Program Information

Evaluation of Technology Using Probabilistic Decision Models

M Phillips*, Univ Washington, Seattle, WA

SU-E-T-221 Sunday 3:00:00 PM - 6:00:00 PM Room: Exhibit Hall

Purpose: Medical physicists are often asked to evaluate or choose appropriate
technology for clinical applications. These are multidimensional
problems that also suffer from different degrees of uncertainty in the
variables. Probabilistic decision models are a robust and
mathematically correct means of handling these issues. The principles
of constructing such models are presented along with practical
examples in the areas of IGRT, IMRT and proton therapy.

Methods: Influence diagrams are used to model the variables and their
uncertainties and include action and reward variables. Influence
diagrams are directed acyclic graphs that use Bayesian probability
calculus to propagate probabilities and to update prior probabilities
in the presence of evidence. An influence diagram was used to model
the question of whether brain tumors are better treated with x-ray
IMRT or proton therapy, with or without CT-guided localization. Data
for the conditional probabilities of the model were obtained from the
literature and included models of TCP, NTCP and induction of second
malignancies, as well as data on the probability density functions for
interfraction patient motion. Dosimetric data were obtained using the
CMS treatment planning system.

Results: Several different tumor types and sites were studied.
The critical variables in the model were identified and
studied using analyses of evidence, parameters and value of
information. The impact of imaging was significant, regardless of the
radiation type. The models used in determining some of the
conditional probabilities parameters also played an important role in
ranking alternatives.

Conclusions: Although such choices are difficult, physicists must
proceed with the best data at hand. Without a rigorous framework on
which to build a model of the process, decisions are likely to be based on
unstated assumptions and incorrect inference. The example of
comparing irradiation modalities for brain tumors shows the power of
influence diagrams in this critical context.

Contact Email