Optimal Treatment Regimes With Quantile Regression

A treatment regime maps observed patient characteristics to a recommended treatment.Because of the technological advances,there is a growing need for powerful estimators of an optimal treatment regime which can be used to both observational and randomized clinical trail data.Traditionally.the optimal regime can be found by assuming a regression model for expected outcome conditional on treatment and covariates.However,this method is suspect if the regression is incorrectly specified.In order to improve this model,we introduce a novel and general framework that transforms the problem of estimating an optimal regime into a classification prob-lem and uses a doubly robust augmented inverse probability weighted estimator to increase precision.Quantile regression model which replaces the original regression model will also improve the robustness of each estimator.Based on each method,we want to derive the causal inference estimator under correctly specified model.Sim-ulation and application to data from Statistical society of Canada demonstrate the performance of those methods.