Network Meta-Analysis for Decision-Making in WinBUGS

Course dates

Online Course Taught Over a Two-Week Period 12th – 21st September 2022

 

For more details see the full course programme.

Overview

This course is for health economists, statisticians, and decision modellers interested in the extension of pair-wise meta-analysis to network meta-analysis (NMA)1. The course is appropriate for anyone involved in health technology assessment who has a strong quantitative background.

The course focuses on Bayesian methods for statistically combining evidence from trials that form a network of treatment comparisons. The assumptions underlying both pair-wise meta-analysis and NMA are critically examined. The course also covers methods for detecting and managing heterogeneity and inconsistency, and how to embed the NMA in a cost-effectiveness analysis.

The Bayesian Markov chain Monte Carlo package WinBUGS is used on the course, and delegates who are new to WinBUGS should, when they have completed the course, be able to understand and code WinBUGS analyses of both NMA and other simple hierarchical models. However, the concepts learned on the course are entirely applicable to other Bayesian software, including OpenBUGS, JAGS and STAN.

This is an informal, hands-on, online course, based on a mixture of pre-recorded lectures, practical exercises with pre-recorded solutions, discussion boards, and live Q&A with practical support sessions. In the live practical support sessions participants will be able to access help from course tutors to complete the exercises in WinBUGS.

The course is a collaboration between the Department of Population Health Sciences, University of Bristol, the Centre for Reviews and Dissemination, University of York, and the Department of Health Sciences, University of Leicester.

The methods taught on the course are designed to be fully compatible with the NICE Guide to the methods of technology appraisal 2013 and the NICE Decision Support Unit Technical Support Documents on Evidence Synthesis, and with the Report of the ISPOR Task Force on Indirect Comparisons.

Participants will receive a free electronic copy of the 2018 book Network meta-analysis for decision making. Material from Chapters 1-4, 7-8 and 12 will be covered on this course.

1 NMA has also been called: mixed treatment comparisons, and multiple treatment meta-analysis; indirect treatment comparisons are another type of NMA.

What you will learn

By the end of the course participants will be able to:

  • conduct pair-wise, indirect comparison and network meta-analysis using WinBUGS Bayesian software;
  • adjust for covariates;
  • integrate statistical evidence synthesis with probabilistic cost-effectiveness analysis;
  • assess the degree of heterogeneity and inconsistency in RCT data;
  • conduct network meta-analysis for binary, continuous, and hazard ratios outcomes
  • understand the assumptions and potential pitfalls in Pair-wise and Network MA;
  • participants will acquire an introductory understanding of Bayesian methods, hierarchical modelling, and be able to use WinBUGS software.

Who is this course for?

The course is suitable for (1) health economists undertaking health technology assessments, including in the context of cost-effectiveness analysis; (2) statisticians, familiar with the principles of meta-analysis, who wish to learn about Bayesian methods for evidence synthesis; and (3) researchers synthesising evidence from reviews of multiple treatments who have a strong quantitative background and experience of statistical or other coding.

The course is popular with those who wish to use widely available NMA WinBUGS codes, and want to make sure they understand what the code is doing, the assumptions being made, and how to make changes to the code and interpret the outputs. Senior managers who are involved in guideline development or health technology assessment, who supervise the work of staff doing the “hands-on” work, but who will not be carrying out analyses themselves, may also wish to take the course, to become better informed about the many kinds of analyses that can be done, the assumptions being made and how to judge if they are met, and what kinds of checks and quality assurance should be carried out.

