Fig. 1

Impacts of data heterogeneity in meta-analyses. A key strength of meta-analyses is the ability to take data from multiple studies to increase sample size. A Methodological heterogeneity risks introducing confounding batch effects in pooled datasets. In this figure, data from three studies have been pooled and analysed using a uniform pipeline that revealed two distinct clusters (group A and group B). However, after accounting for methodological heterogeneity, it becomes evident that all of the data in group A were generated using a single method that was distinct from those that generated the data in group B. This type of batch effect risks incorrect clinical interpretation of the analysis outputs. B Selecting studies for inclusion in meta-analyses based on standardised methodologies can reduce batch effects outlined in A but risks loss of sample size. In respiratory studies, heterogeneity emerges from a multitude of clinic and laboratory factors, from initial differences in diagnostic criteria through to variation in the pipeline used to analyse sequence data. Within a single study (example at top), a given method is applied consistently to all samples and therefore has no impact on sample size. By contrast, within IPD meta-analyses (example at bottom), correction for differences in the methods used at each successive stage progressively erodes the sample size. If the field contains a low level of methodological standardisation, then this reduction could potentially make the statistical power of a meta-analysis no more useful than that of a single study