Robust Response-Guided Dosing

Abstract

In response-guided dosing (RGD), doses are adapted to the uncertain progression of each patient's disease condition. A stochastic dynamic program was recently developed for RGD. We study its robust counterpart, where the dose-response distribution belongs to an uncertainty set. For interval uncertainty, we prove that it is optimal to administer higher doses in worsening disease. When a certain scaling of a nominal distribution describes the interval, optimal doses also increase in the scaling parameter. Theory is illustrated via numerical results.