Subgroup Identification Using the personalized Package

Abstract

A plethora of disparate statistical methods have been proposed for subgroup identification to help tailor treatment decisions for patients. However a majority of them do not have corresponding R packages and the few that do pertain to particular statistical methods or provide little means of evaluating whether meaningful subgroups have been found. Recently, the work of Chen, Tian, Cai, and Yu (2017) unified many of these subgroup identification methods into one general, consistent framework. The goal of the personalized package is to provide a corresponding unified software framework for subgroup identification analyses that provides not only estimation of subgroups, but evaluation of treatment effects within estimated subgroups. The personalized package allows for a variety of subgroup identification methods for many types of outcomes commonly encountered in medical settings. The package is built to incorporate the entire subgroup identification analysis pipeline including propensity score diagnostics, subgroup estimation, analysis of the treatment effects within subgroups, and evaluation of identified subgroups. In this framework, different methods can be accessed with little change in the analysis code. Similarly, new methods can easily be incorporated into the package. Besides familiar statistical models, the package also allows flexible machine learning tools to be leveraged in subgroup identification. Further estimation improvements can be obtained via efficiency augmentation.