Bohning D., Rattanasiri S., Kuhnert R. Meta-analysis of Binary Data Using Profile Likelihood

Chapman and Hall/CRC, 2008. – 208 p. – ISBN: 1584886307, 9781584886303Providing reliable information on an intervention effect, meta-analysis is a powerful statistical tool for analyzing and combining results from individual studies. Meta-Analysis of Binary Data Using Profile Likelihood focuses on the analysis and modeling of a meta-analysis with individually pooled data (MAIPD). It presents a unifying approach to modeling a treatment effect in a meta-analysis of clinical trials with binary outcomes.
After illustrating the meta-analytic situation of an MAIPD with several examples, the authors introduce the profile likelihood model and extend it to cope with unobserved heterogeneity. They describe elements of log-linear modeling, ways for finding the profile maximum likelihood estimator, and alternative approaches to the profile likelihood method. The authors also discuss how to model covariate information and unobserved heterogeneity simultaneously and use the profile likelihood method to estimate odds ratios. The final chapters look at quantifying heterogeneity in an MAIPD and show how meta-analysis can be applied to the surveillance of scrapie.The occurrence of meta-analytic studies with binary outcome
Meta-analytic and multicenter studies
Center or study effect
Sparsity
Some examples of MAIPDs
Choice of effect measure
The Basic Model
Likelihood
Estimation of relative risk in meta-analytic studies using the profile likelihood
The profile likelihood under effect homogeneity
Reliable construction of the profile MLE
A fast converging sequence
Inference under effect homogeneity
Modeling Unobserved Heterogeneity
Unobserved covariate and the marginal profile likelihood
Concavity, the gradient function, and the PNMLE
The PNMLE via the EM algorithm
The EMGFU for the profile likelihood mixture
Likelihood ratio testing and model evaluation
Classification of centers
A reanalysis on the effect of beta-blocker after myocardial infarction
Modeling Covariate Information
Classical methods
Profile likelihood method
Applications of the modelAlternative Approaches
Approximate likelihood model
Multilevel model
Comparing profile and approximate likelihood
Analysis for the MAIPD on selective tract decontamination
Simulation study
Discussion of this comparison
Binomial profile likelihood
Incorporating Covariate Information and Unobserved Heterogeneity
The model for observed and unobserved covariates
Application of the model
Simplification of the model for observed and unobserved covariates
Working with CAMAP
Getting started with CAMAP
Analysis of modelingEstimation of Odds Ratio Using the Profile Likelihood
Profile likelihood under effect homogeneity
Modeling covariate information
Quantification of Heterogeneity in an MAIPD
The problem
The profile likelihood as binomial likelihood
The unconditional variance and its estimation
Testing for heterogeneity in an MAIPD
An analysis of the amount of heterogeneity in MAIPDs: a case study
A simulation study comparing the new estimate and the DerSimonian–Laird estimate of heterogeneity variance
Scrapie in Europe: A Multicountry Surveillance Study as an MAIPD
The problem
The data on scrapie surveillance without covariates
Analysis and results
The data with covariate information on representativenessDerivatives of the binomial profile likelihood
The lower bound procedure for an objective function with a bounded Hesse matrix
Connection between the profile likelihood odds ratio estimation and the Mantel–Haenszel estimator