Online Program

WITHDRAWN: Estimation and Inference for Optimal Treatment Regimes Under Constraints

Shuping Ruan, Department of Statistics, North Carolina State University 
Eric Laber, Department of Statistics, North Carolina State University 

Keywords: Optimal treatment regime, Constrained optimization, Exact penalty method

Treatment regimes formalize clinical decision making as a sequence of maps from current patient information to a recommended treatment. Most current methods define a treatment regime to be optimal if it maximizes the expected value of a single scalar outcome in a population of interest. However, this framework neglects the need to balance several potentially competing outcomes (e.g., treatment efficacy and side effect burden). An existing resolution is to form a composite outcome that attempts to balance these competing outcomes. Unfortunately, it can be difficult to find a composite outcome that closely reflects patient utilities, and the quality of the estimated treatment regime can be severely affected by the misspecification of a composite outcome. Here, we develop a new framework in which the optimal regime maximizes the primary outcome of interest, subject to constraints on secondary outcomes of interest. Estimation and inference are based of reformulating the constrained problem as an unconstrained problem using an exact penalty. The method is illustrated using a series of simulation experiments.