Table 2 shows SNP heritability estimates for AD computed with and without SNPs from the filaggrin region. The figure of 7.8% (9.7%) is low particularly compared to the heritability estimates from twin studies of eczema where figures exceeding 80% are not uncommon (Bataille et al., 2012). This could be for a number of reasons including the fact that genomic control correction in the individual meta-analysis studies causes downward bias, and the fact that LD score regression provides an estimate of the overall proportion of additive genetic variance tagged by SNPs in the GWAS panel (i.e. SNP heritability), rather than total heritability per se. However the greatest contributing factor is likely to be the case definition of AD used in the EAGLE consortium paper (http://www.wikigenes.org/e/art/e/348.html) which is extremely heterogeneous, relying often on self-reported data or retrospective recall which will introduce substantial measurement error into the analysis (and hence decrease heritability estimates). Our results strongly suggest that reanalysis using a more precise definition of eczema would result in a cleaner phenotype and consequently increase the number of genome-wide significant loci detected. Table
