|
Activity Number:
|
476
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #305777 |
|
Title:
|
Multiple Imputation Compared with Single Imputation Methods in the Analysis of Observational Data with Incomplete Covariate Information
|
|
Author(s):
|
Sunni A. Barnes*+ and Dunlei Cheng and David Nicewander and Yahya Daoud
|
|
Companies:
|
Baylor Health Care System and Baylor Health Care System and Baylor Health Care System and Baylor Health Care System
|
|
Address:
|
8080 North Central Expressway, Dallas, TX, 75206,
|
|
Keywords:
|
imputation ; observational study ; propensity score
|
|
Abstract:
|
Propensity score methodology is commonly used in observational studies to account for baseline differences between treatment groups. When covariate information is missing, then the choices are to use only those patients with fully observed data in the analysis or use some imputation method in order to calculate a propensity score for all patients. However, to this date there is little empirical evidence looking at the impact of different imputation methods on propensity score matching. We will present a simulation study where we compare and contrast single and multiple imputation methods under a variety of assumptions and levels of missing data. After imputations, propensity scores are matched based on a greedy matching technique with caliper set as 0.2. Pros and cons of each imputation method are examined through standardized difference values.
|
