|
Activity Number:
|
551
- New Innovations in Handling Incomplete Biomedical Data in the Era of Data Science
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 2, 2017 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
International Indian Statistical Association
|
|
Abstract #321880
|
View Presentation
|
|
Title:
|
Handling Incomplete Correlated Continuous and Binary Outcomes in Meta-Analysis of Individual Participant Data
|
|
Author(s):
|
Manuel Gomes and Laura Anne Hatfield* and Sharon-Lise Normand
|
|
Companies:
|
London School of Hygiene and Tropical Medicine and Harvard Medical School and Harvard Medical School
|
|
Keywords:
|
joint modeling ;
individual patient data meta-analysis ;
comparative effectiveness research ;
missing data ;
multiple imputation ;
Bayesian analysis
|
|
Abstract:
|
Comparative effectiveness research frequently uses hierarchical models to combine individual patient data from multiple studies and estimate treatment effects. When treatments affect multiple health outcomes, the analysis approach should jointly model the outcomes to account for correlations at the individual- and study-level. Two features of real data complicate this modeling task: 1) missing outcomes and covariates and 2) different outcome measurement scales (e.g., continuous and discrete). We compare two approaches to overcome these analysis challenges: a full Bayesian joint model and a sequence of full conditional models. In simulations, we find that the joint model provides better interval coverage and lower mean squared error. We demonstrate the differences by applying the two approaches to meta-analysis of randomized controlled trials for implantable devices used to treat heart failure.
|
Authors who are presenting talks have a * after their name.
