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


36 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 16:00-17:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust is involved in disease processes, including cancer. The purpose of this study was to compare different modeling strategies for methylation data in terms of model performance and performance of downstream hypothesis tests. Specifically, we used the generalized addi- tive models for location, scale and shape (GAMLSS) framework to compare beta regression with Gaussian regression on raw, logit2 and arcsine square root transformed methylation data, with and without modeling a covari- ate effect on the scale parameter. Using simulated and real data, we show that model performance is im- proved in models of location and scale, specifically on logit2-transformed methylation values, as compared to traditional models of location only. Our results further suggest that models of location and scale are specifi- cally sensitive towards violations of the distribution assumption and to- wards outliers in the methylation data. Therefore, a resampling procedure is proposed as a mode of inference and shown to diminish type I error rate in practically relevant settings. We apply the proposed method to ge- nome-wide data from the large population-based KORA study and reveal biologically relevant phenotypic associations with methylation variability.   C17 Adaptive designs I C17.1 Dose-escalation using safety and biomarker data: a Bayesian adaptive approach D Sabanes Bove1 , G Palermo1 1 F. Hoffman-La Roche Ltd., Basel, Switzerland In early clinical dose-escalation studies typically the target is to find a dose with a certain toxicity probability, say between 20 and 35%. Therefore, the dose-escalation is only driven by safety data, ignoring potential biomark- ers for efficacy. This strategy relies on the assumptions that the efficacy increases monotonically with the dose, and that such levels of toxicity can actually be reached. However, for targeted monoclonal antibody thera- pies it is often the case that no dose-limiting toxicity is observed, such that dose selection cannot solely rely on safety, but must take into account pharmacodynamics (PD) data. Therefore we propose a Bayesian adaptive dose escalation framework that also uses a continuous biomarker to find the dose with maximum PD effect within certain safety constraint. Our ap- proach builds on the work by Bekele and Shen (Biometrics, 2005), which uses the probit model to transform the binary safety outcome into a con- tinuous variable, allowing to model safety and biomarker data by a bivari- ate normal distribution. We compare our approach with alternative dual endpoint designs, and illustrate the performance with simulation results.   C17.2 Bayesian adaptive dose-escalation procedures utilizing a gain function with binary and continuous responses WY Yeung1 , T Jaki1 , J Whitehead1 , B Reigner2 , U Beyer2 , C Diack2 1 Lancaster University, Lancaster, United Kingdom, 2 Hoffmann-La Roche Ltd., Basel, Switzerland One of the main aims in early phase clinical trials is to identify a relatively safe dose with an indication of therapeutic benefit to administer to pa- tients in further studies. Therefore, dose-limiting events (DLEs) and effica- cy responses of subjects should be included in the dose-escalation proce- dure. Several methodologies have been suggested for incorporating both DLEs and efficacy responses in escalation trials in oncology. In the presentation, we describe and evaluate a dose-escalation proce- dure for use in non-oncology trials that utilizes measures of efficacy and safety. This is a Bayesian adaptive approach based on one binary response (occurrence of a DLE) and one continuous response (a measure of poten- tial efficacy) per subject. A logistic regression and a linear log-log relation- ship are used respectively to model the binary DLEs and the continuous efficacy responses. A gain function concerning both the DLEs and efficacy responses is used to determine the dose to administer to the next cohort of subjects. Stopping rules are proposed to enable efficient decision mak- ing. Simulation results shows that our approach performs better than one that takes account of DLEs responses only. To assess the robustness of the approach, scenarios where the efficacy responses of subjects are generat- ed from an Emax model, but treated as coming from a linear log-log model are also considered. This evaluation shows that the simpler log-log model leads to robust recommendations even under model misspecification. C17.3 Adaptive dose-finding designs to identify multiple doses that achieve multiple response targets S Bond1,2 , A Mander2 , J Todd3 , L Wicker3 , F Waldron-Lynch3 1 Cambridge Clinical Trials Unit, Cambridge, United Kingdom, 2 MRC Biostatistics Unit Hub for Trials Methodology Research, Cambridge, United Kingdom, 3 JDRF/Wellcome Trust Diabetes and Inflammation Laboratory, Cambridge, United Kingdom The objective of the “Adaptive study of IL-2 dose on regulatory T cells in type 1 diabetes (DILT1D)” (NCT 01827735) was to identify doses of inter- leukin-2 that achieve targeted increases in the T regulatory cell popula- tion in recently diagnosed type 1 diabetes participants. DILT1D aimed to identify a minimally and a maximal effective dose in a limited number of participants (40) that may be repeatedly administered in future late phase trials. The dose was administered subcutaneously so can be chosen from a con- tinuous range up to a maximum determined by tolerability. The design has an initial learning phase where pairs of patients were assigned to five pre-assigned doses. The next phase was fully sequential with an interim analysis after each patient to determine the choice of dose based on the optimality criterion to minimise the determinant of the covariance of the estimated target doses.The dose-choice algorithm assumes that a specific parametric dose-response model is the true relationship, and so a variety of models were considered at the interims and dose determining commit- tee approved all treatment decisions. The estimated dose-response curves and the estimated target doses from the final study data are presented. We consider the statistical lessons learnt during the simulations performed pre-trial and practical lessons learnt whilst conducting the trial and assigning doses. C17.4 Bayesian adaptive designs for biomarker trials with biomarker discovery J Wason1 1 MRC Biostatistics Unit, Cambridge, United Kingdom Response to treatment is highly heterogeneous in many diseases and, in particular cancer. Increased availability of biomarkers and targeted treat- ments has led to the urgent need for new trial designs to efficiently test treatments in patient subgroups. In this presentation I propose a novel Bayesian adaptive randomisation (BAR) design for use in multi-arm phase II trials where there are biomarkers that are thought to be predictive for the effect of different treatments. This design is motivated by a phase II neo-adjuvant breast cancer trial. The proposed design starts by using pre-specified‘pairings’of linked biomarkers and experimental treatments, with patients randomized to the control treatment or to experimental treatments that are paired with biomarkers they are positive for. At interim analyses, the results of patients assessed so far are used to update the al- location probabilities. If the linked treatments are truly effective, the allo-

Pages Overview