Home Publications Research Teaching
In many health applications, the goal is to develop individualized treatment policies that specify which type and/or intensity of treatment should be offered over time. These treatment policies take patient information, such as demographics, preference, intermediate response, adherence as input and output treatment decisions at each decision point. Reinforcement learning (RL) is a cutting-edge area of machine learning concerned with how to take decisions in an environment to optimize an outcome of interest. My research in this area includes the development of novel RL algorithms and investigate their statistical properties to guide intervention decisions in health applications.
To inform clinical decisions or guide future research, results from statistical research need to be reproducible and generalizable. Statistical inference addresses this problem by providing confidence intervals or p-values for the parameters of interest. However, inference is challenging in the presence of non-regularity (e.g., when a nuisance parameter is only defined under an alternative hypothesis, when the parameter of interest under the null hypothesis is on the boundary of the parameter space, post-model-selection inference, etc), in which case standard approaches fail. I have developed statistical methodologies for non-regular inference in various settings.
SMART is a trial design specifically created for comparing sequential treatment policies. The SMART design consists in a sequence of randomizations corresponding to the sequence of decision points in a treatment policy. In this way, SMARTs enable causal comparisons among the policies “embedded” in the trial. In addition, investigators are often interested in using additional information concerning potential moderators collected as part of the SMART study to explore ways to tailor the treatment more deeply. My work in this area includes clinical trial design and analysis methods for SMART studies.
Mobile health is a general term referring to the use of mobile technologies for health management. With the increasing utilization of mobile devices such as smartphones and wearables, there is a great potential for delivering behavioral intervention via mobile technologies as part of a portfolio of available mental health resources. My research in this area focuses on the development of evidence-based personalized mobile clinical decision support systems. More information can be found on our working group website “Research On Adaptive Designs and Mobile Application Platforms (ROADMAP)”.