We performed sequential forward selection to identify independently associated variants in each locus50 for ancestry-stratified and multi-ancestry results. The procedure begins by including only the top association signal into a set of independently associated variants (ϕ) per locus. Conditional analysis is then conducted on the remaining variants, conditioning on variants in ϕ. If any of these conditional signals remained significant (that is, P < 5 × 10−9), we added the top signal to the set ϕ. The process iterates until there are no remaining significantly associated variants. The method requires an external genomic reference panel to estimate LD patterns. For ancestry-stratified conditional analyses, we used ancestry-matched individuals from TOPMed to estimate LD (sample sizes given previously). For multi-ancestry conditional analyses, we used the diverse ancestry TOPMed reference panel (n = 104,976) that matched the ancestry proportions of the included cohorts.
