Estimating Individualized Dosage Rules Using Outcome-Weighted Learning
*Donglin Zeng, University of North Carolina
Keywords: DC algorithm, dose-finding, individualized treatment rule, weighted support vector regression, risk bound
In dose-finding trials, there is a growing recognition of individual-level heterogeneity when searching for optimal doses. An optimal individualized dosage rule should maximize expected clinical benefit. We consider a randomized trial design where candidate dose levels are continuous. To estimate the optimal dosage rule, we propose an outcome-weighted learning method, which converts the estimation into a weighted learning with a truncated L-1 loss. A difference of convex functions algorithm is adopted to efficiently solve the associated nonconvex optimization problem. The consistency and convergence rate for the value under the estimated rule are derived and small-sample performance is evaluated via simulation studies. We demonstrate that the proposed method outperforms competing approaches. Finally, data from a study for Warfarin (an antithrombotic drug) dosing are used to illustrate the method.