Studies should report clearly on all aspects of study design and conduct which impact on harmonisation of analyses across data sources. | Allows assessment of the relative importance of heterogeneity induced by data management and analysis decisions vs heterogeneity inherent in the data. |
Participant characteristics and effect estimates (where applicable) should be reported for each data source. | Assessment of heterogeneity is essential for interpretation, but formal methods for quantifying heterogeneity are inefficient and possibly biased in multi-database settings. |
Where one-stage methods are used, studies should report whether and how analyses accounted for clustering and between database heterogeneity. | Interpretation requires understanding of extent to which heterogeneity might influence study results. |
Where two-stage meta-analysis is used, studies should provide a clear rationale for choice of fixed effect (FE), random effects (RE) or other model. | Interpretation requires understanding of extent to which heterogeneity might influence study results. |
Sensitivity analyses should include alternative methods for combining data. | Comparing the results of one-stage vs two-stage analyses, or FE vs RE models, provides information about potential impact of modelling assumptions. |
Further research is needed to compare performance of one-stage and two-stage approaches for multi-database studies. | Relatively few studies have specifically addressed meta-analysis for multi-database studies. |