One challenge researcher facing regularly is the accurate phenotype which can match with genomic information(Murphy et al., 2009). Our study is best viewed in light of previous GWAS which largely used research populations(Hancock et al., 2015; Nelson et al., 2016). This may introduce greater generalizability to the types of populations seen by practicing clinicians, but also introduces potential misclassification in diagnosis. Our sample size was also only moderate by the standards of previous GWAS, but was large enough to successfully confirm one previous hit and to identify two novel loci that had not been associated with OUD. Moreover, using EHR-derived patient prescription data, we were able to subset the control group on the basis of opioid drug exposure. Recent studies have highlighted the utility of this approach, in which an “opioid exposed” control group can yield significant hits that unexposed controls – who may be less susceptible to genetic tendencies to substance use disorders – may not(Polimanti et al., 2020; Zhou et al., 2019). Our results generally supported this approach, as they revealed one hit that was only identified using opioid-exposed controls. Thus, this EHR-based method supplements but will not replace more traditional GWAS nested in formal research studies.
