Susan Murphy, 2015 Wald Lecturer

Susan Murphy is the H.E. Robbins Distinguished University Professor of Statistics & Professor of Psychiatry and a Research Professor at the Institute for Social Research. Her research focuses on improving sequential, individualized, decision making in health, in particular on clinical trial design and data analysis to inform the development of treatment policies (also known as dynamic treatment regimes and adaptive interventions). She is a leading developer of the “Sequential Multiple Assignment Randomized Trial” design which is being used by clinical researchers to develop treatment policies across multiple health domains (e.g., depression, alcoholism, ADHD, substance abuse, HIV treatment, obesity, diabetes, and autism). Susan is currently working as part of several interdisciplinary teams to develop clinical trial designs and learning algorithms for settings in which patient information is collected in real time (e.g. via smart phones or other wearable devices) and treatments can be provided when and wherever needed. Susan is a Fellow of IMS, ASA and the College on Problems in Drug Dependence; she is a former editor of the Annals of Statistics, President-Elect of the Bernoulli Society, a member of the US National Academy of Medicine and a 2013 MacArthur Fellow.

Wald I:
Sequential Decision-Making and Personalized Treatment: The future is now!

Tuesday, August 11, 4:00pm, Convention Center Ballroom 6E

In the first Wald Lecture, Sequential decision making & personalized treatment: the future is now!, Susan will discuss new experimental designs for use in developing treatment policies in two broad areas: a) guiding expert sequential decision making and b) developing real-time treatment policies delivered via mobile devices. In the former area, each participant may be randomized two or three times, whereas in the latter area each participant may be randomized hundreds or thousands of times during the study. Both of these areas present a number of new challenges to the field of factorial experimental design as well as to the field of clinical trial design.

Wald II:
Offline Data Analysis Methods and Learning Algorithms for Constructing Mobile Treatment Policies

Wednesday, August 12, 10:30am, Convention Center Ballroom 6E

This second Wald Lecture will discuss estimation methods based on experimental or observational data for constructing real-time mobile health treatment policies. Over the last decades much of the research in this area has occurred outside of statistics, namely in the fields of reinforcement learning (an area of machine learning), operations research and in control engineering. As a result many aspects of present methods, in particular estimation methods most useful for health applications as well as inferential methods such as confidence intervals and testing have remained relatively undeveloped. This area presents new challenges for statisticians interested in analysis methods and inferential methods using large, complex data sets (large amounts of data are collected on each participant).

Wald III:
Continual, Online Learning in Sequential Decision-Making

Thursday, August 13, 10:30am, Convention Center Room 4C2

The last Wald Lecture will discuss methods and open problems in on-line personalization of a real-time treatment policy in mobile health. In particular, Susan will discuss proposals for basing the on-line personalization on a “warm-start” real-time treatment policy. These proposals involve the use of stochastic gradient ascent approaches to modifying the warm-start policy so as “personalize” the policy to the individual. She will discuss how learning algorithms may be constrained by scientific demands, the potential role of randomization not only for learning how to improve the warm-start but also for improving treatment effectiveness, how in this setting a learning algorithm is part of the definition of treatment and the challenges this raises. She will also discuss the issue of non-stationarity and the critical need for a new theoretical approach for developing and evaluating learning algorithms in non-stationary settings. This area presents new challenges to the field of statistical sequential analysis.