Course Level: Advanced
Course Limit: 30
Format Requirements: Mix of didactic lectures, interactive sessions, and analysis cases studies. Familiarity with standard meta-analysis methods and multilevel modeling is assumed.
Background: Meta-analysis is the quantitative synthesis of information from independent studies, and one of the major tools in technology assessment. A very practical use of meta-analysis is in the parameterization of decision and economic models. Modelers are often faced with complex data structures that are not amenable to standard meta-analysis methods, such as multiple competing interventions assessed in different trials (networks of treatments), reporting of results at multiple followup times, availability of data both from randomized and observational studies (some of which may be at high risk of bias), studies of test performance using paired designs to compare multiple competing tests. This course discusses synthesis of complex data structures that are often encountered in modeling using aggregate published data. A more introductory treatment of meta-analysis is the focus of the companion short course "INTERMEDIATE META-ANALYSIS FOR HEALTH TECHNOLOGY ASSESSMENT: HANDS ON WORKSHOP".
Description and Objectives: After the course, attendees will be able to
- Recognize when multivariate meta-analysis methods can be applied in place of simpler analyses
- Describe the pros and cons of multivariate compared to univariate approaches
- Describe state-of the art methods for meta-analysis, including:
- Network meta-analysis (and indirect comparisons of interventions)
- Complex associations of multiple correlated exposures (e.g., drug class effects)
- Correlated outcomes, e.g., outcomes reported at successuve follow-up times or outcomes that are mutually exclusive
- Cross-design synthesis for randomized and observational data, including single group studies
- Joint meta-analysis of multiple tests applied to the same group of individuals
- Recognize that likelihood-based inference (maximum likelihood and Bayesian) offers a general framework for meta-analysis of complex datasets