Article type
Year
Abstract
Introduction: Classic methods to investigate heterogeneity in meta-analysis (tests for heterogeneity, random effect model...) do not really provide information on the origins of this heterogeneity which would be very useful for interpretation purposes.
Objectives: To provide a descriptive method to both detect the origins of the heterogeneity and evaluate their contributions to the overall results.
Methods: A graphical method for detecting sources of heterogeneity is presented. Each trial is represented by a dot on a 2D graph. The X-axis represents the contribution of the trial to the overall Cochran Q test for heterogeneity. The Y-axis is the square of the difference between the treatment effect estimated without the trial and the overall treatment effect, weighted by the variance of the treatment effect after exclusion of the trial.
Results: The method has been applied to data from the Meta-Analysis of Chemotherapy in Head and Neck Cancers (MACH-NC) involving 10 741 patients in 63 randomized trials. The graphical method allowed us to identify trials with both a large contribution to the overall heterogeneity and a strong influence on the overall results and provided useful information for the interpretation of the heterogeneity in this meta-analysis.
Discussion: The proposed graphical method provides the advantage of identifying trials which explain most of the heterogeneity without having to explore all possible origins of heterogeneity by subgroup analyses. This method could also be applied to individual data to detect covariate values which explain heterogeneity.
Objectives: To provide a descriptive method to both detect the origins of the heterogeneity and evaluate their contributions to the overall results.
Methods: A graphical method for detecting sources of heterogeneity is presented. Each trial is represented by a dot on a 2D graph. The X-axis represents the contribution of the trial to the overall Cochran Q test for heterogeneity. The Y-axis is the square of the difference between the treatment effect estimated without the trial and the overall treatment effect, weighted by the variance of the treatment effect after exclusion of the trial.
Results: The method has been applied to data from the Meta-Analysis of Chemotherapy in Head and Neck Cancers (MACH-NC) involving 10 741 patients in 63 randomized trials. The graphical method allowed us to identify trials with both a large contribution to the overall heterogeneity and a strong influence on the overall results and provided useful information for the interpretation of the heterogeneity in this meta-analysis.
Discussion: The proposed graphical method provides the advantage of identifying trials which explain most of the heterogeneity without having to explore all possible origins of heterogeneity by subgroup analyses. This method could also be applied to individual data to detect covariate values which explain heterogeneity.