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

ISCB2014_abstract_book

44 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsTuesday, 26th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust within-family correlations for both the genetic and the environmental components. The heritability defined as the relative contribution of the genetic components to the total variance may also be estimated. No time- aspect is included in the liability-threshold model and thus the model does not account for censoring, which may lead to severe bias. In this study, we focus on extending the liability-threshold model for case- control family data to take censoring into account using weighting. We illustrate our work using Danish case-control family data on cancer.   C22 Surrogate and composite endpoints C22.1 A new audit strategy to detect possible bias in the evaluation of progression free survival J Link1 , U von Wangenheim1 , F Fleischer1 1 Boehringer Ingelheim Pharma GmbH & Co. KG, Biberach an der Riß, Germany In oncology trials, an endpoint of increasing use is progression-free sur- vival (PFS), i.e. the time until objective tumour progression or death. To mitigate potential evaluation bias, its assessment is often performed via a blinded independent central review (BICR) in addition to the local evalua- tion by the investigator. However, BICR is not only time consuming, costly and operationally demanding, it may also introduce informative censor- ing of PFS. Thus, specific audit methods have been developed to detect potential bias by only using a subset sample for BICR (Zhang et al 2013 [1]) If no bias is present in that subset one can omit further BICR evaluations. While ex- amining these methods several disadvantages become apparent. For ex- ample: One approach cannot be applied before study end, which causes practical problems like time delay. Additionally, the magnitude of the bias is not taken into consideration if only the treatment effect is large enough. To overcome these constraints, we developed a new strategy based on equivalence testing in a group-sequential approach whereby focusing on testing the potential difference between BICR and local assessments. This approach is able to detect the presence of bias on an ongoing basis during trial conduct, no matter how strong the treatment effect is. Its per- formance is analysed via simulations and an example using real study data shows that it can be used efficiently in practice. Reference: [1] Jenny J Zhang, Lijun Zhang, Huanyu Chen. Assessment of audit meth- odologies for bias evaluation of tumor progression in oncology clinical tri- als.Clinical Cancer Research, 19, 2637-2645, 2013 C22.2 A causal inference/mediation analysis based approach for assessing pseudo end-points applied to to ovarian cancer trials T Lange1 1 University of Copenhagen, Copenhagen, Denmark The use of pseudo end-points in clinical trials is desirable both from a fea- sibility and ethic point of view as replacing the final end-point by an earlier and/or easier to measure end-point, can assist in speeding up treatment development. Of course these benefits can only be harvest if a suitable pseudo end-point can be identified and validated. Until now the validation of a pseudo end-point has been quite cumber- some and the obtained measures of quality of the pseudo end-point not very intuitive, see eg. Eisenhauer (“Optimal assessment of response in ovarian cancer”, Annals of Oncology 22 (Supplement 8): viii49-viii51, 2011). In this work it is explored how to formulate the pseudo end-point problem in a counterfactual framework. In addition it is proposed to employ mediation analysis in the validation of a pseudo end-point since this will provide a direct parameterization of the degree to which the effect of treatment is fact captured in the pseudo end-point. The talk will include both theoretical aspects (which assumptions are made and how can we devise a counterfactual based framework) and practical advice on implementation. All results are applied to ovarian can- cer trials. C22.3 Bias assessment of surrogate threshold effects in simplified correlation based validation approaches C Schürmann1 , W Sieben1 1 Institute for Quality and Efficiency in Health Care (IQWiG), Köln, Germany A well established approach of surrogate endpoint validation is the cor- relation based meta-analysis as outlined in the seminal work of Buyse et al. (Biostat 2000). Surrogacy can be assumed if high values of individual and of study level correlation can be demonstrated. Alternatively, if a true endpoint is to be predicted from a surrogate endpoint, the surrogate threshold effect (STE, Burzykowski and Buyse (PharmStat 2006)) can be used. In practice, as individual patient data are hard to obtain, often only aggregated data are used and simplified analy- ses are performed. We are interested in how much simplified analyses are biased compared to the full model with individual patient data.To this end we conduct a simulation study with individual patient data and compute STEs with full and simplified analyses in various data situations (study siz- es, correlations, variances etc.) with respect to bias. Comparison of the results will help us decide to what extent STEs of the different approaches are suitable if an effect on a true endpoint is to be predicted. C22.4 Extension of win-ratio: analyzing a composite endpoint considering the clinical importance order among components S Witte1 , G Dong2 1 Novartis Pharma AG, Basel, Switzerland, 2 Novartis Pharmaceuticals, East Hanover, United States A composite endpoint consists of multiple endpoints combined in one outcome, and is frequently used as the primary endpoint in randomized clinical trials. During past decade, there are discussions on pros and cons of using a composite endpoint. The event rate in the composite endpoint is higher, therefore the sample size needed for a clinical trial may be re- duced, subsequently length of the study can be shortened and costs can be saved. In addition, with a composite endpoint as a single endpoint, multiplicity issue and competing risk problem may be avoided. However, in the conventional analyses, all components are treated equally important; and in time-to-event analysis, the first event considered may not be the most important component. Recently Pocock [1] published the win ratio method to address these disadvantages. In this new method, they proposed two approaches: matched pair and unmatched pair. In the unmatched pair approach, the confidence interval is constructed based on bootstrap re-sampling, and the hypothesis testing is based on the generalized Wilcoxon test. We extend the unmatched pair approach of Pocock’s win-ratio method to perform hypothesis testing and construct the confidence interval for win ratio based on its asymptotic distribution. This asymptotic distribution is derived via U-statistics following Wei [2]. We illustrate our method with an example from a liver transplant study,

Pages Overview