Seyedeh Solmaz Talebi
1, Kazem Mohammad
1, Aliakbar Rasekhi
2, Mohammad Ali Mansournia
1*1 Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran
2 Department of Biostatistics, Faculty of Medical Sciences, Tarbiat Modares University, Tehran, Iran
Abstract
Longitudinal studies are very common in medical,
behavioral, and interventional sciences. One measure of effect of interest in
longitudinal studies is risk ratio, naturally estimated by log-binomial
regression which suffers from convergence problem. Odds ratio does not well approximate
risk ratio when the outcome is common, so alternative methods have been introduced
in cohort studies with one follow-up visit. In this paper, we illustrate 2
simple methods, COPY method and modified log-Poisson regression for risk ratio
estimation in longitudinal data setting. Our unpublished simulation study on
risk ratio estimation in longitudinal data setting suggests that COPY method performs
well in terms of closeness of the risk ratio estimate and true risk ratio (mean
square error) and so we suggest this method for risk ratio estimation in
longitudinal Longitudinal studies are very common in medical, behavioral, and interventional sciences. One measure of effect of interest in
longitudinal studies is risk ratio, naturally estimated by log-binomial regression which suffers from convergence problems. Odds
ratio (OR) does not approximate risk ratio (RR) well when the outcome is common, so alternative methods have been introduced
in cohort studies with one follow-up visit. In this paper, we illustrate two simple methods: the COPY method and the modified
log-Poisson regression for RR estimation in longitudinal data setting. Our unpublished simulation study on RR estimation in
longitudinal data setting suggests that the COPY method performs well in terms of closeness of the RR estimate and true RR (mean
square error) and so we suggest this method for RR estimation in longitudinal data setting.data setting.