A method for meta-analysis where the level of data aggregation differs between studies.

Abstract

BACKGROUND: This paper presents a statistical method that was used to assess haemoglobin levels world-wide among women, based on a combination of over 400 studies. METHODS: The methodology is easy to implement and is specifically adapted to the case where mean observations are taken, although the subgroups represented by the mean values may differ from study to study. That is, the level of data aggregation is not consistent between studies. RESULTS: In this example some studies report average haemoglobin levels for a sample of the population, while others give averages by urban/rural classification and/or pregnancy status. Though the method is based on likelihood principles, computation is straightforward. CONCLUSIONS: The methodology can be applied to a variety of meta-analytic situations where assessments are based on combining data from several sources. We provide an example of how the method was implemented in a study of haemoglobin levels among women undertaken at the World Health Organisation.