Meta-Analysis of Diagnostic Test Data: Modern Statistical Approaches

Abstract

In the last decades, the amount of published results on clinical diagnostic tests has expanded very rapidly. The counterpart to this accelerated technological development has been the formal evaluation and synthesis of diagnostic results. However, published results can be regarded as so far removed from the classical domain of meta-analysis, that they can provide a rather severe test of classical methods. This work concerns the applications of computer intensive statistical methods in meta-analysis of diagnostic test data. These methods are considered from both the classical and Bayesian perspective. From the classical point of view, bootstrap methods are used to build confidence intervals for complex statistics in meta-analysis. These methods are evaluated extensively by a simulation experiment. Under the Bayesian perspective, a novel statistical model is presented. This model is general enough to include the presence of studies with different designs, unusual accurate results, large amounts of sparsity data, missing reporting data and heterogeneity between studies' population. These multiple sources of variability are modeled with a Bayesian graphical approach. In this approach a complex model is broken up into manageable sub-models. The full model is built up as a network by exploring the local dependency structure of each model component. A schematic description of this process is presented by Directed Acyclic Graph (DAG), which gives a non-algebraic structure and links computations to Markov chain Monte Carlo (MCMC) techniques. Statistical computations are implemented in open source and public domain statistical software (BUGS and R) and illustrated with a complex systematic review which evaluates the diagnostic performance of computer tomography scans in diagnostic of appendicitis.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 5 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.