Personalized Policy Learning Using Longitudinal Mobile Health Data

• Activity Number: 299 - Machine Learning in Causal Inference with Applications in Complicated Settings
• Type: Topic Contributed
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Biometrics Section
• Author(s): Min Qian* and Xinyu Hu and Bin Cheng and Ken Cheung
• Companies: Columbia University and Uber AI Lab and Columbia University and Columbia University
• Keywords: endogenous variables; individualized decision rule,; contextual bandits; generalized linear mixed model

Abstract

Motivated by the fact that in mobile health applications, a person's response to intervention may depend on his/her own circumstances at a certain moment that is difficult to capture or measure, we aims to develop personalized policies, one for each user, to optimize immediate outcomes. The proposed method can be viewed as a generalization of stochastic contextual bandits, which estimates personalized policies through random effects under a generalized linear mixed model with a group lasso type penalty to avoid overfitting of individual deviations from the population model. We examine the conditions under which the proposed method work in the presence of endogenous time-varying covariates, and provide conditional optimality and marginal consistency results of the estimated policies. We apply our method to develop personalized push (``prompt'') schedules in 294 app users, with a goal to maximize the prompt response rate given past app usage and other contextual factors.