Level: Professional (P)
This virtual course will introduce the Bayesian approach to meta-analysis. Attendees will learn practical ways in which they can combine multiple sources of published evidence while accounting for uncertainties such as response bias, publication bias, confounding, and missing information, using either BUGS, JAGS or Stan as software. With Bayesian models, this can be transparent and reproducible.
This course introduces the Bayesian approach to meta-analysis. Attendees will learn practical ways in which they can combine multiple sources of published evidence while accounting for uncertainties such as response bias, publication bias, confounding, and missing information, using either BUGS, JAGS or Stan as software. With Bayesian models, this can be transparent and reproducible.
This two-day course begins by reviewing classic meta-analysis methods and expressing them as statistical models. Once attendees understand meta-analysis is this larger context, they are able to extend the model flexibly to account for common problems such as papers that report only change from baseline. A series of problems will be tackled in this course, and attendees will leave with model code that they can immediately start using with their own projects.
After attending, participants will be able to:
- Write out standard meta-analyses as statistical models
- Use BUGS, JAGS or Stan to fit such models to data
- Recognise several common problems in meta-analysis
- Extend these models to account for these problems
- Understand and communicate their findings
- A review of statistical models of meta-analysis
- Introduction to Bayesian analysisProblems in meta-analysis, and sources of uncertainty
- Models for basic DerSimonian-Laird and Biggerstaff-Tweedie meta-analyses
- Introduction to Bayesian software options: BUGS, JAGS and Stan
- Models for network meta-analysis
- Models for missing statistics
- Models for reporting bias
- Models for publication bias
- Models for a mixture of statistics
- Models for a mixture of study types
- Reporting Bayesian meta-analyses
This course will be of interest to evidence-based healthcare researchers, including those writing guidelines and evaluating policies. Attendees should be comfortable conducting simple meta-analyses in some software but do not have to have experience of Bayesian methods.
Attendees should be comfortable with carrying out a simple meta-analysis using software like RevMan or Stata. They should understand probability distributions, though this can be intuitive and doesn’t have to be mathematically rigorous. They do not have to have any experience of Bayesian modelling or meta-analysis.
Robert Grant is a statistician specialising in Bayesian models and data visualisation, and one of the developers of Stan software. He has twenty years’ experience of research and teaching, and has published more than 60 peer-reviewed papers while working on various healthcare applications. He worked for St George’s Medical School, Kingston University, the Royal College of Physicians and the National Institute for Health and Clinical Excellence. He has provided training for organisations like Harvard Medical School and University College Cork, and has spoken at conferences such as JSM, ICOTS and the RSS conference.
14 September 2021
14 September 2021
RSS CStat/Gradstat also MIS & FIS