With an ever-increasing number of diagnostic tools available, evidence on test accuracy (sensitivity and specificity) is critical. Often multiple, but typically small, studies have evaluated the same test. To inform decisions about use of tests in clinical practice, estimates of sensitivity and specificity are pooled across studies using meta-analysis methods.
Methods for meta-analysing the accuracy of a single test are well established. But these methods are inadequate for answering clinically important questions about which of multiple possible tests for a disease is “best” and about the accuracy of tests used in sequence or combination.
In practice, often a network of evidence on test accuracy is available, with different studies having evaluated one or more tests, by comparing test results with one or more reference standard or "gold standard" test. For evaluation and comparison of the effectiveness of interventions (e.g. pharmaceutical drugs), “network meta-analysis” (NMA) methods are now routinely used to simultaneously analyse all available data in such a network of evidence, to compare and rank competing interventions. A few NMA models for test accuracy have been suggested (1-6), but these methods are complex, highly disparate, and are very rarely used in practice yet.
Aims and objectives
- To compare and evaluate existing methods for network meta-analysis of diagnostic test accuracy
- To perform case study network meta-analyses of test accuracy
- To extend and/or improve statistical methods for NMA of test accuracy, working within a Bayesian statistical framework.
The ultimate aim is to develop flexible and applicable (e.g. reasonable run time) Bayesian statistical models to synthesise any network of evidence on test accuracy. Ideally, these models will also be applicable when none of the "reference standard" tests in the network is in fact perfect, i.e. there is no true gold standard.
This is an excellent opportunity for a student with strong statistical skills to train with internationally leading experts in Bayesian statistical methods for evidence synthesis. We will work within a Bayesian "multi-parameter evidence synthesis" (MPES) framework, in which multiple data sources are modelled together simultaneously, informing different functions of parameters. A key element of the project is development of flexible and user-friendly code and packages (using WinBUGS / JAGS / Stan called through R) to facilitate uptake of methods in practice.
The student will become an expert in meta-analysis methodology, fitting models using Bayesian statistical software, and interpreting and explaining measures of test accuracy. More broadly the models developed will fall within the framework of hierarchical or multilevel modelling, in order to account for heterogeneity across studies. To allow for the possibility of no "gold standard" test in the network, latent class analysis approaches may also be used.
How to apply for this project
This project will be based in Bristol Medical School - Population Health Sciences in the Faculty of Health Sciences at the University of Bristol.
Please visit the Faculty of Health Sciences website for details of how to apply