Abstract
BackgroundDisadvantages have already been pointed out on the use of odds ratio (OR) as a measure of association for designs such as cohort and cross sectional studies, for which relative risk (RR) or prevalence ratio (PR) are preferable. The model that directly estimates RR or PR and correctly specifies the distribution of the outcome as binomial is the log-binomial model, however, convergence problems occur very often. Robust Poisson regression also estimates these measures but it can produce probabilities greater than 1.ResultsIn this paper, the use of Bayesian approach to solve the problem of convergence of the log-binomial model is illustrated. Furthermore, the method is extended to incorporate dependent data, as in cluster clinical trials and studies with multilevel design, and also to analyse polytomous outcomes. Comparisons between methods are made by analysing four data sets.ConclusionsIn all cases analysed, it was observed that Bayesian methods are capable of estimating the measures of interest, always within the correct parametric space of probabilities.Electronic supplementary materialThe online version of this article (doi:10.1186/s12982-015-0030-y) contains supplementary material, which is available to authorized users.
Highlights
Disadvantages have already been pointed out on the use of odds ratio (OR) as a measure of association for designs such as cohort and cross sectional studies, for which relative risk (RR) or prevalence ratio (PR) are preferable
In terms of significance of predictors, the methods diverged only on Pulmonary Artery Systolic Pressure (PASP), which was considered significant by robust Poisson regression and not significant by the Bayesian method
The Bayesian approach was presented in this paper as a unified way to estimate relative risk for situations with binary outcomes and dependent or independent data and for polytomous outcomes in independent data
Summary
Disadvantages have already been pointed out on the use of odds ratio (OR) as a measure of association for designs such as cohort and cross sectional studies, for which relative risk (RR) or prevalence ratio (PR) are preferable. Much has been discussed about disadvantages of the odds ratio (OR) as a measure of association in cross-sectional studies, cohort studies and clinical trials [1,2,3,4,5,6,7,8], and instead of it, relative risk (RR) or prevalence ratio (PR) according to the study design are proposed. Donner [22] and Yelland et al [23] proposed Poisson and log-binomial regression models estimated with Generalized Estimating Equations (GEE), and compared them through simulation. They both verified that the GEE logbinomial model may have convergence problems. Yelland et al [24] noted that there may be convergence problems when using the mixed-effects log-binomial model and that solutions for this problem are lacking
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
