Title: Combining Information From Different Studies with Meta-Analysis and Network Meta-Analysis
Instructors: Professor Chris Schmid, Brown University
Moderator: Alfred Balch, University of Utah
Policymakers, scientists, and health care providers increasingly cite evidence-based decision-making as the basis for their choices. For a defined research question focusing on the effects of interventions, exposures or treatments on defined outcomes, systematic reviews provide a scientifically valid approach to synthesize all of the available evidence from research studies. Meta-analysis applies statistical models to estimate the size and direction of the comparative effects derived from multiple studies designed to determine the effect of a treatment, device or test. This course introduces the major principles and techniques of the statistical analysis of meta-analytic data for both summary data available from reports and individual data from studies.
The first part of the course focuses on comparisons of two treatments under a variety of different outcome types and using a variety of statistical models that incorporate within and between-study heterogeneity. The second part of the course extends the models for data that may involve more than two interventions and more than one outcome measured at different times. Reviews with three or more treatments combine data from studies that may each use only a subset of the treatments. These studies form a treatment network, combining direct evidence from studies with head-to-head comparisons and indirect evidence from studies that compare treatments indirectly through a directed path. The network models provide estimates of the relative effectiveness or harms of all included treatments, and a ranking with associated probability estimates. These methods depend on a crucial assumption that the direct and indirect evidence are compatible (consistency) and that treatments are mutually exchangeable across studies (transitivity).
The presentation will combine principles and intuition about the proper application of the methods as well as technical information about the models employed. Although most of the examples will be taken from healthcare, the methods are applicable in any discipline where meta-analysis is undertaken including education, psychology, economics, etc. Examples in each of these areas will be given and discussion is welcomed.
The short course will
- summarize the parts of a systematic review and the data necessary to carry out meta-analysis
- present different models for analyzing summary data from multiple studies in order to estimate and compare treatment effects in populations and subgroups
- discuss how to model individual participant data from trials
- identify and evaluate concepts and assumptions of network meta-analysis, such as heterogeneity, transitivity, and consistency
- present models for network meta-analysis and how heterogeneity and inconsistency can be explored
- describe efficient tabular and graphical summaries of findings
- include examples from case studies and their interpretation for decision making
- demonstrate how to implement the methods using statistical software
Day 1 Topics Covered
- Systematic Reviews
- Types of Data in Meta-Analysis
- Estimating a Common Effect
- Heterogeneity in Meta-Analysis
- Bayesian Meta-Analysis
- Individual Participant Analysis
- Multivariate Meta-Analysis
Day 2 Topics Covered
- Background for network meta-analysis
- Direct and indirect comparisons
- Models under consistency assumption
- Ranking of Treatments
- Evaluating Network Assumptions: Exchangeability, Consistency
Christopher Schmid is Professor and Chair of Biostatistics at Brown University School of Public Health where he co-founded the Center for Evidence Synthesis in Health. Before that he worked for many years directing the Biostatistics Research Center at Tufts Medical Center in Boston. He has a long record of collaborative research in diverse areas of medicine and health with academia, government and industry and has more than 200 peer-reviewed publications. He has coauthored consensus CONSORT reporting guidelines for N-of-1 trials and single-case designs, and PRISMA guidelines extensions for meta-analysis of individual participant studies and for network meta-analyses as well as the Institute of Medicine report that established US standards for systematic reviews. His research focuses on Bayesian methods for meta-analysis, including networks of treatments and N-of-1 designs, as well as open-source software tools. He has developed predictive models for heart attack risk and the risk of dehydration in children suffering from diseases in the developing world. He also led analyses for the CKD-EPI consortium that developed the most commonly used formulas to estimate kidney function (GFR) based on the biomarkers serum creatinine and serum cystatin. He is lead statistician on several N-of-1 trial consortia. Professor Schmid is a Fellow of the American Statistical Association, founding Editor of the journal Research Synthesis Methods, long-time statistical editor of the American Journal of Kidney Diseases and served for several years on the FDA Drug Safety and Risk Management Committee.