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

ISCB2014_abstract_book

12 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Contributed sessions C01 Randomized clinical trials C01.1 Analysis of clinical trials requiring rescue medication GK Rosenkranz1 1 Novartis Pharma AG, Basel, Switzerland Clinical trials in some indications like asthma or diabetes require to ad- minister rescue medication in case a patient does not sufficiently respond to investigational treatment. The application of additional treatment on an as needed basis causes problems to the analysis and interpretation of the results of these studies since the effect of the drug under study can be confounded by the additional medication. Following-up all patients until study end and capturing all data are not fully addressing the issue. The paper introduces an analysis that takes care of the fact that rescue is a study outcome and not a covariate by considering potential outcomes. This approach allows to clearly define a causal estimand. For normally dis- tributed longitudinal data a practically unbiased causal effect estimator for the randomized treatment can be obtained. The results are compared to those from an ITT analysis and an analysis on all patients not receiving rescue.   C01.2 Assessment of chronological bias in randomized clinical trials M Tamm1 1 RWTH Aachen University, Aachen, Germany In clinical trials patients are usually recruited sequentially over time. Often the recruitment takes place over several years. Especially in clinical trials studying rare diseases, low accrual rates and thus long recruitment phases are common. As a result of the prolonged recruitment time, time trends are suspected to occur.This can be due to several reasons, such as changes in the recruitment policy or learning effects in the application of the new methods. If treatment effects are confounded with time trends, this can result in the so called chronological bias. Even in randomized clinical trials, time trends may impact the results, for instance if a long series of consecutive patients are assigned to the same treatment. To account for this, the ICH E9 guideline recommends the use of randomization in blocks. However, one major drawback of permuted block randomization with short blocks is the increase in the risk of selec- tion bias. Using different time trend models, we evaluate and compare the extent of chronological bias under the random allocation rule and the permuted block randomization with different block sizes. We present theoretical re- sults regarding the extent of bias in the treatment effect estimate under different time trends. Further, the bias in statistical hypothesis testing is discussed considering the empirical type I error rate in simulations. An as- sessment of worst-case-scenarios as well as an overall assessment consid- ering the choice of randomization is given. Acknowledgement: This research is part of the IDeAl EU-FP7 project, Grant-Agreement No. 602 552. C01.3 An analytic framework for randomised trial designs that take patient preferences into account S Walter1 1 McMaster University, Hamilton, Canada Patients in clinical trials often prefer one of the treatments under investi- gation, and such preferences (which are usually unmeasured) may have affect study outcomes. Several trial designs have been proposed to inves- tigate the impacts of patient preferences, something that is not possible in the usual parallel group design. We will describe a model framework to represent the effects of treatment preferences on study outcomes. In particular, we consider the selection effect, which measures how expected outcomes differ between partici- pants who would select one treatment or the other, if they were free to do so. Additionally we can investigate the difference in outcomes for par- ticipants who do or do not receive their preferred treatment, which we designate as a preference effect. We will review several alternative trial designs including: (1) the two-stage design, in which some randomly sampled participants are allowed to choose their treatment; (2) the fully randomised preference design, where preferences are known for all participants, but treatments are nevertheless always randomised; and (3) the partially randomised preference design, where only those participants who are indifferent between treatments are randomised. Based on the model framework, we then determine which effects are estimable with each design. Examples of each of these designs can be found in the medical literature. Preference designs offer potentially greater insight than the conventional parallel group design, by informing investigators about potentially impor- tant effects of preferences on patient outcomes, effects which may some- times exceed the direct effect of treatment itself. C01.4 Covariate adjustment has similar benefits in small and large randomised controlled trials DD Thompson1 , HF Lingsma2 , EW Steyerberg2 1 University of Edinburgh, Edinburgh, United Kingdom, 2 Erasmus Medical Center, Rotterdam, The Netherlands Aim: Covariate adjustment is a standard statistical approach in the analysis of randomised controlled trials, but the benefit in small versus larger stud- ies is not well known. Specifically, chance imbalance in prognostic factors is greater in small trials than in large. We aimed to determine whether the benefit of covariate adjustment on statistical significance and power dif- fered between small and large trials. Methods: We repeatedly (500,000 times) drew random samples of sizes 300 and 5000 per arm from a large trial (GUSTO-I, N=30,510) and simulated 30-day mortality using the control arm. We empirically determined the treatment effects required to fix power at 80% for the unadjusted analyses and calculated the joint probabilities in the discordant cells (i.e., p<0.05 vs. p≥0.05) when cross-classifying adjusted and unadjusted results from logistic regression models. Results: The power gained from an adjusted analysis for small (OR=0.27 for simulated treatment effect) and large (OR=0.77) samples was approxi- mately 5% (from 80% to 85%). Similar discordance irrespective of sample size was noted, with a 2% change from significant unadjusted to non-sig- nificant adjusted treatment effect and a 7% change from non-significant unadjusted to significant adjusted. The Type I error was close to 5% with unadjusted and adjusted analyses, with discordance balanced at 2% for small and large trials. Conclusions: The proportions of change in statistical significance from covariate adjustment were the same for small and large trials with simi- lar gains in statistical power to detect treatment effects. Covariate adjust- ment is hence equally recommendable in small and large trials.

Pages Overview