Hierarchical Random-Effect Meta-Analysis of Binary Events Investigating the Relation between Treatment Effect and Underlying Risk
Year: 2014
Type: review
Abstract: Meta-analysis is a statistical method used to quantitatively synthesize findings from independent but related studies.It has become increasingly important in medical research.Using common moment-based meta-analytical approaches for studies with binary events, such as the Mantel-Haenszel method or the DerSimonian-Laird method, one would first choose a summary metric (e.g., relative risk, odds ratio or risk difference) and then "combine" across studies.The choice among different summary metrics is not straightforward given their different mathematical properties, ease of interpretation and variability across studies.We propose a flexible measure that allows estimation of the measure that "best" fits the data.The accuracy of the moment-based estimators has been questioned by many.Alternative approaches are the likelihood-based methods that model the binomial structure of the data.The difficulty with these approaches lies in the model specification.This dissertation examines current hierarchical models and proposes new methods for hierarchical model selection.By comparing conditional models to joint models, we conclude that the conditional approach that eliminates some nuisance parameters is convenient, but it neither agrees with other conditioning procedures nor captures the structure of the data as a full joint model with meaningful parameters does.5.1.2Implementation and Inference . . . . . . . . . . . . . . . . .
Author: Jingjing Yan
Institution: The Ohio State University
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W2243771056