Variation in baseline risk as an explanation of heterogeneity in meta-analysis

With many research groups in the world studying the same or similar types of intervention, there is an opportunity and a need to collate and synthesize all available evidence on the value of specific interventions. Thus, meta-analysis, the statistical science of doing this, is becoming increasingly important. One phenomenon commonly encountered in meta-analysis is that of heterogeneity in effect. Results may differ substantially between studies; much more than could be expected on the basis of within-study parameter standard errors, the 'measurement errors'. This heterogeneity must impact on the way we analyse and interpret the data. Often because of this heterogeneity no unequivocal conclusion on the value of the intervention can be drawn and there is a need to take study characteristics into account, so that possibly inferences from separate study groups can be made. In a recent study, Walter, following papers by Brand and Kragt, Senn and Brand, addressed the issue of explaining heterogeneity in meta-analysis by a baseline dependent effect of the intervention. As an explicit model he proposed the linear relationship.