Extended Data Table 1 Comparison of λ1000 and λGC for five phenotypes across three association study paradigms
- λGC is the genomic inflation factor, which is a measure used to assess the extent of inflation in association test statistics by measuring the median of the observed chi-square test statistics divided by the expected chi-square distribution under the null hypothesis. Similarly, λ1000 is the genomic inflation factor adjusted for sample size to match the scale of a study of 1,000 cases and 1,000 controls with equivalent inflation. For both metrics, values > 1 indicate inflation from polygenic signal, population stratification, or other confounding. We show these metrics for EUR (the European genetic ancestry group alone), mega-analysis (a single association test across all samples), and meta-analysis (across all available population-specific association results). We show comparisons at full sample sizes as well as an equivalent sample size to EUR, with an equivalent number of EUR individuals removed to assess the effect of sample size. λ1000 values are all very close to 1, consistent with no substantial inflation, indicating that our test statistics are in line with the null hypothesis. λGC values are systematically lower for the meta-analysis versus mega-analysis; since the power of these two approaches should be nearly identical at a given sample size, the lower λGC indicates that we have likely controlled for population stratification better in the meta-analysis (note that 10 PCs differ between meta-analysis–ancestry-specific–versus mega-analysis–computed across the biobank34, but that they are consistent between meta-analysis and EUR). The number of independent genome-wide significant loci for a range of traits is similar in the meta-analysis versus mega-analysis, indicating that we are not losing appreciable power to identify significant associations. As expected, we are gaining power in the multi-ancestry meta-analysis compared to the EUR-only GWAS, as evidenced by the increase in new loci identified, due to larger sample size and increased diversity. The number of independent GWAS significant loci is computed with an LD r2 cutoff of 0.1 and a GWAS significance threshold of 5 × 10−8. The numbers in square brackets show the number of independent loci excluding the Duffy locus (chr1:125,000,000-175,000,000) given its extreme significance and long-range LD.
