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

ISCB2014_abstract_book

ISCB 2014 Vienna, Austria • Abstracts - Oral Presentations 31Monday, 25th August 2014 • 16:00-17:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Contributed sessions C13 Personalized and stratified medicine I C13.1 Interaction of treatment with a continuous variable: simulation study of significance level and power for several methods of analysis W Sauerbrei1 , P Royston2 1 University Medical Center Freiburg, Freiburg, Germany, 2 University College London, MRC Clinical Trials Unit, London, United Kingdom Interactions between treatments and covariates in RCTs are a key topic. Standard methods for modeling treatment-covariate interactions with continuous covariates are categorization or linear functions. Spline based methods and multivariable fractional polynomial interac- tions (MFPI) have been proposed as an alternative which uses full infor- mation of the data. Four variants of MFPI, allowing varying flexibility in functional form, were suggested. In order to work toward guidance strategies in the sprit of the STRATOS initiative we have conducted a large simulation study to investigate sig- nificance level and power of the MFPI approaches, versions based on cat- egorization and on cubic regression splines. We believe that the results provide sufficient evidence to recommend MFPI as a suitable approach to investigate interactions of treatment with a continuous variable. If subject- matter knowledge gives good arguments for a non-monotone treatment effect function, we propose to use a second-degree fractional polynomial (FP2) approach, but otherwise a first-degree fractional polynomial (FP1) function with added flexibility (FLEX3) has a power advantage and there- fore is the method of choice.The FP1 class includes the linear function and the selected functions are simple, understandable and transferable.   C13.2 Comparing a marker based stratified treatment strategy with the standard treatment in a randomized clinical trial H Sun1 , F Bretz2 , O Gerke3 , W Vach1 1 Department of Medical Biometry and Statistics, Freiburg, Germany, 2 Novartis Pharma AG, Basel, Switzerland, 3 Nuclear Medicine, Odense University Hospital, Odense, Denmark The increasing emergence of successful molecularly targeted agents (MTAs) raises the question of how to study treatment strategies suggest- ing a variety of different (combination) therapies based on multiple mark- er information. Here we consider the situation where there already exists a stratified treatment strategy being dependent on a marker pattern, and dividing the whole population into small subpopulations. This strategy has to be compared with the standard treatment in an randomized clinical trial. Due to limited knowledge about the MTAs, we expect no benefit from the new strategy in some subpopulations. In such situation the objective should be trying to demonstrate a treatment effect for a subset of subpopulations, instead of each single subpopulation. We consider a wide class of methods to approach this situation, allowing to select from a variety of significant subsets an optimal one according to different criteria. We present a framework to compare various methods of this class, aiming on measuring not only power (i.e. the probability to find at least one significant subset) but also the actual gain in average im- provement of the outcome. Using the framework, we can observe sub- stantial differences between the methods, which allows to give first rec- ommendations on the choice of adequate methods. C13.3 A framework for comparing methods for marker- based selection of treatment change M Kechel1 , W Vach1 1 Center for Medical Biometry and Medical Informatics, Freiburg, Germany Today, biomarkers often promise to assist in choosing between two dif- ferent therapeutic alternatives. Clinical trials randomizing all patients to the two alternatives and measuring the biomarker in all patients allow to check such a promise by establishing a (qualitative) interaction. Moreover, they allow to determine a cut point or a more complex decision rule to de- cide on the treatment for each patient. Many statistical approaches have been made to determine such cut points. In comparing such methods, of- ten only the power of the methods, i.e. the probability to come to a (differ- ential) decision rule, is considered. In our talk, we develop a more general framework to compare such methods. The framework is suitable for RCTs as described above, when the choice is to be made between a standard therapy and a new therapy, e.g. an add on taking into account the infor- mation given by the biomarker. In this framework we take also the size of the treatment effect in each patient into account, and we define quanti- ties like the expected overall gain, the gain in shifters, the unused gain in stayers, etc. We apply the framework to compare different approaches based on linear and quadratic models for the marker dependent treat- ment effect and the use of simultaneous and pointwise confidence bands. We demonstrate that the framework is useful to obtain more insights than just considering the power.   C13.4 Analyzing treatment-by-subgroup interactions in time-to-event data - comparison of two multivariate approaches H Sommer1 , A-S Stoehlker1 1 University of Freiburg, Freiburg, Germany For determining the differential effects of therapy dependent on indi- vidual patient characteristics, most of the common methods investigate treatment-by-covariate interactions only for one single predictor. We in- troduce and compare two approaches to examine the influence of several covariates simultaneously in context of survival data. Firstly, we focus on a tree-based subgroup identification procedure called Interaction Tree proposed by Su et al. (Applied Statistics, 2011). Perfectly homogeneous subgroup-allocation is assumed implicitly but does not hold naturally. By additionally incorporating specific weights in a locally weighted Cox-regression according to a weighting scheme of Simon (StatMed, 2002) we try to mitigate this problem with a second procedure based on the subgroups from the former. For comparison, this second procedure is also applied to subgroups resulting out of another method. Treatment selection is performed via Selection Impact curves proposed by Song and Pepe (Biometrics, 2004), which we apply on the Interaction Tree subgroups. For this purpose, sufficient heterogeneity of the treatment ef- fect between these subgroups is necessary, which is tested with Cochran´s Q commonly used in meta-analyses. Finally, the approaches are illustrated with data of a randomized clinical trial and we investigated comprehensive simulation study covering vari- ous realistic scenarios to examine strengths and weaknesses.

Pages Overview