Clinical trial simulation: a tool for understanding study failures and preventing them.

Abstract

Execution models describe protocol deviations from a specified study design. When a clinical trial is planned, it is generally supposed that it will be executed according to a specific protocol that defines all aspects of the experimental design, from its beginning to its completion. Adherence to the protocol will allow estimation of the treatment outcome (safety and efficacy) with sufficient statistical power, or at least that is what is assumed. In reality, however, deviations from the protocol may lead to failure of the study to achieve its stated aims. In anticipation of protocol deviations that contribute to inflated residual variability and decreased study statistical power, trial designers tend to overpower studies in a rather arbitrary way. It is difficult to estimate quantitatively the consequences of one protocol deviation on statistical study power and, a fortiori, it is almost impossible to do it for a combination of protocol deviations. One way to study the consequences of model deviations is by using modelling and simulation techniques, and more specifically longitudinal stochastic models that can describe individual behaviours. Thus, execution models are powerful tools for identifying weaknesses or limitations in a proposed study design, which may be anticipated, avoided or resolved in order to increase robustness of the study design prior to implementation of the actual clinical study. As such, they are an integral component of clinical trial simulation and an essential tool in clinical trial design.