Building Multistage Treatment Policy for Depression after Acute Coronary Syndrome Project Summary Depression is not only commonly observed among patients who experienced acute coronary syndrome (ACS), but also has been shown to increase risks for recurrent ACS and mortality. Despite its high prevalence and serious impact, management of post-ACS depression remains poor because of ineciencies in depression screen- ing, limited treatment options of depression after ACS, and lack of e ective procedure if initial treatment fails. To address these issues, clinical researchers have tried to develop personalized stepped care procedures for post-ACS depression patients; this involves o ering patients the choice of receiving psychotherapy and/or antidepressant treatment and adjusting treatment as needed. The treatment decisions are usually based on patient demographics, treatment preference, medical history, progress of disease, and comorbid conditions. With the development of modern technologies, the number of available treatments increases, and more pa- tient information are collected in clinical research. Thus excavating useful information for treatment decisions is becoming more challenging. In this project, we propose to develop a principled way to construct simple interpretable multistage treatment policies from high-dimensional data, that can be used to guide treatment selection throughout the course of the disease.
Aim 1 of the project is devoted to the development of vari- able selection methodology for constructing multistage treatment policies using statistical machine learning techniques. The proposed research seeks to incorporate the popular variable selection technique (LASSO) into existing treatment policy search approaches, namely Q-learning and A-Learning, for developing optimal treatment policies and for identifying patient response status to initial treatment { an important factor for tailoring treatment in the subsequent stages.
Aim 2 evaluates the proposed methods, applies the methods to post-ACS depression data, and addresses some computational challenges. Statistical research in this area has been focused on the development of evidence-based treatment policies using pre-chosen models and variables; few if any discuss how to select models or variables in a principled way. The proposed work aims to ll this gap in methodology using modern machine learning techniques.

Public Health Relevance

s The proposed research aims to answer the following question: How to excavate simple interpretable multistage treatment policies from high-dimensional, longitudinal medical data in a principled way?' This is a crucial step in the management of chronic disease, such as depression, for which a large number of variables are collected over time, by facilitating the construction of a parsimonious clinical decision system.

National Institute of Health (NIH)
National Institute of Mental Health (NIMH)
Exploratory/Developmental Grants (R21)
Project #
Application #
Study Section
Biostatistical Methods and Research Design Study Section (BMRD)
Program Officer
Rupp, Agnes
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Columbia University (N.Y.)
Biostatistics & Other Math Sci
Schools of Public Health
New York
United States
Zip Code
Hu, X; Hsueh, P-Y S; Chen, C-H et al. (2018) An interpretable health behavioral intervention policy for mobile device users. IBM J Res Dev 62:
Wang, Huixia Judy; McKeague, Ian W; Qian, Min (2018) Testing for Marginal Linear Effects in Quantile Regression. J R Stat Soc Series B Stat Methodol 80:433-452
Hu, Xinyu; Hsueh, Pei-Yun S; Chen, Ching-Hua et al. (2017) A First Step Towards Behavioral Coaching for Managing Stress: A Case Study on Optimal Policy Estimation with Multi-stage Threshold Q-learning. AMIA Annu Symp Proc 2017:930-939
Qian, Min (2016) Comment. J Am Stat Assoc 111:1538-1541