Responder Identification in Clinical Trials

The thesis gives an overview of the techniques used up to now for
responder identification and it proposes a new method for
systematic search for responders. The responder identification
method consists of the following three steps: 1. Identification of
prognostic factors (e.g. via Cox-PH model on the standard treatment
arm) 2. Identification of patients in the new treatment arm, who's
survival is badly estimated by the prognostic model (e.g. via
search for outliers in the deviance or martingale residuals) 3.
Identification of predictive factors, which describe common
features of the patients with residual outliers, namely the
positive and negative responders (e.g. via regression tree or bump
hunting analysis, or via the suggested stabilized bump hunting
procedure) The method is evaluated with a simulation study and
applied on the EMIAT data se