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

ISCB2014_abstract_book

ISCB 2014 Vienna, Austria • Abstracts - Oral Presentations 45Tuesday, 26th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust which shows that the confidence interval based on our derivation can be narrower than the bootstrapping confidence interval. References: 1. Pocock SJ et al. Eur Heart J. 2012;33:176-182 2. Wei LJ and Johnson W.E. Biometrika,1985;27:359-364 C22.5 Weighted comparisons of composite endpoints M Wolbers1,2 , A Nguyen Duc1 1 Oxford University Clinical Research Unit, Ho Chi Minh City, Viet Nam, 2 Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom Composite endpoints are widely used as primary endpoints of random- ized controlled trials (RCTs) in many clinical disciplines. For example, a systematic review of published RCTs in cardiovascular medicine and sur- gery showed that 37% of them reported a composite endpoint with a me- dian of 3 individual component outcomes. One limitation of composite endpoints is that they frequently pool component outcomes of varying clinical importance. Thus, several authors have suggested replacing or complementing the standard analysis of composite endpoints which just analyzes whether a subject experiences any component outcome or not by an analysis which weights each component with respect to its clinical importance or associated cost. We suggest interpretable test statistics based on weighted linear combi- nations of absolute risk differences of component outcomes for between- group comparisons of both multinomial and time-to-event composite endpoints. Considerations for choosing component weights are reviewed and it is shown that there is often a conflict between choosing weights that lead to powerful tests and weights that are clinically relevant. One problem of weighted comparisons is that elucidation of quantitative component weights is difficult in practice. However, it is often possible to rank components according to their relative importance. To address this, we introduce methods which control the family-wise error rate across all non-negative weights or across all sets of weights satisfying an order con- straint, respectively.   C23 Design and analysis of clustered studies C23.1 Methods for observed-cluster inference when cluster size is informative SR Seaman1 , M Pavlou2 , AJ Copas3 1 MRC Biostatistics Unit, Cambridge, United Kingdom, 2 University College London, London, United Kingdom, 3 MRC Clinical Trials Unit at University College London, London, United Kingdom Clustered data commonly arise in epidemiology. We assume each clus- ter member has an outcome Y and covariates X. When there are missing data in Y, the distribution of Y given X in all cluster members (`complete clusters´) may be different from the distribution just in members with ob- served Y (`observed clusters´). Often the former is of interest, but when data are missing because in a fundamental sense Y does not exist (e.g. quality of life for a person who has died), the latter may be more meaningful (quality of life conditional on being alive). Weighted and doubly weighted generalised estimating equations and shared random-effects models have been proposed for observed-cluster inference when cluster size is informative, i.e. the distri- bution of Y given X in observed clusters depends on observed cluster size. We show these methods can be seen as actually giving inference for complete clusters and may not also give observed-cluster inference. This is true even if observed clusters are complete in themselves rather than being the observed part of larger complete clusters: here methods may describe imaginary complete clusters rather than the observed clusters. We show under which conditions shared random-effects models pro- posed for observed-cluster inference do actually describe members with observed Y. A psoriatic arthritis dataset is used to illustrate the danger of misinterpreting estimates from shared random-effects models. C23.2 Generalised estimating equation methods for analysing continuous outcomes when cluster size is informative LN Yelland1,2 , TR Sullivan2 , JB Carlin3,4 1 Women’s and Children’s Health Research Institute, North Adelaide, Australia, 2 The University of Adelaide, Adelaide, Australia, 3 Murdoch Children’s Research Institute, Parkville, Australia, 4 University of Melbourne, Melbourne, Australia Generalised estimating equations (GEEs) are a popular method for analys- ing clustered data but parameter estimates may be biased when cluster size is related to the outcome. This type of informative clustering is a com- mon problem in perinatal trials when infants from both single and mul- tiple births are included, since infants from multiple births tend to have worse health outcomes. A cluster weighted GEE approach has been proposed for handling infor- mative cluster size, which estimates parameters with a cluster-level inter- pretation. Alternative methods of analysis are also available, including individually weighted GEEs to estimate individual-level parameters and GEEs with adjustment for cluster size, but these have received limited at- tention. In this presentation, I will report the results of a study comparing these three approaches for analysing clustered continuous outcomes in terms of their theoretical properties, interpretation and finite sample per- formance. I will show why these methods often produce different unadjusted results and demonstrate that adjusting for cluster size does not always solve the problem of informative cluster size. The relative merits of choosing a cluster-level or an individual-level ap- proach in the informative cluster size setting will be discussed and recom- mendations will be made for dealing with informative cluster size in the context of perinatal trials with multiple births. C23.3 Choosing covariates and the effects of covariate adjustment in the analysis of CRTs N Wright1 1 Blizard Institute, Queen Mary University of London, London, United Kingdom Research on the effects of covariate adjustment in the analysis of ran- domised trials has mainly focused on trials in which individuals are ran- domised to treatment arms. This has led to published guidance on choos- ing covariates in the analysis of randomised trials. In cluster randomised trials (CRTs) pre-existing groups (clusters) of indi- viduals are randomised to treatment arms. A valid analysis of a CRT must take into account the additional data structure imposed by cluster ran- domisation, for example by using a mixed effects model. We can adjust for covariates in these models, just as in fixed effects models, by includ- ing variables and parameters for covariates in the linear predictor. There is limited published research on the effects of covariate adjustment in the analysis of CRTs, especially in the analysis of binary outcome variables. We firstly review the published guidance for choosing covariates in ran- domised trials, in the context of analysing CRTs. We then present a selection of results from simulation studies on the ef- fects of covariate adjustment in the analysis of CRTs. Simulations included

Pages Overview