Development and validation of a diagnostic algorithm for Alcohol Use Disorder in the Electronic Health Records

项目总结

PROJECT SUMMARY Alcohol Use Disorder (AUD) is highly prevalent, heterogeneous, heritable and results in an array of negative outcomes. Enhancing our understanding of the genetic basis of AUD can enable the development of new and more effective treatments. Although, AUD Genome Wide Association studies have identified and replicated associations for loci in a number of genes, the sample sizes for AUD GWAS are still relatively small, indicating that there are likely more AUD related genetic loci to be discovered. AUD is also frequently undetected and under-diagnosed, potentially biasing GWAS and follow up analyses. The availability of large, longitudinal datasets associated with Electronic Health Records (EHR) that are linked to clinical and genetic data enables passive collection of data on AUD, across sexes and ancestries, in stark contrast to the costly and labor- intensive processes of traditional ascertainment for AUD. Furthermore, EHR-based phenotyping is a cost- effective strategy that shows strong validity in genetic and epidemiologic findings for other psychiatric conditions. The research will be conducted at Vanderbilt University Medical Center (VUMC), an integrated health system with an EHR including 3.2 million patients linked to BioVU, a genomic resource with genome- wide genotype data for 94,000 patients of diverse ancestry. Our first aim is to develop and validate an algorithm to identify individuals with AUD in the EHR (Aim 1). We will use a combination of structured EHR data (e.g., diagnosis of billing codes, electronic prescriptions, procedures, labs, vital signs) and unstructured data (e.g., clinical notes), to develop a sophisticated algorithm for better phenotypic classification of AUD in the EHR. We will also test the algorithm performance in males and females, and in different races and ethnicities, to ensure that we avoid biasing demographic groups in subsequent research. Our second aim is to determine the utility of EHR-based AUD diagnoses for genomics research (Aim 2). We will test the extent to which an algorithm based solely on billing codes can replicate the AUD related genetic findings, compared to an algorithm that incorporates structured and unstructured data. Also, the GWAS summary statistics created by our analyses will then be meta-analyzed together with other GWAS studies, helping increase the sample sizes and hence the power to detect genetic loci for AUD. Our approach responds to NIAAA’s recent announcement (NOT-AA-20-018) and proposes innovative analyses with existing alcohol research data. Validating the AUD phenotype in Vanderbilt’s EHR is an important first step that will subsequently allow us to perform systematic investigations into the interactions between genetic variation and other AUD-related risk factors.