Research Projects
1. Mixture modelling of SROC
curves in meta-analysis of diagnostic studies
Meta-Analysis (MA) deals with the analysis and integration of empirical findings
from different and primarily independent studies examining the same research
question. This proposal considers the MA of diagnostic studies based on use of
the summary receiver operating characteristic (SROC) curve. It is inarguable
that diagnostic procedures – as any other medical procedure – need to undergo
continuous review and evaluation. Here, as in other areas of medicine, MA can
be very helpful. The benefit of the SROC is that it allows global
characterization of the diagnostic quality of each diagnostic test. However,
current modeling procedures of SROC lack ability to address important questions
in MA (such as heterogeneity) in a compact way. The nature and constraints of
the problem suggest that the Lehmann family is appropriate as a basic model for
the SROC curve as it is a simple, one-parametric family that fulfils the
restrictions of this curve. In this project, to cope with the well-acknowledged
problem of (latent) heterogeneity, we will develop a nonparametric mixture
model which works on the Lehmann family. The benefit of this approach is
manifold. Our method will estimate one, two or many SROCs that will be specific
for the studies involved in the MA. By considering the problem in a Bayesian
framework and using the nonparametric maximum likelihood estimate of the mixing
distribution as a prior, it will be possible to classify the studies into
associated SROC clusters, found by the mixture analysis. This will allow
potentially different interpretations of the found clusters with different
diagnostic accuracy. The approach developed in this project will be compared
with existing SROC techniques and will be illustrated using real examples.