|
Activity Number:
|
408
- SPAAC Poster Competition
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Scientific and Public Affairs Advisory Committee
|
|
Abstract #330203
|
|
|
Title:
|
Online Local Q-Learning
|
|
Author(s):
|
Lili Wu* and Eric Laber
|
|
Companies:
|
NCSU and North Carlina State University
|
|
Keywords:
|
Dynamic treatment regime;
Reinforcement learning;
Local linear regression
|
|
Abstract:
|
Advances in mobile computing technology have made it possible to monitor a patient's health status in real-time and then apply interventions. A dynamic treatment regime formalizes online treatment selection as a sequence of functions mapping up-to-date patient information to a recommended treatment per time point. Estimation of optimal dynamic treatment regimes which maximize expected patient utility over the treatment period has been studied extensively with a small number time points. Existing methodologies rely heavily on pooling across patients and estimated regimes are tailored to potentially coarse subgroups in the population. With mobile technologies, treatment can be adjusted to a large and indefinite number of time points which allows for pooling within a patient and thereby more deeply personalized treatment recommendations. We propose a variant of Q-learning that pools information across patients when the number of time points is low and then adaptively becomes more tailored to individuals with increasing number of time points. We illustrate the proposed method using a suite of simulation experiments and application to mobile-health interventions for type I diabetes.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2018 program
|
