Department of Biostatistics
Adaptive Trials Working Group
2013 - 2014
ABSTRACT: Bayesian adaptive trials allow the probability of allocation to a particular treatment arm to change, as information becomes available about its true worth. Despite some clear advantages over the standard (fixed randomisation probability) approach, they have been criticised in the literature for a host of reasons. For example, the possible frequentist operating characteristics of Bayesian adaptive trials is often raised as a cause for concern. We examine a (slight simplification of a) specific design described by Giles et al (attached) in order to make things concrete, but its a fairly typical example. The questions we would like to answer are:
- What is the direction and magnitude of the bias in the treatment effect estimates from the Giles et al design?
- What factors drive this bias?
- How much of this is due to adaptive randomisation, as opposed to early stopping for efficacy/futility?
- Can it be corrected?
- What are the potential pitfalls of not doing so?
ABSTRACT: None Given.
ABSTRACT: Tumor biomarkers are playing increasing important roles in cancer clinical trial, not only as correlative objectives to characterize disease at the molecular level, but also as assays with clinical utility to inform treatment decisions. Novel clinical trial designs are needed to efficiently answer questions of both drug effects and biomarker performance. Adaptive designs add flexibility to trials in a planned and controlled manner, but require careful consideration to their use based on feasibility, available resources, and the regulatory environment. A recent study in metastatic lung cancer, BATTLE: Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination (PI: Kim), used an adaptive design to assign patients to one of four treatment arms conditional on their molecular profile, and represents an excellent case study on the utility of adaptive-randomization and proper utilization of Bayesian statistics.
ABSTRACT: He will be presenting his research on the sequential parallel comparison design, which was created by researchers at MGH to specifically address minimizing the placebo effect. He will cover three adaptive trial designs that were applied to the SCPD and discuss the characteristics and increased efficiency of the designs based off of simulation data.
ABSTRACT: Motivated by our observation that phase I studies more frequently exceed traditional sample sizes to further assess efficacy endpoints in specific subsets, and the fact that the scientific objectives and designs of expansion cohorts have yet to be systematically examined at the protocol or statistical-design specific level, we conducted a study in which adult therapeutic phase I protocols opened within Dana-Farber/Harvard Cancer Center over the last 25 years were identified, and details of the statistical operating characteristics of the expansion cohorts were examined. Our goal was to assess trends within our institution and to inform recommendations addressing these trends with respect to the research process, ethical concerns and resource burden that they bear.
ABSTRACT: In the last decade, the development of new therapies, or combinations of drugs, in TB studies allows the researchers to use methods able to detect the efficiency of all this agents simultaneously. An important class of designs that provide a way to solve this kind of problems are Multi-Arm Multi-Stage(MAMS) designs. In those studies, a binary response is commonly used as final end-point but it needs since eight-nine months to be observed. Recent studies provided quite positive results about the use of preliminary end-points in order to detect in advance the efficacy, or inefficacy, of therapies. In this work, we propose a Bayesian Adaptive MAMS design that uses joint information of preliminary and final end-points to adapt randomization and make interim analysis ensuring frequentest properties at the end of the trial. Our goal is to show how, in presence of a good correlation between the two end-points, the use of adaptive randomization and preliminary measures provide positive improvements to the trial outcomes.
Research by Matteo Cellamare, Lorenzo Trippa and Steffen Ventz.
ABSTRACT: In the MATCH trial, tumor samples from advanced cancer patients will be screened for presence of a set of mutations, as well as possibly additional markers, and patients will be assigned to targeted therapies based on the molecular characteristics of their tumors. Current plans call for 20-30 treatments, corresponding to different targets, to be evaluated simultaneously. The primary objective is to to determine if each treatment has activity within its targeted population, regardless of the tissue of origin. It is expected that roughly 3000 patients will be screened to enroll at least 1000 patients into the treatment arms. The current plans for the design of the trial will be presented and some related issues discussed. There is currently nothing adaptive about the design, but it is hoped that the presentation will stimulate discussion of additional options for consideration.
ABSTRACT: None Given.
|Back to HSPH Biostatistics||
Maintained by the
Last Update: April 17, 2014