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

ISCB2014_abstract_book

22 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 14:00-15:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Monday, 25th August 2014 – 14:00-15:30 Contributed sessions C07 Clinical trials C07.1 Dealing with challenges in design and analysis of clinical trials in long-term care L Thabane1 1 McMaster University, Hamilton, Canada Conducting trials in long-term care (LTC) can present many implementa- tion, methodological, ethical and analytical challenges. First, the popula- tion is quite frail which can present serious methodological and ethical challenges; second, the environment is hard to work in because of the overcommitted staff and this can create major implementation problems; third, outcome assessment can be challenging; fourth, determining the appropriate unit of randomization, analysis or inference can be compli- cated by several practical factors. In this presentation, I will briefly describe some of the key challenges and use · Our experience from a scoping review of trials of hip-protectors in LTC residents; and · the ViDOS (Vitamin D and Osteoporosis) trial—a cluster randomized con- trolled trial of 40 LTC homes in Ontario—designed to determine the feasi- bility and effectiveness of a multi-faceted knowledge translation interven- tion aimed at improving vitamin D supplementation and other evidence- based osteoporosis/fracture prevention strategies in LTC, to illustrate the issues and to suggest potential solutions. C07.2 Estimating efficacy and effectiveness using data retrieved after treatment non-compliance A-K Leuchs1 1 BfArM, Bonn, Germany Treatment non-compliance (e.g. treatment discontinuation, switch or aug- mentation) and missing data are common issues in clinical trials. In the presence of non-compliance, differentiating between the true benefit of the treatment (efficacy), the effect if the medication is taken as observed (effectiveness), and different estimands for both is essential. In addition to data on non-compliance itself, the retrieval of patients ir- respective of their protocol adherence provides useful information on ef- fectiveness. In longitudinal neuropsychiatric disease trials, in which missing data and non-compliance are frequent, mixed models for repeated measurements (MMRM) are favoured methods to estimate treatment effects. However, depending on the data included (inclusive/exclusive of retrieved data), MMRMs address either efficacy or effectiveness. In various different set- tings, we oppose and compare these standard methods to piecewise linear mixed effects models also including retrieved data regarding their ability to appropriately estimate efficacy and effectiveness, primarily fo- cusing on type 1 errors. Different assumptions on the trend while comply- ing and non-complying are made. It becomes evident that retrieving and including data after protocol vio- lation is crucial to obtain unbiased estimates of effectiveness. A simple MMRM including all data may be severely biased and hence inappropriate, especially if data are only partly retrieved after non-compliance. In some settings, however, MMRMs may still be adequate to estimate efficacy. In conclusion, it is paramount to precisely define the estimation’s objec- tive and the corresponding estimand, to choose an appropriate analysis method addressing the trial’s target estimand and, if effectiveness is of interest, to make any effort to retrieve data. C07.3 Including historical data in the analysis of clinical trials using the modified power prior: practical and theoretical issues J van Rosmalen1 , D Dejardin2 , E Lesaffre1,2 1 Department of Biostatistics, Erasmus MC, Rotterdam, The Netherlands, 2 L-Biostat, KU Leuven, Leuven, Belgium Including historical data in the analysis of clinical trials may improve the precision of the results and reduce the required sample size. In the Bayesian context, Ibrahim and Chen (2000) proposed the power prior to combine historical and current data, where the likelihood of the historical data is raised to a power α (with 0≤α≤1). Using the modified power prior (MPP; see Duan et al. 2006), α is estimated in an adaptive way: if the cur- rent data and historical data are commensurate, α will be high, whereas in case of substantial discrepancy, α will be close to 0, effectively discarding the historical data. We give a methodological overview of the MPP and discuss the relation- ships with other methods for including historical data. Also, we show how the MPP can be used to account for the common situation where the historical data differ substantially from the current data with respect to nuisance parameters (e.g. the baseline hazard in survival models). An im- portant practical issue is the calculation of the normalizing constant of the MPP, which cannot be done using standard MCMC samplers. We propose new algorithms for computing the posterior results, based on Laplace ap- proximation and path sampling (Friel and Pettitt 2008). We illustrate the usefulness of the MPP using data from two randomized controlled trials for progression-free survival in patients with metastatic breast cancer. We find that the MPP is a promising method for incorporat- ing historical data in clinical trials. C07.4 A robust Bayesian meta-analytic-predictive approach to borrow strength from historical information in thorough QT studies K Meiser1 , H Schmidli1 1 Novartis Pharma AG, Basel, Switzerland Thorough QT (tQT) trials are key studies to evaluate the cardiac safety of new test drugs. The primary endpoint is the QT interval measured by elec- trocardiography (ECG). A prolongation of the QT interval is associated with serious cardiac events (ICH E14 guidance, 2005). In tQT studies, the test drug is compared to placebo and an active control known to prolong the QT interval. As these studies are routinely done, many historical studies with healthy subjects are available, providing information on the active control. Borrowing strength from this historical information to reduce the number of patients randomized to the active control would be desirable from an ethical and efficiency perspective. We propose a robust Bayesian meta-analytic-predictive approach to de- rive an informative prior on the active control in a new tQT study from the historical trials. This approach essentially assumes exchangeability of the active control parameter in the historical and the new trial. However, the possibility that the active control parameter is systematically differ- ent from the historical trials is taken into account by adding a weakly in- formative mixture component to the meta-analytic-predictive prior. This provides robustness of the approach to prior-data conflicts. A tQT study will be used to illustrate the methodology.  

Pages Overview