Clinical Trials 1
Presenter: Axel Benner
When: Monday, July 11, 2016 Time: 4:00 PM - 5:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
To randomize or not to randomize the challenge of individualized oncology trials
For estimating causal effects of treatments, randomized trials are generally considered as the gold standard. Unfortunately they are not always feasible for a variety of reasons, including ethical concerns. Consequently, in such situations assessment of causal effects must be derived from non-randomized studies. Especially in clinical cancer trials on personalised therapy randomized treatment assignment is rarely possible. Since many cancer genes are mutated at low frequencies within any histologic tumor subtype, individual patients will have individual patterns of potentially actionable genetic alterations. Based on the molecular diversity and genetic taxonomy of cancer, and clinical actionability, non-randomized trials of "intervention baskets" based on molecular data seem more appropriate. Whereas randomization allows us to make causal inference, the non-randomized selection of participants into treatment groups may be associated with confounding factors, resulting in bias that might occur in naive statistical analyses. However, advances in methodology have provided a toolkit for non-randomized observational studies. We discuss different approaches to reduce selection bias, including propensity score methods and the direct adjustment for confounding in regression models. A simulation study evaluates the different methods in contrast to a randomized experiment. For illustration we present the design and analysis plan for a non-randomized trial within the MASTER (Molecularly Aided Stratification for Tumor Eradication Research) program of the National Center for Tumor Diseases (NCT) in Heidelberg.
Clinical Trials 1
Presenter: Hui-Min Lin
When: Monday, July 11, 2016 Time: 4:00 PM - 5:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Inference on treatment efficacy in subgroups and the mixture population, with an application to time-to-event outcomes
In tailored drug development, the patient population is thought of as a mixture of two or more subgroups that may derive differential treatment efficacy. In order to find the right patient population for the treatment to target, it is necessary to infer treatment efficacy in subgroups and combinations of subgroups. A fundamental consideration in this inference process is that the logical relationships between treatment efficacy in subgroups and their combinations should be respected (for otherwise the consistency assessment of efficacy may become paradoxical). The current statistical practice of estimating the treatment efficacy in a mixture population has serious flaws. We propose a subgroup mixable estimation principle which respects the logical relationships between treatment efficacy in subgroups and their combinations. Focusing on the time-to-event outcomes and ordinal biomarkers, we develop a simultaneous inference procedure, with appropriate efficacy measures, to correctly infer treatment efficacy in a mixture population.
Clinical Trials 1
Presenter: David Oakes
When: Monday, July 11, 2016 Time: 4:00 PM - 5:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Modeling the win ratio in clinical trials with multiple types of event
Pocock (2012), following Finkelstein and Schoenfeld (1999), has popularized the "win ratio" as a simple method of analysis for controlled clinical trials with multiple outcomes. This approach uses pairwise comparisons between patients in the active treatment group and control group based on a primary outcome (say time to death) with ties broken using a secondary outcome (say time to a nonfatal cardiac event). A drawback of this method is that the observed pairwise preferences and the weight they attach to the component rankings will depend on the distribution of potential follow-up time, so that win-ratios across different trials of the same intervention may not be comparable. We propose a modified definition incorporating a time horizon into the comparisons and present conditions under which the modfied win ratio does not depend on this horizon. An important special case is the bivariate Lehmann model in which the logarithms of the joint survivor function of the two survival times in the two groups are proportional. This class includes Hougaard's (1986) model. We show how the modified win ratio may be estimated parametrically or nonparametrically. Extensions to three or more types of events are indicated.
Clinical Trials 1
Presenter: David Schoenfeld
When: Monday, July 11, 2016 Time: 4:00 PM - 5:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
Assessing Survival Benefit When Treatment Delays Disease Progression
Abstract Background For a potentially lethal chronic disease like cancer, it is often infeasible to compare treatments on the basis of overall survival, so a combined outcome such as progression-free survival (which is the time from randomization to progression or death) has become an acceptable primary endpoint. The rationale of using an efficacy measure that is dominated by the time to progression is that an effective treatment will delay progression and when treatment is stopped at progression, the effect of treatment after this time is small. However often trials that show a significant benefit for delaying progression but not on overall survival are not universally viewed as convincing evidence that the drug should become the standard of care. Methods We propose that when there is a significant treatment effect of delaying progression, a Bayesian analysis of overall survival should be undertaken. We suggest using a joint piecewise exponential model, where the treatment effect on the hazard for progression and for death after progression are captured through two distinct parameters. We develop a plot of the overall survival advantage of the new therapy versus the dispersion of prior distribution of relative hazard for death after progression which is assumed to be centered at one. This plot can augment the discussion about whether the new treatment is beneficial on survival, by showing the effect of our prior belief on the treatment effect of death after progression on our posterior belief on the survival benefit of the treatment. Results In the example of an early breast cancer trial for which a new treatment significantly delayed disease recurrence, our Bayesian analysis showed that with very reasonable assumptions on the effects of treatment after recurrence, there is a high probability that the new treatment improves overall survival. Conclusions For a clinical trial for which treatment delays progression, the proposed method can improve the interpretability of the survival comparison using data from the study.
Clinical Trials 1
Presenter: Kevin Kunzmann
When: Monday, July 11, 2016 Time: 4:00 PM - 5:30 PM
Room: Salon C Carson Hall (Level 2)
Session Synopsis:
New Developments in Optimal Adaptive Two-Stage Designs for Single-Arm Trials with Binary Outcome
Minimizing the number of patients exposed to potentially harmful drugs in early oncological trials is a major concern during planning. These trials are often planned with a single arm and binary endpoint. Adaptive designs account for the uncertainty about the true effect size by determining the final sample size within an ongoing trial after an interim look at the data. We formulate the problem of finding adaptive designs minimizing expected sample size under the null hypothesis for single-arm trials with binary outcome as an integer linear program extending previous work [1, 2, 3]. This representation can be used to identify optimal adaptive designs which improve previous designs in two ways: Firstly, the expected sample size under the null hypothesis is reduced and, secondly, we explain how pathologies of previous designs arising from the discrete nature of the underlying statistics can be removed. The resulting designs are both efficient in terms of expected sample size under the null hypothesis and well interpretable. Furthermore, the formulation as integer linear program provides a unified framework for incorporating almost arbitrary additional constraints and solving the resulting problems in an efficient way. Inference after an adaptive trial is challenging. We explain how extensions of traditional approaches from designs with continuous endpoints fail to fulfill elementary consistency requirements between estimation and p values and propose a novel and unified way of deriving point estimates and p values for adaptive designs with binary endpoints. These are consistent in the sense that the test rejects if and only if the p value derived from the ordering of the outcome space which is induced by the point estimator is smaller than the chosen significance level. The bias and MSE profiles of the resulting estimators compare favorable to those of exisitng estimators. [1] Englert, S., Kieser, M. Optimal adaptive two-stage designs for phase II cancer clinical trials. Biometrical Journal 2013; 55:955968 [2] Shan, G., Wilding, G. E., Hutson, A. D., Gerstenberger, S. Optimal adaptive two-stage designs for early phase II clinical trials. Statistics in Medicine 2015; doi:10.1002/sim.6794. [3] Simon, R. Optimal two-stage designs for phase II clinical trials. Controlled Clinical Trials 1989; 10:110
