Heterogeneity testing in meta-analysis of genome searches

Genome searches for identifying susceptibility loci for the same complex disease often give inconclusive or inconsistent results. Genome Search Meta-analysis (GSMA) is an established non-parametric method to identify genetic regions that rank high on average in terms of linkage statistics (e.g., lod scores) across studies. Meta-analysis typically aims not only to obtain average estimates, but also to quantify heterogeneity. However, heterogeneity testing between studies included in GSMA has not been developed yet. Heterogeneity may be produced by differences in study designs, study populations, and chance, and the extent of heterogeneity might influence the conclusions of a meta-analysis. Here, we propose and explore metrics that indicate the extent of heterogeneity for specific loci in GSMA based on Monte Carlo permutation tests. We have also developed software that performs both the GSMA and the heterogeneity testing. To illustrate the concept, the proposed methodology was applied to published data from meta-analyses of rheumatoid arthritis (4 scans) and schizophrenia (20 scans). In the first meta-analysis, we identified 11 bins with statistically low heterogeneity and 8 with statistically high heterogeneity. The respective numbers were 9 and 6 for the schizophrenia meta-analysis. For rheumatoid arthritis, bins 6.2 (the HLA region that is a well-documented susceptibility locus for the disease) and 16.3 (16q12.2-q23.1) had both high average ranks and low between-study heterogeneity. For schizophrenia, this was seen for bin 3.2 (3p25.3-p22.1) and heterogeneity was still significantly low after adjusting for its high average rank. Concordance was high between the proposed metrics and between weighted and unweighted analyses. Data from genome searches should be synthesized and interpreted considering both average ranks and heterogeneity between studies.