Clinical Study Data Request Registered Users, Please Login

Proposal 988

Title of the Proposed Research

PhD Project: Bayesian Multivariate Network Meta-Analysis of Ordered Categorical Data

Lead Researcher

Fabrizio Messina


School of Health and Related Research (ScHARR)
University of Sheffield

Funding Source

The funding for the PhD project is provided jointly be the Centre for Bayesian Statistics in Health Economics (CHEBS) and the School of Mathematics and Statistics (SoMaS) at the University Of Sheffield.

Potential Conflicts of Interest


Data Sharing Agreement Date

09 October 2014

Lay Summary

In the UK, a health technology assessment (HTA) conducted on behalf the National Institute for Health and Care Excellence (NICE), involves a comparison of the population mean costs and population mean benefits quantified in terms of quality-adjusted life years (QALY) of two or more medical technologies. A health technology assessment is typically conducted using a decision analytic model that incorporates all relevant costs and consequences, and decisions regarding reimbursement are made based on the incremental cost-effectiveness ratio (ICER) or the net (monetary) benefit.

Inputs to decision analytic models are population parameters, although the true values of the parameters are unknown. Uncertainty about population parameters is represented by probability distributions, and the output uncertainty (i.e. uncertainty about the population mean costs and benefits) is derived by propagating the input uncertainty through the model. The need to characterise uncertainty about population parameters necessitates a Bayesian approach to synthesising all available evidence about treatment effects.

Evidence about population parameters can arise from multiple sources, including randomised controlled trials (RCTs). Outcome measures that comprise a set of different categories give rise to categorical data. When the categories have no natural ordering they are referred to as nominal, whereas when the categories are ordered the outcome measure is referred to as ordinal.

Ordered categorical data arise in many disease areas with the simplest example being a classification of the severity of a patients’ disease as None, Mild Moderate and Severe. Two outcome measures that are often used in the development of new treatments for rheumatoid arthritis are the EULAR (European League Against Rheumatism) response and the ACR (American College of Rheumatology) response.

A health technology assessment of Adalimumab, etanercept, infliximab, certolizumab pegol, golimumab, tocilizumab and abatacept was conducted for the treatment of patients with rheumatoid arthritis who were not previously treated with disease-modifying anti rheumatic drugs and after the failure of conventional disease-modifying anti-rheumatic drugs only. Whilst most trials provided information on ACR response, fewer provided information on EULAR response. Univariate network meta-analyses were performed separately for each outcome measure with a differing numbers of studies and different treatments being included in the analyses. However, it would have been advantageous to perform a multivariate network meta-analysis in order to include all studies in the analysis of each outcome and to borrow strength about correlation between outcomes from those studies that provided data on both outcomes.

Study Data Provided

Roche Study MRA009JP: A randomized, double-blind, parallel-group, comparison study of three doses (placebo, MRA 4 mg/kg, or MRA 8 mg/kg) in patients with rheumatoid arthritis
Roche Study MRA012JP: A phase III randomised, parallel-group study of MRA in patients with rheumatoid arthritis

Statistical Analysis Plan

This will be added after the research is published.

Publication Citation