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

ISCB2014_abstract_book

16 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Results: We selected 49 trials (1735 patients treated at 4919 cycles) in which mild, moderate and severe grades were observed. The PO assump- tion for dose effect at cycle 1 was rejected at the 5% level only in 4 trials. On repeated cycles, PO were observed for both the dose and the cycle variables in 3 and 5 out of the 44 studies respectively. In mixed effect models, variance of the random-intercept ranged from 1.5 to 13. Markov chain modeling revealed significant association between the risk of severe toxicity at a cycle and a moderate or mild toxicity at the previous cycle. In studies with no time effect detected from PO mixed model, common pattern of serial correlation was observed across several trials. Conclusions: PO assumption is reasonable and can be implemented in single agents phase I trials. Large inter-patient variability can be expected and modeled using mildly informative priors in dose-finding trials incor- porating repeated toxicity measurements.   C04.2 A Bayesian approach to oncology combination dose-finding H Shen1 , M Whiley2 , B Neuenschwander2 1 China Novartis Institute of Biomedical Research, Shanghai, China, 2 Novartis Pharma AG, Basel, Switzerland In oncology the primary objective of Phase I trials for new compounds is to determine the maximum tolerated dose (MTD) or recommended Phase II dose (RP2D) based on severe safety events called dose limiting toxicities (DLTs). Bayesian approaches have been increasingly used in this setting and the flexibility they bring to dose escalation study design enables us to better balance statistical, clinical and operational considerations. However, efforts to further improve the efficiency and effectiveness of drug devel- opment in early phase oncology have led to increasingly complex trials, and adapting statistical methods to support such complex development strategies is challenging. We will present a practical and comprehensive Bayesian approach used in a dual-agent combination dose escalation study. In this approach a 5-pa- rameter logistic regression model is used to characterize the dose-toxicity relationship, and Bayesian inference is used to guide the dose escalation process. We will discuss the use of historical data from previous single agent dose escalation studies to inform the combination dose escalation. We will further demonstrate how this methodology can be used to ad- dress emerging drug development issues. In our example we will show how to support both a change to the dosing schedule and to the drug formulation during the course of the dose escalation. In each case we will demonstrate how the Bayesian approach allows current study information to inform the continued dose escalation. C04.3 Bayesian optimal clinical trial design for monoclonal antibodies MR Lange1 , H Schmidli1 1 Novartis Pharma AG, Basel, Switzerland Monoclonal antibodies are increasingly and successfully used for the treatment of many chronic diseases. Semi-mechanistic nonlinear models are needed to adequately describe the dose-time-response relationship for such drugs. We consider the optimal allocation of patients to doses in a planned clinical trial for a monoclonal antibody, in order to learn most about the nonlinear dose-time-response model. To characterize a design as optimal, Lindley (Ann. Statist. 1956) introduced an information theoretic approach which aims to maximize the expected Kullback-Leibler difference between prior and posterior of the model parameters. Later, Bernardo (Ann. Statist. 1979) showed that this criterion can also be justi- fied within a decision theoretic framework and emphasized its compatibil- ity with the concept of Bayesian inference. Despite the strong theoretical motivation, this criterion has not often been used in practice due to the heavy computational challenges for complex models. We will demon- strate the applicability of these concepts for finding optimal clinical trial designs for monoclonal antibodies. For illustration, we will discuss the de- sign of a clinical trial in patients with urticaria. C04.4 Dose-escalation strategies which utilise subgroup information A Cotterill1 , T Jaki1 1 Lancaster University, Lancaster, United Kingdom In determining the maximum tolerated dose (MTD) of a drug, phase I on- cology trials commonly assume that the patient population is homoge- neous. A single MTD is therefore identified for the entire population. Strict inclusion criteria can be used to justify this assumption but may lead to a treatment effect being missed in the excluded patients. Conversely, inclu- sion of a subgroup in which the treatment is inefficacious could mask a treatment effect in the remaining population. Removing the underlying assumption of a homogeneous population by investigating suspected subgroup effects in dose-escalation to identify, when necessary, an MTD in each subgroup could therefore be beneficial. A simple example involves a drug approved in adults and to be tested in children. In the adult trial, patients with a certain biomarker appeared more likely to experience side-effects than the remaining population. This information is of interest here as it could be used to aid development of patient-specific dosing in the paediatric population. A Bayesian decision theoretic approach assuming a homogeneous popu- lation is used as the baseline for comparison of three designs which ac- count for a subgroup effect. The first design escalates independently within subgroups while in the second design information is shared across subgroups. Finally, an approach using sparsity inducing priors is presented. Simulations comparing these designs were carried out in the setting of a phase I study of temozolomide in children and adolescents with recurrent solid tumours. Simulation results indicate that accounting for potential differences in tolerance between subgroups can be beneficial. C04.5 Continual reassessment method for dose escalation clinical trials in oncology: a comparison of prior approaches using AZD3514 data G James1 , S Symeonides2 , J Marshall2 , J Young2 , S Smith2 , G Clack2 1 PHASTAR, London, United Kingdom, 2 AstraZeneca, Macclesfield, United Kingdom Background: The continual reassessment method (CRM) is considered more efficient and ethical than standard methods for dose-escalation trials in oncology, but requires an underlying model of the dose-toxicity relationship (“priors”) and there is limited guidance of what this should be when little is known about this association. Aim: Compare the impact of applying the CRM with different prior ap- proaches and the 3+3 method in terms of ability to determine the true maximum tolerated dose (MTD) and number of patients allocated to sub- optimal and toxic doses. Methods: Post-hoc dose-escalation analyses on real-life clinical trial data on an early oncology drug (AZD3514) using the 3+3 method and CRM us- ing five prior approaches: conservative, aggressive, dose-linear, sigmoidal and O’Quigley.The prior probability values were further examined by add- ing 10% to each prior within each method. Dose limiting toxicity (DLT) was retrospectively defined as moderate or greater nausea/vomiting. Results: All methods correctly identified the true MTD. The 3+3 method allocated six patients to both sub-optimal and toxic doses. All CRM ap- proaches allocated four patients to sub-optimal doses. No patients were

Pages Overview