< Back to previous page

Project

Bayesian methods for the inclusion of historical data in Phase I and Phase II clinical studies (R-8141)

Standard RCTs are analyzed usually by only current data. But, they are costly with respect to trial duration, effort and money, especially due to the number of patients being treated. Moreover, ethical issues are in question that patients who receive placebo are at risk. However, similar historical trials that have been carried out with the same control treatment are available and can be included in the analysis of a new trial. For this reason, the formal inclusion of historical data in the analysis of a clinical trial has recently gained increasing interest. Including historical controls has ethical and economic advantages, as (if done appropriately) less patients need to be put at risk with the control treatment and conclusions may be reached earlier. In MAP approach a Bayesian meta-analysis is performed by assuming the control parameters of all trials are exchangeable and are drawn from the same distribution. However, often the inference using this method is not robust when the historical and current data are incompatible. The robustified version of MAP prior was then proposed so that the historical information would be less credible and then the prior could be discarded when there is a conflict between the historical and current data. This has been handeled by adding weakly informative component to the MAP prior. However, the proportion of the added robust component is often set to be fixed based on the degree of confidence of the clinical trial team, which is not estimated based on the compatibility of the historical and current data. The MAP approach is used to incorporate when several historical studies are available, but for a single historical trial other methods like the power prior is preferred due to a challenge in the setting for the variability of between studies. The power prior provides a simple way to incorporate and downweight historical data by raising the historical likelihood to a power. But, the formulation of this prior violates the likelihood principle and has a property that discourages borrowing of information from the historical data. For this reason, the modified power prior (MPP) has later been developed by multiplying a scaling constant, for incorporating a single/multiple historical study/studies in the analysis of current data. The computation of the scaling constant however is quite challenging; and it can neither be implemented by the Bayesian softwars like Winbugs. For multiple historical studies, the MPP is performed by assuming different weight (power) parameters with independent distributions to account for heterogeneity among historical studies. However, similar patient populations with the same disease are more likely to take the same control treatment. For this reason, in this study we will consider the MPP for multiple historical data with dependent weight parameters in a hierarchical Bayesian framework by assuming the same parent distribution for the weight parameters in the analysis of clinical trial. The robust version of the MPP is not yet studied and it could play an important role to account for the possibility that the historical data and the current data are in conflict. In this study, we consider the robustified version of the the MPP that could discount the historical data through their power parameters when there is a conflict between the historical and current data
Date:1 Oct 2017 →  31 Dec 2022
Keywords:MULTIVARIATE CATEGORICAL DATA
Disciplines:Applied mathematics in specific fields, Statistics and numerical methods