|
Activity Number:
|
34
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 3, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #302272 |
|
Title:
|
Orthogonal Experimental Designs Can Disentangle Confounding in Database Studies
|
|
Author(s):
|
Charlie H. Goldsmith*+ and Lehana Thabane and Gary Foster and Eleanor M. Pullenayegum
|
|
Companies:
|
McMaster University and McMaster University and St. Joseph's Healthcare Hamilton and St. Joseph's Healthcare
|
|
Address:
|
Biostatistics Unit, Martha H322, Hamilton, ON, L8N 4A6, Canada
|
|
Keywords:
|
Orthogonal designs ; confounding ; databases ; bootstrapping
|
|
Abstract:
|
To determine whether confounding in databases studies can be disentangled using orthogonal factorial designs. Factorial designs permit estimation of interactions and main effects. Replicates of cases were sampled according to a bootstrap scheme to create a factorial design, which was then used to estimate the impact of main effects and interactions. All estimable main effects and interactions are combined like imputation to provide an analysis of the factors and interactions better than with the analysis of the database because of confounding. Ideas will be shown using an osteoporosis database, a 2**3 complete factorial, 10 replicates and bootstrapping. A comparison with the usual analysis, feasibility and interpretations of these bootstrapped effects will be shown. Orthogonal designs created via bootstrapping can disentangle interactions between factors in database studies.
|
