|
Activity Number:
|
67
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #310055 |
|
Title:
|
Analyzing Correlated Data with Partial and Small Clusters
|
|
Author(s):
|
Xianqun Luan*+ and Avital Cnaan
|
|
Companies:
|
The Children's Hospital of Philadelphia and The Children's Hospital of Philadelphia
|
|
Address:
|
34th St and Civic Ctr Blvd, Philadelphia, PA, 19104,
|
|
Keywords:
|
Multiple outputation ; Generalized estimating equations ; within-cluster resampling
|
|
Abstract:
|
Studies in clinical research may have a setup where only a small proportion of the study sample has clusters, and those clusters are small. For example, studies in children when all siblings within an age range are in the study. One may use correlated analysis techniques: mixed effects models or generalized estimating equations. However, these approaches generally assume that most of the data are clustered, and that the clusters mostly are of size n>2. An alternative approach is multiple outputation or within-cluster resampling. We compare these various methods with simulation studies for binary outcomes. We vary: percentage of clustered data; size and patterns of the actual clusters; correlation within the cluster; and proportion of successes under H0 and H1. We explore cluster size correlated with outcome. We define guidelines of when to use which approach based on these simulations.
|