Course pre-requisites

Participants should:

  •  be familiar with the basic principles of meta-analysis and have a good working knowledge of logistic regression and statistical interaction
  • have experience programming in standard languages or statistical software
  • ideally have experience with probabilistic decision analysis in cost-effectiveness analysis (although this is not essential)

Before the course all participants must:

  1. Download WinBUGS 1.4.3 from the BUGS Project website (https://www.mrc-bsu.cam.ac.uk/software/bugs/the-bugs-project-winbugs/ ) onto the computer or laptop that you will be using during the course. The best way to do this is to download a zipped version of the whole file structure and unzip it into Program Files or wherever you want it. This includes the key for unrestricted use. Then, download and install the patch for version 1.4.3. To check that you have the patched version: open WinBUGS, go to the “Help” menu and click “About WinBUGS”. This should show the current version as 1.4.3. The software, patch and key are free.
  2. For those using a Mac: To run WinBUGS on a Mac you will first need to install a Windows emulator (eg WINE). You can then install WinBUGS on that and it should work. For more precise instructions please check for advice online.
  3. Make sure you run the example in the Tutorial.  Go To "Help", User Manual, then click on "Introduction".
  4. Try to run the BLOCKER example in the online Help, (Examples Vol 1_Blocker).  NB you need to use initial values Inits2 for this to run.
  5. Delegates will also need calculators or spreadsheet software for some of the course exercises.
  6. Note there are some optional sessions using/demonstrating R and MetaInsight. Those interested in the optional practical in R should install a recent version of R.

Note on OpenBUGS: We strongly recommend that you use WinBUGS 1.4.3 during the course. Participants wishing to use OpenBUGS should be aware that although the basic code should run, some files may not open properly and are advised to have a copy of WinBUGS 1.4.3 on their laptops to deal with such files. 

The course is suitable for (1) health economists undertaking health technology assessments, including in the context of cost-effectiveness analysis; (2) statisticians, familiar with the principles of meta-analysis, who wish to learn about Bayesian methods for evidence synthesis; and (3) researchers synthesising evidence from reviews of multiple treatments who have a strong quantitative background and experience of statistical or other coding.

The course is popular with those who wish to use widely available NMA WinBUGS codes, and want to make sure they understand what the code is doing, the assumptions being made, and how to make changes to the code and interpret the outputs. Senior managers who are involved in guideline development or health technology assessment, who supervise the work of staff doing the “hands-on” work, but who will not be carrying out analyses themselves, may also wish to take the course, to become better informed about the many kinds of analyses that can be done, the assumptions being made and how to judge if they are met, and what kinds of checks and quality assurance should be carried out.

Course format

This is an informal, hands-on online course, based on a mixture of lectures and practical work using published datasets. The course will be taught online in 7 sessions over a two-week period using a mixture of pre-recorded lectures, practical exercises with pre-recorded solutions, discussion boards, live Q&A, and practical support sessions. Participants can work through the materials in their own time, post queries to the discussion board, and attend the live sessions for discussions on the materials covered in specific sessions. In the live practical support sessions participants will be able to access help from course tutors with the exercises in WinBUGS. All materials will be made available to participants 1 week in advance of the Q&A and practical sessions. There will be an opportunity to book a short slot with a tutor for advice on participants own evidence synthesis project.

Live Q&A and Practical Support Sessions: 2pm – 4:30pm BST

Mon 12th, Tue 13th, Wed 14th, Thurs 15th,  Mon 19th, Tue 20th, Wed 21st September

Time commitment

We expect that the time taken to complete the core material is between 23 - 28 hours depending on participants experience. There are approximately 3 hours of further optional material.

Course programme

The course covers the following sessions.

SESSION 1: Introductions: NMA, Pairwise M-A, and Bayesian Inference

  • Introduction to Indirect Comparisons and Network Meta-Analyses
  • Pairwise Meta-Analysis: A Refresher
  • Introduction to Bayesian Inference

SESSION 2: Bayesian Meta-Analysis in WinBUGS

  • Bayesian Meta-Analysis in WinBUGS: binary outcomes

SESSION 3 NMA in WinBUGS

  • Network meta-analysis in WinBUGS: binary outcomes
  • **OPTIONAL** Multivariate Normal code for studies with 3 or more arms
  • Assessing Model Fit in Bayesian Models
  • **OPTIONAL** Outlier Detection Using Cross Validation

SESSION 4: Integrating NMA and Decision Modelling, and Streamlining Analyses

  • Integrating NMA and Decision Modelling
  • **OPTIONAL** Streamlining Analyses: R2WinBUGS
  • **OPTIONAL** Streamlining Analyses: MetaInsight

