[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: GEE for relative risk regression models

From   Raoul Reulen <[email protected]>
To   <[email protected]>
Subject   st: GEE for relative risk regression models
Date   Thu, 24 Jan 2008 17:01:45 +0000

Dear all,


I've got a cohort of pregnant females for whom I want to investigate the risk of being exposed to radiation on miscarriage.  One group has been exposed to radiation before their pregnancy the other not. The unit of analysis is the pregnancy and the outcome is miscarriage. There can be multiple pregnancies in the cohort of the same female. I could use logistic regression with a generalized estimating equation (GEE) modification to take into account the clustering of pregnancies of the same woman, like this:


. xtlogit miscarriage radiation , pa corr(exch) i(indexno) robust eform


And this would give me Odds Ratios. However, because the outcome is common in this cohort (>10%) these odds ratios cannot be interpreted as relative risks.  I can calculate relative risks by using log-binomial regression or a Poisson regression model with a robust error variance. These two methods have been described nicely here:

. glm miscarriage radiation, nolog fam(binom) link(log)   eform 

. glm miscarriage radiation, nolog fam(poisson) link(log) eform robust


However, how do I take into account the clustering of pregnancies of the same woman?  Can I use GEE with these two models? If so how? Thanks. 



Raoul Reulen
Cancer Research UK Graduate Training Fellow
Centre for Childhood Cancer Survivor Studies
Department of Public Health & Epidemiology
University of Birmingham

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index