Bayesian Adaptive Designs for Early Phase Clinical Trials
Abstract
My dissertation work primarily focuses on Bayesian adaptive design for phase I and phase II clinical trials. Specifically, it includes three topics: 1) developing an adaptive dose modification design for phase I clinical trials, 2) proposing a calibrated Bayesian hierarchical model for basket trial design, and 3) originating a Bayesian latent subgroup design for basket trials. Most phase I dose-finding methods in oncology aim to find the maximum-tolerated dose (MTD) from a set of prespecified doses. However, in practice, due to a lack of understanding of the true dose-toxicity relationship, it is likely that none of these prespecified doses is equal or reasonably close to the true MTD. To handle this issue, we propose an adaptive dose modification (ADM) method that can be coupled with any existing dose-finding method to adaptively modify the dose, when it is needed, during the course of dose finding. To reflect clinical practice, we divide the toxicity probability into three regions: underdosing, acceptable and overdosing regions. We adaptively add a new dose whenever the observed data suggest that none of the investigational doses is likely to be located in the acceptable region. The new dose is estimated via a nonparametric dose-toxicity model based on local polynomial regression. The simulation study shows that ADM substantially outperforms the similar existing method. We applied ADM to a phase I cancer trial. The basket trial evaluates the treatment effect of a targeted therapy in patients with the same genetic or molecular aberration, regardless of their cancer types. Bayesian hierarchical modeling has been proposed to adaptively borrow information across cancer types to improve the statistical power of basket trials. Although conceptually attractive, research has shown that Bayesian hierarchical models cannot appropriately determine the degree of information borrowing and may lead to substantially inflated type I error rates. We propose a novel calibrated Bayesian hierarchical model approach to evaluate the treatment effect in basket trials. In our approach, the shrinkage parameter that controls information borrowing is not regarded as an unknown parameter. Instead, it is dened as a function of a similarity measure of the treatment effect across tumor subgroups. The key is that the function is calibrated using simulation such that information is strongly borrowed across subgroups if their treatment effects are similar, and barely borrowed if the treatment effects are heterogeneous. The simulation study shows that our method is more powerful than the independent approach, and has substantially better controlled type I error rates than the Bayesian hierarchical model. Motivated by the heterogeneity feature of basket trials, we propose a Bayesian latent subgroup trial (BLAST) design to accommodate such treatment heterogeneity across cancer types. We assume that a cancer type may belong to the sensitive subgroup, which is responsive to the treatment, or the insensitive subgroup, which is not responsive to the treatment. Conditional on the latent subgroup membership of the cancer type, we jointly model the binary treatment response and the longitudinal biomarker measurement that represents the biological activity of the targeted agent. The BLAST design makes the interim go/no-go treatment decision in a group sequential fashion for each cancer type based on accumulating data. The simulation study shows that the BLAST design outperforms existing trial designs. It yields high power to detect the treatment effect for sensitive cancer types that are responsive to the treatment, and maintains a reasonable type I error rate for insensitive cancer types that are not responsive to the treatment.
Subject Area
Biostatistics|Statistics
Recommended Citation
Chu, Yiyi, "Bayesian Adaptive Designs for Early Phase Clinical Trials" (2017). Texas Medical Center Dissertations (via ProQuest). AAI10605865.
https://digitalcommons.library.tmc.edu/dissertations/AAI10605865