SESSION 5: Inconsistency and Meta-Regression

  • Inconsistency: methods and implications
  • **OPTIONAL** Node-splitting using gemtc in R
  • Incorporating Covariate Effects in pairwise meta-analysis

SESSEION 6: Other Outcomes

  • NMA with Continuous Outcomes
  • NMA with Hazard Ratios
  • Choosing the right code/model flow chart

SESSION 7: Further Topics

  • Network Meta-Regression
  • Modelling Treatment Class Effects
  • **OPTIONAL**: Further Topics
  • GUEST LECTURE: population adjusted indirect comparisons
  • **OPTIONAL** Book a session to discuss participants own evidence synthesis project

Course faculty

Tony Ades PhD, Professor of Public Health Science, originated the programme in Multi-Parameter Evidence Synthesis in epidemiology and decision making at the University of Bristol. His particular interests are in epidemiological applications, particularly in screening and infectious disease. He is a past member of a NICE Technology Appraisals committee.

Nicola Cooper PhD is a Professor of Healthcare Evaluation Research at the University of Leicester. Her primary research interest is the integration of statistical evidence synthesis with probabilistic economic decision modelling.

Sofia Dias PhD is a Professor of Health Technology Assessment at the University of York. She is a statistician with extensive experience of systematic review and meta-analysis. Her recent interests include using NMA evidence structures to estimate bias in RCTs, and developing methods for assessing inconsistency in NMAs. She is a member of a NICE Technology Appraisals committee.

Alex Sutton PhD is a Professor in Medical Statistics at the University of Leicester. He has a long-standing interest in methods for evidence synthesis; particularly when used for decision making. His current research includes development of interactive visualisation tools.

Nicky Welton PhD is a Professor in Statistical and Health Economic Modelling at the University of Bristol where she leads the Multi-Parameter Evidence Synthesis research group. She is director of the NICE Guidelines Technical Support Unit in Bristol, co-director of the NIHR Bristol Technology Assessment Group, and is a member of a NICE Technology Appraisals committee. Nicky is a statistician with extensive experience of evidence synthesis with different types of outcomes, and has a particular interest in the role of evidence synthesis in decision modelling, identifying future research priorities, and study design.

David Phillippo PhD is a Senior Research Associate in Evidence Synthesis at the University of Bristol. His research focuses on methods for evidence synthesis, Bayesian network meta-analysis, population adjustment methods for indirect comparisons, and assessing the impact of bias in clinical guidelines and decision making. He is the lead author of a recent Technical Support Document published by the NICE Decision Support Unit on population-adjusted indirect comparisons, providing guidance on the use of this new class of methods in NICE Technology Appraisals. He is a member of the NIHR Bristol Technology Assessment Group, and he supports the development of NICE Clinical Guidelines through his involvement with the NICE Technical Support Unit.

Hugo Pedder is a Senior Research Associate in Statistical Modelling at the University of Bristol. He is a member of the NIHR Bristol Technology Assessment Group and of a NICE Technology Appraisal Committee, and he supports NICE guideline development through his work with the NICE Guidelines Technical Support Unit. Hugo has experience implementing complex evidence synthesis techniques to a variety of different data types, and has worked on developing methods for the incorporation of dose-response and time-course information into NMA.

Bea Downing PhD is a Senior Research Associate in Evidence Synthesis and Biostatistics at the University of Bristol and Scientific Coordinator with the NICE Guidelines Technical Support Unit in Bristol. Her research interests include using evidence synthesis of indirect sources to bridge knowledge gaps and in communicating uncertainty.

 

Registration

Bookings are open

Make a booking

Fees

  • Student: £550*
  • Academic, public sector: £900
  • Commercial sector: £1,600

*Evidence of current student status will be required. If you have questions about this please contact the course administrator.

See our booking terms & conditions.

Contacts

For further information please email mpes-admin@bristol.ac.uk

For enquiries about course content, Nicky Welton Nicky.Welton@bristol.ac.uk or Sofia Dias Sofia.Dias@york.ac.uk

Edit this page