complete transparency in relation to protocol changes in our SAP where we devoted a whole chapter to the subject and the chosen principal model, as the primary analysis strategy outlined in the protocol which was purely based on hypothesis testing instead of specifying a target parameter of interest was no longer appropriate.
Besides, corresponding sensitivity analyses to assess the robustness of the longitudinal, subject-specific modelling approach used to address the scientific question were prospectively planned in the SAP as well.

The parameters of primary interest as a basis for a comparison between treatment conditions were 1.) the rate ratios (RR) - LD vs PL and HD vs PL - together with 2.) the overall decay rate of attacks over time on placebo, estimated from the mixed effects regression model. The underlying model assumed a group-specific decline in the number of attacks per 30 days over the nine time intervals, as well as individual variations of the baseline level and the speed of change ("speed of efficacy") for the number of attack per 30 days for the LD and HD group, respectively.

Measure of the intervention effect and pre-specified target estimate:
The applied modelling approach can be used to derive the treatment effect within the assessment period month 7 to 9 (time period of primary interest) pre-specified in the trial protocol. The estimated population-averaged (marginal) mean number of attacks per 30 days at month 7 to 9 can be interpreted as the population-level summary measure (treatment effect measure) addressing the scientific question of primary interest targeting the last 3 months of the 9-month treatment period.

REFERENCES

[1] Adrion C, Fischer CS, Wagner J, Gurkov R, Mansmann U, Strupp M On behalf of the BEMED study group. Efficacy and safety of betahistine treatment in patients with Meniere's disease: primary results of a long term, multicentre, double blind, randomised, placebo controlled, dose defining trial (BEMED trial). BMJ 2016; 352: h6816.

[2] White IR, Carpenter J, Horton NJ. Including all individuals is not enough: lessons for intention-to-treat analysis. Clin Trials 2012; 9(4):396-407.

[3] National Research Council. The Prevention and Treatment of Missing Data in Clinical Trials. Panel on Handling Missing Data in Clinical Trials. Committee on National Statistics, Division of Behavioral and Social Sciences and Education. The National Academies Press, Washington, DC, 2010. URL www.nap.edu/catalog/12955.html.

[4] CHMP. Guideline on Missing Data in Confirmatory Clinical Trials. European Medicines Agency, Committee for Medicinal Products for Human Use (CHMP), London, UK, 2010. URL http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guidelin.... EMA/CPMP/EWP/1776/99 Rev. 1. Date for coming into effect: 1 January 2011.

[5] Keene ON, Calverley PM, Jones PW, Vestbo J, Anderson JA. Statistical analysis of exacerbation rates in COPD: TRISTAN and ISOLDE revisited. Eur Respir J 2008;32(1):17-24.

[6] Suissa S. Statistical treatment of exacerbations in therapeutic trials of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2006;173(8):842-6.

[7] Keene ON, Jones MR, Lane PW, Anderson J. Analysis of exacerbation rates in asthma and chronic obstructive pulmonary disease: example from the TRISTAN study. Pharm Stat 2007;6(2):89-97.

Competing interests: No competing interests

24 November 2017
Christine Adrion
biostatistician
Ulrich Mansmann, director and chair of biostatistics and bioinformatics
Institute for Medical Informatics, Biometry, and Epidemiology, University of Munich
Marchioninistr. 15, 81377 Munich, Germany