Abstract:
|
Precision medicine is an emerging scientific area for disease treatment and prevention by considering individual variability. In clinical trial studies, one of the main goals is to obtain an optimal individual treatment rule (ITR) which can help treatment selection according to each patient's attributes. Recently, outcome weighted learning (OWL) was proposed to estimate such an optimal ITR in the two-arm setting by maximizing the expected clinical outcome. However, it is unclear how OWL would work for the ordinal treatment settings such as dose finding. Furthermore, OWL requires data transformation when the outcome has negative values. In this paper, we propose a new method to estimate ITR with ordinal treatments. In particular, we use a data duplication technique with a piecewise convex loss function. We establish Fisher consistency for the estimated ITR and convergence properties for the risk bound. Numerical examples show the highly competitive performance of the new method.
|