Committees
Important dates
Abstract
Submission
CLOSED
Acceptance
note
July
1
Reduced early
payment fee
July 15
|
Dynamic treatment regimes as a prediction problem
Elja Arjas
University of Helsinki and University of Oslo
Adequate medical treatment of many diseases, including different types of
cancer, involves a sequence of treatment assignments over time.
For an optimal allocation, each assignment of a new treatment should be
allowed to depend adaptively on how the patient responded to the ones that
were administered previously. The task of establishing a close-to-optimal
dynamic treatment regime of this type for patients with different individual
characteristics, based on the accrued follow-up information, offers several
challenges which are both conceptual and technical. The purpose of this talk
is to consider their solution from a Bayesian perspective, by combining
tools from constrained nonparametric modeling of stochastic processes and
their inference, and the consequent predictive distributions, with dynamic
programming for establishing the optimum. While the ideas underlying this
approach are general, practical considerations will often restrict their
direct applicability, and require that certain simplifying assumptions in
the modeling are employed. An example based on HIV data is considered as an
illustration.
|
|
