Please activate JavaScript!
Please install Adobe Flash Player, click here for download


46 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsTuesday, 26th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust comparisons of the use of individual and cluster level covariates, cluster aggregation of covariate data, and the use of separate individual and clus- ter level covariate effect parameters. To conclude this work, we suggest some additional guidance on choosing covariates that is specific to the analysis of CRTs. C23.4 Sample size and analysis considerations for cluster randomised crossover trials with unbalanced cluster sizes and binary data AB Forbes1 , M Akram1 , R Bellomo1,2 1 Monash University, Melbourne, Australia, 2 Australian and New Zealand Intensive Care Society, Melbourne, Australia Cluster randomised crossover trials are a class of multiple-period cluster designs that have been increasingly used in clinical and public health re- search. These trials gain efficiency by incorporating treatment crossover across observation periods within each cluster. However, the development and assessment of these designs to date has been fairly limited. In this presentation we report on our recent design and analysis work: We present expressions for the variance of treatment effect estimators which take into account period effects, within- and between-period intra- cluster correlations, as well as unbalanced within-and between-period cluster sizes. Using these expressions, we present sample size formulae for unbalanced cluster sizes as would typically occur in practice. We then discuss extensions for use with binary outcomes. Using an underlying marginal model for binary outcomes, we present results of a simulation exercise with varying numbers of clusters, cluster sizes and outcome prev- alences, and discuss problematic parameter configurations. We illustrate all methods with an application involving a proposed large cluster randomised crossover trial to evaluate interventions to reduce mortality in the intensive care research setting. We also discuss the po- tential for extension to multiple-period-multiple-treatment designs and conditions for their feasibility in particular research settings. C23.5 A multidisciplinary approach to benefits and drawbacks of the stepped wedge cluster randomized design E de Hoop1 , H Koffijberg1 , R van der Graaf1 , JB Reitsma1 , I van der Tweel1 1 University Medical Center Utrecht, Utrecht, The Netherlands Stepped wedge cluster randomized designs are increasingly being used over the last couple of years. Some reviews have listed reasons for the use of this design. Furthermore, benefits and drawbacks have been men- tioned in several papers. However, there is extensive debate whether this design is useful or not. Uncertainties concern the usefulness of this design for different types of research questions, but also various design issues such as risk of bias, ef- ficiency, ethical issues, data analysis and economic evaluations are topics of debate. Although these issues have been mentioned before, it remains unclear whether and to what extent these are in favour or against the use of the stepped wedge design. Therefore, a comprehensive overview of characteristics, benefits and drawbacks of the stepped wedge design is required. We took a multidisciplinary approach where input from statisticians, methodologists, ethicists and health economists led to such an overview. For this overview we compared the stepped wedge design to the paral- lel group clustered trial design. We will present our findings and explain which aspects are unique to the stepped wedge design and what its ad- vantages and disadvantages are compared to a parallel group cluster ran- domized trial. Besides, we will illustrate our findings with a trial that used the stepped wedge design. This will help researchers and research ethics committees to decide on the appropriateness of a stepped wedge design in a cluster randomized trial for their research question.   C24 Group-sequential designs C24.1 Student Conference Award Group sequential monitoring of response- adaptive randomised clinical trials with censored survival data HY Liu1 1 Queen Mary University of London, London, United Kingdom Methods for combining group sequential tests with a response-adaptive randomisation design in clinical trials with immediate responses have been studied. However, application of this combined approach to survival responses with inherent right-censoring has not yet been investigated. Such an approach that does not require the number of interim analyses to be pre-specified is applied to censored survival times. It utilises an error spending function to compute the critical values at any point during the course of the trial. The approach is based on the canonical joint distribution of the sequence of test statistics, which generalises to group sequential response- adaptive randomisation designs. Since the design is adaptive, both the sample size and the treatment allocation proportions are random at interim analyses. In this paper, the mean of the treatment allocation proportions and the corresponding standard deviations for two response-adaptive randomi- sation designs are compared with complete randomisation under group sequential monitoring. Simulation results show that the combined ap- proach can not only reduce the total number of patients, but also increase the power compared with that of a group sequential non-adaptive ran- domised design. Moreover, more patients are assigned to the more promising treatment using the response-adaptive randomisation designs. But the difference in the means and standard deviations of the treatment allocation propor- tions between the two response-adaptive designs is generally less than 1%. In conclusion, the combined approach can achieve higher power and have ethical benefits. C24.2 Group-sequential designs for cross-over trials MJ Grayling1 , J Wason1 , A Mander1 1 MRC Biostatistics Unit, Cambridge, United Kingdom Group-sequential procedures have assisted greatly in reducing the ex- pected sample size of parallel clinical trial designs, which remain the con- ventional means by which to estimate the treatment effect of an experi- mental drug in many settings. However, particularly within the context of treatments for chronic diseases, the cross-over trial remains the design of choice; exploiting the opportunity to treat patients with multiple ex- perimental drugs to substantially reduce the variance of the estimated treatment effects. Theoretically, group-sequential approaches to cross- over trials promise to bring the same advantages as in a parallel setting. Here, we discuss our work to date on establishing a framework for such designs. By determining the joint distribution of the test statistics, within a linear mixed model setting for data analysis, optimal designs in-terms of minimising the expected sample size or expected number of observa- tions, subject to type-I and type-II error constraints, can easily be deter- mined using a simple search over possible sample sizes at each stage of

Pages Overview