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


98 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsMonday, 25th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust In small sample settings, the two-sided test size of all methods was higher than the nominal significance level. For all methods, heterogeneous clus- tering led to small sample bias and asymmetry of the one-sided test-sizes in some scenarios. Conclusion: The random coefficient model is not suitable for the analysis of cluster randomised trials with binary outcomes. Care is need interpret- ing other analysis methods where there is heterogeneous clustering for binary data. P1.2.121 Treatment crossovers in time-to-event non-inferiority randomized trials of radiotherapy in subjects with breast cancer S Parpia1 , JA Julian1 , L Thabane2,3 , C Gu1 , TJ Whelan1,4 , MN Levine1,4 1 Ontario Clinical Oncology Group, McMaster University, Hamilton, Canada, 2 Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada, 3 Centre for Evaluation of Medicine, Hamilton, Canada, 4 Juravinski Cancer Centre, Hamilton, Canada   The non-inferiority randomized trial design is commonly used to compare novel experimental breast radiation regimens to standard breast irradia- tion for the prevention of local recurrence in patients with breast cancer who have undergone breast conserving surgery. Prior to beginning ra- diation therapy (RT) the patient will undergo a planning process to estab- lish the treatment fields to target the tumour and avoid normal tissues. Planning generally occurs after randomization. Sometimes it is not pos- sible to deliver the experimental regimen and the patient will cross over to the standard RT. In addition, sometimes the patient decides to receive the usual RT. Although the intention-to-treat (ITT) analysis is the preferred approach for superiority trials, its role in non-inferiority trials is still under debate. The ITT is generally perceived to produce a diluted treatment ef- fect and, therefore, is anti-conservative in demonstrating non-inferiority. This has led to the use of alternative approaches such as the per-protocol (PP) analysis or the as-treated (AS) analysis, despite the inherent biases of such approaches. Using simulation, we investigate the effect of random and non-random crossovers, under various scenarios, on the ITT, PP, AS, and the combina- tion of the ITT and PP analyses, with respect to type I error in trials with time-to-event outcomes. We also evaluate bias and standard error of the estimates from the ITT, PP and AS approaches. Results will be presented. Our research will guide methodologists in the analysis of non-inferiority trials with crossovers.   P1.2.122 An application of non-parametric factorial MANOVA in health research O Pasin1 , H Ankarali1 , S Cangur1 , MA Sungur1 1 Duzce University, Institute of Health Sciences, Biostatistics, Duzce, Turkey   In health research, factors can have effects on multiple correlated out- comes. Analyzing these related outcomes separately can cause an in- crease in type I error. However in practical applications, this error is made frequently by using univariate analyses, which are simple and easily inter- pretable. In addition a majority of variables used in studies do not exhibit normal distributions or homogeneity of variance which are assumptions of univariate parametric models. For these reasons, introduction and ap- plication of new approaches, which provide more reliable results, are be- coming important. Our aim is to introduce the theoretical characteristics of a nonparametric MANOVA (PERMANOVA) and to discuss its application to a data set from a clinical trial. The PERMANOVA model uses distance or dissimilarity measures between pairs of subjects or variables. The null hypotheses regarding factor effects are tested with Pseudo F statistics and type I error of these statistics are calculated by a permutation approach or Monte Carlo simulation. In this study, we investigated the relationships among presence of the Hasimoto illness, gender and lipid profile by PERMANOVA. Data analysis was per- formed using the algorithm developed by Marti J. Anderson. We observed that there was no significant interaction between gender groups, but the lipid values in Hasimoto patients are significantly higher than in healthy individuals (P=0.0177). As conclusion we can say that in our example the true biological structure is better reflected, if related variables are simultaneously evaluated. P1.2.142 Comparison of different methods for controlling false positives in adverse event reports analysis S Romio1,2 , L Scotti1 , A Zambon1 , L Hazell3 , N Schmedt4 , M Sturkenboom2 , G Corrao1 1 University of Milano-Bicocca, Milan, Italy, 2 Erasmus University Medical Center, Rotterdam, The Netherlands, 3 Drug Safety Research Unit, Southampton, United Kingdom, 4 Leibniz Institute-BIPS GmbH, Bremen, Germany   Adverse event reports databases can be used to generate hypotheses of associations between drug use and adverse events. False positive (FP) associations need to be minimized when multiple association tests are performed simultaneously. Several methods to control for multiple com- parisons in hypothesis testing have been developed but few methods are available for testing one-sided hypothesis. I this context, we compared two modifications of the False Discovery Rate (FDR) approach: pFDR proposed by Storey and robust FDR proposed by Pounds and Cheng (rFDR) to control for FP. Within the SAFEGUARD project, adverse events reports available in FDA’s Adverse Event Report System and EudraVigilance databases in 2004-2012 were analyzed. The Proportional Reporting Ratio (PRR) was used to evalu- ate associations between non-insulin blood glucose lowering drugs and selected outcomes. Two identification criteria (broad/narrow) were used for each event. Data were analysed applying pFDR and rFDR approaches to the hypothesis tests on the PRR using proc multtest in SASv9.3 and a specific R script developed by Pounds and Cheng respectively. Statistical significance threshold was 0.05. The pFDR showed a reduction of the number of statistically significant associations from 12% to 17% in both databases and event definitions systematically higher than that from rFDR approach (0% to 2%). Different performance in controlling FP could depend on the sensitivity of rFDR to non-monotone distribution of original p-values. From these early findings the pFDR seems to be a robust tool for control- ling FP in pharmacovigilance.   P1.2.146 Adaptive crossover designs for phase II dose-finding trials S Simpson1 , LV Hampson1 , J Whitehead1 , B Jones2 1 Lancaster University, Lancaster, United Kingdom, 2 Novartis Pharma AG, Basel, Switzerland   Finding the optimal dose of a new medicine is an important part of drug development. Adaptive dose-finding procedures use accumulating data to determine which doses should be allocated to each new cohort of pa- tients recruited to the trial, with the aim of obtaining an accurate estimate of the target dose. In this presentation, we explore adaptive crossover de- signs for estimating the dose which provides a proportion, π, of the maxi- mum effect of a drug, i.e., the ED100π. We restrict attention to designs where each patient receives placebo and three active doses of the drug in a sequence determined by a Williams square.

Pages Overview