|
Activity Number:
|
132
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #304377 |
|
Title:
|
Modeling Prior Knowledge in Developing a Bayesian Network
|
|
Author(s):
|
Futoshi Yumoto*+ and Rochelle E. Tractenberg
|
|
Companies:
|
American Institutes for Research and Georgetown University Medical Center
|
|
Address:
|
1000 Thomas Jefferson Street, NW, Washington, DC, 20007,
|
|
Keywords:
|
bayesian network ; directed graphs ; longitudinal modeling ; collaboration
|
|
Abstract:
|
Bayesian networks (BN) are a class of directed graphs that model information that can feed into decisions, based on data/prior knowledge. A BN models state-based problems; the result of the modeling is a better understanding, given the variables and data that the network represents, of outcomes that are of greatest interest. This poster describes a five-step procedure for building a BN (identify/summarize variables for network; model change in continuous variables over time; translate model of continuous change into categorical-changes-in-variables over time; build the network; test the network) in an example BN to identify 'clinically meaningful' changes in the severity of Alzheimer's disease (AD). The 5-step 'conservative' approach, requiring close collaboration between statistics & content experts, is contrasted with a more typical 'liberal' approach involving less intensive modeling.
|
