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

ISCB2014_abstract_book

ISCB 2014 Vienna, Austria • Abstracts - Oral Presentations 49Tuesday, 26th August 2014 • 11:00-12:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Contributed sessions C25 Personalized and stratified medicine II C25.1 Mastering variation: variance components and personalised medicine S Senn1 1 CRP Santé, Strassen, Luxembourg It ought to be clear to any statistician that there are at least four potential sources of variation in clinical trials: the main effect of treatment, the main effects of patients, treatment-by-patient interaction and within-patient variation. It should also be obvious that identification of interactive effects requires replication at the level at which interaction is claimed. Hence treatment-by-patient interaction is only fully identifiable in multi-period cross-over trials, or, which amounts to the same thing, series of n-of-1 tri- als. The medical literature, however, pays scant attention to these realities and it is plausible that much of the belief that the personal component in response is important is based on a misunderstanding that apparent observed difference in outcome must reflect differences in the effective- ness of treatment. Ironically, there is a fifth important source of variation: difference in medi- cal practice that is nearly always overlooked. It will be argued here that the key to improving the treatment of patients is to master variation and that this involves the following elements. 1. Better communication of the problems by statisticians to their colleagues (some graphical approaches will be suggested). 2. Application of decision analysis to determine when personalisation is worth pursuing.. 3. Appropriate design for teasing out components of variation. 4. Application of random effect methodology for improving estimates. 5. Translating from additive to relevant scales. 6. Application of Deming´s ideas to understanding the system. 7. Realistic monitoring and feedback. Some suggestions for addressing these issues are given.   C25.2 Subgroup analyses: time to be specific about their goals J Tanniou1,2 , I van der Tweel1 , S Teerenstra2,3 , KCB Roes1,2 1 Julius Center, Department of Biostatistics, UMC, Utrecht, The Netherlands, 2 Medicines Evaluation Board, Utrecht, The Netherlands, 3 Department of Biostatistics, Radboud UMC, Nijmegen, The Netherlands The growing interest in personalised medicines and targeted therapies in- creases the attention for subgroup analysis, illustrated by the draft EMA’s “guideline on the investigation of subgroups in confirmatory clinical trials”. In light of the need to further develop our understanding and improve methodology and practice, a literature review of existing statistical and methodological methods on subgroup analysis was undertaken. At least five different objectives for subgroup analysis could be considered: 1. Confirm consistency of the treatment benefits across all subgroups, 2. Identify safety problems limited to one or few subgroups, 3. Identify subgroups with larger effect when the study reaches an over- all significant effect, 4. Check specific subgroups that apriori are suspected to show less or no treatment effect, and 5. Identify a statistically positive subgroup in case of an overall non-sig- nificant effect. Based on this classification, most papers we reviewed fall in more than one category, but most authors were not explicit about subgroup analyses objectives when presenting methodology. Moreover, limited attention was given to objectives 2, 3 and 4 illustrating that thinking about subgroup analyses in terms of distinctive objectives is not commonplace. Thus, research to improve statistical and methodologi- cal aspects depending on the subgroup’s objective is still clearly needed. Subgroup analysis methodology are too often undertaken without prior thoughts or knowledge about their goals, and are therefore inadequately incorporated into trial designs and methodologies. Adequate and efficient trials should be designed not only for the main analyses but also for subgroup analyses depending of their objective. C25.3 On the evaluation of predictive biomarkers with dichotomous endpoints: a comparison of the linear and the logistic probability models N Heßler1 , A Ziegler2 1 Inst of Med Biom & Stat, Lübeck University, Lübeck, Germany, 2 Inst of Med Biom & Stat & Center for Clinical Trials, Lübeck University, Lübeck, Germany   The standard statistical approach for analyzing dichotomous endpoints is the logistic regression model which has major statistical advantages. However, some researchers prefer the linear probability model over the logistic model in randomized trials for evaluating predictive biomarkers. The main reason seems to be the interpretation of effect estimates as ab- solute risk reductions which can be directly related to the number needed to treat. In the first part of our presentation, we provide a comprehensive comparison of the two different models for the investigation of treatment and biomarker effects. Using the logistic regression model, Kraft et al. (2007, Hum Hered) showed that the combined 2 degrees of freedom (2df) gene, gene-environment interaction test should be the test of choice for testing genetic effects. In the biomarker treatment setting a gene corresponds to the treatment and environment to biomarker. Using this analogy we extend the study of Kraft et al. in the second part of our presentation. We compare several test statistics including the 2df combination test using the linear probability model. The pros and cons of the combined test are discussed in detail. We demonstrate substantial power loss of the combination test in comparison with either the test for treatment or the test for treatment-biomarker interaction in many scenarios. Although the combination test has reasonable power in all situations considered, its power loss compared to a specialized 1df test can be large. Therefore, the combined test cannot be recommended as the standard approach in studies of treatment-biomarker interaction.   C25.4 Design dilemmas in the multi-drug, genetic-marker-directed, non-comparative, multi-centre, multi-arm phase II National Lung Matrix Trial L Billingham1,2 , L Crack1 , K Brock1 , S Popat3 , G Middleton4 1 CRUK Clinical Trials Unit, University of Birmingham, Birmingham, United Kingdom, 2 MRC Midland Hub for Trials Methodology Research, Birmingham, United Kingdom, 3 Royal Marsden Hospital and Imperial College, London, United Kingdom, 4 School of Cancer Sciences, University of Birmingham, Birmingham, United Kingdom   Stratified medicine aims to tailor treatment decisions to individual pa- tients, typically using molecular information to predict treatment benefit. The potential impact to benefit patients is considerable and recognised as strategically important. Cancer Research UK has made major investments into their Stratified Medicine Programme which provides a significant step

Pages Overview