Multi-parameter Evidence Synthesis Research Group

Formal methods for multi-parameter evidence synthesis and their role in epidemiology and economic evaluation

We run two courses in collaboration with Keith Abrams, Alex Sutton, and Nicola Cooper from Leicester University.

We do not have courses scheduled at present and we are not taking bookings. However, if you wish to express an interest email mpes-admin@bristol.ac.uk, and we will give you early warning when we are taking bookings, and we will prioritise your registration.

If you have any difficulties downloading the files on these pages, please contact mpes-admin@bristol.ac.uk or S.Dias@bristol.ac.uk

This collaborative programme centred in Bristol and York attempts to develop and evaluate new statistical methods for complex evidence synthesis with applications in both epidemiology and medical decision-making. This can be characterised as a multi-parameter generalisation of meta-analysis, but broadened and extended so that it can combine not only evidence from different sources and different research designs, but also data relating to complex functions of parameters.

The approach builds on the Confidence Profile Method (CPM) of David Eddy and colleagues, but with some major differences. First, there is a commitment to develop richer, more complex, and more realistic applications than were attempted in the CPM literature; to develop robust statistical methods with standard software; to focus particularly on issues of evidence consistency - which have not been addressed in previous work; and to embed the programme firmly in the context of economic evaluation.

The central challenge is statistical: the need to develop models of a realistic level of complexity, estimated using all the data available, including data on arbitrarily complex functions of parameters. The issue of evidence consistency is a major emphasis of the programme, particularly when indirect evidence might be combined with information from direct ‘head to head’ comparisons on which most epidemiologists prefer to rely.

The impact of statistical evidence synthesis on decision-making and research prioritisation is a driving force behind the current work, and the rationale for the collaboration with Mark Sculpher and Karl Claxton at York. In a standard decision analysis, each model parameter would be informed independently. What is being advocated, however, calls on the decision analyst to incorporate evidence on model outputs as well as model inputs, and to incorporate more items of data, if available, than there are parameters. Neither requirement can be accommodated within standard decision analysis methodology. Further, if the purpose of collecting data is to reduce uncertainty in decision, then accurate estimation of uncertainty must be a key objective of decision analysis. The decision-making context therefore generates a clear rationale for multi-parameter evidence synthesis, and links the programme to Bayesian decision theory and Expected Value of Information analysis.

The programme has been funded in Bristol since July 2001 as part of the Medical Research Council Health Services Research Collaboration. Following the MRC's Quinquenial review in May 2002, additional funding for the sister programme in York was agreed. In October 2007 the programme was transferred to the University of Bristol. It is fully funded to March 2009.