Abstract:
|
One goal of early phase clinical studies is to estimate the dose-response relationship with sufficient precision to make an informed dose selection for future studies. For such estimation, the simplest case is a study with fixed doses and patients in the study adhere to their assigned dose regimen. However, in many clinical studies, it is inevitable for the study doses to be titrated up or down due to tolerability, safety or efficacy considerations. The existing methods, based on either the intent-to-treat principle that only takes into accounts the original doses, or the as-treated analysis that utilizes actual doses taken, ignore the potential confounding effect brought by dose titrations and thus may lead to severe bias for the estimation of dose-response curve. To this end, we propose a two-step estimation approach to provide consistent estimation for the dose-response curve. This approach first estimates the probability of dose titration to obtain the inverse probability weights, and then constructs an unbiased estimating equation based on the naïve Fisher's score function. We demonstrate the strengths of the new approach over existing methods through simulation studies.
|