Activity Number:
|
428
|
Type:
|
Contributed
|
Date/Time:
|
Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Biometrics Section
|
Abstract - #306243 |
Title:
|
Handling Missing Data for Smoking Cessation with Bootstrap, Trees, and Multiple Imputation
|
Author(s):
|
Jeff Thostenson*+ and Lowell C. Dale and Darrell Schroeder and Heike Hofmann
|
Companies:
|
University of Arkansas for Medical Sciences and Mayo Clinic College of Medicine and Mayo Clinic College of Medicine and Iowa State University
|
Address:
|
4301 W. Markham Street, # 781, Little Rock, AR, 77205-7199,
|
Keywords:
|
missing data ; multiple imputation ; bootstrap ; classification trees
|
Abstract:
|
Due to incomplete follow-up data, smoking intervention studies can suffer missing outcomes. The current accepted approach for missing outcomes assumes treatment failure (i.e. smoking) in place of missing outcomes, with secondary analyses excluding incomplete cases. Analysis of a pilot study to assess the efficacy of a tobacco quitline to increase abstinence from smoking following an initial consultation at the Mayo Clinic Nicotine Dependence Center suffered missing data problems. Due to differing rates of missing data in the study arms, the two analyses produced conflicting results. Multiple Imputation (MI) analyses gave unreasonable predictions. A method of bootstrapping classification trees gave more believable predictions. Thus that prediction method was used as the imputation step of a new MI analysis, giving more reasonable predictions and using Rubin's Rules to assess variability.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2006 program |