Bayesian Modeling of Follow-up Studies with Missing Data
Received 17 September 2008; accepted 9 January 2009. published online 20 April 2009.
Purpose
The purpose of this study is to illustrate the impact of ignoring missing data in follow-up studies and to provide a hierarchical Bayesian approach to simultaneously estimate rates and missing data probabilities.
Methods
To account for missing data in follow up studies, a hierarchical Bayesian procedure is proposed and investigated via simulation.
Results
A simulation study demonstrates the impact of ignoring missing data on inferences in terms of bias and in ranking populations in terms of risk. An example of rates of disabilities for various German construction worker professions also illustrates the usefulness of the method.
Conclusions
Use of a hierarchical Bayesian approach allows for flexible modeling of rates and data availability.
From the Department of Statistical Science, Baylor University, Waco (J.D.S., S.P.) and the Department of Biostatistics, University of Texas M. D. Anderson Cancer Center, Houston (N.B.), TX
Address correspondence to: James D. Stamey. Tel.: (254) 710-7405; fax: (254) 710-1669.
ABSTRACT