Analysis of Clustered Binominal Data Using Logistic-GLMMs

• 발행기관: 한국자료분석학회
• 저자명: Geon-Ho Cho
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.11 No.1, 33~41쪽, 전체 9쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2009.02.28

영문 초록

Correlated or heterogeneous response data which are usually clustered by trials, hospitals, families or subjects can be analyzed using random-effect models including generalized linear mixed models(GLMMs). In this paper we propose how to model the clustered meta-binominal data using logistic-GLMMs and to fit the models via SAS PROC NLMIXED. Here, the correlated random effects are allowed to model the variations in both the odds ratio and the treatment effect across trials, which can be at least two possible sources of inter-trial heterogeneity in outcomes. The proposed method is illustrated with a meta-analysis data set on 22 trials for respiratory tract infections. We also demonstrate that an appropriate model can be confirmed via model-selection criteria. From the final model we find that there is little difference in the treatment effects across trials and the treatment is shown to be effective, while there appears to be substantial variation in the log-odds ratio for infections across trials.

목차

1. Introduction

2. Example : Respiratory Infection Data

3. Model Estimation

4. Analysis

References

키워드

Generalized linear mixed models Meta-analysis Random effects Random treatment-by-trial interaction Trial effect