|
Activity Number:
|
466
- Personalized/Precision Medicine I
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
|
Abstract #304504
|
Presentation
|
|
Title:
|
Robust Estimation for Optimal Dynamic Treatment Regimes with Restricted Arms Using Observational Data
|
|
Author(s):
|
Nina Zhou* and Lu Wang and Daniel Almirall
|
|
Companies:
|
University of Michigan and University of Michigan and University of Michigan
|
|
Keywords:
|
Dynamic Treatment Regime;
Restricted optimization;
Observational studies;
Tree-based reinforcement learning
|
|
Abstract:
|
As we enter the big data era, massive amounts of health records are available for scientific discoveries. Observational studies become one of the primary data sources for optimizing the dynamic treatment regime (DTR), which offers a capable vehicle for personalized medicine of chronic diseases with time-varying treatments. A natural challenge arises often since not all observed treatment routes are applicable for future patients. Traditional methods may lead to unrealistic recommendations in this situation and thus cannot be applied directly. Deleting the patient records with restricted arms is not applicable either due to selection bias. We develop a new tree-based method to estimate DTR with restricted arms from observational data in a multi-stage multi-treatment setting. Our algorithm provides sub-optimal treatment recommendations for patients among the restricted DTRs and utilizes all valid observations to avoid selection bias and improve efficiency. Since backward induction is utilized, partially valid observations from patients in the restricted arms are used for DTR estimation. The effectiveness and robustness of our proposed method are evaluated through simulation studies.
|
Authors who are presenting talks have a * after their name.
