|
Activity Number:
|
211
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #304193 |
|
Title:
|
Comparison of Methods for Meta-Analysis with Sparse Events Incorporating Exposure Time
|
|
Author(s):
|
Yang Zhao and Hongwei Wang*+ and Arlene Swern
|
|
Companies:
|
Purdue University and Merck & Co., Inc. and Merck & Co., Inc.
|
|
Address:
|
RY34-A312, PO Box 2000, Rahway, NJ, 07065,
|
|
Keywords:
|
meta-analysis ; sparse event ; relative risk ; continuity correction ; exact Poisson ; exposure time
|
|
Abstract:
|
Routinely used methods for meta-analysis break down with sparse event dichotomous data. We conducted extensive simulations to evaluate different analytical methods with different continuity corrections for pooling the relative risk estimates that incorporates exposure time in the case of sparse events. The simulation study suggests that Mantel-Haenszel method, asymptotic Poisson and generalized linear mixed effect model behave similarly and perform fairly well in terms of bias, coverage of confidence interval (CI) and statistical power. The exact Poisson regression controls Type-I error rate and maintains coverage of CI while often being conservative. Large amount of continuity correction is not recommended which usually introduces bias and destroys coverage probability of CI. Marginal table method, inverse variance method with or without continuity correction are also to be avoided.
|
