Article type
Year
Abstract
Background: The methodology described here was developed for a systematic review and individual subject level meta-analysis of home safety education and the provision of safety equipment for the prevention of childhood poisoning. This review had a particular emphasis on exploring whether effectiveness was related to socio-demographic characteristics previously shown to be associated with injury risk. Individual subject data were only made available to us for a proportion of the studies. This resulted in the need for developing new methodology to combine data most efficiently.
Objectives: To develop a (random effects) meta-analysis model that could meta-analyse both individual subject and aggregate level outcome data while exploring the effects of covariates also available in a combination of individual subject and aggregate level. To add further complication, the studies to be combined were a mixture of cluster and individual subject allocated designs.
Methods: A number of possible analysis strategies of increasing sophistication are discussed, initially ignoring covariates, and, then exploring their effect on between study heterogeneity. We argue that the most appropriate and complete analysis approach uses a Bayesian model, which allows different likelihood functions to be fitted to the different designs and data formats available. Using the aggregate analysis as prior information in an analysis of the individual subject level data will also be given careful consideration.
Results: It will be demonstrated through this example that it is possible to develop 'tailor made' evidence synthesis models using the freely available WinBUGS software which allow data from studies with different designs and reported in different formats to be appropriately and efficiently combined. While keeping technical details to a minimum we will describe the 'building block' philosophy behind the approach that enables complex models to be built up using a number of simpler code modules.
Conclusion & Discussion: We have demonstrated a natural way of meta-analysing evidence structures which would not be possible using a more traditional approach. Such approaches are currently being keenly researched and we highlight other types of evidence synthesis problems that may benefit from such an approach.
Objectives: To develop a (random effects) meta-analysis model that could meta-analyse both individual subject and aggregate level outcome data while exploring the effects of covariates also available in a combination of individual subject and aggregate level. To add further complication, the studies to be combined were a mixture of cluster and individual subject allocated designs.
Methods: A number of possible analysis strategies of increasing sophistication are discussed, initially ignoring covariates, and, then exploring their effect on between study heterogeneity. We argue that the most appropriate and complete analysis approach uses a Bayesian model, which allows different likelihood functions to be fitted to the different designs and data formats available. Using the aggregate analysis as prior information in an analysis of the individual subject level data will also be given careful consideration.
Results: It will be demonstrated through this example that it is possible to develop 'tailor made' evidence synthesis models using the freely available WinBUGS software which allow data from studies with different designs and reported in different formats to be appropriately and efficiently combined. While keeping technical details to a minimum we will describe the 'building block' philosophy behind the approach that enables complex models to be built up using a number of simpler code modules.
Conclusion & Discussion: We have demonstrated a natural way of meta-analysing evidence structures which would not be possible using a more traditional approach. Such approaches are currently being keenly researched and we highlight other types of evidence synthesis problems that may benefit from such an approach.
PDF