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

ISCB2014_abstract_book

94 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsMonday, 25th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P1.2.57 Trial-situations where stratification in randomization is advantageous - results of a simulation study A Glass1 , G Kundt1 1 University of Rostock, Institute for Biostatistics, Rostock, Germany   Objective: Randomization of patients has been established to a gold stan- dard in clinical trials. But pre-stratification of randomization has been dis- cussed controversially in this context. Randomization of patients in strata might help to prevent type I and type II errors via reduction of variance. But, stratification forces administrative efforts, and an increasing number of strata bears some problem. To support investigators decision concerning stratification we investigat- ed the impact with respect to the risk of prognostic imbalance under H0 by a simulation approach, where differently designed hypothetical trials were simulated (at least 1,000 times) for two therapy groups and two strata. Results and conclusions: The risk of prognostic imbalance could be quantified with maximum 60% for complete randomization. In larger tri- als, and/or with a factor of less prevalence this risk decreases. Restricted randomization as blocking (alone) can reduce this risk marginally from 32% to 29.6% for a factor of high prevalence (50%) in small studies (N = 100). Stating this, stratification of randomization can be helpful to provide comparable groups with higher probability for certain trial constellations. Decreased type I error by maximum 16pps due to stratification were de- tected for prognostic factors with large differences (80pps) of success rates between strata concerning small trials (N = 100). For trials with less than 400 patients and differences in success rates ≥30pps, subgroup and interim analyses stratification is recommended to reduce the expected risk of prognostic imbalance. For large trials with N≥400 patients relevant imbalance were not observed, independently of any factor prevalence.   P1.2.59 CompARE: web platform to choose the primary endpoint of a randomized clinical trial M Gómez-Mateu1 , G Gómez Melis1 1 Universitat Politècnica de Catalunya, Barcelona, Spain   CompARE is a free web-based interface. Its main aim is to help investiga- tors to choose the primary endpoint at the design stage of a trial. This tool provides a quantitative measure to decide between a relevant endpoint (RE) and a composite endpoint (CE) defined as the union of the RE and an additional endpoint (AE). Calculations in CompARE are based on the ARE method developed by Gómez and Lagakos. The ARE (Asymptotic Relative Efficiency) can be interpreted as the ratio of the required sample sizes to detect a specific treatment effect to attain the same power for a given significance level. The method is internally programmed and executed in R and run in the server, returning results in a dynamically generated webpage. R does not have to be installed in your computer, neither knowledge of R is required. CompARE is run through interactive HTML forms. Users are required to en- ter a list of candidate endpoints together with the probability of observing them in the control group and the relative treatment effect given by the hazard ratio. Results in CompARE are shown immediately with tables and intuitive plots in terms of ARE values for different scenarios. We encourage investigators to try CompARE, which is currently accessible as a beta version, by freely registering into the website (http://composite. upc.edu/CompARE). Furthermore, you can refer to the user’s guide to see how CompARE works.   P1.2.63 Effect of one-patient clusters on power in cluster-randomized trials S Hayoz1 , D Klingbiel1 1 SAKK Coordinating Center, Statistics Unit, Bern, Switzerland   Aims: The aim of this research was to assess the effect of one-patient clus- ters on the power in cluster-randomized trials with small cluster sizes. Methods: The assumptions for the simulations were based on the trial SAKK 95/06 where 40 physicians (clusters) with equal patient numbers of 4 (160 patients in total) were planned. A linear mixed model with physician as random effect should yield a power of 81.9% with alpha 5%. In reality the cluster sizes varied between 1 and 10, with 22.6% one-patient clusters. Simulations were performed with 10000 repetitions per scenario. Results: With 40 physicians, cluster sizes between 1 and 10 and 5.0%- 25.0% one-patient clusters, the achieved power varied between 79.6% and 80.7%. If the number of patients per cluster was restricted to a maxi- mum of 6 and the number of physicians increased to 41-55 with 4.9%- 36.4% one-patient clusters, the achieved power varied between 81.8% and 82.6%. Resampling from the cluster sizes of SAKK 95/06 with 7%-43% one-patient clusters yielded a power between 78.1% and 79.3%.When the one-patient clusters were excluded from the analysis the achieved power was 0.3%-0.9% lower. The proportion of models with numerical problems was never higher than 0.2%. Conclusion: In all considered scenarios the decrease in power in compari- son to the theoretical model with equal cluster sizes was negligible. When the number of patients per cluster was restricted and instead the number of physicians was increased there was no relevant loss in power even with a high proportion of one-patient clusters. P1.2.64 Comparison of design options for phase IB clinical trials in oncology: simulation results M Bigler1 , S Hayoz1 , A Xyrafas1 , D Klingbiel1 1 Swiss Group for Clinical Cancer Research (SAKK), Bern, Switzerland Background: Phase I trials play an essential role in the development of new drugs in cancer research. Many designs and many comparisons of these designs with simulations exist. To the best of our knowledge, no simulation results exist when there are only few dose levels to compare, as it is often the case in phase IB trials, where for example toxicity data in other indications are available. Methods: We compared the 3+3 design, the continual reassessment method (CRM), the modified toxicity probability interval method (mTPI) and the rolling-6 design in a dose-finding study with only 2-4 dose levels of a drug. We explored the properties of the designs in several scenarios. Results: The 3+3 design generally performed poorly. When the prior probabilities for toxicity were close to the true values, CRM was the best method in terms of selecting the correct dose as maxi- mum tolerated dose (MTD) and number of patients treated at the MTD. However, it suffered from misspecification of the priors. In all scenarios the rolling-6 design required the shortest time to complete the trial. The mTPI method was—depending on the design parameters—slightly better than the 3+3 and rolling-6 designs as it selected the correct dose more often and treated more patients at the MTD. Conclusion: CRM has the best performance in terms of different metrics for phase IB trials in oncology given that the priors are well specified. As investigators seem to feel uncomfortable with this complicated method, there may be alternatives like mTPI to consider.  

Pages Overview