EHR data reflect real clinical use patterns, but some data can be missing or inaccurate. We addressed this potential limitation by implementing a carefully curated CAD phenotype, and conducted sensitivity analyses excluding people with unknown covariate values. In addition, the VUMC inpatient and outpatient population included in our study may be sicker than the general population. Nonetheless, we replicated our findings in a prospective epidemiological cohort ascertained from the general population. Even though MDD was not assessed in ARIC participants, which hampered our ability to replicate results from aim 2, we replicated all other aims and findings, including that polygenic scores for MDD and loneliness were associated with CAD even after adjusting for conventional heart disease risk factors. Another limitation of our study was its restriction to individuals of European ancestry, but this decision was necessitated by the ancestry of patients in the meta-GWAS used to build the polygenic scores. The relevance of our findings to individuals of other ancestries, therefore, is unknown [44, 45].
