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


ISCB 2014 Vienna, Austria • Abstracts - Poster Presentations 91Monday, 25th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P1.1.110 On adapting the sample size in a Bayesian clinical trial in small populations T Brakenhoff1 , S Nikolakopoulos1 , KCB Roes1 , I van der Tweel1 1 UMC Utrecht, Utrecht, The Netherlands   In the design stage of a clinical trial in small populations there are sev- eral methodological challenges. An obvious fundamental obstacle is the limited available number of patients to be included in such a trial. As a consequence, limited information will be available, concerning design pa- rameters, on which sample size calculations can be based. Thus, framework of sample size re-estimation which is thoroughly stud- ied in the frequentist paradigm, is an obvious alternative. We extend proposed Bayesian sample size estimation methodology to the situation where the sample size can be reevaluated at one or more interim stages. Working with normally distributed outcomes, such an approach handles the scenario where the variance observed in the trial is different from the one anticipated in the design stage. By using a fully Bayesian predictive approach, our method can handle sample size re-estimation in combination with imposing a maximum sample size as a realistic constraint when conducting research in small populations. The approach is illustrated by reanalyzing data from a real randomized trial in the field of pediatrics. P1.1.118 Bayesian analysis of parametric frailty models for repeated event data: estimating unreported event times using interval data RK Owen1 , DG Tincello1 , PC Lambert1,2 , S Bujkiewicz1 , KR Abrams1 1 University of Leicester, Leicester, United Kingdom, 2 Karolinska Institutet, Stockholm, Sweden   Background: We are often interested in analysing the time to recurrent events associated with chronic diseases following repeated treatments, but patient follow-up can be intermittent and event times are frequently unreported. Motivated by a clinical trial in overactive bladder syndrome, we want to analyse the potential diminishing effect of repeated injections of botulinum toxin, adjusting for severity status, on patient reported re- currence of symptoms. Methods: We used a Bayesian framework to fit aWeibull proportional haz- ards model for repeated event data to obtain posterior predictive distribu- tions from which to sample unreported event times. To further account for the correlation between repeated events within the same individual, we incorporate a shared frailty term. We applied this methodology to a clinical trial of patients receiving a maximum of 3 repeated injections of botulinum toxin for overactive bladder over a 5 year follow-up period. Results: Bayesian approaches had an improved fit to the data compared to the frequentist alternative, and including posterior predictions of the missing severity covariate increased precision in the estimates. Patients with severe symptom severity had a considerably higher rate of symptom recurrence (HR: 3.07, 95%CrI: 1.56,7.15) compared to patients with normal severity status. Repeated injections appear to reduce the rate of symptom recurrence in severe patients. Conclusions: With an increasing need to assess the time to symptom re- currence in chronic conditions, and the difficulties faced with intermittent follow-up, the use of a flexible Bayesian framework would appear to be advantageous.   P1.1.124 Phase II study to assess the safety of bevacizumab with neoadjuvant chemotherapy in ovarian cancer using a Bayesian approach S Zohar1 , F Joly2 , R Rouzier3 , Y Ghazi4 , V Menguy4 , D Pau4 1 INSERM UMRS 1138, Team 22, Paris, France, 2 INSERM U1086, Centre François Baclesse, Caen, France, 3 Institut Curie, Paris, France, 4 Roche, Boulogne-Billancourt, France   Background: Bayesian analysis is rarely used in randomized phase II clini- cal trials. In contrast to the frequentist approach, the Bayesian approach has no consideration of Type I error and no power calculation because the inference is based on the posterior or predictive distributions. Due to regulatory constraints, Bayesian analyses are typically only used in addi- tion to the primary frequentist. Objectives: This randomized phase II study will evaluate the feasibility of using two types of inferences: the safety is evaluated using Bayesian infer- ence and the efficacy is evaluated using frequentist inference.The primary endpoint is to evaluate the benefit of neoadjuvant bevacizumab and che- motherapy assessed by the complete resection at surgery in patients with advanced ovarian cancer. Methods: For both efficacy and safety endpoints sequential analyses are performed. For the efficacy endpoint a frequentist hypothesis testing is used and for the safety and point a Bayesian approach, based on a beta- binomial model with 3 prior distributions is performed. The parameters of each prior distribution were selected from expert’s elicitation. Analyses are performed sequentially but more frequently for the safety endpoint as the investigators wished to stop early the trial if the treatment is estimated to be too toxic (for minimum 2 prior distributions out of three). Conclusions: Bayesian approach provides flexibility in decision making process regarding the continuation or discontinuation of patient accrual regarding the safety endpoint. As the inference is not influenced by the number of interim analyses it allows stopping the trial as early as needed. P1.1.130 Some inferential results in branching processes in random environments M Molina1 , M Mota1 , A Ramos2 1 University of Extremadura, Badajoz, Spain, 2 University of Extremadura, Cáceres, Spain   This work deals with mathematical modelling through branching pro- cesses. We are interested in developing stochastic models to describe the demographic dynamics of populations.We focus on the class of branching processes with progenitor couples in a random environment introduced in Molina etal.(2012).We provide several results concerning the extinction of the population and, under a parametric and nonparametric setting, we derive estimators for the offspring distribution and for its main moments. In order to determine the corresponding highest posterior density cred- ibility sets, we also propose a computational method. By way of illustra- tion, we include a simulated example in population dynamics. P1.1.164 Bayesian methods in adaptive dose finding C Tirodkar1 , S Solanki1 1 Cytel Statistical Software and Services Pvt. Ltd., Pune, India   Biostatistics has seen a phenomenal growth in recent years and has been marked by considerable technical innovation in both methodology and computational practicality. The complexity and amount of data produced by biomedical studies is increasing at a staggering pace every year and is one of the most significant challenges experienced by clinical researchers. Hence, devising strategies to effectively design and obtain useful informa-

Pages Overview