• University of Surrey Featured PhD Programmes
  • University of Bristol Featured PhD Programmes
  • Queen’s University Belfast Featured PhD Programmes
  • University of Stirling Featured PhD Programmes
  • University of Glasgow Featured PhD Programmes
  • Northumbria University Featured PhD Programmes
  • University of Southampton Featured PhD Programmes
  • University of Manchester Featured PhD Programmes
University of Southampton Featured PhD Programmes
Engineering and Physical Sciences Research Council Featured PhD Programmes
University of Leicester Featured PhD Programmes
FindA University Ltd Featured PhD Programmes
Lancaster University Featured PhD Programmes

Complex treatment by covariate interactions in network meta-analysis

This project is no longer listed in the FindAPhD
database and may not be available.

Click here to search the FindAPhD database
for PhD studentship opportunities
  • Full or part time
    Dr S Donegan
    Prof C Tudur Smith
  • Application Deadline
    Applications accepted all year round
  • Self-Funded PhD Students Only
    Self-Funded PhD Students Only

Project Description

For most diseases, many treatments exist. Network meta-analysis (NMA) can estimate the relative effects of all treatment pairings even when treatments are not compared in the same trial. Therefore, NMA has huge potential because it is useful for all clinical fields.

It is common to explore treatment by covariate interactions in meta-analyses. Interactions can be included in an NMA model to evaluate whether each treatment effect varies with a covariate (e.g. a patient or methodological characteristic, such as, disease severity or allocation concealment).

The benefits of including interactions in NMA can be substantial. The model can produce the relative effects of all treatment pairings for each covariate value. For example, including an interaction for disease severity (i.e. severe or non-severe) could give one set of the relative effects for patients with severe disease, and another set for patients with non-severe disease. This allows different recommendations to be made for different patient groups; personalising treatment in this way can benefit patients. For example, for the treatment of epilespy, sodium valproate is recommended for patients with generalized seizures whereas carbamazepine is advised for patients with partial seizures. Furthermore, when heterogeneity (i.e. variability across trials) or inconsistency (i.e. variability between direct and indirect evidence) is detected in the NMA without interactions, including interactions provides the opportunity to assess whether the covariate reduces this variabiltiy; this can help analysts understand how best to analyse data and draw valid clinical inferences.

Currently, research has focused on including interactions for a single covariate (e.g. disease severity), rather than numerous covariates simultaneously (e.g. disease severity and dose). However, it is unlikely that one covariate would cause all heterogeneity or inconsistency if it exists, but instead several covariates would contribute. Therefore, when the purpose of including interactions is to explain variability, it seems sensible to include multiple covariates simultaneously.

Additionally, current NMA publications explore linear interactions, rather than non-linear interactions. A non-linear interaction is observed when the graph of treatment effect versus a covariate is not a straight line. For example, a review showed that the graph of relative risk of mortality versus BMI was a j-shaped curve. In such cases, if a linear interaction is fitted, the analyst may fail to detect that an interaction exists and this could lead to incorrect clinical guidance.

The overarching aim of the project is to develop methodology for including multiple, complex treatment by covariate interactions in NMA. The student could develop NMA models including interactions for several covariates and non-linear interactions; highlight the underlying modelling assumptions; develop methods to assess the assumptions; and demonstrate methods using real and/or simulated individual patient data and aggregate data.

Person specification

The successful candidate is likely to hold a 1st or 2:1 degree in a relevant discipline (statistics or mathematics). A Master degree in Statistics would be desirable. An understanding of statistical models is essential. Experience of coding in a statistical package (e.g. R, SAS, stata, Winbugs) is desirable.

Applicants should send a CV, academic transcripts, a letter of motivation and two names of referees who can send letters of recommendation to Nyree Collinson [email protected]

Funding Notes

This is a self funded opportunity.

Cookie Policy    X