An additional potential limitation is the variability in OA case definitions (e.g., diagnostic and frequency of use) and types of controls (e.g., exposed, unexposed, and population-based) used to define OA phenotypes across the cohorts within the GENOA and across the other contributing GWAS. However, the genetic correlations across phenotypes were uniformly high (rg > 0.9) and resulted in a well-fitting single latent factor gSEM model. An important caveat to the high correlations observed across cohorts with exposed, unexposed, and public controls is that the exposure to opioids was often based on prescribed medication (MVP and PH), which differs in risk of OA from exposure to illicit heroin use. Fine-grained comparison of large samples with different types of exposure to opioids will be needed to resolve this question. Because we have incorporated GENOA and previously published GWAS of OA for our discovery analyses, we do not have independent replication cohorts available in which to test the identified associations. However, OA associations with the intron 1 locus have been previously reported29,39,51,52, the chromosome 16 (rs13333582) is an eQTL for a gene previously
